diff --git a/TraceLens/PerfModel/kernel_name_parser.py b/TraceLens/PerfModel/kernel_name_parser.py index f009f95e9..87e59266c 100644 --- a/TraceLens/PerfModel/kernel_name_parser.py +++ b/TraceLens/PerfModel/kernel_name_parser.py @@ -6,6 +6,30 @@ import re +import itanium_demangler + + +def _demangle_ck(kernel_name): + """ + Demangle a mangled CK kernel name using itanium_demangler. + Returns the normalized demangled string (with (int)/(bool)/enum qualifiers + stripped so bare integers remain), or None if parsing fails. + """ + try: + ast = itanium_demangler.parse(kernel_name) + if ast is None: + return None + demangled = str(ast) + # Strip (int) and (bool) type qualifiers from integer/bool literals + demangled = re.sub(r"\(int\)\s*", "", demangled) + demangled = re.sub(r"\(bool\)\s*", "", demangled) + # Normalize enum casts: (ck::SomeEnum)7 -> SomeEnum)7 + # This preserves the EnumName)value pattern that parse_ck_gemm uses as an anchor. + demangled = re.sub(r"\([^)]*::(\w+)\)", r"\1)", demangled) + return demangled + except Exception: + return None + def gemm_name_parser(kernel_name): """ @@ -15,14 +39,23 @@ def gemm_name_parser(kernel_name): Returns: dict: A dictionary containing the GEMM operation details. """ - if is_rocm_gemm(kernel_name): - return parse_rocm_gemm(kernel_name) - # NVIDIA trace: check if any of the known NVIDIA markers exist in the kernel name - elif is_cuda_gemm(kernel_name): - return parse_cuda_gemm(kernel_name) + if is_tensile_gemm(kernel_name): + return parse_tensile_gemm(kernel_name) + elif is_triton_gemm(kernel_name): + return parse_triton_gemm(kernel_name) + elif is_ck_gemm(kernel_name): + return parse_ck_gemm(kernel_name) + elif is_igemm(kernel_name): + return parse_igemm(kernel_name) + elif is_cublas_xmma_gemm(kernel_name): + return parse_cublas_xmma_gemm(kernel_name) + elif is_cutlass_gemm(kernel_name): + return parse_cutlass_gemm(kernel_name) + elif is_nvjet_gemm(kernel_name): + return parse_nvjet_gemm(kernel_name) -def is_rocm_gemm(kernel_name): +def is_tensile_gemm(kernel_name): """ Check if a kernel name matches the more general ROCm GEMM naming pattern. Allows an arbitrary prefix before 'Cijk_Alik_Bljk_...' where each of C/A/B @@ -35,7 +68,7 @@ def is_rocm_gemm(kernel_name): return bool(re.match(pattern, kernel_name)) -def parse_rocm_gemm(kernel_name): +def parse_tensile_gemm(kernel_name): # 1. Parse the transpose flags from the kernel name trans_a, trans_b = None, None @@ -57,9 +90,9 @@ def parse_rocm_gemm(kernel_name): if macro_tile_match: mt_m = int(macro_tile_match.group(1)) mt_n = int(macro_tile_match.group(2)) - depth_u = int(macro_tile_match.group(3)) + mt_k = int(macro_tile_match.group(3)) else: - mt_m, mt_n, depth_u = None, None, None # Fallback in case pattern is not found + mt_m, mt_n, mt_k = None, None, None # Fallback in case pattern is not found # Feel free to add more details as needed. # https://github.com/ROCm/Tensile/wiki/Kernel-Parameters#kernel-names @@ -68,20 +101,278 @@ def parse_rocm_gemm(kernel_name): "transpose": (trans_a, trans_b), "mt_m": mt_m, "mt_n": mt_n, - "depth_u": depth_u, + "mt_k": mt_k, + } + + +def is_ck_gemm(kernel_name): + """ + Check if a kernel name is a Composable Kernel (CK) GEMM. + Matches demangled (void ck::kernel_*) and mangled (_ZN2ck*kernel_*) forms. + """ + if "void ck::kernel_" in kernel_name: + return True + if "_ZN2ck" in kernel_name and "kernel_" in kernel_name: + return True + return False + + +def _parse_gridwise_params(kernel_name, class_name): + """ + Parse the top-level template parameters of a CK Gridwise class by name. + Returns a list of parameter strings, or [] if the class is not found. + """ + idx = kernel_name.find(class_name + "<") + if idx < 0: + return [] + start = idx + len(class_name) + 1 + depth = 1 + params = [] + current = [] + i = start + while i < len(kernel_name) and depth > 0: + c = kernel_name[i] + if c == "<": + depth += 1 + current.append(c) + elif c == ">": + depth -= 1 + if depth == 0: + params.append("".join(current).strip()) + else: + current.append(c) + elif c == "," and depth == 1: + params.append("".join(current).strip()) + current = [] + else: + current.append(c) + i += 1 + return params + + +def parse_ck_gemm(kernel_name): + """ + Parse tile sizes from CK kernel names (demangled form). + For mangled names, demangle first via _demangle_ck then recurse. + + Absolute template parameter positions are verified from CK source headers and + doxygen (docs-7.1.0). All positions are 0-indexed top-level params + of the Gridwise class template. + """ + + # ── Mangled CK kernels: demangle first, then recurse ── + if kernel_name.startswith("_ZN2ck"): + demangled = _demangle_ck(kernel_name) + if demangled is not None: + return parse_ck_gemm(demangled) + return None + + # ── Demangled CK kernels: dispatch on Gridwise class name ── + + # Map of Gridwise class substring → (M_pos, N_pos, K_pos, trans_A_pos, trans_B_pos) + # Positions are absolute indices in the top-level template parameter list. + # trans positions are indices of the RowMajor/ColumnMajor layout type params + # + # Layout params (RowMajor/ColumnMajor) are the first two type params for + # classes that have them; None for classes where layout is not type-parameterized. + + GRIDWISE_CONFIG = { + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_moe_gemm_m_x.html + "MoeGemmMX": (18, 19, 20, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_gemm_multi_d___a_b_scale__xdl__cshuffle__v3.html + "GridwiseGemmMultiD_ABScale": (18, 19, 20, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_gemm_multi_d__blockscale__xdl__cshuffle__v3__b__preshuffle.html + "GridwiseGemmMultiD_blockscale": (18, 19, 20, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_moe_gemm_block_scale.html + "MoeGemmBlockScale": (18, 19, 20, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_gemm_multi_d__xdl__cshuffle__v3__b__preshuffle.html + "GridwiseGemmMultiD_xdl_cshuffle_v3_b_preshuffle": (15, 16, 17, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_gemm_multi_d__xdl__cshuffle__v3.html + "GemmMultiD_xdl": (15, 16, 17, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_moe_gemm.html + "GridwiseMoeGemm": (15, 16, 17, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_gemm__xdl__cshuffle__v3.html + "GridwiseGemm_xdl_cshuffle_v3": (13, 14, 15, True), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_gemm_multiple_d__xdl__cshuffle.html + "GridwiseGemmMultipleD_xdl_cshuffle": (13, 14, 15, False), + # https://rocm.docs.amd.com/projects/composable_kernel/en/docs-7.1.0/doxygen/html/structck_1_1_gridwise_gemm__bk0mk1__bk0nk1__mn__xdlops__bwd__weight.html + "GridwiseGemm_bk0mk1": (12, 13, 14, False), + } + + for class_substr, (m_pos, n_pos, k_pos, has_layout) in GRIDWISE_CONFIG.items(): + if class_substr not in kernel_name: + continue + + # Find the actual class name (longest match wins to avoid prefix collisions) + cls_match = re.search( + r"(Gridwise\w*" + re.escape(class_substr.lstrip("Gridwise")) + r"\w*)<", + kernel_name, + ) + if not cls_match: + # Fallback: search directly for the substring followed by < + m = re.search(re.escape(class_substr) + r"(\w*)<", kernel_name) + if not m: + continue + actual_class = class_substr + m.group(1) + else: + actual_class = cls_match.group(0).rstrip("<") + + params = _parse_gridwise_params(kernel_name, actual_class) + if len(params) <= max(m_pos, n_pos, k_pos): + continue + + def to_int(p): + p = p.strip() + m = re.match(r"^-?\d+$", p) + return int(p) if m else None + + mt_m = to_int(params[m_pos]) + mt_n = to_int(params[n_pos]) + mt_k = to_int(params[k_pos]) + if mt_m is None or mt_n is None or mt_k is None: + continue + + # A and B layouts: first two RowMajor/ColumnMajor type params in the name. + # RowMajor = not transposed, ColumnMajor = transposed. + # https://rocm.docs.amd.com/en/latest/how-to/rocm-for-ai/inference-optimization/optimizing-with-composable-kernel.html + if has_layout: + layouts = re.findall(r"(RowMajor|ColumnMajor)", kernel_name) + trans_a = layouts[0] == "ColumnMajor" if len(layouts) >= 2 else None + trans_b = layouts[1] == "ColumnMajor" if len(layouts) >= 2 else None + else: + trans_a, trans_b = None, None + + return { + "transpose": (trans_a, trans_b), + "mt_m": mt_m, + "mt_n": mt_n, + "mt_k": mt_k, + } + + return None + + +def is_triton_gemm(kernel_name): + """Check if a kernel name is a Triton GEMM (AITER/vLLM FP8 block-scale, batched).""" + return "BLOCK_SIZE_M_" in kernel_name and "BLOCK_SIZE_N_" in kernel_name + + +def parse_triton_gemm(kernel_name): + m_m = re.search(r"BLOCK_SIZE_M_(\d+)", kernel_name) + m_n = re.search(r"BLOCK_SIZE_N_(\d+)", kernel_name) + m_k = re.search(r"BLOCK_SIZE_K_(\d+)", kernel_name) + mt_m = int(m_m.group(1)) if m_m else None + mt_n = int(m_n.group(1)) if m_n else None + mt_k = int(m_k.group(1)) if m_k else None + return {"transpose": (None, None), "mt_m": mt_m, "mt_n": mt_n, "mt_k": mt_k} + + +def is_igemm(kernel_name): + """Check if a kernel name is a MIOpen implicit GEMM convolution kernel.""" + return kernel_name.startswith("igemm_") + + +def parse_igemm(kernel_name): + trans_a, trans_b = None, None + bt = re.search(r"bt(\d+)x(\d+)x(\d+)", kernel_name) + mt_m = int(bt.group(1)) if bt else None + mt_n = int(bt.group(2)) if bt else None + mt_k = int(bt.group(3)) if bt else None + return { + "transpose": (trans_a, trans_b), + "mt_m": mt_m, + "mt_n": mt_n, + "mt_k": mt_k, + } + + +def is_cublas_xmma_gemm(kernel_name): + """Check if a kernel name is a cuBLAS XMMA GEMM (sm80/sm90 Hopper TMA/WGMMA path).""" + return "xmma_gemm_" in kernel_name or "xmma_fprop_implicit_gemm_" in kernel_name + + +def parse_cublas_xmma_gemm(kernel_name): + ts = re.search(r"tilesize(\d+)x(\d+)x(\d+)", kernel_name) + mt_m = int(ts.group(1)) if ts else None + mt_n = int(ts.group(2)) if ts else None + mt_k = int(ts.group(3)) if ts else None + + trans_a, trans_b = None, None + tr = re.search(r"_([tn])([tn])_", kernel_name) + if tr: + trans_a = tr.group(1) == "t" + trans_b = tr.group(2) == "t" + + return { + "transpose": (trans_a, trans_b), + "mt_m": mt_m, + "mt_n": mt_n, + "mt_k": mt_k, + } + + +def is_cutlass_gemm(kernel_name): + """Check if a kernel name is a CUTLASS GEMM/conv kernel.""" + return "cutlass" in kernel_name and re.search( + r"gemm|fprop|dgrad|wgrad", kernel_name + ) + + +def parse_cutlass_gemm(kernel_name): + # Format 1: {M}x{N}_{K}x{stages} (e.g. 256x128_64x3) + m = re.search( + r"(?:gemm|fprop|dgrad|wgrad)\w*_(\d+)x(\d+)_(\d+)x(\d+)", + kernel_name, + ) + if m: + mt_m = int(m.group(1)) + mt_n = int(m.group(2)) + mt_k = int(m.group(3)) + else: + # Format 2: {M}x{N}x{K} (e.g. 128x128x64) + m = re.search( + r"(?:gemm|fprop|dgrad|wgrad)\w*_(\d+)x(\d+)x(\d+)", + kernel_name, + ) + if m: + mt_m = int(m.group(1)) + mt_n = int(m.group(2)) + mt_k = int(m.group(3)) + else: + mt_m, mt_n, mt_k = None, None, None + + trans_a, trans_b = None, None + tr = re.search(r"_([tn]{2})_", kernel_name) + if tr: + trans_a = tr.group(1)[0] == "t" + trans_b = tr.group(1)[1] == "t" + + return { + "transpose": (trans_a, trans_b), + "mt_m": mt_m, + "mt_n": mt_n, + "mt_k": mt_k, } -def is_cuda_gemm(kernel_name): +def is_nvjet_gemm(kernel_name): """ Check if a kernel name matches the NVIDIA GEMM naming pattern: """ - # Right now, we only check if the kernel name starts with 'nvjet'. - # This is a temporary solution and will be expanded in the future. return kernel_name.startswith("nvjet") -def parse_cuda_gemm(kernel_name): +def parse_nvjet_gemm(kernel_name): transpose_chars = kernel_name.split("_")[-1] transpose = transpose_chars[0] == "T", transpose_chars[1] == "T" - return {"transpose": transpose} + + # Tile format: nvjet_{dtype}_{MxN}_{KxStages}_... + # e.g. nvjet_tst_64x64_64x16_... → M=64, N=64, K=64 + mt_m, mt_n, mt_k = None, None, None + mn = re.search(r"nvjet_\w+_(\d+)x(\d+)_(\d+)x\d+", kernel_name) + if mn: + mt_m = int(mn.group(1)) + mt_n = int(mn.group(2)) + mt_k = int(mn.group(3)) + + return {"transpose": transpose, "mt_m": mt_m, "mt_n": mt_n, "mt_k": mt_k} diff --git a/TraceLens/PerfModel/perf_model.py b/TraceLens/PerfModel/perf_model.py index d7a670051..2a44a2aa5 100644 --- a/TraceLens/PerfModel/perf_model.py +++ b/TraceLens/PerfModel/perf_model.py @@ -52,7 +52,16 @@ def __init__(self, event, arch=None, python_path=None, enable_origami=False): self.param_details["B"] = 1 if self.parsed_kernel_info is not None: - self.param_details["transpose"] = self.parsed_kernel_info["transpose"] + self.param_details["transpose"] = ( + self.parsed_kernel_info.get("transpose") or "Unknown" + ) + self.impl_param = { + key: self.parsed_kernel_info.get(key) or "Unknown" + for key in ("mt_m", "mt_n", "mt_k") + } + else: + self.param_details["transpose"] = "Unknown" + self.impl_param = {key: "Unknown" for key in ("mt_m", "mt_n", "mt_k")} self.B, self.M, self.N, self.K = ( self.param_details["B"], diff --git a/TraceLens/TreePerf/jax_analyses.py b/TraceLens/TreePerf/jax_analyses.py index 13af6d0f2..c09259879 100644 --- a/TraceLens/TreePerf/jax_analyses.py +++ b/TraceLens/TreePerf/jax_analyses.py @@ -595,4 +595,8 @@ def gemm_perf_metrics( for key, value in perf_model.param_details.items(): dict_metrics[f"param: {key}"] = value + if hasattr(perf_model, "impl_param"): + for key, value in perf_model.impl_param.items(): + dict_metrics[f"impl_param: {key}"] = value + return dict_metrics diff --git a/TraceLens/TreePerf/tree_perf.py b/TraceLens/TreePerf/tree_perf.py index 4ae2cf269..c95e47736 100644 --- a/TraceLens/TreePerf/tree_perf.py +++ b/TraceLens/TreePerf/tree_perf.py @@ -522,6 +522,10 @@ def compute_perf_metrics( for key, value in perf_model.param_details.items(): dict_metrics[f"param: {key}"] = value + if hasattr(perf_model, "impl_param"): + for key, value in perf_model.impl_param.items(): + dict_metrics[f"impl_param: {key}"] = value + return dict_metrics def compute_fwd_perf_metrics(self, event, non_data_mov=False): @@ -733,9 +737,13 @@ def summarize_df_perf_metrics( param_cols = [ col for col in df_perf_metrics.columns if col.startswith("param: ") ] - if group_by_num_kernels and "num_kernels" in df_perf_metrics.columns: - param_cols.append("num_kernels") groupby_cols = ["name"] + param_cols + if group_by_num_kernels and "num_kernels" in df_perf_metrics.columns: + groupby_cols.append("num_kernels") + + for col in df_perf_metrics.columns: + if col.startswith("impl_param: "): + dict_agg[col] = "first" # Convert parameter columns to strings to avoid type comparison issues df_perf_metrics = df_perf_metrics.copy() if ( @@ -765,6 +773,9 @@ def summarize_df_perf_metrics( if "Compute Spec_first" in df_perf_metrics_summary.columns: rename_map["Compute Spec_first"] = "Compute Spec" + for col in df_perf_metrics_summary.columns: + if col.startswith("impl_param: ") and col.endswith("_first"): + rename_map[col] = col[: -len("_first")] if rename_map: df_perf_metrics_summary.rename(columns=rename_map, inplace=True) @@ -780,14 +791,21 @@ def summarize_df_perf_metrics( priority_cols.append("process_label") if "thread_name" in df_perf_metrics_summary.columns: priority_cols.append("thread_name") + impl_param_cols = [ + col + for col in df_perf_metrics_summary.columns + if col.startswith("impl_param: ") + ] other_cols = [ col for col in df_perf_metrics_summary.columns - if col not in priority_cols and col not in param_cols + if col not in priority_cols + and col not in param_cols + and col not in impl_param_cols ] df_perf_metrics_summary = df_perf_metrics_summary[ - priority_cols + param_cols + other_cols + priority_cols + param_cols + impl_param_cols + other_cols ] if "Kernel Time (µs)_sum" in df_perf_metrics_summary.columns: @@ -3635,6 +3653,10 @@ def compute_perf_metrics(self, event, bwd=False): for key, value in perf_model.param_details.items(): dict_metrics[f"param: {key}"] = value + if hasattr(perf_model, "impl_param"): + for key, value in perf_model.impl_param.items(): + dict_metrics[f"impl_param: {key}"] = value + return dict_metrics def build_df_perf_metrics( diff --git a/setup.py b/setup.py index 3cb4fb98c..63a09e899 100755 --- a/setup.py +++ b/setup.py @@ -54,7 +54,8 @@ def _wheel_version(): "orjson", "matplotlib", "xprof==2.20.1", # Last version with HLO sidecar generation; supports JAX 0.8+ (with benign INT_MAX warnings) - "protobuf>=6.31.1,<7.0.0", # Required by xprof's grpcio-status dependency + "protobuf>=6.31.1,<7.0.0", # Required by xprof's grpcio-status dependency, + "itanium-demangler", # 'openpyxl', # 'tensorflow', ], diff --git a/tests/test_kernel_name_parser.py b/tests/test_kernel_name_parser.py new file mode 100644 index 000000000..7a44e6dc2 --- /dev/null +++ b/tests/test_kernel_name_parser.py @@ -0,0 +1,765 @@ +############################################################################### +# Copyright (c) 2026 Advanced Micro Devices, Inc. All rights reserved. +# +# See LICENSE for license information. +############################################################################### + +""" +Integration tests using verbatim kernel names extracted from trace files. + +Trace sources: + Tensile — MI300 Qwen inference trace + iGEMM — MI300 ResNet training trace + cuBLAS — H100 Qwen, Wan-AI, Falconsai inference traces + CUTLASS — H100 BERT, Falconsai, Wan-AI inference traces + CK — MI300 vLLM FP8 inference trace + Triton — MI300 SGLang and vLLM FP8 inference traces + nvjet — B200 and H100 graph capture traces +""" + +from TraceLens.PerfModel.kernel_name_parser import gemm_name_parser + + +class TestRealTensile: + # Source: MI300 Qwen inference trace + + def test_mt256x128x64_tn(self): + name = ( + "Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_" + "AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_" + "GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_" + "LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_" + "ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_" + "USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 256 + assert result["mt_n"] == 128 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_mt128x224x64_tn(self): + name = ( + "Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x16x1_SN_LDSB1_" + "AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_" + "GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_" + "LWPMn1_MIAV0_MIWT4_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_" + "ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_" + "USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 224 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_mt16x16x128_nn(self): + name = ( + "Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_SN_LDSB1_AFC1_" + "AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_" + "GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB8_LPM0_LRVW4_LWPMn1_" + "MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_" + "PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_" + "UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 16 + assert result["mt_k"] == 128 + assert result["transpose"] == (False, False) + + def test_mt16x16x256_tn(self): + # Source: MI300 Qwen inference trace + name = ( + "Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_" + "AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_" + "GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_" + "LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_" + "ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_" + "USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 16 + assert result["mt_k"] == 256 + assert result["transpose"] == (True, False) + + def test_mt64x32x256_tn(self): + # Source: MI300 Qwen inference trace + name = ( + "Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16x1_SN_LDSB1_" + "AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_" + "GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_" + "LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_" + "ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_" + "USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 32 + assert result["mt_k"] == 256 + assert result["transpose"] == (True, False) + + +class TestRealIgemm: + # Source: MI300 ResNet training trace + + def test_wrw_64x32x32(self): + name = ( + "igemm_wrw_gtcx3_nhwc_bf16_bx0_ex1_bt64x32x32_wt16x16x16_ws1x1_wr2x1" + "_ta1x4x1x2_1x8x1x32_tb1x4x1x1_1x8x1x32_vs1_gkgs" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 32 + assert result["mt_k"] == 32 + assert result["transpose"] == (None, None) + + def test_fwd_256x64x8(self): + name = ( + "igemm_fwd_gtcx3_nhwc_bf16_bx0_ex1_bt256x64x8_wt64x16x4_ws1x1_wr2x2" + "_ta1x1x8x1_1x8x1x32_tb1x1x2x1_1x8x1x32_me" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 256 + assert result["mt_n"] == 64 + assert result["mt_k"] == 8 + assert result["transpose"] == (None, None) + + def test_fwd_64x128x32(self): + name = ( + "igemm_fwd_gtcx3_nhwc_bf16_bx0_ex1_bt64x128x32_wt32x32x8_ws1x1_wr2x1" + "_ta1x8x1x1_1x4x1x64_tb1x8x2x1_1x4x1x64_gkgs" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 128 + assert result["mt_k"] == 32 + assert result["transpose"] == (None, None) + + def test_fwd_128x128x32(self): + name = ( + "igemm_fwd_gtcx3_nhwc_bf16_bx0_ex1_bt128x128x32_wt32x32x8_ws1x1_wr2x2" + "_ta1x8x2x1_1x4x1x64_tb1x8x2x1_1x4x1x64_gkgs" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 128 + assert result["mt_k"] == 32 + assert result["transpose"] == (None, None) + + def test_wrw_256x128x16(self): + name = ( + "igemm_wrw_gtcx3_nhwc_bf16_bx0_ex1_bt256x128x16_wt32x32x8_ws2x1_wr2x2" + "_ta1x4x1x4_1x4x1x64_tb1x4x1x2_1x4x1x64_gkgs" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 256 + assert result["mt_n"] == 128 + assert result["mt_k"] == 16 + assert result["transpose"] == (None, None) + + +class TestRealCublasXmma: + # Source: H100 Qwen, Wan-AI, and Falconsai inference traces + + def test_sm90_128x128x64_tn(self): + name = ( + "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n" + "_tilesize128x128x64_warpgroupsize1x1x1" + "_execute_segment_k_off_kernel__5x_cublas" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 128 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_sm90_64x64x64_tn(self): + name = ( + "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n" + "_tilesize64x64x64_warpgroupsize1x1x1" + "_execute_segment_k_off_kernel__5x_cublas" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 64 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_sm90_128x64x64_tn(self): + name = ( + "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n" + "_tilesize128x64x64_warpgroupsize1x1x1" + "_execute_segment_k_off_kernel__5x_cublas" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 64 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_sm90_256x128x64_tn(self): + # Source: H100 Wan-AI inference trace + name = ( + "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n" + "_tilesize256x128x64_warpgroupsize2x1x1" + "_execute_segment_k_off_kernel__5x_cublas" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 256 + assert result["mt_n"] == 128 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_sm90_implicit_gemm_128x128x64(self): + # Source: H100 Wan-AI inference trace + name = ( + "sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc" + "_tilesize128x128x64_warpgroupsize1x1x1_g1" + "_execute_segment_k_off_kernel__5x_cudnn" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 128 + assert result["mt_k"] == 64 + + +class TestRealCutlass: + # Source: H100 BERT, Falconsai, and Wan-AI inference traces + + def test_wmma_32x32x128_tn(self): + name = ( + "void cutlass::Kernel2" + "(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align8::Params)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 32 + assert result["mt_n"] == 32 + assert result["mt_k"] == 128 + assert result["transpose"] == (True, False) + + def test_tensorop_64x64x64_tn(self): + name = ( + "void cutlass::Kernel2" + "(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 64 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_tensorop_256x128x64_tn(self): + name = ( + "void cutlass::Kernel2" + "(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_64x3_tn_align2::Params)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 256 + assert result["mt_n"] == 128 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + def test_cudnn_fprop_256x64x32(self): + # cuDNN conv variant — no transpose in name + name = ( + "void cutlass__5x_cudnn::Kernel" + "(cutlass_tensorop_bf16_s16816fprop_optimized_bf16_256x64_32x4_nhwc_align8::Params)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 256 + assert result["mt_n"] == 64 + assert result["mt_k"] == 32 + + def test_tensorop_128x64x64_tn(self): + # Source: H100 Falconsai inference trace + name = ( + "void cutlass::Kernel2" + "(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x64_64x6_tn_align8::Params)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 64 + assert result["mt_k"] == 64 + assert result["transpose"] == (True, False) + + +class TestRealCKMangled: + # Source: MI300 ResNet training trace (tests/traces/mi300/resnet_act_checkpoint.json.gz) + + def test_anchor1_conv_fwd_16x32x64(self): + # kernel_grouped_conv_fwd_xdl_cshuffle_v3 / GridwiseGemm_xdl_cshuffle_v3 + # GemmSpecializationE7ELi128ELi16ELi32ELi64E -> BlockSize=128, M=16, N=32, K=64 + name = ( + "_ZN2ck16tensor_operation6device12_GLOBAL__N_139kernel_grouped_conv_fwd_xdl_cshuffle_v3" + "INS_28GridwiseGemm_xdl_cshuffle_v3INS_13tensor_layout4gemm8RowMajorENS6_11ColumnMajorE" + "S7_ttfttNS0_12element_wise11PassThroughESA_SA_LNS1_18GemmSpecializationE7ELi128ELi16ELi32E" + "Li64ELi8ELi8ELi16ELi16ELi1ELi1ENS_8SequenceIJLi8ELi16ELi1EEEENSC_IJLi1ELi0ELi2EEEESE_Li2E" + "Li8ELi8ELb0ELi0ESD_SE_SE_Li2ELi8ELi8ELb0ELi0ELi1ELi1ENSC_IJLi1ELi16ELi1ELi8EEEELi4ELNS_26" + "BlockGemmPipelineSchedulerE1ELNS_24BlockGemmPipelineVersionE0EttLb0ELb0EEENS_16TensorDescriptor" + "INS_5TupleIJNS_5EmbedINSK_IJiiiiEEESM_Lb0EEENS_11PassThroughIiEENS_3PadIiiiLb0EEESR_SP_SP_NSL_" + "INSK_IJiiEEESS_Lb0EEEST_SP_NS_23Merge_v2_magic_divisionINSK_IJiiiEEEEESW_NS_8RightPadIiiLb0EEESY_" + "NS_7UnMergeISS_Lb0EEESP_EEENSK_IJNSC_IJLi0EEEENSC_IJLi1EEEENSC_IJLi2EEEENSC_IJLi3EEEENSC_IJLi4E" + "EEENSC_IJLi5EEEENSC_IJLi6EEEENSC_IJLi7EEEENSC_IJLi8EEEENSC_IJLi9ELi11ELi13EEEENSC_IJLi10ELi12ELi" + "14EEEENSC_IJLi15EEEENSC_IJLi16EEEENSC_IJLi18EEEENSC_IJLi17EEEEEEENSK_IJNSC_IJLi1ELi2ELi3ELi4EEEES" + "17_S18_S19_S1A_NSC_IJLi9EEEENSC_IJLi10ELi11EEEENSC_IJLi12ELi13EEEENSC_IJLi14EEEES1D_S1E_S1G_S1F_NS" + "C_IJLi19ELi20EEEENSC_IJLi21EEEEEEENSC_IJLi19ELi21ELi20EEEElEENSJ_INSK_IJS10_SY_SY_S10_SP_EEENSK_IJS" + "12_S13_S14_S16_S15_EEENSK_IJNSC_IJLi1ELi2EEEES15_S16_NSC_IJLi5ELi6EEEES19_EEENSC_IJLi5ELi7ELi6EEEE" + "lEENSJ_INSK_IJST_SY_SY_NSZ_INSK_IJiNS_17integral_constantIiLi16EEEEEELb0EEENSZ_INSK_IJiNS1Z_IiLi32E" + "EEEEELb0EEEEEENSK_IJS12_S13_S14_S15_S16_EEENSK_IJS1U_S15_S16_S1V_NSC_IJLi7ELi8EEEEEEENSC_IJLi5ELi6E" + "Li7ELi8EEEElEENS1_30ComputePtrOffsetOfStridedBatchILi1ELi1ELi0EvEELb1ELNS_25InMemoryDataOperationEnum" + "E0ELi2ELNS_10TailNumberE10EEEvNT_8ArgumentET0_T1_T2_T3_S2L_i" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 32 + assert result["mt_k"] == 64 + + def test_anchor1_conv_fwd_64x64x64(self): + # GemmSpecializationE7ELi256ELi64ELi64ELi64E -> BlockSize=256, M=64, N=64, K=64 + name = ( + "_ZN2ck16tensor_operation6device12_GLOBAL__N_139kernel_grouped_conv_fwd_xdl_cshuffle_v3" + "INS_28GridwiseGemm_xdl_cshuffle_v3INS_13tensor_layout4gemm8RowMajorENS6_11ColumnMajorE" + "S7_ttfttNS0_12element_wise11PassThroughESA_SA_LNS1_18GemmSpecializationE7ELi256ELi64ELi64E" + "Li64ELi8ELi8ELi32ELi32ELi1ELi1ENS_8SequenceIJLi8ELi32ELi1EEEENSC_IJLi1ELi0ELi2EEEESE_Li2E" + "Li8ELi8ELb0ELi0ESD_SE_SE_Li2ELi8ELi8ELb0ELi0ELi1ELi1ENSC_IJLi1ELi32ELi1ELi8EEEELi8ELNS_26" + "BlockGemmPipelineSchedulerE0ELNS_24BlockGemmPipelineVersionE2EttLb0ELb0EEENS_16TensorDescriptor" + "INS_5TupleIJNS_5EmbedINSK_IJiiiiEEESM_Lb0EEENS_11PassThroughIiEENS_3PadIiiiLb0EEESR_SP_SP_NSL_" + "INSK_IJiiEEESS_Lb0EEEST_SP_NS_23Merge_v2_magic_divisionINSK_IJiiiEEEEESW_NS_8RightPadIiiLb0EEESY_" + "NS_7UnMergeISS_Lb0EEESP_EEENSK_IJNSC_IJLi0EEEENSC_IJLi1EEEENSC_IJLi2EEEENSC_IJLi3EEEENSC_IJLi4E" + "EEENSC_IJLi5EEEENSC_IJLi6EEEENSC_IJLi7EEEENSC_IJLi8EEEENSC_IJLi9ELi11ELi13EEEENSC_IJLi10ELi12ELi" + "14EEEENSC_IJLi15EEEENSC_IJLi16EEEENSC_IJLi18EEEENSC_IJLi17EEEEEEENSK_IJNSC_IJLi1ELi2ELi3ELi4EEEES" + "17_S18_S19_S1A_NSC_IJLi9EEEENSC_IJLi10ELi11EEEENSC_IJLi12ELi13EEEENSC_IJLi14EEEES1D_S1E_S1G_S1F_NS" + "C_IJLi19ELi20EEEENSC_IJLi21EEEEEEENSC_IJLi19ELi21ELi20EEEElEENSJ_INSK_IJS10_SY_SY_S10_SP_EEENSK_IJS" + "12_S13_S14_S16_S15_EEENSK_IJNSC_IJLi1ELi2EEEES15_S16_NSC_IJLi5ELi6EEEES19_EEENSC_IJLi5ELi7ELi6EEEE" + "lEENSJ_INSK_IJST_SY_SY_NSZ_INSK_IJiNS_17integral_constantIiLi64EEEEEELb0EEES22_EEENSK_IJS12_S13_S14_" + "S15_S16_EEENSK_IJS1U_S15_S16_S1V_NSC_IJLi7ELi8EEEEEEENSC_IJLi5ELi6ELi7ELi8EEEElEENS1_30ComputePtrOf" + "fsetOfStridedBatchILi1ELi1ELi0EvEELb1ELNS_25InMemoryDataOperationEnumE0ELi1ELNS_10TailNumberE10EEEvNT_" + "8ArgumentET0_T1_T2_T3_S2I_i" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 64 + assert result["mt_k"] == 64 + + def test_anchor2_conv_fwd_multiple_abd_128x64x32(self): + # kernel_grouped_conv_fwd_multiple_abd_xdl_cshuffle / GridwiseGemmMultipleD_xdl_cshuffle + # InMemoryDataOperationEnumE0ELi1ELi256ELi128ELi64ELi32E -> M=128, N=64, K=32 + name = ( + "_ZN2ck16tensor_operation6device12_GLOBAL__N_149kernel_grouped_conv_fwd_multiple_abd_xdl_cshuffle" + "INS_34GridwiseGemmMultipleD_xdl_cshuffleItttftNS_5TupleIJEEEtNS0_12element_wise11PassThroughES8_S8_" + "LNS_25InMemoryDataOperationEnumE0ELi1ELi256ELi128ELi64ELi32ELi8ELi8ELi32ELi32ELi2ELi1ENS_8Sequence" + "IJLi4ELi64ELi1EEEENSA_IJLi1ELi0ELi2EEEESC_Li2ELi8ELi8ELb0ELi1ESB_SC_SC_Li2ELi8ELi8ELb0ELi1ELi1ELi1" + "ENSA_IJLi1ELi32ELi1ELi8EEEELi8ELNS_13LoopSchedulerE0ELNS_15PipelineVersionE0EtEEPKtSI_S6_tS8_S8_S8_" + "NS_16TensorDescriptorINS5_IJNS_5EmbedINS5_IJiiiiEEESL_Lb0EEENS_11PassThroughIiEENS_3PadIiiiLb0EEESQ_SO_" + "SO_NSK_INS5_IJiiEEESR_Lb0EEESS_SO_NS_23Merge_v2_magic_divisionINS5_IJiiiEEEEESV_NS_8RightPadIiiLb0EEESX_" + "NS_7UnMergeINS5_IJiNS_17integral_constantIiLi8EEEEEELb0EEESO_EEENS5_IJNSA_IJLi0EEEENSA_IJLi1EEEENSA_IJLi2" + "EEEENSA_IJLi3EEEENSA_IJLi4EEEENSA_IJLi5EEEENSA_IJLi6EEEENSA_IJLi7EEEENSA_IJLi8EEEENSA_IJLi9ELi11ELi13EEEE" + "NSA_IJLi10ELi12ELi14EEEENSA_IJLi15EEEENSA_IJLi16EEEENSA_IJLi18EEEENSA_IJLi17EEEEEEENS5_IJNSA_IJLi1ELi2ELi3" + "ELi4EEEES19_S1A_S1B_S1C_NSA_IJLi9EEEENSA_IJLi10ELi11EEEENSA_IJLi12ELi13EEEENSA_IJLi14EEEES1F_S1G_S1I_S1H_" + "NSA_IJLi19ELi20EEEENSA_IJLi21EEEEEEENSA_IJLi19ELi21ELi20EEEElEENSJ_INS5_IJNSY_ISR_Lb0EEESX_SX_S12_SO_EEENS5_" + "IJS14_S15_S16_S18_S17_EEENS5_IJNSA_IJLi1ELi2EEEES17_S18_NSA_IJLi5ELi6EEEES1B_EEENSA_IJLi5ELi7ELi6EEEElEES6_" + "NSJ_INS5_IJSS_SX_SX_NSY_INS5_IJiNSZ_IiLi128EEEEEELb0EEENSY_INS5_IJiNSZ_IiLi64EEEEEELb0EEEEEENS5_IJS14_S15_" + "S16_S17_S18_EEENS5_IJS1X_S17_S18_S1Y_NSA_IJLi7ELi8EEEEEEENSA_IJLi5ELi6ELi7ELi8EEEElEENS_31BlockToCTileMap_" + "M00_N0_M01AdaptILi128ELi64ENSJ_INS5_IJSS_SX_SX_EEENS5_IJS14_S15_S16_EEENS5_IJS1X_S17_S18_EEENSA_IJLi3ELi4EEE" + "ElEEEENS1_30ComputePtrOffsetOfStridedBatchILi1ELi1ELi0EvEES2M_Lb1ELb0ELb0EEEvT0_T1_T2_PT3_T4_T5_T6_T7_T8_T9_" + "T10_T11_T12_T13_" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 64 + assert result["mt_k"] == 32 + + def test_anchor3_bwd_weight_64x64x4(self): + # kernel_batched_gemm_xdlops_bwd_weight / GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight + # 11PassThroughES26_S26_Li64ELi64ELi4E -> M=64, N=64, K=4 + name = ( + "_ZN2ck16tensor_operation6device37kernel_batched_gemm_xdlops_bwd_weightINS_47GridwiseGemm_bk0mk1_bk0nk1" + "_mn_xdlops_bwd_weightILi64EttftLNS_25InMemoryDataOperationEnumE1ENS_16TensorDescriptorINS_5TupleIJNS_5Embed" + "INS6_IJiiEEENS6_IJiNS_17integral_constantIiLi1EEEEEELb0EEENS_8RightPadIiiLb0EEENS_11PassThroughIiEENS_7UnMerge" + "INS6_IJiiiEEELb0EEESG_SG_SG_SE_SG_EEENS6_IJNS_8SequenceIJLi0EEEENSL_IJLi1EEEENSL_IJLi2EEEENSL_IJLi3EEEENSL_" + "IJLi4EEEENSL_IJLi5EEEENSL_IJLi6EEEENSL_IJLi8EEEENSL_IJLi7EEEEEEENS6_IJNSL_IJLi1ELi2EEEESP_SQ_NSL_IJLi5ELi6E" + "Li7EEEEST_NSL_IJLi9EEEENSL_IJLi10EEEENSL_IJLi11EEEENSL_IJLi12EEEEEEENSL_IJLi9ELi10ELi11ELi12EEEElEENS5_INS6_" + "IJNS7_INS6_IJiiiiEEES15_Lb0EEESG_NS_3PadIiiiLb0EEES18_SG_SG_NS7_IS8_S8_Lb0EEES19_SG_NS_23Merge_v2_magic_divis" + "ionISI_EES1B_SE_SG_SJ_SG_SG_SG_SE_SG_EEENS6_IJSM_SN_SO_SP_SQ_SR_SS_SU_ST_NSL_IJLi10ELi12ELi14EEEENSL_IJLi9ELi" + "11ELi13EEEENSL_IJLi16EEEENSL_IJLi15EEEENSL_IJLi17EEEENSL_IJLi18EEEENSL_IJLi19EEEENSL_IJLi20EEEENSL_IJLi22EEEEN" + "SL_IJLi21EEEEEEENS6_IJNSL_IJLi1ELi2ELi3ELi4EEEESR_SS_SU_ST_SY_NSL_IJLi10ELi11EEEENSL_IJLi12ELi13EEEENSL_IJLi14" + "EEEES1G_S1F_S1H_S1I_NSL_IJLi19ELi20ELi21EEEES1L_NSL_IJLi23EEEENSL_IJLi24EEEENSL_IJLi25EEEENSL_IJLi26EEEEEEENSL_" + "IJLi23ELi24ELi25ELi26EEEElEENS5_INS6_IJSC_SE_SE_EEENS6_IJSM_SN_SO_EEENS6_IJSW_SP_SQ_EEENSL_IJLi3ELi4EEEElEENS0_" + "12element_wise11PassThroughES26_S26_Li64ELi64ELi4ELi32ELi32ELi8ELi2ELi2ENSL_IJLi1ELi4ELi8ELi2EEEENSL_IJLi0ELi3ELi" + "1ELi2EEEENSL_IJLi0ELi2ELi1ELi3EEEELi2ELi8ELi4ELb0ELb1ELi8ELi8ELi4ES27_S28_S29_Li2ELi8ELi4ELb0ELb1ELi8ELi8ELi4E" + "Li1ELi1ELi8ENSL_IJLi1ELi16ELi1ELi4EEEELb1ELb1ELi1ELNS_15PipelineVersionE0EttEEtttS26_S26_S26_S14_S1Z_NS5_INS6_IJ" + "SC_SE_SE_NSH_INS6_IJiNS9_IiLi64EEEEEELb0EEES2F_EEENS6_IJSM_SN_SO_SP_SQ_EEENS6_IJSW_SP_SQ_NSL_IJLi5ELi6EEEENSL_IJ" + "Li7ELi8EEEEEEENSL_IJLi5ELi6ELi7ELi8EEEElEENS_38BlockToCTileMap_KSplit_M00_N00_M01_N01ILi64ELi64ES24_Lb0EEENS1_30Co" + "mputePtrOffsetOfStridedBatchILi1ELi1ELi0EvEELb1EEEvPKT0_PKT1_PT2_T3_T4_T5_iT6_T7_T8_T9_T10_" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 64 + assert result["mt_k"] == 4 + + def test_anchor3_bwd_weight_64x32x4(self): + # 11PassThroughES26_S26_Li64ELi32ELi4E -> M=64, N=32, K=4 + name = ( + "_ZN2ck16tensor_operation6device37kernel_batched_gemm_xdlops_bwd_weightINS_47GridwiseGemm_bk0mk1_bk0nk1" + "_mn_xdlops_bwd_weightILi64EttftLNS_25InMemoryDataOperationEnumE1ENS_16TensorDescriptorINS_5TupleIJNS_5Embed" + "INS6_IJiiEEENS6_IJiNS_17integral_constantIiLi1EEEEEELb0EEENS_8RightPadIiiLb0EEENS_11PassThroughIiEENS_7UnMerge" + "INS6_IJiiiEEELb0EEESG_SG_SG_SE_SG_EEENS6_IJNS_8SequenceIJLi0EEEENSL_IJLi1EEEENSL_IJLi2EEEENSL_IJLi3EEEENSL_" + "IJLi4EEEENSL_IJLi5EEEENSL_IJLi6EEEENSL_IJLi8EEEENSL_IJLi7EEEEEEENS6_IJNSL_IJLi1ELi2EEEESP_SQ_NSL_IJLi5ELi6E" + "Li7EEEEST_NSL_IJLi9EEEENSL_IJLi10EEEENSL_IJLi11EEEENSL_IJLi12EEEEEEENSL_IJLi9ELi10ELi11ELi12EEEElEENS5_INS6_" + "IJNS7_INS6_IJiiiiEEES15_Lb0EEESG_NS_3PadIiiiLb0EEES18_SG_SG_NS7_IS8_S8_Lb0EEES19_SG_NS_23Merge_v2_magic_divis" + "ionISI_EES1B_SE_SG_SJ_SG_SG_SG_SE_SG_EEENS6_IJSM_SN_SO_SP_SQ_SR_SS_SU_ST_NSL_IJLi10ELi12ELi14EEEENSL_IJLi9ELi" + "11ELi13EEEENSL_IJLi16EEEENSL_IJLi15EEEENSL_IJLi17EEEENSL_IJLi18EEEENSL_IJLi19EEEENSL_IJLi20EEEENSL_IJLi22EEEEN" + "SL_IJLi21EEEEEEENS6_IJNSL_IJLi1ELi2ELi3ELi4EEEESR_SS_SU_ST_SY_NSL_IJLi10ELi11EEEENSL_IJLi12ELi13EEEENSL_IJLi14" + "EEEES1G_S1F_S1H_S1I_NSL_IJLi19ELi20ELi21EEEES1L_NSL_IJLi23EEEENSL_IJLi24EEEENSL_IJLi25EEEENSL_IJLi26EEEEEEENSL_" + "IJLi23ELi24ELi25ELi26EEEElEENS5_INS6_IJSC_SE_SE_EEENS6_IJSM_SN_SO_EEENS6_IJSW_SP_SQ_EEENSL_IJLi3ELi4EEEElEENS0_" + "12element_wise11PassThroughES26_S26_Li64ELi32ELi4ELi32ELi32ELi8ELi2ELi1ENSL_IJLi1ELi4ELi8ELi2EEEENSL_IJLi0ELi3ELi" + "1ELi2EEEENSL_IJLi0ELi2ELi1ELi3EEEELi2ELi8ELi4ELb0ELb1ELi8ELi8ELi4ENSL_IJLi1ELi4ELi4ELi4EEEES28_S29_Li2ELi8ELi2E" + "Lb0ELb1ELi8ELi4ELi4ELi1ELi1ELi8ENSL_IJLi1ELi16ELi1ELi4EEEELb1ELb1ELi1ELNS_15PipelineVersionE0EttEEtttS26_S26_S26_" + "S14_S1Z_NS5_INS6_IJSC_SE_SE_NSH_INS6_IJiNS9_IiLi64EEEEEELb0EEENSH_INS6_IJiNS9_IiLi32EEEEEELb0EEEEEENS6_IJSM_SN_SO_" + "SP_SQ_EEENS6_IJSW_SP_SQ_NSL_IJLi5ELi6EEEENSL_IJLi7ELi8EEEEEEENSL_IJLi5ELi6ELi7ELi8EEEElEENS_38BlockToCTileMap_KSplit_" + "M00_N00_M01_N01ILi64ELi32ES24_Lb0EEENS1_30ComputePtrOffsetOfStridedBatchILi1ELi1ELi0EvEELb1EEEvPKT0_PKT1_PT2_T3_T4_T5_" + "iT6_T7_T8_T9_T10_" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 64 + assert result["mt_n"] == 32 + assert result["mt_k"] == 4 + + +class TestRealCKDemangled: + # Source: MI300 vLLM FP8 inference trace + + def test_kernel_gemm_xdl_abscale(self): + # ABScale layout: BlockSize=256, ScaleBlockM=1, ScaleBlockN=128, ScaleBlockK=128, + # MPerBlock=16, NPerBlock=128, KPerBlock=256 + name = ( + "void ck::kernel_gemm_xdl_cshuffle_v3, ck::tensor_layout::gemm::RowMajor, ck::f8_fnuz_t, ck::f8_fnuz_t, " + "float, float, ck::Tuple<>, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "(ck::tensor_operation::device::GemmSpecialization)2, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 16, 1, 16>, ck::Sequence<8>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t>, " + "true, (ck::InMemoryDataOperationEnum)0, 2, (ck::TailNumber)10>" + "(ck::GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3<" + "ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::ColumnMajor, " + "ck::Tuple<>, ck::tensor_layout::gemm::RowMajor, ck::f8_fnuz_t, ck::f8_fnuz_t, " + "float, float, ck::Tuple<>, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "(ck::tensor_operation::device::GemmSpecialization)2, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 16, 1, 16>, ck::Sequence<8>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t>::Argument)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 128 + assert result["mt_k"] == 256 + assert result["transpose"] == (False, True) # RowMajor A, ColumnMajor B + + def test_kernel_gemm_xdl_abscale_gemspec0(self): + # Same class, different GemmSpecialization enum value (0 vs 2) + name = ( + "void ck::kernel_gemm_xdl_cshuffle_v3, ck::tensor_layout::gemm::RowMajor, ck::f8_fnuz_t, ck::f8_fnuz_t, " + "float, float, ck::Tuple<>, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "(ck::tensor_operation::device::GemmSpecialization)0, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 16, 1, 16>, ck::Sequence<8>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t>, " + "true, (ck::InMemoryDataOperationEnum)0, 2, (ck::TailNumber)10>" + "(ck::GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3<" + "ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::ColumnMajor, " + "ck::Tuple<>, ck::tensor_layout::gemm::RowMajor, ck::f8_fnuz_t, ck::f8_fnuz_t, " + "float, float, ck::Tuple<>, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "(ck::tensor_operation::device::GemmSpecialization)0, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 16, 1, 16>, ck::Sequence<8>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t>::Argument)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 128 + assert result["mt_k"] == 256 + + def test_kernel_moe_gemm_blockscale(self): + # MoeGemmBlockScale: BlockSize=256, ScaleBlockM=1, ScaleBlockN=128, ScaleBlockK=128, + # MPerBlock=16, NPerBlock=128, KPerBlock=256 + name = ( + "void ck::kernel_moe_gemm, ck::tensor_layout::gemm::RowMajor, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, float, float, ck::Tuple, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "MulABScaleExpertWeightA8W8blkscale, " + "(ck::tensor_operation::device::GemmSpecialization)0, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 16, 1, 16>, ck::Sequence<2, 1, 1, 1>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "1, false, true, false, false, int, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, true>, " + "true, (ck::InMemoryDataOperationEnum)0, 2, (ck::TailNumber)1>" + "(ck::GridwiseMoeGemmBlockScale<" + "ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::ColumnMajor, " + "ck::Tuple, ck::tensor_layout::gemm::RowMajor, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, float, float, ck::Tuple, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "MulABScaleExpertWeightA8W8blkscale, " + "(ck::tensor_operation::device::GemmSpecialization)0, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 16, 1, 16>, ck::Sequence<2, 1, 1, 1>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "1, false, true, false, false, int, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, true>::Argument)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 128 + assert result["mt_k"] == 256 + assert result["transpose"] == (False, True) # RowMajor A, ColumnMajor B + + def test_kernel_moe_gemm_blockscale_tailnumber0(self): + # Same class, different TailNumber (0 vs 1) and InMemoryDataOperationEnum (1 vs 0) + name = ( + "void ck::kernel_moe_gemm, ck::tensor_layout::gemm::RowMajor, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, float, float, ck::Tuple, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "MulABScaleExpertWeightA8W8blkscale, " + "(ck::tensor_operation::device::GemmSpecialization)0, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 4, 1, 64>, ck::Sequence<2, 1, 1, 1>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "0, false, false, false, true, int, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, true>, " + "false, (ck::InMemoryDataOperationEnum)1, 2, (ck::TailNumber)0>" + "(ck::GridwiseMoeGemmBlockScale<" + "ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::ColumnMajor, " + "ck::Tuple, ck::tensor_layout::gemm::RowMajor, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, float, float, ck::Tuple, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "MulABScaleExpertWeightA8W8blkscale, " + "(ck::tensor_operation::device::GemmSpecialization)0, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 4, 1, 64>, ck::Sequence<2, 1, 1, 1>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "0, false, false, false, true, int, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, true>::Argument)" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 128 + assert result["mt_k"] == 256 + + def test_kernel_gemm_xdl_abscale_via_sglang(self): + # Source: MI300 SGLang FP8 inference trace — different GemmSpec variant + name = ( + "void ck::kernel_gemm_xdl_cshuffle_v3, ck::tensor_layout::gemm::RowMajor, ck::f8_fnuz_t, ck::f8_fnuz_t, " + "float, float, ck::Tuple<>, unsigned short, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "ck::tensor_operation::element_wise::PassThrough, " + "(ck::tensor_operation::device::GemmSpecialization)0, " + "256, 1, 128, 128, 16, 128, 256, 16, 16, 16, 16, 1, 2, " + "ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, ck::Sequence<1, 0, 2>, " + "2, 16, 16, false, 0, ck::Sequence<16, 16, 1>, ck::Sequence<1, 0, 2>, " + "ck::Sequence<1, 0, 2>, 2, 16, 16, false, 0, 1, 2, " + "ck::Sequence<1, 16, 1, 16>, ck::Sequence<8>, " + "(ck::BlockGemmPipelineScheduler)0, (ck::BlockGemmPipelineVersion)0, " + "ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t, ck::f8_fnuz_t>, " + "true, (ck::InMemoryDataOperationEnum)0, 2, (ck::TailNumber)10>" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 128 + assert result["mt_k"] == 256 + + +class TestRealTriton: + # Source: MI300 SGLang and vLLM FP8 inference traces + + def test_blockscale_128x128x128_splitk_7168_grid17(self): + name = ( + "_gemm_a8w8_blockscale_kernel_GROUP_K_128_GROUP_N_128" + "_BLOCK_SIZE_M_128_BLOCK_SIZE_N_128_BLOCK_SIZE_K_128" + "_GROUP_SIZE_M_1_NUM_KSPLIT_1_SPLITK_BLOCK_SIZE_7168" + "_EVEN_K_1_GRID_MN_17_cache_modifier_CG" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 128 + assert result["mt_k"] == 128 + assert result["transpose"] == (None, None) + + def test_blockscale_128x128x128_splitk_1536_grid24(self): + name = ( + "_gemm_a8w8_blockscale_kernel_GROUP_K_128_GROUP_N_128" + "_BLOCK_SIZE_M_128_BLOCK_SIZE_N_128_BLOCK_SIZE_K_128" + "_GROUP_SIZE_M_1_NUM_KSPLIT_1_SPLITK_BLOCK_SIZE_1536" + "_EVEN_K_1_GRID_MN_24_cache_modifier_CG" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 128 + assert result["mt_k"] == 128 + + def test_blockscale_128x128x128_splitk_2048_grid56(self): + name = ( + "_gemm_a8w8_blockscale_kernel_GROUP_K_128_GROUP_N_128" + "_BLOCK_SIZE_M_128_BLOCK_SIZE_N_128_BLOCK_SIZE_K_128" + "_GROUP_SIZE_M_1_NUM_KSPLIT_1_SPLITK_BLOCK_SIZE_2048" + "_EVEN_K_1_GRID_MN_56_cache_modifier_CG" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 128 + assert result["mt_n"] == 128 + assert result["mt_k"] == 128 + + def test_batched_gemm_16x128x128(self): + # Source: MI300 vLLM FP8 inference trace + name = ( + "_batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant_kernel" + "_HAS_BIAS_0_BLOCK_SIZE_M_16_BLOCK_SIZE_N_128_BLOCK_SIZE_K_128" + "_GROUP_SIZE_M_1_EVEN_K_1_cache_modifier_CG_GRID_MN_8" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 128 + assert result["mt_k"] == 128 + + def test_batched_gemm_16x64x128(self): + # Source: MI300 vLLM FP8 inference trace + name = ( + "_batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant_kernel" + "_HAS_BIAS_0_BLOCK_SIZE_M_16_BLOCK_SIZE_N_64_BLOCK_SIZE_K_128" + "_GROUP_SIZE_M_1_EVEN_K_1_cache_modifier_CG_GRID_MN_4" + ) + result = gemm_name_parser(name) + assert result is not None + assert result["mt_m"] == 16 + assert result["mt_n"] == 64 + assert result["mt_k"] == 128 + + +class TestRealNvjet: + # Source: B200 eager inference trace and H100 graph capture trace + # Transpose is the last 3 chars: T=transposed, N=not transposed, order = A, B, C. + + def test_tst_bias_tnt(self): + # Source: B200 eager inference trace + result = gemm_name_parser("nvjet_tst_64x64_64x16_2x1_2cta_v_bz_bias_TNT") + assert result is not None + assert result["transpose"] == (True, False) + + def test_tst_bias_tnn(self): + # Source: B200 eager inference trace + result = gemm_name_parser("nvjet_tst_144x128_64x6_2x1_v_bz_bias_TNN") + assert result is not None + assert result["transpose"] == (True, False) + + def test_tst_no_bias_tnt(self): + # Source: B200 eager inference trace + result = gemm_name_parser("nvjet_tst_192x8_64x8_4x1_v_bz_TNT") + assert result is not None + assert result["transpose"] == (True, False) + + def test_hsh_coopa_tnt(self): + # Source: H100 graph capture trace + result = gemm_name_parser("nvjet_hsh_256x144_64x4_1x2_h_bz_coopA_TNT") + assert result is not None + assert result["transpose"] == (True, False) + + def test_hsh_coopa_bias_tnn(self): + # Source: H100 graph capture trace + result = gemm_name_parser("nvjet_hsh_128x256_64x4_2x1_v_bz_coopA_bias_TNN") + assert result is not None + assert result["transpose"] == (True, False) diff --git a/tests/traces/compare_test_e2e/reference/compare_h100_mi300_qwen_csvs/GEMM_intersect_mi300.csv b/tests/traces/compare_test_e2e/reference/compare_h100_mi300_qwen_csvs/GEMM_intersect_mi300.csv index 42f8da6d6..b566cc52d 100644 --- a/tests/traces/compare_test_e2e/reference/compare_h100_mi300_qwen_csvs/GEMM_intersect_mi300.csv +++ b/tests/traces/compare_test_e2e/reference/compare_h100_mi300_qwen_csvs/GEMM_intersect_mi300.csv @@ -1,6 +1,6 @@ -name,param: B,param: K,param: M,param: N,param: bias,param: dtype_A_B,param: stride_A,param: stride_B,Kernel Time (µs)_sum__mi300_diff,Kernel Time (µs)_sum__mi300_pct,Kernel Time (µs)_mean__mi300_diff,Kernel Time (µs)_mean__mi300_pct,name_count__mi300_diff,name_count__mi300_pct,TFLOPS/s_mean__mi300_diff,TFLOPS/s_mean__mi300_pct,TB/s_mean__mi300_diff,TB/s_mean__mi300_pct,h100::GFLOPS_first,h100::Data Moved (MB)_first,h100::FLOPS/Byte_first,h100::TB/s_mean,h100::TB/s_median,h100::TB/s_std,h100::TB/s_min,h100::TB/s_max,h100::TFLOPS/s_mean,h100::TFLOPS/s_median,h100::TFLOPS/s_std,h100::TFLOPS/s_min,h100::TFLOPS/s_max,h100::process_name_first,h100::process_label_first,h100::thread_name_first,h100::Compute Spec,h100::kernel_details__summarize_kernel_stats,h100::trunc_kernel_details,h100::Input Dims_first,h100::Input type_first,h100::Input Strides_first,h100::Concrete Inputs_first,h100::Kernel Time (µs)_mean,h100::Kernel Time (µs)_median,h100::Kernel Time (µs)_std,h100::Kernel Time (µs)_min,h100::Kernel Time (µs)_max,h100::Kernel Time (µs)_sum,h100::name_count,h100::UID_first,mi300::param: transpose,mi300::TB/s_mean,mi300::TB/s_median,mi300::TB/s_std,mi300::TB/s_min,mi300::TB/s_max,mi300::TFLOPS/s_mean,mi300::TFLOPS/s_median,mi300::TFLOPS/s_std,mi300::TFLOPS/s_min,mi300::TFLOPS/s_max,mi300::process_name_first,mi300::process_label_first,mi300::thread_name_first,mi300::Compute Spec,mi300::kernel_details__summarize_kernel_stats,mi300::trunc_kernel_details,mi300::Kernel Time (µs)_mean,mi300::Kernel Time (µs)_median,mi300::Kernel Time (µs)_std,mi300::Kernel Time (µs)_min,mi300::Kernel Time (µs)_max,mi300::Kernel Time (µs)_sum,mi300::name_count,mi300::UID_first -aten::mm,1,1024,8944,2816,False,"('c10::BFloat16', 'c10::BFloat16')","(1024, 1)","(1, 1024)",1253.02099609375,30.898717507688296,26.104604085286482,30.89871750768832,0,0.0,-143.84136455424243,-23.554845497884315,-0.20763255496852728,-23.554845497884322,51.581550592,71.0078125,692.7688414567059,0.8814855312346527,0.8815297729018408,0.0125245493923166,0.8566940617991415,0.9050115804454784,610.6657102342793,610.6963594828014,8.676617572282526,593.490952675431,626.9638240901165,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(39.75), 'mean_duration_us': np.float64(0.828125), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.07937112851450877), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.057)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(4015.5009999999997), 'mean_duration_us': np.float64(83.65627083333332), 'median_duration_us': np.float64(83.679), 'std_dev_duration_us': np.float64(1.1791844593401435), 'min_duration_us': np.float64(81.472), 'max_duration_us': np.float64(86.016)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(83.66)}]","[[8944, 1024], [1024, 2816]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",84.4844258626302,84.4635009765625,1.1997237297511991,82.27197265625,86.912109375,4055.25244140625,48,215,"(True, False)",0.6738529762661254,0.674108665064226,0.0198295327654339,0.62806071132492,0.7091151238095238,466.8243456800369,467.0014789124705,13.737282440537475,435.1008913490395,491.2528627809524,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(5308.273), 'mean_duration_us': np.float64(110.58902083333334), 'median_duration_us': np.float64(110.453), 'std_dev_duration_us': np.float64(3.2462516941953186), 'min_duration_us': np.float64(105.0), 'max_duration_us': np.float64(118.551)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(110.59)}]",110.58902994791669,110.452880859375,3.280588398059727,105.0,118.55078125,5308.2734375,48,231 -aten::addmm,1,1024,8944,1024,True,"('c10::BFloat16', 'c10::BFloat16')","(1024, 1)","(1, 1024)",625.686767578125,26.864163599284897,8.690093994140632,26.864163599284925,0,0.0,-122.57930111386281,-21.12764373989647,-0.2530077416758212,-21.127643739896445,18.766086144,36.939453125,484.4883413525089,1.1975199165160728,1.1966690312454464,0.0120594282927426,1.1683671458638645,1.2214187300624355,580.1844380894671,579.77219409602,5.842652411210404,566.0602605903484,591.7631346248373,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(62.783), 'mean_duration_us': np.float64(0.8719861111111111), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.11567921318307071), 'min_duration_us': np.float64(0.767), 'max_duration_us': np.float64(1.216)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2266.292), 'mean_duration_us': np.float64(31.476277777777778), 'median_duration_us': np.float64(31.456), 'std_dev_duration_us': np.float64(0.31881400107892144), 'min_duration_us': np.float64(30.912), 'max_duration_us': np.float64(32.256)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.87)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.48)}]","[[1024], [8944, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",32.34827677408854,32.3680419921875,0.3264361462154341,31.712158203125,33.152099609375,2329.075927734375,72,89,"(True, False)",0.9445121748402516,0.951401332232734,0.024813139318591,0.8603128998015335,0.9788727841162896,457.6051369756043,460.94285341400456,12.02167671221286,416.81156986901215,474.2524515716137,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(2954.763), 'mean_duration_us': np.float64(41.038375), 'median_duration_us': np.float64(40.7125), 'std_dev_duration_us': np.float64(1.1135532272552386), 'min_duration_us': np.float64(39.57), 'max_duration_us': np.float64(45.023)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(41.04)}]",41.03837076822917,40.71240234375,1.1213617423390887,39.56982421875,45.02294921875,2954.7626953125,72,90 -aten::mm,1,2816,8944,1024,False,"('c10::BFloat16', 'c10::BFloat16')","(2816, 1)","(1, 2816)",539.5068359375,29.39631805735436,22.47945149739583,29.396318057354353,0,0.0,-152.684246690696,-22.634945173465233,-0.22039710442179017,-22.634945173465262,51.581550592,71.0078125,692.7688414567059,0.9737028419232032,0.9733527752266448,0.0054710444705342,0.9622886978998384,0.9821816193669104,674.5509897222395,674.3084744224323,3.790169139410141,666.6436263909532,680.4248225488859,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.512999999999998), 'mean_duration_us': np.float64(0.8547083333333333), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.10426427926454754), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.088)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(1814.774), 'mean_duration_us': np.float64(75.61558333333333), 'median_duration_us': np.float64(75.6315), 'std_dev_duration_us': np.float64(0.4239531142184896), 'min_duration_us': np.float64(74.752), 'max_duration_us': np.float64(76.384)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.85)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(75.62)}]","[[8944, 2816], [2816, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[2816, 1], [1, 2816]]","['', '']",76.47029622395833,76.4954833984375,0.4304329398221454,75.807861328125,77.375,1835.287109375,24,235,"(True, False)",0.753305737501413,0.7572560566039392,0.0253954121366892,0.703202779015716,0.7950246671011402,521.8667430315435,524.6034010195848,17.593150244249767,487.1569745278536,550.7683175571602,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(2374.795), 'mean_duration_us': np.float64(98.94979166666667), 'median_duration_us': np.float64(98.32499999999999), 'std_dev_duration_us': np.float64(3.3136632546066833), 'min_duration_us': np.float64(93.654), 'max_duration_us': np.float64(105.883)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(98.95)}]",98.94974772135416,98.324951171875,3.3849417047301227,93.65380859375,105.8828125,2374.7939453125,24,253 -aten::mm,1,1024,8944,1024,False,"('c10::BFloat16', 'c10::BFloat16')","(1024, 1)","(1, 1024)",202.40576171875,26.01141415694309,8.433573404947929,26.01141415694313,0,0.0,-118.94614002917274,-20.560202306938443,-0.24561566677815094,-20.560202306938425,18.756927488,36.9375,484.2774957698816,1.1946169746358155,1.1954245756104465,0.0055271729586538,1.1843094336647857,1.2031439226446432,578.5261168808248,578.9172198584004,2.6766854791039054,573.5344067518289,582.6555259090998,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","[[8944, 1024], [1024, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",32.422587076822914,32.4000244140625,0.1504456175239329,32.192138671875,32.7041015625,778.14208984375,24,189,"(True, False)",0.9490013078576646,0.9555973894680864,0.0308235475155676,0.861800215638343,0.9878291064508096,459.5799768516521,462.774310735841,14.927150401583084,417.35045028328057,478.38340592059774,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(980.549), 'mean_duration_us': np.float64(40.856208333333335), 'median_duration_us': np.float64(40.5325), 'std_dev_duration_us': np.float64(1.3524004269928913), 'min_duration_us': np.float64(39.209), 'max_duration_us': np.float64(44.943)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(40.86)}]",40.85616048177084,40.532470703125,1.3815012260133064,39.208984375,44.94287109375,980.5478515625,24,204 -aten::bmm,1,1,32,559,False,"('float', 'float')","(32, 1, 1)","(559, 559, 1)",23.82177734375,676.7512831183243,23.82177734375,676.7512831183243,0,0.0,-0.0088551070785618,-87.12586613329592,-0.018295340306881,-87.12586613329559,3.5776e-05,0.0704917907714843,0.4840088749391201,0.0209987471216534,0.0209987471216534,,0.0209987471216534,0.0209987471216534,0.0101635799694825,0.0101635799694825,,0.0101635799694825,0.0101635799694825,python3,CPU,thread 25286 (python3),matrix_fp32,"[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2 >, false, true, false, false, false, true, false>(pytorch_flash::Flash_fwd_params)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(2722.705), 'mean_duration_us': np.float64(113.44604166666666), 'median_duration_us': np.float64(113.775), 'std_dev_duration_us': np.float64(2.127694555913488), 'min_duration_us': np.float64(109.28), 'max_duration_us': np.float64(119.647)}]","[{'name': 'void pytorch_flash::flash_fwd_kernel, std::array >(int, at::native::FillFunctor, std::array)', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(107.545), 'mean_duration_us': np.float64(4.481041666666667), 'median_duration_us': np.float64(4.69), 'std_dev_duration_us': np.float64(0.9520142890719772), 'min_duration_us': np.float64(0.681), 'max_duration_us': np.float64(5.411)}, {'name': 'attn_fwd', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(3908.798), 'mean_duration_us': np.float64(162.86658333333332), 'median_duration_us': np.float64(164.23700000000002), 'std_dev_duration_us': np.float64(9.253315184825505), 'min_duration_us': np.float64(138.557), 'max_duration_us': np.float64(173.077)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Fi...', 'stream': 0, 'mean_duration_us': np.float64(4.48)}, {'name': 'attn_fwd', 'stream': 0, 'mean_duration_us': np.float64(162.87)}]","{'B': 16, 'N_Q': 559, 'H_Q': 16, 'N_KV': 559, 'H_KV': 16, 'd_h_qk': 64, 'd_h_v': 64, 'dropout': 0.0, 'causal': True, 'flash_impl': True, 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16')}",True,13.503069593183117,31.34964817858372 -aten::addmm,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1024,), (8944, 1024), (1024, 1024), (), ())","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",625.686767578125,26.864163599284897,8.690093994140632,26.864163599284925,0,0.0,GEMM,python3,CPU,thread 25286 (python3),89,72,3146.172,43.69683333333333,7.040913515074629,18.766086144,36.939453125,484.4883413525089,matrix_bf16,1.1975199165160728,0.0120594282927426,580.1844380894671,5.842652411210404,32.34827677408854,0.3264361462154341,2329.075927734375,40.986,35.55,77.871,1.1966690312454464,1.1683671458638645,1.2214187300624355,579.77219409602,566.0602605903484,591.7631346248373,32.3680419921875,31.712158203125,33.152099609375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(62.783), 'mean_duration_us': np.float64(0.8719861111111111), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.11567921318307071), 'min_duration_us': np.float64(0.767), 'max_duration_us': np.float64(1.216)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2266.292), 'mean_duration_us': np.float64(31.476277777777778), 'median_duration_us': np.float64(31.456), 'std_dev_duration_us': np.float64(0.31881400107892144), 'min_duration_us': np.float64(30.912), 'max_duration_us': np.float64(32.256)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.87)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.48)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,7.299923704244882,28.543785627071003,GEMM,python3,CPU,thread 23232 (python3),90,72,2597.598,36.07775,23.478277126697943,18.766086144,36.939453125,484.4883413525089,matrix_bf16,0.9445121748402516,0.024813139318591,457.6051369756043,12.02167671221286,41.03837076822917,1.1213617423390887,2954.7626953125,31.1365,26.076,218.26,0.951401332232734,0.8603128998015335,0.9788727841162896,460.94285341400456,416.81156986901215,474.2524515716137,40.71240234375,39.56982421875,45.02294921875,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(2954.763), 'mean_duration_us': np.float64(41.038375), 'median_duration_us': np.float64(40.7125), 'std_dev_duration_us': np.float64(1.1135532272552386), 'min_duration_us': np.float64(39.57), 'max_duration_us': np.float64(45.023)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(41.04)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,9.93400307350023,41.28365125208395 +aten::addmm,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1024,), (8944, 1024), (1024, 1024), (), ())","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",625.686767578125,26.864163599284897,8.690093994140632,26.864163599284925,0,0.0,GEMM,python3,CPU,thread 25286 (python3),89,72,3146.172,43.69683333333333,7.040913515074629,18.766086144,36.939453125,484.4883413525089,matrix_bf16,1.1975199165160728,0.0120594282927426,580.1844380894671,5.842652411210404,32.34827677408854,0.3264361462154341,2329.075927734375,40.986,35.55,77.871,1.1966690312454464,1.1683671458638645,1.2214187300624355,579.77219409602,566.0602605903484,591.7631346248373,32.3680419921875,31.712158203125,33.152099609375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(62.783), 'mean_duration_us': np.float64(0.8719861111111111), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.11567921318307071), 'min_duration_us': np.float64(0.767), 'max_duration_us': np.float64(1.216)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2266.292), 'mean_duration_us': np.float64(31.476277777777778), 'median_duration_us': np.float64(31.456), 'std_dev_duration_us': np.float64(0.31881400107892144), 'min_duration_us': np.float64(30.912), 'max_duration_us': np.float64(32.256)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.87)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.48)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,7.299923704244882,28.543785627071003,GEMM,python3,CPU,thread 23232 (python3),90,72,2597.598,36.07775,23.478277126697943,18.766086144,36.939453125,484.4883413525089,matrix_bf16,0.9445121748402516,0.024813139318591,457.6051369756043,12.02167671221286,41.03837076822917,1.1213617423390887,2954.7626953125,31.1365,26.076,218.26,0.951401332232734,0.8603128998015335,0.9788727841162896,460.94285341400456,416.81156986901215,474.2524515716137,40.71240234375,39.56982421875,45.02294921875,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(2954.763), 'mean_duration_us': np.float64(41.038375), 'median_duration_us': np.float64(40.7125), 'std_dev_duration_us': np.float64(1.1135532272552386), 'min_duration_us': np.float64(39.57), 'max_duration_us': np.float64(45.023)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(41.04)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,9.93400307350023,41.28365125208395 aten::copy_,"('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((16, 16, 559, 64), (16, 16, 559, 64), ())","((572416, 35776, 64, 1), (572416, 64, 1024, 1), ())","('', '', 'False')",-1020.003173828125,-48.3829382208991,-14.166710747612846,-48.3829382208991,0,0.0,elementwise,python3,CPU,thread 25286 (python3),145,72,513.764,7.135611111111111,2.1930469417887277,0.009158656,34.9375,0.25,vector_bf16,1.2513164151278826,0.0138102190092724,0.3128291037819706,0.0034525547523181,29.280385335286457,0.3233270218510626,2108.187744140625,5.84,5.44,17.08,1.2498160942179874,1.2191995247202971,1.2777089764562632,0.3124540235544968,0.3047998811800742,0.3194272441140657,29.31201171875,28.672119140625,30.048095703125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2108.187), 'mean_duration_us': np.float64(29.280375), 'median_duration_us': np.float64(29.312), 'std_dev_duration_us': np.float64(0.3210814381594732), 'min_duration_us': np.float64(28.672), 'max_duration_us': np.float64(30.048)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(29.28)}]","{'op_shape': (16, 16, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 35776, 64, 1), 'stride_output': (572416, 64, 1024, 1)}",True,6.607603257237321,35.151388884308325,elementwise,python3,CPU,thread 23232 (python3),146,72,519.891,7.220708333333333,1.6625682613329145,0.009158656,34.9375,0.25,vector_bf16,2.427453889163802,0.0953509917895861,0.6068634722909504,0.0238377479473965,15.11367458767361,0.5665176534834554,1088.1845703125,6.822,5.069,14.67,2.411303549799132,2.250853807098071,2.8206966409263505,0.6028258874497829,0.5627134517745177,0.7051741602315875,15.19287109375,12.98779296875,16.27587890625,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(1088.184), 'mean_duration_us': np.float64(15.113666666666667), 'median_duration_us': np.float64(15.193), 'std_dev_duration_us': np.float64(0.5625623422242988), 'min_duration_us': np.float64(12.988), 'max_duration_us': np.float64(16.276)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::na...', 'stream': 0, 'mean_duration_us': np.float64(15.11)}]","{'op_shape': (16, 16, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 35776, 64, 1), 'stride_output': (572416, 64, 1024, 1)}",True,3.658509999185101,61.344488249341325 aten::cat,"('TensorList', 'Scalar')","(((16, 16, 559, 32), (16, 16, 559, 32)), ())","(((286208, 32, 512, 1), (572416, 64, 1024, 1)), ())","('', '-1')",-664.590087890625,-32.99315947540343,-13.845626831054684,-32.993159475403424,0,0.0,other,python3,CPU,thread 25286 (python3),129,48,554.286,11.547624999999998,1.8371072615461328,,,,,,,,,41.96514383951823,0.2933502450882166,2014.326904296875,12.645,9.26,16.121,,,,,,,41.9678955078125,41.3759765625,42.464111328125,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 4, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(2014.326), 'mean_duration_us': np.float64(41.965125), 'median_duration_us': np.float64(41.968), 'std_dev_duration_us': np.float64(0.290296301575362), 'min_duration_us': np.float64(41.376), 'max_duration_us': np.float64(42.464)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 4, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(1349.738), 'mean_duration_us': np.float64(28.119541666666667), 'median_duration_us': np.float64(27.903), 'std_dev_duration_us': np.float64(0.8283528575415325), 'min_duration_us': np.float64(26.82), 'max_duration_us': np.float64(30.348)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(1671.096), 'mean_duration_us': np.float64(34.8145), 'median_duration_us': np.float64(34.784), 'std_dev_duration_us': np.float64(0.3216488121124247), 'min_duration_us': np.float64(34.271), 'max_duration_us': np.float64(35.615)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(34.81)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (16, 16, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (572416, 35776, 64, 1), 'stride_output': None}",True,5.237645012877888,52.454715507632,elementwise,python3,CPU,thread 23232 (python3),132,48,453.22,9.442083333333334,3.9467750089954703,0.009158656,52.40625,0.1666666666666666,vector_bf16,2.689082598561989,0.1203802188890865,0.4481804330936647,0.0200633698148477,20.47682698567708,0.9517316900632534,982.8876953125,8.437000000000001,7.003,28.739,2.692939782921997,2.375597689196606,2.922702044564484,0.4488232971536661,0.3959329481994342,0.4871170074274139,20.406005859375,18.8017578125,23.1318359375,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(982.889), 'mean_duration_us': np.float64(20.476854166666666), 'median_duration_us': np.float64(20.406), 'std_dev_duration_us': np.float64(0.94175673587502), 'min_duration_us': np.float64(18.802), 'max_duration_us': np.float64(23.132)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::na...', 'stream': 0, 'mean_duration_us': np.float64(20.48)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (16, 16, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (572416, 35776, 64, 1), 'stride_output': None}",True,3.3044986663835205,68.19747577154502 aten::mul,"('float', 'float')","((16, 559, 1024), (16, 559, 1))","((572416, 1024, 1), (559, 1, 1))","('', '')",-454.756591796875,-28.264673994274787,-9.280746771364804,-28.26467399427481,0,0.0,elementwise,python3,CPU,thread 25286 (python3),77,49,436.2,8.902040816326531,0.7633815384847344,0.009158656,69.90911865234375,0.1249389946315275,vector_fp32,2.232889871737575,0.0290138600557387,0.2789750156978132,0.0036249625057438,32.83514528858419,0.4306580406298848,1608.922119140625,8.69,8.28,12.321,2.24587209729754,2.163222010677156,2.27259389729112,0.2805970019073548,0.2702707831787955,0.2839355967332975,32.639892578125,32.256103515625,33.886962890625,"[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(1608.9229999999998), 'mean_duration_us': np.float64(32.835163265306115), 'median_duration_us': np.float64(32.64), 'std_dev_duration_us': np.float64(0.42625121207266925), 'min_duration_us': np.float64(32.256), 'max_duration_us': np.float64(33.887)}]","[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(32.84)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1), 'dtype_in1_in2_out': ('float', 'float', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (559, 1, 1), 'stride_output': None}",True,5.042776225515153,57.49749173314716,elementwise,python3,CPU,thread 23232 (python3),78,49,442.585,9.03234693877551,2.095407354499959,0.009158656,69.90911865234375,0.1249389946315275,vector_fp32,3.128810266926054,0.2357826710054091,0.3909104091425425,0.029458449866952,23.554398517219383,1.7106005219749216,1154.16552734375,8.786,7.227,20.135,3.125597292471686,2.6929395890868,3.975128793708793,0.390508983344437,0.3364531648639437,0.4966485950168135,23.453125,18.44091796875,27.22119140625,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 49, 'total_duration_us': np.float64(1154.1640000000002), 'mean_duration_us': np.float64(23.55436734693878), 'median_duration_us': np.float64(23.453), 'std_dev_duration_us': np.float64(1.6930433736488344), 'min_duration_us': np.float64(18.441), 'max_duration_us': np.float64(27.221)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 4, at::na...', 'stream': 0, 'mean_duration_us': np.float64(23.55)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1), 'dtype_in1_in2_out': ('float', 'float', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (559, 1, 1), 'stride_output': None}",True,3.880340006373413,57.68597825015623 aten::mul,"('c10::BFloat16', 'c10::BFloat16')","((16, 16, 559, 64), (1, 1, 559, 64))","((572416, 64, 1024, 1), (35776, 35776, 64, 1))","('', '')",-582.743408203125,-37.893423857367175,-12.140487670898434,-37.89342385736717,0,0.0,elementwise,python3,CPU,thread 25286 (python3),123,48,562.537,11.719520833333334,4.2320548702833065,0.009158656,35.0057373046875,0.2495126705653021,vector_bf16,1.145799975045878,0.0113659258518683,0.2858916117073536,0.0028359425127468,32.03850809733073,0.3186166119751925,1537.848388671875,14.5,6.89,21.42,1.1470691458092843,1.1234709276742014,1.1633548975602963,0.2862082858939344,0.2803202314664674,0.2902717873054929,32.0,31.552001953125,32.672119140625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(1537.849), 'mean_duration_us': np.float64(32.03852083333333), 'median_duration_us': np.float64(32.0), 'std_dev_duration_us': np.float64(0.31526120054007917), 'min_duration_us': np.float64(31.552), 'max_duration_us': np.float64(32.672)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(32.04)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (1, 1, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (35776, 35776, 64, 1), 'stride_output': None}",True,4.820012852445286,62.317504585592445,elementwise,python3,CPU,thread 23232 (python3),124,48,419.069,8.730604166666668,1.7582366860686272,0.009158656,35.0057373046875,0.2495126705653021,vector_bf16,1.8501508764556909,0.1012320060205431,0.4616360861331937,0.0252586681688684,19.898020426432293,1.0929181120532645,955.10498046875,8.2185,6.529,18.027,1.831224988062527,1.614809968165317,2.025604883811166,0.4569138371773946,0.4029155476123987,0.5054140840698426,20.044921875,18.12109375,22.73095703125,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(955.105), 'mean_duration_us': np.float64(19.898020833333334), 'median_duration_us': np.float64(20.045), 'std_dev_duration_us': np.float64(1.0814598060334801), 'min_duration_us': np.float64(18.121), 'max_duration_us': np.float64(22.731)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::na...', 'stream': 0, 'mean_duration_us': np.float64(19.9)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (1, 1, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (35776, 35776, 64, 1), 'stride_output': None}",True,3.21109232445094,74.70520013571573 @@ -16,7 +16,7 @@ aten::mul,"('c10::BFloat16', 'c10::BFloat16')","((16, 16, 559, 64), (1, 1, 559, aten::silu,"('c10::BFloat16',)","((16, 559, 2816),)","((1574144, 2816, 1),)","('',)",-147.47216796875,-15.940959604356511,-6.144673665364586,-15.94095960435652,0,0.0,elementwise,python3,CPU,thread 25286 (python3),217,24,306.045,12.751875,1.1219036737888284,0.025186304,96.078125,0.25,vector_bf16,2.613825645550667,0.0244029009512421,0.6534564113876667,0.0061007252378105,38.54644775390625,0.3625162870903211,925.11474609375,12.41,12.09,17.14,2.6224883540301014,2.5658473790517644,2.641178232798679,0.6556220885075253,0.6414618447629411,0.6602945581996698,38.4158935546875,38.14404296875,39.263916015625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(925.115), 'mean_duration_us': np.float64(38.546458333333334), 'median_duration_us': np.float64(38.416), 'std_dev_duration_us': np.float64(0.35488331171042203), 'min_duration_us': np.float64(38.144), 'max_duration_us': np.float64(39.264)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(a...', 'stream': 7, 'mean_duration_us': np.float64(38.55)}]","{'op_shape': (16, 559, 2816), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (1574144, 2816, 1), 'stride_output': None}",True,2.899547835147452,85.07401687597186,elementwise,python3,CPU,thread 23232 (python3),234,24,220.281,9.178375,1.9646155462495,0.025186304,96.078125,0.25,vector_bf16,3.111357887512081,0.0812583466327194,0.7778394718780203,0.0203145866581798,32.401774088541664,0.8776191306910622,777.642578125,8.828,7.464,17.731,3.1216660858225627,2.871864071710929,3.201188499650909,0.7804165214556407,0.7179660179277323,0.8002971249127272,32.27294921875,31.47119140625,35.080078125,"[{'name': 'void at::native::vectorized_elementwise_kernel<8, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(777.643), 'mean_duration_us': np.float64(32.40179166666667), 'median_duration_us': np.float64(32.272999999999996), 'std_dev_duration_us': np.float64(0.8591138835629157), 'min_duration_us': np.float64(31.471), 'max_duration_us': np.float64(35.08)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<8, at::native::(a...', 'stream': 0, 'mean_duration_us': np.float64(32.4)}]","{'op_shape': (16, 559, 2816), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (1574144, 2816, 1), 'stride_output': None}",True,2.61445826882601,86.17658698850758 aten::copy_,"('c10::BFloat16', 'float', 'Scalar')","((16, 559, 1024), (16, 559, 1024), ())","((572416, 1024, 1), (572416, 1024, 1), ())","('', '', 'False')",-216.952392578125,-24.148013605586367,-4.427599848533161,-24.148013605586353,0,0.0,elementwise,python3,CPU,thread 25286 (python3),81,49,325.833,6.64965306122449,0.2256175369052318,0.009158656,52.40625,0.1666666666666666,vector_bf16,2.9981934625636244,0.0579875708530817,0.4996989104272706,0.0096645951421802,18.335254902742346,0.3646247760047354,898.427490234375,6.62,6.38,7.5,3.0127174024708547,2.815157838957401,3.077681103946181,0.5021195670784757,0.4691929731595667,0.5129468506576967,18.239990234375,17.85498046875,19.52001953125,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(898.427), 'mean_duration_us': np.float64(18.335244897959186), 'median_duration_us': np.float64(18.24), 'std_dev_duration_us': np.float64(0.36088720215331926), 'min_duration_us': np.float64(17.855), 'max_duration_us': np.float64(19.52)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 7, 'mean_duration_us': np.float64(18.34)}]","{'op_shape': (16, 559, 1024), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (572416, 1024, 1), 'stride_output': (572416, 1024, 1)}",True,2.81590310320494,87.88991997917681,elementwise,python3,CPU,thread 23232 (python3),82,49,340.724,6.953551020408163,2.196181706026248,0.009158656,52.40625,0.1666666666666666,vector_bf16,3.968993721320353,0.2869884871057401,0.6614989535533919,0.0478314145176233,13.907655054209185,0.8931457341508564,681.47509765625,6.578,5.213,20.417,3.984900677289144,3.496817205070843,5.418206390063069,0.6641501128815239,0.5828028675118071,0.9030343983438446,13.7900390625,10.14208984375,15.71484375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 0, 'count': 49, 'total_duration_us': np.float64(681.476), 'mean_duration_us': np.float64(13.907673469387754), 'median_duration_us': np.float64(13.79), 'std_dev_duration_us': np.float64(0.8839741751642698), 'min_duration_us': np.float64(10.142), 'max_duration_us': np.float64(15.715)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 0, 'mean_duration_us': np.float64(13.91)}]","{'op_shape': (16, 559, 1024), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (572416, 1024, 1), 'stride_output': (572416, 1024, 1)}",True,2.291140241269046,93.2004470975595 aten::add,"('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((16, 559, 1024), (16, 559, 1024), ())","((572416, 1024, 1), (572416, 1024, 1), ())","('', '', '1')",-165.207275390625,-19.331093857171545,-3.4418182373046857,-19.33109385717154,0,0.0,elementwise,python3,CPU,thread 25286 (python3),191,48,573.702,11.952125,1.30855332861515,0.009158656,52.40625,0.1666666666666666,vector_bf16,3.0889613339158086,0.08968722206012,0.5148268889859681,0.0149478703433533,17.80457051595052,0.5205274181052388,854.619384765625,11.9,10.66,19.271,3.077512782083186,2.8909812843546505,3.22187099892644,0.5129187970138642,0.4818302140591083,0.5369784998210732,17.85595703125,17.055908203125,19.008056640625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(854.619), 'mean_duration_us': np.float64(17.8045625), 'median_duration_us': np.float64(17.856), 'std_dev_duration_us': np.float64(0.5150970744371872), 'min_duration_us': np.float64(17.056), 'max_duration_us': np.float64(19.008)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(17.8)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1024), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (572416, 1024, 1), 'stride_output': None}",True,2.67859721989676,90.56851719907355,elementwise,python3,CPU,thread 23232 (python3),207,48,398.657,8.305354166666666,1.7518041215866007,0.009158656,52.40625,0.1666666666666666,vector_bf16,3.831243379011916,0.1397165464887619,0.6385405631686526,0.0232860910814603,14.362752278645834,0.5517010273075365,689.412109375,8.114,6.398,17.4,3.828856017691288,3.2870367699047844,4.031724759188937,0.6381426696152145,0.5478394616507973,0.6719541265314894,14.35205078125,13.6298828125,16.7177734375,"[{'name': 'void at::native::vectorized_elementwise_kernel<8, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(689.413), 'mean_duration_us': np.float64(14.362770833333334), 'median_duration_us': np.float64(14.352), 'std_dev_duration_us': np.float64(0.5459252558570988), 'min_duration_us': np.float64(13.63), 'max_duration_us': np.float64(16.718)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<8, at::native::CU...', 'stream': 0, 'mean_duration_us': np.float64(14.36)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1024), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (572416, 1024, 1), 'stride_output': None}",True,2.317824718818987,90.90930685629046 -aten::mm,"('c10::BFloat16', 'c10::BFloat16')","((8944, 1024), (1024, 1024))","((1024, 1), (1, 1024))","('', '')",202.40576171875,26.01141415694309,8.433573404947929,26.01141415694313,0,0.0,GEMM,python3,CPU,thread 25286 (python3),189,24,1085.031,45.209625,1.5749301934150726,18.756927488,36.9375,484.2774957698816,matrix_bf16,1.1946169746358155,0.0055271729586538,578.5261168808248,2.6766854791039054,32.422587076822914,0.1504456175239329,778.14208984375,44.821,43.621,49.271,1.1954245756104465,1.1843094336647857,1.2031439226446432,578.9172198584004,573.5344067518289,582.6555259090998,32.4000244140625,32.192138671875,32.7041015625,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,2.438897684390508,93.00741488346408,GEMM,python3,CPU,thread 23232 (python3),204,24,1041.089,43.37870833333333,19.59709899888505,18.756927488,36.9375,484.2774957698816,matrix_bf16,0.9490013078576646,0.0308235475155676,459.5799768516521,14.927150401583084,40.85616048177084,1.3815012260133064,980.5478515625,38.6435,36.638,134.457,0.9555973894680864,0.861800215638343,0.9878291064508096,462.774310735841,417.35045028328057,478.38340592059774,40.532470703125,39.208984375,44.94287109375,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(980.549), 'mean_duration_us': np.float64(40.856208333333335), 'median_duration_us': np.float64(40.5325), 'std_dev_duration_us': np.float64(1.3524004269928913), 'min_duration_us': np.float64(39.209), 'max_duration_us': np.float64(44.943)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(40.86)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,3.2966320397197677,71.49410781126478 +aten::mm,"('c10::BFloat16', 'c10::BFloat16')","((8944, 1024), (1024, 1024))","((1024, 1), (1, 1024))","('', '')",202.40576171875,26.01141415694309,8.433573404947929,26.01141415694313,0,0.0,GEMM,python3,CPU,thread 25286 (python3),189,24,1085.031,45.209625,1.5749301934150726,18.756927488,36.9375,484.2774957698816,matrix_bf16,1.1946169746358155,0.0055271729586538,578.5261168808248,2.6766854791039054,32.422587076822914,0.1504456175239329,778.14208984375,44.821,43.621,49.271,1.1954245756104465,1.1843094336647857,1.2031439226446432,578.9172198584004,573.5344067518289,582.6555259090998,32.4000244140625,32.192138671875,32.7041015625,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,2.438897684390508,93.00741488346408,GEMM,python3,CPU,thread 23232 (python3),204,24,1041.089,43.37870833333333,19.59709899888505,18.756927488,36.9375,484.2774957698816,matrix_bf16,0.9490013078576646,0.0308235475155676,459.5799768516521,14.927150401583084,40.85616048177084,1.3815012260133064,980.5478515625,38.6435,36.638,134.457,0.9555973894680864,0.861800215638343,0.9878291064508096,462.774310735841,417.35045028328057,478.38340592059774,40.532470703125,39.208984375,44.94287109375,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(980.549), 'mean_duration_us': np.float64(40.856208333333335), 'median_duration_us': np.float64(40.5325), 'std_dev_duration_us': np.float64(1.3524004269928913), 'min_duration_us': np.float64(39.209), 'max_duration_us': np.float64(44.943)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(40.86)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,3.2966320397197677,71.49410781126478 aten::mean,"('float', 'ScalarList', 'Scalar', '')","((16, 559, 1024), (), (), ())","((572416, 1024, 1), (), (), ())","('', '[-1]', 'True', '')",-124.275146484375,-16.718763189036313,-2.5362274792729593,-16.718763189036316,0,0.0,reduce,python3,CPU,thread 25286 (python3),74,49,583.168,11.90138775510204,0.8492257948745272,0.009158656,34.937503814697266,0.2499999727034214,vector_fp32,2.4171547016614285,0.0729462051494216,0.6042886094353039,0.0182365492961735,15.169946787308673,0.4689977659830294,743.327392578125,11.72,11.05,14.96,2.430637989600713,2.227167885536178,2.584266533849996,0.6076594310520774,0.5567919105899813,0.6460665629208645,15.072021484375,14.176025390625,16.448974609375,"[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, float, 4> >(at::native::ReduceOp, unsigned int, float, 4>)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(743.3270000000001), 'mean_duration_us': np.float64(15.169938775510206), 'median_duration_us': np.float64(15.072), 'std_dev_duration_us': np.float64(0.46419596531071383), 'min_duration_us': np.float64(14.176), 'max_duration_us': np.float64(16.449)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, float, 4, 4> >(at::native::ReduceOp, unsigned int, float, 4, 4>)', 'stream': 0, 'count': 49, 'total_duration_us': np.float64(619.053), 'mean_duration_us': np.float64(12.633734693877551), 'median_duration_us': np.float64(12.828), 'std_dev_duration_us': np.float64(0.9567734633782996), 'min_duration_us': np.float64(8.017), 'max_duration_us': np.float64(13.871)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(658.0799999999999), 'mean_duration_us': np.float64(27.419999999999998), 'median_duration_us': np.float64(27.456), 'std_dev_duration_us': np.float64(0.2218289506203673), 'min_duration_us': np.float64(26.913), 'max_duration_us': np.float64(27.872)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(27.42)}]","{'op_shape': (16, 559, 16, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 1024, 64, 1), 'stride_output': (572416, 64, 35776, 1)}",True,2.062593188396829,97.3997875300918,elementwise,python3,CPU,thread 23232 (python3),194,24,169.22,7.050833333333333,1.0625545783083603,0.009158656,34.9375,0.25,vector_bf16,2.448720524362377,0.0777443849185206,0.612180131090594,0.0194360962296301,14.97515869140625,0.4748121556323811,359.40380859375,7.0015,5.659,11.217,2.4433422356977093,2.3193925421046124,2.610749180597119,0.6108355589244272,0.579848135526153,0.6526872951492796,14.99365234375,14.0322265625,15.794921875,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(359.404), 'mean_duration_us': np.float64(14.975166666666667), 'median_duration_us': np.float64(14.993500000000001), 'std_dev_duration_us': np.float64(0.464862315338878), 'min_duration_us': np.float64(14.032), 'max_duration_us': np.float64(15.795)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::na...', 'stream': 0, 'mean_duration_us': np.float64(14.98)}]","{'op_shape': (16, 559, 16, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 1024, 64, 1), 'stride_output': (572416, 64, 35776, 1)}",True,1.2083266601619251,98.22854632028032 aten::neg,"('c10::BFloat16',)","((16, 16, 559, 32),)","((572416, 64, 1024, 1),)","('',)",-70.22900390625,-11.957391054216417,-1.4631042480468768,-11.95739105421643,0,0.0,elementwise,python3,CPU,thread 25286 (python3),128,48,424.407,8.8418125,1.3407239177661494,0.004579328,17.46875,0.25,vector_bf16,1.4981145850997697,0.0409643573140178,0.3745286462749422,0.0102410893285044,12.235982259114584,0.3391273724226469,587.3271484375,9.37,7.14,12.69,1.5024171963634907,1.379312619762846,1.576908088694592,0.3756042990908726,0.3448281549407114,0.3942270221736479,12.19189453125,11.615966796875,13.280029296875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(587.328), 'mean_duration_us': np.float64(12.235999999999999), 'median_duration_us': np.float64(12.192), 'std_dev_duration_us': np.float64(0.3355813363900522), 'min_duration_us': np.float64(11.616), 'max_duration_us': np.float64(13.28)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(12.24)}]","{'op_shape': (16, 16, 559, 32), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (572416, 64, 1024, 1), 'stride_output': None}",True,1.8408345223833464,99.24062205247516,elementwise,python3,CPU,thread 23232 (python3),129,48,427.61,8.908541666666666,1.64628317615752,0.004579328,17.46875,0.25,vector_bf16,1.7086441588833885,0.1276272127087474,0.4271610397208469,0.0319068031771868,10.772878011067707,0.7221992437648039,517.09814453125,8.667,6.44,17.176,1.6799003616497248,1.5383357244320512,2.186122084848485,0.4199750904124311,0.3845839311080127,0.5465305212121211,10.90380859375,8.37890625,11.9072265625,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(517.0999999999999), 'mean_duration_us': np.float64(10.772916666666665), 'median_duration_us': np.float64(10.904), 'std_dev_duration_us': np.float64(0.7146025070313955), 'min_duration_us': np.float64(8.379), 'max_duration_us': np.float64(11.907)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::na...', 'stream': 0, 'mean_duration_us': np.float64(10.77)}]","{'op_shape': (16, 16, 559, 32), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (572416, 64, 1024, 1), 'stride_output': None}",True,1.738499868440847,97.0202196601184 @@ -24,7 +24,7 @@ aten::add,"('float', 'double', 'Scalar')","((16, 559, 1), (), ())","((559, 1, 1) aten::rsqrt,"('float',)","((16, 559, 1),)","((559, 1, 1),)","('',)",136.371337890625,172.45513234145423,2.783088528380102,172.45513234145423,0,0.0,elementwise,python3,CPU,thread 25286 (python3),76,49,395.084,8.062938775510204,1.234398477129706,8.944e-06,0.0682373046875,0.125,vector_fp32,0.0444819894962948,0.0023373962616093,0.0055602486870368,0.0002921745327011,1.6138044084821428,0.1026021369823694,79.076416015625,7.69,7.37,14.47,0.0447172706743973,0.0328819692583866,0.046586709903036,0.0055896588342996,0.0041102461572983,0.0058233387378795,1.60009765625,1.535888671875,2.176025390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array >(int, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(79.073), 'mean_duration_us': np.float64(1.613734693877551), 'median_duration_us': np.float64(1.6), 'std_dev_duration_us': np.float64(0.10155782016738636), 'min_duration_us': np.float64(1.536), 'max_duration_us': np.float64(2.176)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rs...', 'stream': 7, 'mean_duration_us': np.float64(1.61)}]","{'op_shape': (16, 559, 1), 'dtype_in_out': ('float', None), 'stride_input': (559, 1, 1), 'stride_output': None}",True,0.2478458502985414,99.74484955445378,elementwise,python3,CPU,thread 23232 (python3),77,49,424.39,8.661020408163266,3.4995645591618247,8.944e-06,0.0682373046875,0.125,vector_fp32,0.0169178580351796,0.0050691900222846,0.0021147322543974,0.0006336487527855,4.396892936862245,0.5811734917147134,215.44775390625,8.061,6.122,26.691,0.0160783954355935,0.0145130727938991,0.0415240850099178,0.0020097994294491,0.0018141340992373,0.0051905106262397,4.4501953125,1.72314453125,4.93017578125,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array >(int, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array)', 'stream': 0, 'count': 49, 'total_duration_us': np.float64(215.44599999999997), 'mean_duration_us': np.float64(4.396857142857142), 'median_duration_us': np.float64(4.45), 'std_dev_duration_us': np.float64(0.5752471341137846), 'min_duration_us': np.float64(1.723), 'max_duration_us': np.float64(4.93)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rs...', 'stream': 0, 'mean_duration_us': np.float64(4.4)}]","{'op_shape': (16, 559, 1), 'dtype_in_out': ('float', None), 'stride_input': (559, 1, 1), 'stride_output': None}",True,0.7243419760506526,99.70364224157883 aten::index_select,"('c10::BFloat16', 'Scalar', 'long int')","((151936, 1024), (), (8944,))","((1024, 1), (), (1,))","('', '0', '')",-30.76806640625,-59.425483557626116,-30.76806640625,-59.425483557626116,0,0.0,other,python3,CPU,thread 25286 (python3),4,1,73.341,73.341,,,,,,,,,,51.77587890625,,51.77587890625,73.341,73.341,73.341,,,,,,,51.77587890625,51.77587890625,51.77587890625,"[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex(at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, int, int, unsigned int, unsigned int, long)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(51.776), 'mean_duration_us': np.float64(51.776), 'median_duration_us': np.float64(51.776), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(51.776), 'max_duration_us': np.float64(51.776)}]","[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex(at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, int, int, unsigned int, unsigned int, long)', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(21.008), 'mean_duration_us': np.float64(21.008), 'median_duration_us': np.float64(21.008), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(21.008), 'max_duration_us': np.float64(21.008)}]","[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex, unsigned int, bool, 4> >(at::native::ReduceOp, unsigned int, bool, 4>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(4.128), 'mean_duration_us': np.float64(4.128), 'median_duration_us': np.float64(4.128), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(4.128), 'max_duration_us': np.float64(4.128)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, bool, 4, 4> >(at::native::ReduceOp, unsigned int, bool, 4, 4>)', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(2.325), 'mean_duration_us': np.float64(2.325), 'median_duration_us': np.float64(2.325), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(2.325), 'max_duration_us': np.float64(2.325)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<4u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.456), 'mean_duration_us': np.float64(3.456), 'median_duration_us': np.float64(3.456), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.456), 'max_duration_us': np.float64(3.456)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<4u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(6.333), 'mean_duration_us': np.float64(6.333), 'median_duration_us': np.float64(6.333), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(6.333), 'max_duration_us': np.float64(6.333)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 2, 'total_duration_us': np.float64(3.136), 'mean_duration_us': np.float64(1.568), 'median_duration_us': np.float64(1.568), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(1.568), 'max_duration_us': np.float64(1.568)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(1.57)}]","{'shape_in1': (1, 559, 64), 'shape_in2': (), 'dtype_in1_in2_out': ('float', 'double', None), 'stride_input1': (35776, 64, 1), 'stride_input2': (), 'stride_output': None}",True,0.0098297538814347,99.9517610529006,elementwise,python3,CPU,thread 23232 (python3),58,2,38.464,19.232,10.40436917837886,3.5776e-05,0.2729568481445312,0.1249965061352265,vector_fp32,0.0772556482055671,0.0029641423686581,0.009656686104908,0.0003705074397696,3.70751953125,0.1422499969965125,7.4150390625,19.232,11.875,26.589,0.0772556482055671,0.0751596830362866,0.0793516133748477,0.009656686104908,0.0093946977817668,0.0099186744280492,3.70751953125,3.60693359375,3.80810546875,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 0, 'count': 2, 'total_duration_us': np.float64(7.415), 'mean_duration_us': np.float64(3.7075), 'median_duration_us': np.float64(3.7075), 'std_dev_duration_us': np.float64(0.10049999999999981), 'min_duration_us': np.float64(3.607), 'max_duration_us': np.float64(3.808)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 0, 'mean_duration_us': np.float64(3.71)}]","{'shape_in1': (1, 559, 64), 'shape_in2': (), 'dtype_in1_in2_out': ('float', 'double', None), 'stride_input1': (35776, 64, 1), 'stride_input2': (), 'stride_output': None}",True,0.0249295894231562,99.8911247050097 aten::copy_,"('c10::BFloat16', 'float', 'Scalar')","((1, 559, 64), (1, 559, 64), ())","((35776, 64, 1), (35776, 64, 1), ())","('', '', 'False')",0.110107421875,3.701272055806319,0.0550537109375,3.701272055806319,0,0.0,elementwise,python3,CPU,thread 25286 (python3),62,2,13.74,6.87,2.319310242291876,3.5776e-05,0.2047119140625,0.1666666666666666,vector_bf16,0.144331376873417,0.0022614343650225,0.0240552294789028,0.0003769057275037,1.4874267578125,0.0233055213525839,2.974853515625,6.87,5.23,8.51,0.144331376873417,0.1427323012987013,0.1459304524481328,0.0240552294789028,0.0237887168831168,0.0243217420746887,1.4874267578125,1.470947265625,1.50390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 2, 'total_duration_us': np.float64(2.975), 'mean_duration_us': np.float64(1.4875), 'median_duration_us': np.float64(1.4875), 'std_dev_duration_us': np.float64(0.01649999999999996), 'min_duration_us': np.float64(1.471), 'max_duration_us': np.float64(1.504)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 7, 'mean_duration_us': np.float64(1.49)}]","{'op_shape': (1, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (35776, 64, 1), 'stride_output': (35776, 64, 1)}",True,0.0093239569551052,99.9610850098557,elementwise,python3,CPU,thread 23232 (python3),63,2,27.387,13.6935,4.347999597516082,3.5776e-05,0.2047119140625,0.1666666666666666,vector_bf16,0.1391874678431517,0.0026170667296354,0.0231979113071919,0.0004361777882725,1.54248046875,0.0290024265721044,3.0849609375,13.6935,10.619,16.768,0.1391874678431517,0.1373369222118088,0.1410380134744947,0.0231979113071919,0.0228894870353014,0.0235063355790824,1.54248046875,1.52197265625,1.56298828125,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 0, 'count': 2, 'total_duration_us': np.float64(3.085), 'mean_duration_us': np.float64(1.5425), 'median_duration_us': np.float64(1.5425), 'std_dev_duration_us': np.float64(0.020499999999999963), 'min_duration_us': np.float64(1.522), 'max_duration_us': np.float64(1.563)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 0, 'mean_duration_us': np.float64(1.54)}]","{'op_shape': (1, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (35776, 64, 1), 'stride_output': (35776, 64, 1)}",True,0.0103717335687805,99.98194878405788 diff --git a/tests/traces/compare_test_e2e/reference/h100_perf_report_csvs/GEMM.csv b/tests/traces/compare_test_e2e/reference/h100_perf_report_csvs/GEMM.csv index c69658de7..4fba6b709 100644 --- a/tests/traces/compare_test_e2e/reference/h100_perf_report_csvs/GEMM.csv +++ b/tests/traces/compare_test_e2e/reference/h100_perf_report_csvs/GEMM.csv @@ -1,6 +1,6 @@ -name,param: B,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::mm,1,8944,2816,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",51.581550592,71.0078125,692.7688414567059,0.8814855312346527,0.8815297729018408,0.012524549392316679,0.8566940617991415,0.9050115804454784,610.6657102342793,610.6963594828014,8.676617572282526,593.490952675431,626.9638240901165,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(39.75), 'mean_duration_us': np.float64(0.828125), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.07937112851450877), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.057)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(4015.5009999999997), 'mean_duration_us': np.float64(83.65627083333332), 'median_duration_us': np.float64(83.679), 'std_dev_duration_us': np.float64(1.1791844593401435), 'min_duration_us': np.float64(81.472), 'max_duration_us': np.float64(86.016)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(83.66)}]","[[8944, 1024], [1024, 2816]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",84.4844258626302,84.4635009765625,1.1997237297511993,82.27197265625,86.912109375,4055.25244140625,48,215 -aten::addmm,1,8944,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",18.766086144,36.939453125,484.4883413525089,1.1975199165160728,1.1966690312454464,0.012059428292742667,1.1683671458638643,1.2214187300624357,580.1844380894671,579.77219409602,5.842652411210404,566.0602605903484,591.7631346248373,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(62.783), 'mean_duration_us': np.float64(0.8719861111111111), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.11567921318307071), 'min_duration_us': np.float64(0.767), 'max_duration_us': np.float64(1.216)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2266.292), 'mean_duration_us': np.float64(31.476277777777778), 'median_duration_us': np.float64(31.456), 'std_dev_duration_us': np.float64(0.31881400107892144), 'min_duration_us': np.float64(30.912), 'max_duration_us': np.float64(32.256)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.87)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.48)}]","[[1024], [8944, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",32.34827677408854,32.3680419921875,0.32643614621543415,31.712158203125,33.152099609375,2329.075927734375,72,89 -aten::mm,1,8944,1024,2816,False,"(2816, 1)","(1, 2816)","('c10::BFloat16', 'c10::BFloat16')",51.581550592,71.0078125,692.7688414567059,0.9737028419232031,0.9733527752266449,0.00547104447053427,0.9622886978998384,0.9821816193669105,674.5509897222395,674.3084744224323,3.790169139410141,666.6436263909532,680.4248225488859,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.512999999999998), 'mean_duration_us': np.float64(0.8547083333333333), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.10426427926454754), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.088)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(1814.774), 'mean_duration_us': np.float64(75.61558333333333), 'median_duration_us': np.float64(75.6315), 'std_dev_duration_us': np.float64(0.4239531142184896), 'min_duration_us': np.float64(74.752), 'max_duration_us': np.float64(76.384)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.85)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(75.62)}]","[[8944, 2816], [2816, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[2816, 1], [1, 2816]]","['', '']",76.47029622395833,76.4954833984375,0.4304329398221454,75.807861328125,77.375,1835.287109375,24,235 -aten::mm,1,8944,1024,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",18.756927488,36.9375,484.27749576988157,1.1946169746358157,1.1954245756104465,0.005527172958653849,1.1843094336647855,1.2031439226446432,578.5261168808248,578.9172198584004,2.6766854791039054,573.5344067518289,582.6555259090998,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","[[8944, 1024], [1024, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",32.422587076822914,32.4000244140625,0.1504456175239329,32.192138671875,32.7041015625,778.14208984375,24,189 -aten::bmm,1,32,559,1,False,"(32, 1, 1)","(559, 559, 1)","('float', 'float')",3.5776e-05,0.07049179077148438,0.4840088749391201,0.020998747121653488,0.020998747121653488,,0.020998747121653488,0.020998747121653488,0.01016357996948259,0.01016357996948259,,0.01016357996948259,0.01016357996948259,python3,CPU,thread 25286 (python3),matrix_fp32,"[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2 >, false, true, false, false, false, true, false>(pytorch_flash::Flash_fwd_params)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(2722.705), 'mean_duration_us': np.float64(113.44604166666666), 'median_duration_us': np.float64(113.775), 'std_dev_duration_us': np.float64(2.127694555913488), 'min_duration_us': np.float64(109.28), 'max_duration_us': np.float64(119.647)}]","[{'name': 'void pytorch_flash::flash_fwd_kernel(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2108.187), 'mean_duration_us': np.float64(29.280375), 'median_duration_us': np.float64(29.312), 'std_dev_duration_us': np.float64(0.3210814381594732), 'min_duration_us': np.float64(28.672), 'max_duration_us': np.float64(30.048)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(29.28)}]","{'op_shape': (16, 16, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 35776, 64, 1), 'stride_output': (572416, 64, 1024, 1)}",True,6.607603257237321,35.151388884308325 aten::cat,other,python3,CPU,thread 25286 (python3),"(((16, 16, 559, 32), (16, 16, 559, 32)), ())","('TensorList', 'Scalar')","(((286208, 32, 512, 1), (572416, 64, 1024, 1)), ())","('', '-1')",129,48,554.286,11.547624999999998,1.8371072615461332,,,,,,,,,41.96514383951823,0.2933502450882166,2014.326904296875,12.645,9.26,16.121,,,,,,,41.9678955078125,41.3759765625,42.464111328125,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 4, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(2014.326), 'mean_duration_us': np.float64(41.965125), 'median_duration_us': np.float64(41.968), 'std_dev_duration_us': np.float64(0.290296301575362), 'min_duration_us': np.float64(41.376), 'max_duration_us': np.float64(42.464)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(1671.096), 'mean_duration_us': np.float64(34.8145), 'median_duration_us': np.float64(34.784), 'std_dev_duration_us': np.float64(0.3216488121124247), 'min_duration_us': np.float64(34.271), 'max_duration_us': np.float64(35.615)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(34.81)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (16, 16, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (572416, 35776, 64, 1), 'stride_output': None}",True,5.237645012877888,52.454715507632 aten::mul,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (16, 559, 1))","('float', 'float')","((572416, 1024, 1), (559, 1, 1))","('', '')",77,49,436.2,8.902040816326531,0.7633815384847344,0.009158656,69.90911865234375,0.12493899463152758,vector_fp32,2.232889871737575,0.029013860055738732,0.2789750156978132,0.003624962505743834,32.83514528858419,0.4306580406298848,1608.922119140625,8.69,8.28,12.321,2.2458720972975397,2.1632220106771562,2.2725938972911197,0.28059700190735487,0.2702707831787955,0.2839355967332975,32.639892578125,32.256103515625,33.886962890625,"[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(1608.9229999999998), 'mean_duration_us': np.float64(32.835163265306115), 'median_duration_us': np.float64(32.64), 'std_dev_duration_us': np.float64(0.42625121207266925), 'min_duration_us': np.float64(32.256), 'max_duration_us': np.float64(33.887)}]","[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(32.84)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1), 'dtype_in1_in2_out': ('float', 'float', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (559, 1, 1), 'stride_output': None}",True,5.042776225515153,57.49749173314716 aten::mul,elementwise,python3,CPU,thread 25286 (python3),"((16, 16, 559, 64), (1, 1, 559, 64))","('c10::BFloat16', 'c10::BFloat16')","((572416, 64, 1024, 1), (35776, 35776, 64, 1))","('', '')",123,48,562.537,11.719520833333334,4.2320548702833065,0.009158656,35.0057373046875,0.24951267056530213,vector_bf16,1.145799975045878,0.011365925851868387,0.2858916117073536,0.002835942512746873,32.03850809733073,0.3186166119751925,1537.848388671875,14.5,6.89,21.42,1.1470691458092843,1.1234709276742014,1.1633548975602963,0.28620828589393443,0.2803202314664674,0.29027178730549297,32.0,31.552001953125,32.672119140625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(1537.849), 'mean_duration_us': np.float64(32.03852083333333), 'median_duration_us': np.float64(32.0), 'std_dev_duration_us': np.float64(0.31526120054007917), 'min_duration_us': np.float64(31.552), 'max_duration_us': np.float64(32.672)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(32.04)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (1, 1, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (35776, 35776, 64, 1), 'stride_output': None}",True,4.820012852445286,62.317504585592445 @@ -16,7 +16,7 @@ aten::mul,elementwise,python3,CPU,thread 25286 (python3),"((16, 16, 559, 64), (1 aten::silu,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 2816),)","('c10::BFloat16',)","((1574144, 2816, 1),)","('',)",217,24,306.045,12.751875,1.1219036737888284,0.025186304,96.078125,0.25,vector_bf16,2.613825645550667,0.024402900951242156,0.6534564113876667,0.006100725237810539,38.54644775390625,0.3625162870903211,925.11474609375,12.41,12.09,17.14,2.6224883540301014,2.5658473790517644,2.641178232798679,0.6556220885075253,0.6414618447629411,0.6602945581996698,38.4158935546875,38.14404296875,39.263916015625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(925.115), 'mean_duration_us': np.float64(38.546458333333334), 'median_duration_us': np.float64(38.416), 'std_dev_duration_us': np.float64(0.35488331171042203), 'min_duration_us': np.float64(38.144), 'max_duration_us': np.float64(39.264)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(a...', 'stream': 7, 'mean_duration_us': np.float64(38.55)}]","{'op_shape': (16, 559, 2816), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (1574144, 2816, 1), 'stride_output': None}",True,2.8995478351474517,85.07401687597186 aten::copy_,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (16, 559, 1024), ())","('c10::BFloat16', 'float', 'Scalar')","((572416, 1024, 1), (572416, 1024, 1), ())","('', '', 'False')",81,49,325.833,6.64965306122449,0.22561753690523184,0.009158656,52.40625,0.16666666666666666,vector_bf16,2.9981934625636244,0.05798757085308172,0.49969891042727066,0.009664595142180274,18.335254902742346,0.3646247760047354,898.427490234375,6.62,6.38,7.5,3.0127174024708547,2.8151578389574006,3.077681103946181,0.5021195670784757,0.4691929731595667,0.5129468506576967,18.239990234375,17.85498046875,19.52001953125,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(898.427), 'mean_duration_us': np.float64(18.335244897959186), 'median_duration_us': np.float64(18.24), 'std_dev_duration_us': np.float64(0.36088720215331926), 'min_duration_us': np.float64(17.855), 'max_duration_us': np.float64(19.52)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 7, 'mean_duration_us': np.float64(18.34)}]","{'op_shape': (16, 559, 1024), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (572416, 1024, 1), 'stride_output': (572416, 1024, 1)}",True,2.8159031032049398,87.88991997917681 aten::add,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (16, 559, 1024), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((572416, 1024, 1), (572416, 1024, 1), ())","('', '', '1')",191,48,573.702,11.952125,1.30855332861515,0.009158656,52.40625,0.16666666666666666,vector_bf16,3.0889613339158086,0.08968722206012,0.5148268889859681,0.01494787034335336,17.80457051595052,0.5205274181052388,854.619384765625,11.899999999999999,10.66,19.271,3.077512782083186,2.8909812843546505,3.2218709989264394,0.5129187970138642,0.4818302140591083,0.5369784998210732,17.85595703125,17.055908203125,19.008056640625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(854.619), 'mean_duration_us': np.float64(17.8045625), 'median_duration_us': np.float64(17.856), 'std_dev_duration_us': np.float64(0.5150970744371872), 'min_duration_us': np.float64(17.056), 'max_duration_us': np.float64(19.008)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(17.8)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1024), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (572416, 1024, 1), 'stride_output': None}",True,2.6785972198967594,90.56851719907357 -aten::mm,GEMM,python3,CPU,thread 25286 (python3),"((8944, 1024), (1024, 1024))","('c10::BFloat16', 'c10::BFloat16')","((1024, 1), (1, 1024))","('', '')",189,24,1085.031,45.209624999999996,1.5749301934150726,18.756927488,36.9375,484.27749576988157,matrix_bf16,1.1946169746358157,0.005527172958653849,578.5261168808248,2.6766854791039054,32.422587076822914,0.1504456175239329,778.14208984375,44.821,43.621,49.271,1.1954245756104465,1.1843094336647855,1.2031439226446432,578.9172198584004,573.5344067518289,582.6555259090998,32.4000244140625,32.192138671875,32.7041015625,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,2.4388976843905077,93.00741488346408 +aten::mm,GEMM,python3,CPU,thread 25286 (python3),"((8944, 1024), (1024, 1024))","('c10::BFloat16', 'c10::BFloat16')","((1024, 1), (1, 1024))","('', '')",189,24,1085.031,45.209624999999996,1.5749301934150726,18.756927488,36.9375,484.27749576988157,matrix_bf16,1.1946169746358157,0.005527172958653849,578.5261168808248,2.6766854791039054,32.422587076822914,0.1504456175239329,778.14208984375,44.821,43.621,49.271,1.1954245756104465,1.1843094336647855,1.2031439226446432,578.9172198584004,573.5344067518289,582.6555259090998,32.4000244140625,32.192138671875,32.7041015625,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,2.4388976843905077,93.00741488346408 aten::mean,reduce,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (), (), ())","('float', 'ScalarList', 'Scalar', '')","((572416, 1024, 1), (), (), ())","('', '[-1]', 'True', '')",74,49,583.168,11.901387755102041,0.8492257948745272,0.009158656,34.937503814697266,0.24999997270342147,vector_fp32,2.4171547016614285,0.07294620514942167,0.6042886094353039,0.018236549296173573,15.169946787308673,0.4689977659830294,743.327392578125,11.72,11.05,14.96,2.430637989600713,2.227167885536178,2.5842665338499957,0.6076594310520774,0.5567919105899813,0.6460665629208645,15.072021484375,14.176025390625,16.448974609375,"[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, float, 4> >(at::native::ReduceOp, unsigned int, float, 4>)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(743.3270000000001), 'mean_duration_us': np.float64(15.169938775510206), 'median_duration_us': np.float64(15.072), 'std_dev_duration_us': np.float64(0.46419596531071383), 'min_duration_us': np.float64(14.176), 'max_duration_us': np.float64(16.449)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(658.0799999999999), 'mean_duration_us': np.float64(27.419999999999998), 'median_duration_us': np.float64(27.456), 'std_dev_duration_us': np.float64(0.2218289506203673), 'min_duration_us': np.float64(26.913), 'max_duration_us': np.float64(27.872)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(27.42)}]","{'op_shape': (16, 559, 16, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 1024, 64, 1), 'stride_output': (572416, 64, 35776, 1)}",True,2.062593188396829,97.3997875300918 aten::neg,elementwise,python3,CPU,thread 25286 (python3),"((16, 16, 559, 32),)","('c10::BFloat16',)","((572416, 64, 1024, 1),)","('',)",128,48,424.407,8.8418125,1.3407239177661494,0.004579328,17.46875,0.25,vector_bf16,1.4981145850997695,0.040964357314017895,0.37452864627494226,0.010241089328504474,12.235982259114584,0.33912737242264696,587.3271484375,9.370000000000001,7.14,12.69,1.5024171963634907,1.379312619762846,1.576908088694592,0.3756042990908726,0.34482815494071145,0.39422702217364797,12.19189453125,11.615966796875,13.280029296875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(587.328), 'mean_duration_us': np.float64(12.235999999999999), 'median_duration_us': np.float64(12.192), 'std_dev_duration_us': np.float64(0.3355813363900522), 'min_duration_us': np.float64(11.616), 'max_duration_us': np.float64(13.28)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(12.24)}]","{'op_shape': (16, 16, 559, 32), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (572416, 64, 1024, 1), 'stride_output': None}",True,1.8408345223833464,99.24062205247515 @@ -24,7 +24,7 @@ aten::add,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1), (), ()) aten::rsqrt,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1),)","('float',)","((559, 1, 1),)","('',)",76,49,395.084,8.062938775510204,1.2343984771297059,8.944e-06,0.0682373046875,0.125,vector_fp32,0.044481989496294824,0.0023373962616093716,0.005560248687036851,0.00029217453270117123,1.6138044084821428,0.10260213698236942,79.076416015625,7.69,7.37,14.47,0.04471727067439732,0.03288196925838663,0.04658670990303609,0.005589658834299664,0.004110246157298328,0.00582333873787951,1.60009765625,1.535888671875,2.176025390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array >(int, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(79.073), 'mean_duration_us': np.float64(1.613734693877551), 'median_duration_us': np.float64(1.6), 'std_dev_duration_us': np.float64(0.10155782016738636), 'min_duration_us': np.float64(1.536), 'max_duration_us': np.float64(2.176)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rs...', 'stream': 7, 'mean_duration_us': np.float64(1.61)}]","{'op_shape': (16, 559, 1), 'dtype_in_out': ('float', None), 'stride_input': (559, 1, 1), 'stride_output': None}",True,0.24784585029854148,99.74484955445378 aten::index_select,other,python3,CPU,thread 25286 (python3),"((151936, 1024), (), (8944,))","('c10::BFloat16', 'Scalar', 'long int')","((1024, 1), (), (1,))","('', '0', '')",4,1,73.341,73.341,,,,,,,,,,51.77587890625,,51.77587890625,73.341,73.341,73.341,,,,,,,51.77587890625,51.77587890625,51.77587890625,"[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex(at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, int, int, unsigned int, unsigned int, long)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(51.776), 'mean_duration_us': np.float64(51.776), 'median_duration_us': np.float64(51.776), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(51.776), 'max_duration_us': np.float64(51.776)}]","[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex, unsigned int, bool, 4> >(at::native::ReduceOp, unsigned int, bool, 4>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(4.128), 'mean_duration_us': np.float64(4.128), 'median_duration_us': np.float64(4.128), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(4.128), 'max_duration_us': np.float64(4.128)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<4u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.456), 'mean_duration_us': np.float64(3.456), 'median_duration_us': np.float64(3.456), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.456), 'max_duration_us': np.float64(3.456)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 2, 'total_duration_us': np.float64(3.136), 'mean_duration_us': np.float64(1.568), 'median_duration_us': np.float64(1.568), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(1.568), 'max_duration_us': np.float64(1.568)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(1.57)}]","{'shape_in1': (1, 559, 64), 'shape_in2': (), 'dtype_in1_in2_out': ('float', 'double', None), 'stride_input1': (35776, 64, 1), 'stride_input2': (), 'stride_output': None}",True,0.009829753881434726,99.9517610529006 aten::copy_,elementwise,python3,CPU,thread 25286 (python3),"((1, 559, 64), (1, 559, 64), ())","('c10::BFloat16', 'float', 'Scalar')","((35776, 64, 1), (35776, 64, 1), ())","('', '', 'False')",62,2,13.74,6.87,2.3193102422918757,3.5776e-05,0.2047119140625,0.16666666666666666,vector_bf16,0.14433137687341707,0.002261434365022539,0.02405522947890284,0.00037690572750375653,1.4874267578125,0.023305521352583965,2.974853515625,6.87,5.23,8.51,0.14433137687341707,0.14273230129870132,0.1459304524481328,0.02405522947890284,0.023788716883116883,0.024321742074688794,1.4874267578125,1.470947265625,1.50390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 2, 'total_duration_us': np.float64(2.975), 'mean_duration_us': np.float64(1.4875), 'median_duration_us': np.float64(1.4875), 'std_dev_duration_us': np.float64(0.01649999999999996), 'min_duration_us': np.float64(1.471), 'max_duration_us': np.float64(1.504)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 7, 'mean_duration_us': np.float64(1.49)}]","{'op_shape': (1, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (35776, 64, 1), 'stride_output': (35776, 64, 1)}",True,0.009323956955105258,99.9610850098557 diff --git a/tests/traces/compare_test_e2e/reference/mi300_perf_report_csvs/GEMM.csv b/tests/traces/compare_test_e2e/reference/mi300_perf_report_csvs/GEMM.csv index 3d41c3d7e..190b00cdf 100644 --- a/tests/traces/compare_test_e2e/reference/mi300_perf_report_csvs/GEMM.csv +++ b/tests/traces/compare_test_e2e/reference/mi300_perf_report_csvs/GEMM.csv @@ -1,6 +1,6 @@ -name,param: B,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: transpose,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::mm,1,8944,2816,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",51.581550592,71.0078125,692.7688414567059,0.6738529762661254,0.674108665064226,0.019829532765433946,0.62806071132492,0.7091151238095238,466.8243456800369,467.0014789124705,13.737282440537477,435.1008913490395,491.2528627809524,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(5308.273), 'mean_duration_us': np.float64(110.58902083333334), 'median_duration_us': np.float64(110.453), 'std_dev_duration_us': np.float64(3.2462516941953186), 'min_duration_us': np.float64(105.0), 'max_duration_us': np.float64(118.551)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(110.59)}]","[[8944, 1024], [1024, 2816]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",110.58902994791667,110.452880859375,3.280588398059727,105.0,118.55078125,5308.2734375,48,231 -aten::addmm,1,8944,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",18.766086144,36.939453125,484.4883413525089,0.9445121748402515,0.951401332232734,0.024813139318591027,0.8603128998015335,0.9788727841162896,457.6051369756043,460.94285341400456,12.02167671221286,416.81156986901215,474.2524515716137,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(2954.763), 'mean_duration_us': np.float64(41.038375), 'median_duration_us': np.float64(40.7125), 'std_dev_duration_us': np.float64(1.1135532272552386), 'min_duration_us': np.float64(39.57), 'max_duration_us': np.float64(45.023)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(41.04)}]","[[1024], [8944, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",41.038370768229164,40.71240234375,1.1213617423390887,39.56982421875,45.02294921875,2954.7626953125,72,90 -aten::mm,1,8944,1024,2816,False,"(2816, 1)","(1, 2816)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",51.581550592,71.0078125,692.7688414567059,0.753305737501413,0.7572560566039392,0.025395412136689215,0.703202779015716,0.7950246671011402,521.8667430315435,524.6034010195848,17.593150244249767,487.1569745278536,550.7683175571602,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(2374.795), 'mean_duration_us': np.float64(98.94979166666667), 'median_duration_us': np.float64(98.32499999999999), 'std_dev_duration_us': np.float64(3.3136632546066833), 'min_duration_us': np.float64(93.654), 'max_duration_us': np.float64(105.883)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(98.95)}]","[[8944, 2816], [2816, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[2816, 1], [1, 2816]]","['', '']",98.94974772135417,98.324951171875,3.3849417047301227,93.65380859375,105.8828125,2374.7939453125,24,253 -aten::mm,1,8944,1024,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",18.756927488,36.9375,484.27749576988157,0.9490013078576646,0.9555973894680865,0.030823547515567696,0.861800215638343,0.9878291064508096,459.57997685165213,462.774310735841,14.927150401583084,417.35045028328057,478.38340592059774,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(980.549), 'mean_duration_us': np.float64(40.856208333333335), 'median_duration_us': np.float64(40.5325), 'std_dev_duration_us': np.float64(1.3524004269928913), 'min_duration_us': np.float64(39.209), 'max_duration_us': np.float64(44.943)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(40.86)}]","[[8944, 1024], [1024, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",40.856160481770836,40.532470703125,1.3815012260133064,39.208984375,44.94287109375,980.5478515625,24,204 -aten::bmm,1,32,559,1,False,"(32, 1, 1)","(559, 559, 1)","('float', 'float')","(False, False)",3.5776e-05,0.07049179077148438,0.4840088749391201,0.0027034068147724836,0.0027034068147724836,,0.0027034068147724836,0.0027034068147724836,0.00130847289092078,0.00130847289092078,,0.00130847289092078,0.00130847289092078,python3,CPU,thread 23232 (python3),matrix_fp32,"[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB8_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(27.342), 'mean_duration_us': np.float64(27.342), 'median_duration_us': np.float64(27.342), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(27.342), 'max_duration_us': np.float64(27.342)}]","[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_...', 'stream': 0, 'mean_duration_us': np.float64(27.34)}]","[[1, 32, 1], [1, 1, 559]]","['float', 'float']","[[32, 1, 1], [559, 559, 1]]","['', '']",27.341796875,27.341796875,,27.341796875,27.341796875,27.341796875,1,50 +name,param: B,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: transpose,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aten::mm,1,8944,2816,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",128,224,64,51.581550592,71.0078125,692.7688414567059,0.6738529762661254,0.674108665064226,0.019829532765433946,0.62806071132492,0.7091151238095238,466.8243456800369,467.0014789124705,13.737282440537477,435.1008913490395,491.2528627809524,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(5308.273), 'mean_duration_us': np.float64(110.58902083333334), 'median_duration_us': np.float64(110.453), 'std_dev_duration_us': np.float64(3.2462516941953186), 'min_duration_us': np.float64(105.0), 'max_duration_us': np.float64(118.551)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(110.59)}]","[[8944, 1024], [1024, 2816]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",110.58902994791667,110.452880859375,3.280588398059727,105.0,118.55078125,5308.2734375,48,231 +aten::addmm,1,8944,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",256,128,64,18.766086144,36.939453125,484.4883413525089,0.9445121748402515,0.951401332232734,0.024813139318591027,0.8603128998015335,0.9788727841162896,457.6051369756043,460.94285341400456,12.02167671221286,416.81156986901215,474.2524515716137,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(2954.763), 'mean_duration_us': np.float64(41.038375), 'median_duration_us': np.float64(40.7125), 'std_dev_duration_us': np.float64(1.1135532272552386), 'min_duration_us': np.float64(39.57), 'max_duration_us': np.float64(45.023)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(41.04)}]","[[1024], [8944, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",41.038370768229164,40.71240234375,1.1213617423390887,39.56982421875,45.02294921875,2954.7626953125,72,90 +aten::mm,1,8944,1024,2816,False,"(2816, 1)","(1, 2816)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",256,128,64,51.581550592,71.0078125,692.7688414567059,0.753305737501413,0.7572560566039392,0.025395412136689215,0.703202779015716,0.7950246671011402,521.8667430315435,524.6034010195848,17.593150244249767,487.1569745278536,550.7683175571602,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(2374.795), 'mean_duration_us': np.float64(98.94979166666667), 'median_duration_us': np.float64(98.32499999999999), 'std_dev_duration_us': np.float64(3.3136632546066833), 'min_duration_us': np.float64(93.654), 'max_duration_us': np.float64(105.883)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(98.95)}]","[[8944, 2816], [2816, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[2816, 1], [1, 2816]]","['', '']",98.94974772135417,98.324951171875,3.3849417047301227,93.65380859375,105.8828125,2374.7939453125,24,253 +aten::mm,1,8944,1024,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",256,128,64,18.756927488,36.9375,484.27749576988157,0.9490013078576646,0.9555973894680865,0.030823547515567696,0.861800215638343,0.9878291064508096,459.57997685165213,462.774310735841,14.927150401583084,417.35045028328057,478.38340592059774,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(980.549), 'mean_duration_us': np.float64(40.856208333333335), 'median_duration_us': np.float64(40.5325), 'std_dev_duration_us': np.float64(1.3524004269928913), 'min_duration_us': np.float64(39.209), 'max_duration_us': np.float64(44.943)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(40.86)}]","[[8944, 1024], [1024, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",40.856160481770836,40.532470703125,1.3815012260133064,39.208984375,44.94287109375,980.5478515625,24,204 +aten::bmm,1,32,559,1,False,"(32, 1, 1)","(559, 559, 1)","('float', 'float')","(False, False)",16,16,128,3.5776e-05,0.07049179077148438,0.4840088749391201,0.0027034068147724836,0.0027034068147724836,,0.0027034068147724836,0.0027034068147724836,0.00130847289092078,0.00130847289092078,,0.00130847289092078,0.00130847289092078,python3,CPU,thread 23232 (python3),matrix_fp32,"[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB8_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(27.342), 'mean_duration_us': np.float64(27.342), 'median_duration_us': np.float64(27.342), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(27.342), 'max_duration_us': np.float64(27.342)}]","[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_...', 'stream': 0, 'mean_duration_us': np.float64(27.34)}]","[[1, 32, 1], [1, 1, 559]]","['float', 'float']","[[32, 1, 1], [559, 559, 1]]","['', '']",27.341796875,27.341796875,,27.341796875,27.341796875,27.341796875,1,50 diff --git a/tests/traces/h100/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv b/tests/traces/h100/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv index f45f84bcf..f0ec824b4 100644 --- a/tests/traces/h100/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv +++ b/tests/traces/h100/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv @@ -1,5 +1,5 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::addmm,788,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,0.930167808,3.43505859375,258.2422174840085,0.43327482905502873,0.43627359966880563,0.006921872667601486,0.4168940721693182,0.4431534398654331,111.88985263517533,112.66426180820298,1.7875197468233597,107.65964965294302,114.44092699651567,python3,CPU,thread 5617 (python3),matrix_bf16,"[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(399.136), 'mean_duration_us': np.float64(8.315333333333333), 'median_duration_us': np.float64(8.256), 'std_dev_duration_us': np.float64(0.13305517861231703), 'min_duration_us': np.float64(8.128), 'max_duration_us': np.float64(8.64)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(6.879), 'mean_duration_us': np.float64(6.879), 'median_duration_us': np.float64(6.879), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(6.879), 'max_duration_us': np.float64(6.879)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(399.136), 'mean_duration_us': np.float64(8.315333333333333), 'median_duration_us': np.float64(8.256), 'std_dev_duration_us': np.float64(0.13305517861231703), 'min_duration_us': np.float64(8.128), 'max_duration_us': np.float64(8.64)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(6.879), 'mean_duration_us': np.float64(6.879), 'median_duration_us': np.float64(6.879), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(6.879), 'max_duration_us': np.float64(6.879)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(399.136), 'mean_duration_us': np.float64(8.315333333333333), 'median_duration_us': np.float64(8.256), 'std_dev_duration_us': np.float64(0.13305517861231703), 'min_duration_us': np.float64(8.128), 'max_duration_us': np.float64(8.64)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(399.136), 'mean_duration_us': np.float64(8.315333333333333), 'median_duration_us': np.float64(8.256), 'std_dev_duration_us': np.float64(0.13305517861231703), 'min_duration_us': np.float64(8.128), 'max_duration_us': np.float64(8.64)}]","[{'name': 'void cutlass::Kernel2(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(2.208), 'mean_duration_us': np.float64(2.208), 'median_duration_us': np.float64(2.208), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(2.208), 'max_duration_us': np.float64(2.208)}, {'name': 'Memset (Device)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(2.272), 'mean_duration_us': np.float64(2.272), 'median_duration_us': np.float64(2.272), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(2.272), 'max_duration_us': np.float64(2.272)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.968), 'mean_duration_us': np.float64(3.968), 'median_duration_us': np.float64(3.968), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.968), 'max_duration_us': np.float64(3.968)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(10.4), 'mean_duration_us': np.float64(10.4), 'median_duration_us': np.float64(10.4), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(10.4), 'max_duration_us': np.float64(10.4)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(10.464), 'mean_duration_us': np.float64(10.464), 'median_duration_us': np.float64(10.464), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(10.464), 'max_duration_us': np.float64(10.464)}, {'name': 'void cutlass__5x_cudnn::Kernel(cutlass_tensorop_bf16_s16816fprop_optimized_bf16_256x64_32x4_nhwc_align8::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(98.176), 'mean_duration_us': np.float64(98.176), 'median_duration_us': np.float64(98.176), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(98.176), 'max_duration_us': np.float64(98.176)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(2.11)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(2.21)}, {'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(2.27)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(3.97)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(10.4)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(10.46)}, {'name': 'void cutlass__5x_cudnn::Kernel >, false, false, false, false, false, true, false>(pytorch_flash::Flash_fwd_params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(122.848), 'mean_duration_us': np.float64(10.237333333333334), 'median_duration_us': np.float64(10.224), 'std_dev_duration_us': np.float64(0.21878045819699932), 'min_duration_us': np.float64(10.016), 'max_duration_us': np.float64(10.848)}]","[{'name': 'void pytorch_flash::flash_fwd_kernel(int, float, c10::BFloat16 const*, c10::BFloat16 const*, c10::BFloat16 const*, float*, float*, c10::BFloat16*)', 'stream': 7, 'count': 25, 'total_duration_us': np.float64(111.20100000000001), 'mean_duration_us': np.float64(4.448040000000001), 'median_duration_us': np.float64(4.448), 'std_dev_duration_us': np.float64(0.10934037863479323), 'min_duration_us': np.float64(4.256), 'max_duration_us': np.float64(4.64)}]","[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_ke...', 'stream': 7, 'mean_duration_us': np.float64(4.45)}]","{'op_shape': (4, 197, 768), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (151296, 768, 1), 'stride_output': None, 'num_channels': 768, 'has_bias': True, 'is_affine': True, 'is_training': True}",True,9.075338746907917,89.10462823765931 aten::gelu,elementwise,python3,CPU,thread 5617 (python3),"((4, 197, 3072), ())","('c10::BFloat16', '')","((605184, 3072, 1), ())","('', '')",107,12,153.19299999999998,12.766083333333333,2.9325678929855483,,,,,,,,,5.04510498046875,0.039437883181483786,60.541259765625,11.9605,11.42,22.04,,,,,,,5.055419921875,4.990966796875,5.1201171875,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::GeluCUDAKernelImpl(at::TensorIteratorBase&, at::native::GeluType)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::GeluCUDAKernelImpl(at::TensorIteratorBase&, at::native::GeluType)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(60.541999999999994), 'mean_duration_us': np.float64(5.045166666666666), 'median_duration_us': np.float64(5.0555), 'std_dev_duration_us': np.float64(0.037739972914080884), 'min_duration_us': np.float64(4.991), 'max_duration_us': np.float64(5.12)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Ge...', 'stream': 7, 'mean_duration_us': np.float64(5.05)}]",,False,4.940897991876649,94.04552622953597 aten::add,elementwise,python3,CPU,thread 5617 (python3),"((4, 197, 768), (4, 197, 768), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((151296, 768, 1), (151296, 768, 1), ())","('', '', '1')",90,24,298.853,12.452208333333333,0.43644940731421283,0.000605184,3.462890625,0.16666666666666666,vector_bf16,1.6281847288941593,0.024976088698786075,0.2713641214823599,0.00416268144979767,2.2306620279947915,0.03450867810431143,53.535888671875,12.405000000000001,11.76,13.47,1.6324122702199348,1.5760307284094524,1.6686864113093234,0.27206871170332253,0.26267178806824204,0.27811440188488723,2.2244873046875,2.176025390625,2.303955078125,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(53.536), 'mean_duration_us': np.float64(2.2306666666666666), 'median_duration_us': np.float64(2.2245), 'std_dev_duration_us': np.float64(0.033798011118341686), 'min_duration_us': np.float64(2.176), 'max_duration_us': np.float64(2.304)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(2.23)}]","{'shape_in1': (4, 197, 768), 'shape_in2': (4, 197, 768), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (151296, 768, 1), 'stride_input2': (151296, 768, 1), 'stride_output': None}",True,4.369175102338875,98.41470133187484 aten::cat,other,python3,CPU,thread 5617 (python3),"(((4, 1, 768), (4, 196, 768)), ())","('TensorList', 'Scalar')","(((0, 768, 1), (150528, 1, 196)), ())","('', '1')",14,1,16.66,16.66,,,,,,,,,,7.263916015625,,7.263916015625,16.66,16.66,16.66,,,,,,,7.263916015625,7.263916015625,7.263916015625,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(7.264), 'mean_duration_us': np.float64(7.264), 'median_duration_us': np.float64(7.264), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(7.264), 'max_duration_us': np.float64(7.264)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(6.879), 'mean_duration_us': np.float64(6.879), 'median_duration_us': np.float64(6.879), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(6.879), 'max_duration_us': np.float64(6.879)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(6.879), 'mean_duration_us': np.float64(6.879), 'median_duration_us': np.float64(6.879), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(6.879), 'max_duration_us': np.float64(6.879)}]","[{'name': 'void cutlass::Kernel2 >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.777), 'mean_duration_us': np.float64(3.777), 'median_duration_us': np.float64(3.777), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.777), 'max_duration_us': np.float64(3.777)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(3.78)}]","{'shape_in1': (4, 197, 768), 'shape_in2': (1, 197, 768), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (151296, 768, 1), 'stride_input2': (151296, 768, 1), 'stride_output': None}",True,0.3082569465407019,99.87718338708055 aten::tanh,elementwise,python3,CPU,thread 5617 (python3),"((4, 768),)","('c10::BFloat16',)","((768, 1),)","('',)",1258,1,12.47,12.47,,,,,,,,,,1.5048828125,,1.5048828125,12.47,12.47,12.47,,,,,,,1.5048828125,1.5048828125,1.5048828125,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::tanh_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::tanh_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(1.505), 'mean_duration_us': np.float64(1.505), 'median_duration_us': np.float64(1.505), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(1.505), 'max_duration_us': np.float64(1.505)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::ta...', 'stream': 7, 'mean_duration_us': np.float64(1.5)}]",,False,0.12281661291945489,100.0 diff --git a/tests/traces/h100/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv b/tests/traces/h100/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv index c69658de7..4fba6b709 100644 --- a/tests/traces/h100/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv +++ b/tests/traces/h100/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv @@ -1,6 +1,6 @@ -name,param: B,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::mm,1,8944,2816,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",51.581550592,71.0078125,692.7688414567059,0.8814855312346527,0.8815297729018408,0.012524549392316679,0.8566940617991415,0.9050115804454784,610.6657102342793,610.6963594828014,8.676617572282526,593.490952675431,626.9638240901165,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(39.75), 'mean_duration_us': np.float64(0.828125), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.07937112851450877), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.057)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(4015.5009999999997), 'mean_duration_us': np.float64(83.65627083333332), 'median_duration_us': np.float64(83.679), 'std_dev_duration_us': np.float64(1.1791844593401435), 'min_duration_us': np.float64(81.472), 'max_duration_us': np.float64(86.016)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(83.66)}]","[[8944, 1024], [1024, 2816]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",84.4844258626302,84.4635009765625,1.1997237297511993,82.27197265625,86.912109375,4055.25244140625,48,215 -aten::addmm,1,8944,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",18.766086144,36.939453125,484.4883413525089,1.1975199165160728,1.1966690312454464,0.012059428292742667,1.1683671458638643,1.2214187300624357,580.1844380894671,579.77219409602,5.842652411210404,566.0602605903484,591.7631346248373,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(62.783), 'mean_duration_us': np.float64(0.8719861111111111), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.11567921318307071), 'min_duration_us': np.float64(0.767), 'max_duration_us': np.float64(1.216)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2266.292), 'mean_duration_us': np.float64(31.476277777777778), 'median_duration_us': np.float64(31.456), 'std_dev_duration_us': np.float64(0.31881400107892144), 'min_duration_us': np.float64(30.912), 'max_duration_us': np.float64(32.256)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.87)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.48)}]","[[1024], [8944, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",32.34827677408854,32.3680419921875,0.32643614621543415,31.712158203125,33.152099609375,2329.075927734375,72,89 -aten::mm,1,8944,1024,2816,False,"(2816, 1)","(1, 2816)","('c10::BFloat16', 'c10::BFloat16')",51.581550592,71.0078125,692.7688414567059,0.9737028419232031,0.9733527752266449,0.00547104447053427,0.9622886978998384,0.9821816193669105,674.5509897222395,674.3084744224323,3.790169139410141,666.6436263909532,680.4248225488859,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.512999999999998), 'mean_duration_us': np.float64(0.8547083333333333), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.10426427926454754), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.088)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(1814.774), 'mean_duration_us': np.float64(75.61558333333333), 'median_duration_us': np.float64(75.6315), 'std_dev_duration_us': np.float64(0.4239531142184896), 'min_duration_us': np.float64(74.752), 'max_duration_us': np.float64(76.384)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.85)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(75.62)}]","[[8944, 2816], [2816, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[2816, 1], [1, 2816]]","['', '']",76.47029622395833,76.4954833984375,0.4304329398221454,75.807861328125,77.375,1835.287109375,24,235 -aten::mm,1,8944,1024,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",18.756927488,36.9375,484.27749576988157,1.1946169746358157,1.1954245756104465,0.005527172958653849,1.1843094336647855,1.2031439226446432,578.5261168808248,578.9172198584004,2.6766854791039054,573.5344067518289,582.6555259090998,python3,CPU,thread 25286 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","[[8944, 1024], [1024, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",32.422587076822914,32.4000244140625,0.1504456175239329,32.192138671875,32.7041015625,778.14208984375,24,189 -aten::bmm,1,32,559,1,False,"(32, 1, 1)","(559, 559, 1)","('float', 'float')",3.5776e-05,0.07049179077148438,0.4840088749391201,0.020998747121653488,0.020998747121653488,,0.020998747121653488,0.020998747121653488,0.01016357996948259,0.01016357996948259,,0.01016357996948259,0.01016357996948259,python3,CPU,thread 25286 (python3),matrix_fp32,"[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2 >, false, true, false, false, false, true, false>(pytorch_flash::Flash_fwd_params)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(2722.705), 'mean_duration_us': np.float64(113.44604166666666), 'median_duration_us': np.float64(113.775), 'std_dev_duration_us': np.float64(2.127694555913488), 'min_duration_us': np.float64(109.28), 'max_duration_us': np.float64(119.647)}]","[{'name': 'void pytorch_flash::flash_fwd_kernel(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 72, 'total_duration_us': np.float64(2108.187), 'mean_duration_us': np.float64(29.280375), 'median_duration_us': np.float64(29.312), 'std_dev_duration_us': np.float64(0.3210814381594732), 'min_duration_us': np.float64(28.672), 'max_duration_us': np.float64(30.048)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(29.28)}]","{'op_shape': (16, 16, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 35776, 64, 1), 'stride_output': (572416, 64, 1024, 1)}",True,6.607603257237321,35.151388884308325 aten::cat,other,python3,CPU,thread 25286 (python3),"(((16, 16, 559, 32), (16, 16, 559, 32)), ())","('TensorList', 'Scalar')","(((286208, 32, 512, 1), (572416, 64, 1024, 1)), ())","('', '-1')",129,48,554.286,11.547624999999998,1.8371072615461332,,,,,,,,,41.96514383951823,0.2933502450882166,2014.326904296875,12.645,9.26,16.121,,,,,,,41.9678955078125,41.3759765625,42.464111328125,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 4, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(2014.326), 'mean_duration_us': np.float64(41.965125), 'median_duration_us': np.float64(41.968), 'std_dev_duration_us': np.float64(0.290296301575362), 'min_duration_us': np.float64(41.376), 'max_duration_us': np.float64(42.464)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(1671.096), 'mean_duration_us': np.float64(34.8145), 'median_duration_us': np.float64(34.784), 'std_dev_duration_us': np.float64(0.3216488121124247), 'min_duration_us': np.float64(34.271), 'max_duration_us': np.float64(35.615)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(34.81)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (16, 16, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (572416, 35776, 64, 1), 'stride_output': None}",True,5.237645012877888,52.454715507632 aten::mul,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (16, 559, 1))","('float', 'float')","((572416, 1024, 1), (559, 1, 1))","('', '')",77,49,436.2,8.902040816326531,0.7633815384847344,0.009158656,69.90911865234375,0.12493899463152758,vector_fp32,2.232889871737575,0.029013860055738732,0.2789750156978132,0.003624962505743834,32.83514528858419,0.4306580406298848,1608.922119140625,8.69,8.28,12.321,2.2458720972975397,2.1632220106771562,2.2725938972911197,0.28059700190735487,0.2702707831787955,0.2839355967332975,32.639892578125,32.256103515625,33.886962890625,"[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(1608.9229999999998), 'mean_duration_us': np.float64(32.835163265306115), 'median_duration_us': np.float64(32.64), 'std_dev_duration_us': np.float64(0.42625121207266925), 'min_duration_us': np.float64(32.256), 'max_duration_us': np.float64(33.887)}]","[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(32.84)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1), 'dtype_in1_in2_out': ('float', 'float', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (559, 1, 1), 'stride_output': None}",True,5.042776225515153,57.49749173314716 aten::mul,elementwise,python3,CPU,thread 25286 (python3),"((16, 16, 559, 64), (1, 1, 559, 64))","('c10::BFloat16', 'c10::BFloat16')","((572416, 64, 1024, 1), (35776, 35776, 64, 1))","('', '')",123,48,562.537,11.719520833333334,4.2320548702833065,0.009158656,35.0057373046875,0.24951267056530213,vector_bf16,1.145799975045878,0.011365925851868387,0.2858916117073536,0.002835942512746873,32.03850809733073,0.3186166119751925,1537.848388671875,14.5,6.89,21.42,1.1470691458092843,1.1234709276742014,1.1633548975602963,0.28620828589393443,0.2803202314664674,0.29027178730549297,32.0,31.552001953125,32.672119140625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(1537.849), 'mean_duration_us': np.float64(32.03852083333333), 'median_duration_us': np.float64(32.0), 'std_dev_duration_us': np.float64(0.31526120054007917), 'min_duration_us': np.float64(31.552), 'max_duration_us': np.float64(32.672)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(32.04)}]","{'shape_in1': (16, 16, 559, 64), 'shape_in2': (1, 1, 559, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 64, 1024, 1), 'stride_input2': (35776, 35776, 64, 1), 'stride_output': None}",True,4.820012852445286,62.317504585592445 @@ -16,7 +16,7 @@ aten::mul,elementwise,python3,CPU,thread 25286 (python3),"((16, 16, 559, 64), (1 aten::silu,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 2816),)","('c10::BFloat16',)","((1574144, 2816, 1),)","('',)",217,24,306.045,12.751875,1.1219036737888284,0.025186304,96.078125,0.25,vector_bf16,2.613825645550667,0.024402900951242156,0.6534564113876667,0.006100725237810539,38.54644775390625,0.3625162870903211,925.11474609375,12.41,12.09,17.14,2.6224883540301014,2.5658473790517644,2.641178232798679,0.6556220885075253,0.6414618447629411,0.6602945581996698,38.4158935546875,38.14404296875,39.263916015625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(925.115), 'mean_duration_us': np.float64(38.546458333333334), 'median_duration_us': np.float64(38.416), 'std_dev_duration_us': np.float64(0.35488331171042203), 'min_duration_us': np.float64(38.144), 'max_duration_us': np.float64(39.264)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(a...', 'stream': 7, 'mean_duration_us': np.float64(38.55)}]","{'op_shape': (16, 559, 2816), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (1574144, 2816, 1), 'stride_output': None}",True,2.8995478351474517,85.07401687597186 aten::copy_,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (16, 559, 1024), ())","('c10::BFloat16', 'float', 'Scalar')","((572416, 1024, 1), (572416, 1024, 1), ())","('', '', 'False')",81,49,325.833,6.64965306122449,0.22561753690523184,0.009158656,52.40625,0.16666666666666666,vector_bf16,2.9981934625636244,0.05798757085308172,0.49969891042727066,0.009664595142180274,18.335254902742346,0.3646247760047354,898.427490234375,6.62,6.38,7.5,3.0127174024708547,2.8151578389574006,3.077681103946181,0.5021195670784757,0.4691929731595667,0.5129468506576967,18.239990234375,17.85498046875,19.52001953125,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(898.427), 'mean_duration_us': np.float64(18.335244897959186), 'median_duration_us': np.float64(18.24), 'std_dev_duration_us': np.float64(0.36088720215331926), 'min_duration_us': np.float64(17.855), 'max_duration_us': np.float64(19.52)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 7, 'mean_duration_us': np.float64(18.34)}]","{'op_shape': (16, 559, 1024), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (572416, 1024, 1), 'stride_output': (572416, 1024, 1)}",True,2.8159031032049398,87.88991997917681 aten::add,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (16, 559, 1024), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((572416, 1024, 1), (572416, 1024, 1), ())","('', '', '1')",191,48,573.702,11.952125,1.30855332861515,0.009158656,52.40625,0.16666666666666666,vector_bf16,3.0889613339158086,0.08968722206012,0.5148268889859681,0.01494787034335336,17.80457051595052,0.5205274181052388,854.619384765625,11.899999999999999,10.66,19.271,3.077512782083186,2.8909812843546505,3.2218709989264394,0.5129187970138642,0.4818302140591083,0.5369784998210732,17.85595703125,17.055908203125,19.008056640625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(854.619), 'mean_duration_us': np.float64(17.8045625), 'median_duration_us': np.float64(17.856), 'std_dev_duration_us': np.float64(0.5150970744371872), 'min_duration_us': np.float64(17.056), 'max_duration_us': np.float64(19.008)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(17.8)}]","{'shape_in1': (16, 559, 1024), 'shape_in2': (16, 559, 1024), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (572416, 1024, 1), 'stride_input2': (572416, 1024, 1), 'stride_output': None}",True,2.6785972198967594,90.56851719907357 -aten::mm,GEMM,python3,CPU,thread 25286 (python3),"((8944, 1024), (1024, 1024))","('c10::BFloat16', 'c10::BFloat16')","((1024, 1), (1, 1024))","('', '')",189,24,1085.031,45.209624999999996,1.5749301934150726,18.756927488,36.9375,484.27749576988157,matrix_bf16,1.1946169746358157,0.005527172958653849,578.5261168808248,2.6766854791039054,32.422587076822914,0.1504456175239329,778.14208984375,44.821,43.621,49.271,1.1954245756104465,1.1843094336647855,1.2031439226446432,578.9172198584004,573.5344067518289,582.6555259090998,32.4000244140625,32.192138671875,32.7041015625,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,2.4388976843905077,93.00741488346408 +aten::mm,GEMM,python3,CPU,thread 25286 (python3),"((8944, 1024), (1024, 1024))","('c10::BFloat16', 'c10::BFloat16')","((1024, 1), (1, 1024))","('', '')",189,24,1085.031,45.209624999999996,1.5749301934150726,18.756927488,36.9375,484.27749576988157,matrix_bf16,1.1946169746358157,0.005527172958653849,578.5261168808248,2.6766854791039054,32.422587076822914,0.1504456175239329,778.14208984375,44.821,43.621,49.271,1.1954245756104465,1.1843094336647855,1.2031439226446432,578.9172198584004,573.5344067518289,582.6555259090998,32.4000244140625,32.192138671875,32.7041015625,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(20.001000000000005), 'mean_duration_us': np.float64(0.8333750000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.08185037797713582), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.056)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(758.1399999999999), 'mean_duration_us': np.float64(31.58916666666666), 'median_duration_us': np.float64(31.552), 'std_dev_duration_us': np.float64(0.15096182372448408), 'min_duration_us': np.float64(31.392), 'max_duration_us': np.float64(31.936)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.83)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.59)}]","{'M': 8944, 'N': 1024, 'K': 1024, 'bias': False, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,2.4388976843905077,93.00741488346408 aten::mean,reduce,python3,CPU,thread 25286 (python3),"((16, 559, 1024), (), (), ())","('float', 'ScalarList', 'Scalar', '')","((572416, 1024, 1), (), (), ())","('', '[-1]', 'True', '')",74,49,583.168,11.901387755102041,0.8492257948745272,0.009158656,34.937503814697266,0.24999997270342147,vector_fp32,2.4171547016614285,0.07294620514942167,0.6042886094353039,0.018236549296173573,15.169946787308673,0.4689977659830294,743.327392578125,11.72,11.05,14.96,2.430637989600713,2.227167885536178,2.5842665338499957,0.6076594310520774,0.5567919105899813,0.6460665629208645,15.072021484375,14.176025390625,16.448974609375,"[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, float, 4> >(at::native::ReduceOp, unsigned int, float, 4>)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(743.3270000000001), 'mean_duration_us': np.float64(15.169938775510206), 'median_duration_us': np.float64(15.072), 'std_dev_duration_us': np.float64(0.46419596531071383), 'min_duration_us': np.float64(14.176), 'max_duration_us': np.float64(16.449)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(658.0799999999999), 'mean_duration_us': np.float64(27.419999999999998), 'median_duration_us': np.float64(27.456), 'std_dev_duration_us': np.float64(0.2218289506203673), 'min_duration_us': np.float64(26.913), 'max_duration_us': np.float64(27.872)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(27.42)}]","{'op_shape': (16, 559, 16, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (572416, 1024, 64, 1), 'stride_output': (572416, 64, 35776, 1)}",True,2.062593188396829,97.3997875300918 aten::neg,elementwise,python3,CPU,thread 25286 (python3),"((16, 16, 559, 32),)","('c10::BFloat16',)","((572416, 64, 1024, 1),)","('',)",128,48,424.407,8.8418125,1.3407239177661494,0.004579328,17.46875,0.25,vector_bf16,1.4981145850997695,0.040964357314017895,0.37452864627494226,0.010241089328504474,12.235982259114584,0.33912737242264696,587.3271484375,9.370000000000001,7.14,12.69,1.5024171963634907,1.379312619762846,1.576908088694592,0.3756042990908726,0.34482815494071145,0.39422702217364797,12.19189453125,11.615966796875,13.280029296875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(587.328), 'mean_duration_us': np.float64(12.235999999999999), 'median_duration_us': np.float64(12.192), 'std_dev_duration_us': np.float64(0.3355813363900522), 'min_duration_us': np.float64(11.616), 'max_duration_us': np.float64(13.28)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(12.24)}]","{'op_shape': (16, 16, 559, 32), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (572416, 64, 1024, 1), 'stride_output': None}",True,1.8408345223833464,99.24062205247515 @@ -24,7 +24,7 @@ aten::add,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1), (), ()) aten::rsqrt,elementwise,python3,CPU,thread 25286 (python3),"((16, 559, 1),)","('float',)","((559, 1, 1),)","('',)",76,49,395.084,8.062938775510204,1.2343984771297059,8.944e-06,0.0682373046875,0.125,vector_fp32,0.044481989496294824,0.0023373962616093716,0.005560248687036851,0.00029217453270117123,1.6138044084821428,0.10260213698236942,79.076416015625,7.69,7.37,14.47,0.04471727067439732,0.03288196925838663,0.04658670990303609,0.005589658834299664,0.004110246157298328,0.00582333873787951,1.60009765625,1.535888671875,2.176025390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array >(int, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 49, 'total_duration_us': np.float64(79.073), 'mean_duration_us': np.float64(1.613734693877551), 'median_duration_us': np.float64(1.6), 'std_dev_duration_us': np.float64(0.10155782016738636), 'min_duration_us': np.float64(1.536), 'max_duration_us': np.float64(2.176)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rs...', 'stream': 7, 'mean_duration_us': np.float64(1.61)}]","{'op_shape': (16, 559, 1), 'dtype_in_out': ('float', None), 'stride_input': (559, 1, 1), 'stride_output': None}",True,0.24784585029854148,99.74484955445378 aten::index_select,other,python3,CPU,thread 25286 (python3),"((151936, 1024), (), (8944,))","('c10::BFloat16', 'Scalar', 'long int')","((1024, 1), (), (1,))","('', '0', '')",4,1,73.341,73.341,,,,,,,,,,51.77587890625,,51.77587890625,73.341,73.341,73.341,,,,,,,51.77587890625,51.77587890625,51.77587890625,"[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex(at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, int, int, unsigned int, unsigned int, long)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(51.776), 'mean_duration_us': np.float64(51.776), 'median_duration_us': np.float64(51.776), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(51.776), 'max_duration_us': np.float64(51.776)}]","[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex, unsigned int, bool, 4> >(at::native::ReduceOp, unsigned int, bool, 4>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(4.128), 'mean_duration_us': np.float64(4.128), 'median_duration_us': np.float64(4.128), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(4.128), 'max_duration_us': np.float64(4.128)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_simt_sgemm_128x32_8x5_nn_align1::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.52), 'mean_duration_us': np.float64(3.52), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.52), 'max_duration_us': np.float64(3.52)}]","[{'name': 'void cutlass::Kernel2, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<4u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.456), 'mean_duration_us': np.float64(3.456), 'median_duration_us': np.float64(3.456), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.456), 'max_duration_us': np.float64(3.456)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 2, 'total_duration_us': np.float64(3.136), 'mean_duration_us': np.float64(1.568), 'median_duration_us': np.float64(1.568), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(1.568), 'max_duration_us': np.float64(1.568)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(1.57)}]","{'shape_in1': (1, 559, 64), 'shape_in2': (), 'dtype_in1_in2_out': ('float', 'double', None), 'stride_input1': (35776, 64, 1), 'stride_input2': (), 'stride_output': None}",True,0.009829753881434726,99.9517610529006 aten::copy_,elementwise,python3,CPU,thread 25286 (python3),"((1, 559, 64), (1, 559, 64), ())","('c10::BFloat16', 'float', 'Scalar')","((35776, 64, 1), (35776, 64, 1), ())","('', '', 'False')",62,2,13.74,6.87,2.3193102422918757,3.5776e-05,0.2047119140625,0.16666666666666666,vector_bf16,0.14433137687341707,0.002261434365022539,0.02405522947890284,0.00037690572750375653,1.4874267578125,0.023305521352583965,2.974853515625,6.87,5.23,8.51,0.14433137687341707,0.14273230129870132,0.1459304524481328,0.02405522947890284,0.023788716883116883,0.024321742074688794,1.4874267578125,1.470947265625,1.50390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 2, 'total_duration_us': np.float64(2.975), 'mean_duration_us': np.float64(1.4875), 'median_duration_us': np.float64(1.4875), 'std_dev_duration_us': np.float64(0.01649999999999996), 'min_duration_us': np.float64(1.471), 'max_duration_us': np.float64(1.504)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 7, 'mean_duration_us': np.float64(1.49)}]","{'op_shape': (1, 559, 64), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (35776, 64, 1), 'stride_output': (35776, 64, 1)}",True,0.009323956955105258,99.9610850098557 diff --git a/tests/traces/h100/Wan-AI_Wan2.1-T2V-1.3B-Diffusers__1016009_perf_report_csvs/GEMM.csv b/tests/traces/h100/Wan-AI_Wan2.1-T2V-1.3B-Diffusers__1016009_perf_report_csvs/GEMM.csv index daf889a84..7072542b5 100644 --- a/tests/traces/h100/Wan-AI_Wan2.1-T2V-1.3B-Diffusers__1016009_perf_report_csvs/GEMM.csv +++ b/tests/traces/h100/Wan-AI_Wan2.1-T2V-1.3B-Diffusers__1016009_perf_report_csvs/GEMM.csv @@ -1,15 +1,15 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::addmm,32256,1536,1536,True,"(1536, 1)","(1, 1536)","('c10::BFloat16', 'c10::BFloat16')",1,152.252448768,193.5029296875,750.3723599146089,0.9961870730638054,0.9961897768889316,0.009165655649765165,0.9676120402884131,1.0164743660409066,747.5112449313145,747.5132738069551,6.87765466007899,726.069330153006,762.734268838821,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 1800, 'total_duration_us': np.float64(1709.3609999999999), 'mean_duration_us': np.float64(0.949645), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.21114300598172794), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.76)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 1800, 'total_duration_us': np.float64(364944.137), 'mean_duration_us': np.float64(202.74674277777777), 'median_duration_us': np.float64(202.7345), 'std_dev_duration_us': np.float64(1.8754461904520678), 'min_duration_us': np.float64(198.718), 'max_duration_us': np.float64(208.735)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.95)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(202.75)}]","[[1536], [32256, 1536], [1536, 1536], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1536, 1], [1, 1536], [], []]","['', '', '', '1', '1']",203.69642388237847,203.6785888671875,1.8795173958208344,199.614013671875,209.694091796875,366653.56298828125,1800,11917 -aten::addmm,32256,8960,1536,True,"(1536, 1)","(1, 1536)","('c10::BFloat16', 'c10::BFloat16')",1,888.13928448,672.01708984375,1260.3781031773076,0.6130246274525368,0.6136162762465612,0.009278216572878261,0.4536644041637975,0.6161746427320722,772.642817149604,773.3885183343638,11.69406100499256,571.7886811990305,776.6130274326044,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(284.666), 'mean_duration_us': np.float64(0.9488866666666667), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.21814426531286327), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.536)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(344666.045), 'mean_duration_us': np.float64(1148.8868166666666), 'median_duration_us': np.float64(1147.4615), 'std_dev_duration_us': np.float64(23.393421280844223), 'min_duration_us': np.float64(1142.646), 'max_duration_us': np.float64(1552.273)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.95)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(1148.89)}]","[[8960], [32256, 1536], [1536, 8960], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1536, 1], [1, 1536], [], []]","['', '', '', '1', '1']",1149.8357413736978,1148.3740234375,23.43479543431236,1143.60595703125,1553.264892578125,344950.7224121094,300,12226 -aten::addmm,32256,1536,8960,True,"(8960, 1)","(1, 8960)","('c10::BFloat16', 'c10::BFloat16')",1,887.899815936,672.0029296875,1260.0648190533489,0.6280223769901393,0.6281309463124868,0.0011147025918556313,0.6239660806244178,0.6314452104953299,791.3489028235339,791.4857072070524,1.404597519704882,786.2377064774342,795.6618949049015,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(283.045), 'mean_duration_us': np.float64(0.9434833333333333), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.20174815089335738), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.6)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(336320.40800000005), 'mean_duration_us': np.float64(1121.0680266666668), 'median_duration_us': np.float64(1120.854), 'std_dev_duration_us': np.float64(1.9656534280714149), 'min_duration_us': np.float64(1115.222), 'max_duration_us': np.float64(1128.246)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.94)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(1121.07)}]","[[1536], [32256, 8960], [8960, 1536], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [8960, 1], [1, 8960], [], []]","['', '', '', '1', '1']",1122.0115462239582,1121.8140869140625,1.9930724283064778,1115.926025390625,1129.302001953125,336603.4638671875,300,12237 -aten::mm,512,10240,4096,False,"(4096, 1)","(1, 4096)","('c10::BFloat16', 'c10::BFloat16')",1,42.94967296,94.0,435.74468085106383,1.4296001882974725,1.4303193657714175,0.015685558431838626,1.386853511813404,1.4730903491980094,622.9406777943029,623.2540555531624,6.834898652852257,604.3140408923089,641.8912840760688,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(72.356), 'mean_duration_us': np.float64(0.7537083333333333), 'median_duration_us': np.float64(0.736), 'std_dev_duration_us': np.float64(0.13248647074533898), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.76)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(6547.305), 'mean_duration_us': np.float64(68.20109375), 'median_duration_us': np.float64(68.15950000000001), 'std_dev_duration_us': np.float64(0.7345655473096121), 'min_duration_us': np.float64(66.176), 'max_duration_us': np.float64(69.887)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.75)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(68.2)}]","[[512, 4096], [4096, 10240]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",68.95487213134766,68.9119873046875,0.757329355493187,66.9111328125,71.07177734375,6619.667724609375,96,259 -aten::addmm,512,1536,1536,True,"(1536, 1)","(1, 1536)","('c10::BFloat16', 'c10::BFloat16')",1,2.416705536,7.5029296875,307.1800078094494,0.7516805042352582,0.7683082274922163,0.06526312615574323,0.6419162493177426,0.8657005596389427,230.9012231611975,236.00892732112325,20.04752760219027,197.18383847843666,265.92590467053515,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 610, 'total_duration_us': np.float64(6432.974999999999), 'mean_duration_us': np.float64(10.545860655737703), 'median_duration_us': np.float64(10.24), 'std_dev_duration_us': np.float64(0.916629834968112), 'min_duration_us': np.float64(9.088), 'max_duration_us': np.float64(12.256)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpg...', 'stream': 7, 'mean_duration_us': np.float64(10.55)}]","[[1536], [512, 1536], [1536, 1536], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1536, 1], [1, 1536], [], []]","['', '', '', '1', '1']",10.545858014216188,10.239990234375,0.917382381720403,9.087890625,12.256103515625,6432.973388671875,610,11855 -aten::mm,512,4096,4096,False,"(4096, 1)","(1, 4096)","('c10::BFloat16', 'c10::BFloat16')",1,17.179869184,40.0,409.6,1.3387112394781984,1.351256031461381,0.04694098358016499,1.2136275720199494,1.4201173121719364,548.33612369027,553.4744704865817,19.22702687443559,497.1018534993713,581.6800510656251,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 192, 'total_duration_us': np.float64(6023.076), 'mean_duration_us': np.float64(31.3701875), 'median_duration_us': np.float64(31.04), 'std_dev_duration_us': np.float64(1.1237826868855696), 'min_duration_us': np.float64(29.535), 'max_duration_us': np.float64(34.56)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.37)}]","[[512, 4096], [4096, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",31.37018585205078,31.0400390625,1.126722931713206,29.534912109375,34.56005859375,6023.07568359375,192,77 -aten::mm,512,4096,10240,False,"(10240, 1)","(1, 10240)","('c10::BFloat16', 'c10::BFloat16')",1,42.94967296,94.0,435.74468085106383,1.3724764570786148,1.3711251186526003,0.01613094400980978,1.3328587891965415,1.4019999854982375,598.0493157653199,597.4604772341544,7.028973049380965,580.7861277179825,610.9140362341257,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(3447.648), 'mean_duration_us': np.float64(71.82600000000001), 'median_duration_us': np.float64(71.887), 'std_dev_duration_us': np.float64(0.8376598354940977), 'min_duration_us': np.float64(70.304), 'max_duration_us': np.float64(73.951)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(71.83)}]","[[512, 10240], [10240, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[10240, 1], [1, 10240]]","['', '']",71.82601420084636,71.8870849609375,0.846523518853255,70.303955078125,73.950927734375,3447.648681640625,48,289 -aten::bmm,512,64,512,False,"(262144, 512, 1)","(64, 4096, 1)","('c10::BFloat16', 'c10::BFloat16')",64,2.147483648,40.0,51.2,2.291045759224885,2.309636210368003,0.062142092073413965,2.114055150925983,2.3874524637640877,117.30154287231413,118.25337397084175,3.181675114158794,108.23962372741032,122.2375661447213,python3,CPU,thread 10586 (python3),matrix_bf16,"[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_32x6_nn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(879.4110000000001), 'mean_duration_us': np.float64(18.3210625), 'median_duration_us': np.float64(18.159999999999997), 'std_dev_duration_us': np.float64(0.5100926143950885), 'min_duration_us': np.float64(17.568), 'max_duration_us': np.float64(19.84)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x4_tn_align8::Params)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(190.655), 'mean_duration_us': np.float64(19.0655), 'median_duration_us': np.float64(19.024), 'std_dev_duration_us': np.float64(0.22117380043757456), 'min_duration_us': np.float64(18.72), 'max_duration_us': np.float64(19.424)}]","[{'name': 'void cutlass::Kernel2::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(64.446), 'mean_duration_us': np.float64(6.444599999999999), 'median_duration_us': np.float64(6.4635), 'std_dev_duration_us': np.float64(0.08959374978200195), 'min_duration_us': np.float64(6.24), 'max_duration_us': np.float64(6.56)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(37.024), 'mean_duration_us': np.float64(3.7024), 'median_duration_us': np.float64(3.696), 'std_dev_duration_us': np.float64(0.13277665457451465), 'min_duration_us': np.float64(3.519), 'max_duration_us': np.float64(3.904)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_32x6_nn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(879.4110000000001), 'mean_duration_us': np.float64(18.3210625), 'median_duration_us': np.float64(18.159999999999997), 'std_dev_duration_us': np.float64(0.5100926143950885), 'min_duration_us': np.float64(17.568), 'max_duration_us': np.float64(19.84)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x4_tn_align8::Params)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(190.655), 'mean_duration_us': np.float64(19.0655), 'median_duration_us': np.float64(19.024), 'std_dev_duration_us': np.float64(0.22117380043757456), 'min_duration_us': np.float64(18.72), 'max_duration_us': np.float64(19.424)}]","[{'name': 'void cutlass::Kernel2::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(64.446), 'mean_duration_us': np.float64(6.444599999999999), 'median_duration_us': np.float64(6.4635), 'std_dev_duration_us': np.float64(0.08959374978200195), 'min_duration_us': np.float64(6.24), 'max_duration_us': np.float64(6.56)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(37.024), 'mean_duration_us': np.float64(3.7024), 'median_duration_us': np.float64(3.696), 'std_dev_duration_us': np.float64(0.13277665457451465), 'min_duration_us': np.float64(3.519), 'max_duration_us': np.float64(3.904)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel >, false, false, false, false, true, true, false>(pytorch_flash::Flash_fwd_params)', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(5328692.5649999995), 'mean_duration_us': np.float64(17762.308549999998), 'median_duration_us': np.float64(17749.3595), 'std_dev_duration_us': np.float64(234.40312254453335), 'min_duration_us': np.float64(17253.507), 'max_duration_us': np.float64(18602.391)}]","[{'name': 'void pytorch_flash::flash_fwd_kernel(int, float, float const*, float const*, float const*, float*, float*, float*)', 'stream': 7, 'count': 610, 'total_duration_us': np.float64(94543.31599999999), 'mean_duration_us': np.float64(154.98904262295082), 'median_duration_us': np.float64(154.943), 'std_dev_duration_us': np.float64(0.9304707718137512), 'min_duration_us': np.float64(152.286), 'max_duration_us': np.float64(158.335)}, {'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 610, 'total_duration_us': np.float64(245193.71000000002), 'mean_duration_us': np.float64(401.9569016393443), 'median_duration_us': np.float64(401.98), 'std_dev_duration_us': np.float64(0.5264631478678129), 'min_duration_us': np.float64(400.733), 'max_duration_us': np.float64(403.228)}]","[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_ke...', 'stream': 7, 'mean_duration_us': np.float64(154.99)}, {'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(401.96)}]","{'op_shape': (1, 32256, 1536), 'dtype_in_out': ('float', None), 'stride_input': (49545216, 1, 32256), 'stride_output': None, 'num_channels': 1536, 'has_bias': True, 'is_affine': True, 'is_training': True}",True,2.9334496203528926,55.088219713923266 aten::convolution,CONV_fwd,python3,CPU,thread 10586 (python3),"((1, 96, 6, 258, 258), (96, 96, 3, 3, 3), (96,), (), (), (), (), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'ScalarList', 'ScalarList', 'ScalarList', 'Scalar', 'ScalarList', 'Scalar')","((38340864, 399384, 66564, 258, 1), (2592, 27, 9, 3, 1), (1,), (), (), (), (), (), ())","('', '', '', '[1, 1, 1]', '[0, 0, 0]', '[1, 1, 1]', 'False', '[0, 0, 0]', '1')",137093,750,66923.343,89.231124,80.87066060096622,130.459631616,121.60400390625,1023.1242064687104,matrix_bf16,0.2825786840015168,0.0006022781490769081,289.11309183402443,0.6162053533477364,451.242876953125,0.9615305761214227,338432.15771484375,82.696,73.951,1662.24,0.2825665209737053,0.2811115063277445,0.2844425817856833,289.1006475458465,287.61198684079744,291.02009077538855,451.26025390625,448.283935546875,453.595947265625,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(721.184), 'mean_duration_us': np.float64(0.9615786666666666), 'median_duration_us': np.float64(0.992), 'std_dev_duration_us': np.float64(0.20368544657115023), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.535)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(2586.933), 'mean_duration_us': np.float64(3.449244), 'median_duration_us': np.float64(3.424), 'std_dev_duration_us': np.float64(0.1322517112579897), 'min_duration_us': np.float64(3.231), 'max_duration_us': np.float64(4.16)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16, __nv_bfloat16, float, true, false, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(37789.115000000005), 'mean_duration_us': np.float64(50.38548666666667), 'median_duration_us': np.float64(50.336), 'std_dev_duration_us': np.float64(0.3434009888680517), 'min_duration_us': np.float64(49.664), 'max_duration_us': np.float64(51.456)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(57705.129), 'mean_duration_us': np.float64(76.940172), 'median_duration_us': np.float64(76.991), 'std_dev_duration_us': np.float64(0.33444091219426697), 'min_duration_us': np.float64(75.935), 'max_duration_us': np.float64(77.919)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(62149.172999999995), 'mean_duration_us': np.float64(82.86556399999999), 'median_duration_us': np.float64(82.88), 'std_dev_duration_us': np.float64(0.46354650026076166), 'min_duration_us': np.float64(81.344), 'max_duration_us': np.float64(83.808)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(177480.581), 'mean_duration_us': np.float64(236.6407746666667), 'median_duration_us': np.float64(236.638), 'std_dev_duration_us': np.float64(0.7877751717494584), 'min_duration_us': np.float64(233.822), 'max_duration_us': np.float64(239.39)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.96)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(3.45)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(50.39)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(76.94)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(82.87)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc...', 'stream': 7, 'mean_duration_us': np.float64(236.64)}]","{'convNd': 'conv3d', 'input_shape': (1, 96, 6, 258, 258), 'filter_shape': (96, 96, 3, 3, 3), 'dtype_input_weight': ('c10::BFloat16', 'c10::BFloat16'), 'input_stride': (38340864, 399384, 66564, 258, 1), 'weight_stride': (2592, 27, 9, 3, 1), 'bias': False, 'stride': (1, 1, 1), 'padding': (0, 0, 0), 'dilation': (1, 1, 1), 'transposed_conv': False, 'output_padding': (0, 0, 0), 'groups': 1}",True,2.9221827693600573,58.01040248328332 -aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (32256, 8960), (8960, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (8960, 1), (1, 8960), (), ())","('', '', '', '1', '1')",12237,300,14617.529,48.725096666666666,3.622101279587241,887.899815936,672.0029296875,1260.0648190533489,matrix_bf16,0.6280223769901393,0.0011147025918556313,791.3489028235339,1.404597519704882,1122.0115462239582,1.9930724283064778,336603.4638671875,48.460499999999996,41.731,64.921,0.6281309463124868,0.6239660806244178,0.6314452104953299,791.4857072070524,786.2377064774342,795.6618949049015,1121.8140869140625,1115.926025390625,1129.302001953125,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(283.045), 'mean_duration_us': np.float64(0.9434833333333333), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.20174815089335738), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.6)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(336320.40800000005), 'mean_duration_us': np.float64(1121.0680266666668), 'median_duration_us': np.float64(1120.854), 'std_dev_duration_us': np.float64(1.9656534280714149), 'min_duration_us': np.float64(1115.222), 'max_duration_us': np.float64(1128.246)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.94)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(1121.07)}]","{'M': 32256, 'N': 1536, 'K': 8960, 'bias': True, 'stride_A': (8960, 1), 'stride_B': (1, 8960), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,2.9063929647264253,60.91679544800975 +aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (32256, 8960), (8960, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (8960, 1), (1, 8960), (), ())","('', '', '', '1', '1')",12237,300,14617.529,48.725096666666666,3.622101279587241,887.899815936,672.0029296875,1260.0648190533489,matrix_bf16,0.6280223769901393,0.0011147025918556313,791.3489028235339,1.404597519704882,1122.0115462239582,1.9930724283064778,336603.4638671875,48.460499999999996,41.731,64.921,0.6281309463124868,0.6239660806244178,0.6314452104953299,791.4857072070524,786.2377064774342,795.6618949049015,1121.8140869140625,1115.926025390625,1129.302001953125,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(283.045), 'mean_duration_us': np.float64(0.9434833333333333), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.20174815089335738), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.6)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 300, 'total_duration_us': np.float64(336320.40800000005), 'mean_duration_us': np.float64(1121.0680266666668), 'median_duration_us': np.float64(1120.854), 'std_dev_duration_us': np.float64(1.9656534280714149), 'min_duration_us': np.float64(1115.222), 'max_duration_us': np.float64(1128.246)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.94)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(1121.07)}]","{'M': 32256, 'N': 1536, 'K': 8960, 'bias': True, 'stride_A': (8960, 1), 'stride_B': (1, 8960), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,2.9063929647264253,60.91679544800975 aten::add,elementwise,python3,CPU,thread 10586 (python3),"((1, 32256, 1536), (1, 32256, 1536), ())","('float', 'float', 'Scalar')","((49545216, 1, 32256), (49545216, 1536, 1), ())","('', '', '1')",12052,600,22725.826,37.876376666666665,93.00535312156448,0.049545216,567.0,0.08333333333333333,vector_fp32,1.3695784197911651,0.002731917611581861,0.11413153498259711,0.00022765980096514993,434.1080045572917,0.8663701575617777,260464.802734375,9.44,8.43,1517.998,1.3698760468264561,1.3622422738092774,1.3768777251826652,0.11415633723553803,0.11352018948410647,0.11473981043188879,434.011962890625,431.804931640625,436.444091796875,"[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 600, 'total_duration_us': np.float64(260464.80200000003), 'mean_duration_us': np.float64(434.1080033333334), 'median_duration_us': np.float64(434.012), 'std_dev_duration_us': np.float64(0.8656460304240348), 'min_duration_us': np.float64(431.805), 'max_duration_us': np.float64(436.444)}]","[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(434.11)}]","{'shape_in1': (1, 32256, 1536), 'shape_in2': (1, 32256, 1536), 'dtype_in1_in2_out': ('float', 'float', None), 'stride_input1': (49545216, 1, 32256), 'stride_input2': (49545216, 1536, 1), 'stride_output': None}",True,2.248975876625963,63.16577132463571 aten::convolution,CONV_fwd,python3,CPU,thread 10586 (python3),"((1, 192, 6, 130, 130), (192, 192, 3, 3, 3), (192,), (), (), (), (), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'ScalarList', 'ScalarList', 'ScalarList', 'Scalar', 'ScalarList', 'Scalar')","((19468800, 101400, 16900, 130, 1), (5184, 27, 9, 3, 1), (1,), (), (), (), (), (), ())","('', '', '', '[1, 1, 1]', '[0, 0, 0]', '[1, 1, 1]', 'False', '[0, 0, 0]', '1')",136704,750,64940.835,86.58778,19.385352322182328,130.459631616,63.0322265625,1973.8474552637695,matrix_bf16,0.19276502569912513,0.0011759237299539993,380.48875544007325,2.3210940619539846,342.886572265625,2.093920309276013,257164.92919921875,83.05600000000001,73.301,357.294,0.19281666887221474,0.18812910161714333,0.19590743628264023,380.5906911858579,371.33814848805747,386.6913945737385,342.781982421875,337.3740234375,351.322998046875,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(726.1790000000001), 'mean_duration_us': np.float64(0.9682386666666668), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.2008908701382143), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.568)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(3685.314), 'mean_duration_us': np.float64(4.913752), 'median_duration_us': np.float64(4.896), 'std_dev_duration_us': np.float64(0.14547149032026865), 'min_duration_us': np.float64(4.64), 'max_duration_us': np.float64(5.664)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16, __nv_bfloat16, float, true, false, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(18797.244), 'mean_duration_us': np.float64(25.062991999999998), 'median_duration_us': np.float64(25.024), 'std_dev_duration_us': np.float64(0.21605943611886072), 'min_duration_us': np.float64(24.639), 'max_duration_us': np.float64(25.951)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(26742.756), 'mean_duration_us': np.float64(35.657008000000005), 'median_duration_us': np.float64(35.647), 'std_dev_duration_us': np.float64(0.2934462902633686), 'min_duration_us': np.float64(34.848), 'max_duration_us': np.float64(36.512)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(29195.826999999997), 'mean_duration_us': np.float64(38.92776933333333), 'median_duration_us': np.float64(38.815), 'std_dev_duration_us': np.float64(0.3583380993692351), 'min_duration_us': np.float64(38.304), 'max_duration_us': np.float64(39.84)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize128x128x64_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn', 'stream': 7, 'count': 750, 'total_duration_us': np.float64(178017.588), 'mean_duration_us': np.float64(237.35678399999998), 'median_duration_us': np.float64(237.19799999999998), 'std_dev_duration_us': np.float64(2.040857747454241), 'min_duration_us': np.float64(231.614), 'max_duration_us': np.float64(245.054)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.97)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(4.91)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(25.06)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(35.66)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(38.93)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc...', 'stream': 7, 'mean_duration_us': np.float64(237.36)}]","{'convNd': 'conv3d', 'input_shape': (1, 192, 6, 130, 130), 'filter_shape': (192, 192, 3, 3, 3), 'dtype_input_weight': ('c10::BFloat16', 'c10::BFloat16'), 'input_stride': (19468800, 101400, 16900, 130, 1), 'weight_stride': (5184, 27, 9, 3, 1), 'bias': False, 'stride': (1, 1, 1), 'padding': (0, 0, 0), 'dilation': (1, 1, 1), 'transposed_conv': False, 'output_padding': (0, 0, 0), 'groups': 1}",True,2.220483213131421,65.38625453776713 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((1, 32256, 1536), (1, 32256, 1536), ())","('float', 'c10::BFloat16', 'Scalar')","((49545216, 1, 32256), (49545216, 1, 32256), ())","('', '', 'False')",11888,1510,55791.176,36.947798675496685,51.39609857974198,0.049545216,283.5,0.16666666666666666,vector_fp32,1.933926056609385,0.012908670993090208,0.3223210094348975,0.0021514451655150375,153.72073303626863,1.0259995853644428,232118.30688476562,11.405000000000001,5.57,154.452,1.9343703063418067,1.8974369326435208,1.9740429145370508,0.3223950510569678,0.31623948877392016,0.32900715242284184,153.6785888671875,150.590087890625,156.669921875,"[{'name': 'void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)', 'stream': 7, 'count': 1510, 'total_duration_us': np.float64(232118.304), 'mean_duration_us': np.float64(153.72073112582783), 'median_duration_us': np.float64(153.67849999999999), 'std_dev_duration_us': np.float64(1.0256629543464066), 'min_duration_us': np.float64(150.59), 'max_duration_us': np.float64(156.67)}]","[{'name': 'void at::native::unrolled_elementwise_kernel, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp, unsigned int, c10::BFloat16, 4>)', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(7529.325000000001), 'mean_duration_us': np.float64(12.046920000000002), 'median_duration_us': np.float64(10.4), 'std_dev_duration_us': np.float64(2.242482709141812), 'min_duration_us': np.float64(9.888), 'max_duration_us': np.float64(16.448)}]","[{'name': 'void at::native::reduce_kernel<128, 4, at::native::ReduceOp, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp, unsigned int, c10::BFloat16, 4>)', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(7346.969999999999), 'mean_duration_us': np.float64(11.755151999999999), 'median_duration_us': np.float64(11.744), 'std_dev_duration_us': np.float64(0.16544031218539224), 'min_duration_us': np.float64(11.36), 'max_duration_us': np.float64(12.224)}]","[{'name': 'void at::native::reduce_kernel<128, 4, at::native::ReduceOp(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 124, 'total_duration_us': np.float64(2522.916), 'mean_duration_us': np.float64(20.34609677419355), 'median_duration_us': np.float64(20.255), 'std_dev_duration_us': np.float64(0.27235459036487725), 'min_duration_us': np.float64(19.936), 'max_duration_us': np.float64(21.024)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 124, 'total_duration_us': np.float64(4549.335999999999), 'mean_duration_us': np.float64(36.68819354838709), 'median_duration_us': np.float64(36.544), 'std_dev_duration_us': np.float64(0.35302091863037643), 'min_duration_us': np.float64(36.159), 'max_duration_us': np.float64(37.535)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(20.35)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(36.69)}]",,False,0.06106515475395358,97.63096524799587 -aten::mm,GEMM,python3,CPU,thread 10586 (python3),"((512, 4096), (4096, 10240))","('c10::BFloat16', 'c10::BFloat16')","((4096, 1), (1, 4096))","('', '')",259,96,3918.391,40.816572916666665,3.239533906516384,42.94967296,94.0,435.74468085106383,matrix_bf16,1.4296001882974725,0.015685558431838626,622.9406777943029,6.834898652852257,68.95487213134766,0.757329355493187,6619.667724609375,40.7555,35.67,51.151,1.4303193657714175,1.386853511813404,1.4730903491980094,623.2540555531624,604.3140408923089,641.8912840760688,68.9119873046875,66.9111328125,71.07177734375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(72.356), 'mean_duration_us': np.float64(0.7537083333333333), 'median_duration_us': np.float64(0.736), 'std_dev_duration_us': np.float64(0.13248647074533898), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.76)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(6547.305), 'mean_duration_us': np.float64(68.20109375), 'median_duration_us': np.float64(68.15950000000001), 'std_dev_duration_us': np.float64(0.7345655473096121), 'min_duration_us': np.float64(66.176), 'max_duration_us': np.float64(69.887)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.75)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(68.2)}]","{'M': 512, 'N': 10240, 'K': 4096, 'bias': False, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.057157331307863435,97.68812257930374 +aten::mm,GEMM,python3,CPU,thread 10586 (python3),"((512, 4096), (4096, 10240))","('c10::BFloat16', 'c10::BFloat16')","((4096, 1), (1, 4096))","('', '')",259,96,3918.391,40.816572916666665,3.239533906516384,42.94967296,94.0,435.74468085106383,matrix_bf16,1.4296001882974725,0.015685558431838626,622.9406777943029,6.834898652852257,68.95487213134766,0.757329355493187,6619.667724609375,40.7555,35.67,51.151,1.4303193657714175,1.386853511813404,1.4730903491980094,623.2540555531624,604.3140408923089,641.8912840760688,68.9119873046875,66.9111328125,71.07177734375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(72.356), 'mean_duration_us': np.float64(0.7537083333333333), 'median_duration_us': np.float64(0.736), 'std_dev_duration_us': np.float64(0.13248647074533898), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(1.76)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(6547.305), 'mean_duration_us': np.float64(68.20109375), 'median_duration_us': np.float64(68.15950000000001), 'std_dev_duration_us': np.float64(0.7345655473096121), 'min_duration_us': np.float64(66.176), 'max_duration_us': np.float64(69.887)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.75)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(68.2)}]","{'M': 512, 'N': 10240, 'K': 4096, 'bias': False, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.057157331307863435,97.68812257930374 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((4, 384, 128, 128), (4, 384, 128, 128), ())","('c10::BFloat16', 'float', 'Scalar')","((6291456, 16384, 128, 1), (6291456, 16384, 128, 1), ())","('', '', 'False')",136641,125,1316.144,10.529152,3.9147063335993124,0.025165824,144.0,0.16666666666666666,vector_bf16,2.92788110329652,0.02877491554535805,0.4879801838827533,0.00479581925755968,51.5763515625,0.5073307716576265,6447.0439453125,9.83,6.06,21.151,2.9253534007066535,2.87019468272988,2.978967172849615,0.48755890011777564,0.47836578045498,0.4964945288082692,51.615966796875,50.68701171875,52.60791015625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(6447.043999999998), 'mean_duration_us': np.float64(51.576351999999986), 'median_duration_us': np.float64(51.616), 'std_dev_duration_us': np.float64(0.5053116623391942), 'min_duration_us': np.float64(50.687), 'max_duration_us': np.float64(52.608)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::bf...', 'stream': 7, 'mean_duration_us': np.float64(51.58)}]","{'op_shape': (4, 384, 128, 128), 'dtype_in_out': ('c10::BFloat16', 'float'), 'stride_input': (6291456, 16384, 128, 1), 'stride_output': (6291456, 16384, 128, 1)}",True,0.055666816231373054,97.74378939553512 -aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (512, 1536), (1536, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1536, 1), (1, 1536), (), ())","('', '', '', '1', '1')",11855,610,84411.132,138.3789049180328,96.2345841762914,2.416705536,7.5029296875,307.1800078094494,matrix_bf16,0.7516805042352582,0.06526312615574323,230.9012231611975,20.04752760219027,10.545858014216188,0.917382381720403,6432.973388671875,203.15699999999998,29.53,252.763,0.7683082274922163,0.6419162493177426,0.8657005596389427,236.00892732112325,197.18383847843666,265.92590467053515,10.239990234375,9.087890625,12.256103515625,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 610, 'total_duration_us': np.float64(6432.974999999999), 'mean_duration_us': np.float64(10.545860655737703), 'median_duration_us': np.float64(10.24), 'std_dev_duration_us': np.float64(0.916629834968112), 'min_duration_us': np.float64(9.088), 'max_duration_us': np.float64(12.256)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpg...', 'stream': 7, 'mean_duration_us': np.float64(10.55)}]","{'M': 512, 'N': 1536, 'K': 1536, 'bias': True, 'stride_A': (1536, 1), 'stride_B': (1, 1536), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.05554532441319548,97.79933471994832 +aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (512, 1536), (1536, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1536, 1), (1, 1536), (), ())","('', '', '', '1', '1')",11855,610,84411.132,138.3789049180328,96.2345841762914,2.416705536,7.5029296875,307.1800078094494,matrix_bf16,0.7516805042352582,0.06526312615574323,230.9012231611975,20.04752760219027,10.545858014216188,0.917382381720403,6432.973388671875,203.15699999999998,29.53,252.763,0.7683082274922163,0.6419162493177426,0.8657005596389427,236.00892732112325,197.18383847843666,265.92590467053515,10.239990234375,9.087890625,12.256103515625,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 610, 'total_duration_us': np.float64(6432.974999999999), 'mean_duration_us': np.float64(10.545860655737703), 'median_duration_us': np.float64(10.24), 'std_dev_duration_us': np.float64(0.916629834968112), 'min_duration_us': np.float64(9.088), 'max_duration_us': np.float64(12.256)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpg...', 'stream': 7, 'mean_duration_us': np.float64(10.55)}]","{'M': 512, 'N': 1536, 'K': 1536, 'bias': True, 'stride_A': (1536, 1), 'stride_B': (1, 1536), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.05554532441319548,97.79933471994832 aten::div,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 2, 64, 64), (1, 384, 2, 64, 64))","('c10::BFloat16', 'c10::BFloat16')","((3145728, 8192, 4096, 64, 1), (8192, 0, 4096, 64, 1))","('', '')",136266,625,9465.978,15.145564799999999,58.79092328700498,0.003145728,18.0,0.16666666666666666,vector_bf16,1.864674935791814,0.036392794112675576,0.3107791559653023,0.006065465685445945,10.12598359375,0.20095624411106822,6328.73974609375,9.68,8.99,1472.027,1.8726240511578334,1.7145958288717869,1.9277232028725313,0.3121040085263056,0.2857659714786312,0.32128720047875525,10.0791015625,9.791015625,11.008056640625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(6328.746), 'mean_duration_us': np.float64(10.125993600000001), 'median_duration_us': np.float64(10.079), 'std_dev_duration_us': np.float64(0.20077886731187633), 'min_duration_us': np.float64(9.791), 'max_duration_us': np.float64(11.008)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(10.13)}]","{'shape_in1': (1, 384, 2, 64, 64), 'shape_in2': (1, 384, 2, 64, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (3145728, 8192, 4096, 64, 1), 'stride_input2': (8192, 0, 4096, 64, 1), 'stride_output': None}",True,0.05464532201275552,97.85398004196108 aten::add,elementwise,python3,CPU,thread 10586 (python3),"((1, 192, 4, 128, 128), (1, 192, 4, 128, 128), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((12582912, 65536, 16384, 128, 1), (12582912, 65536, 16384, 128, 1), ())","('', '', '1')",136890,250,2377.8269999999998,9.511308,1.4857245190552948,0.012582912,72.0,0.16666666666666666,vector_bf16,3.0842689745598553,0.06303210252987679,0.5140448290933093,0.010505350421646157,24.4884892578125,0.5037953218003651,6122.122314453125,9.23,8.53,21.781,3.088084016337291,2.863231996444543,3.240941197618848,0.5146806693895485,0.4772053327407572,0.540156866269808,24.447998046875,23.294921875,26.367919921875,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 250, 'total_duration_us': np.float64(6122.124), 'mean_duration_us': np.float64(24.488495999999998), 'median_duration_us': np.float64(24.448), 'std_dev_duration_us': np.float64(0.5027776625746219), 'min_duration_us': np.float64(23.295), 'max_duration_us': np.float64(26.368)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(24.49)}]","{'shape_in1': (1, 192, 4, 128, 128), 'shape_in2': (1, 192, 4, 128, 128), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (12582912, 65536, 16384, 128, 1), 'stride_input2': (12582912, 65536, 16384, 128, 1), 'stride_output': None}",True,0.05286128971905608,97.90684133168014 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 3, 32, 32), (1, 384, 3, 32, 32), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((1331712, 3468, 1156, 34, 1), (1179648, 3072, 1024, 32, 1), ())","('', '', 'False')",137609,1240,7758.13,6.256556451612903,0.5288344700659493,0.001179648,4.5,0.25,vector_bf16,0.9646919909012525,0.014855199007441462,0.24117299772531312,0.0037137997518603655,4.892459795551916,0.0757944864681429,6066.650146484375,6.18,5.73,14.06,0.9637654748179915,0.9102078191579542,1.0030803836412705,0.24094136870449787,0.22755195478948856,0.2507700959103176,4.89599609375,4.7041015625,5.18408203125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1240, 'total_duration_us': np.float64(6066.648999999999), 'mean_duration_us': np.float64(4.892458870967742), 'median_duration_us': np.float64(4.896), 'std_dev_duration_us': np.float64(0.07580069315379813), 'min_duration_us': np.float64(4.704), 'max_duration_us': np.float64(5.184)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(4.89)}]","{'op_shape': (1, 384, 3, 32, 32), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (1331712, 3468, 1156, 34, 1), 'stride_output': (1179648, 3072, 1024, 32, 1)}",True,0.05238231687406447,97.9592236485542 -aten::mm,GEMM,python3,CPU,thread 10586 (python3),"((512, 4096), (4096, 4096))","('c10::BFloat16', 'c10::BFloat16')","((4096, 1), (1, 4096))","('', '')",77,192,8040.803,41.879182291666666,106.47227864690899,17.179869184,40.0,409.6,matrix_bf16,1.3387112394781984,0.04694098358016499,548.33612369027,19.22702687443559,31.37018585205078,1.126722931713206,6023.07568359375,35.7605,22.18,1498.848,1.351256031461381,1.2136275720199494,1.4201173121719364,553.4744704865817,497.1018534993713,581.6800510656251,31.0400390625,29.534912109375,34.56005859375,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 192, 'total_duration_us': np.float64(6023.076), 'mean_duration_us': np.float64(31.3701875), 'median_duration_us': np.float64(31.04), 'std_dev_duration_us': np.float64(1.1237826868855696), 'min_duration_us': np.float64(29.535), 'max_duration_us': np.float64(34.56)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.37)}]","{'M': 512, 'N': 4096, 'K': 4096, 'bias': False, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.05200607442268847,98.01122972297689 +aten::mm,GEMM,python3,CPU,thread 10586 (python3),"((512, 4096), (4096, 4096))","('c10::BFloat16', 'c10::BFloat16')","((4096, 1), (1, 4096))","('', '')",77,192,8040.803,41.879182291666666,106.47227864690899,17.179869184,40.0,409.6,matrix_bf16,1.3387112394781984,0.04694098358016499,548.33612369027,19.22702687443559,31.37018585205078,1.126722931713206,6023.07568359375,35.7605,22.18,1498.848,1.351256031461381,1.2136275720199494,1.4201173121719364,553.4744704865817,497.1018534993713,581.6800510656251,31.0400390625,29.534912109375,34.56005859375,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 192, 'total_duration_us': np.float64(6023.076), 'mean_duration_us': np.float64(31.3701875), 'median_duration_us': np.float64(31.04), 'std_dev_duration_us': np.float64(1.1237826868855696), 'min_duration_us': np.float64(29.535), 'max_duration_us': np.float64(34.56)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(31.37)}]","{'M': 512, 'N': 4096, 'K': 4096, 'bias': False, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.05200607442268847,98.01122972297689 aten::mul,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 2, 64, 64), (384, 1, 1, 1))","('c10::BFloat16', 'c10::BFloat16')","((3145728, 8192, 4096, 64, 1), (1, 1, 1, 1))","('', '')",136268,625,9075.711,14.5211376,14.709127691313626,0.003145728,12.000732421875,0.24998474214220323,vector_bf16,1.3670114335850339,0.012991124790125945,0.34173200073019827,0.003247582980796759,9.20608359375,0.08802088745502001,5753.80224609375,8.25,7.63,70.121,1.3701619777765963,1.310786665988505,1.3946304800043292,0.34251958870753363,0.3276766667005747,0.3486363409275394,9.18408203125,9.02294921875,9.60009765625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(5753.777), 'mean_duration_us': np.float64(9.2060432), 'median_duration_us': np.float64(9.184), 'std_dev_duration_us': np.float64(0.08795475049001056), 'min_duration_us': np.float64(9.023), 'max_duration_us': np.float64(9.6)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(9.21)}]","{'shape_in1': (1, 384, 2, 64, 64), 'shape_in2': (384, 1, 1, 1), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (3145728, 8192, 4096, 64, 1), 'stride_input2': (1, 1, 1, 1), 'stride_output': None}",True,0.049681040641555145,98.06091076361845 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((4, 192, 128, 128), (4, 192, 128, 128), ())","('float', 'float', 'Scalar')","((3145728, 16384, 128, 1), (16384, 65536, 128, 1), ())","('', '', 'False')",137025,125,1052.134,8.417072000000001,1.0289282803308089,0.012582912,96.0,0.125,vector_fp32,2.255789829457515,0.02112702295482662,0.2819737286821894,0.0026408778693533274,44.62830859375,0.4193536706912899,5578.53857421875,8.22,7.5,15.8,2.266423673709901,2.209073065283664,2.2844811254944983,0.2833029592137376,0.276134133160458,0.2855601406868123,44.4150390625,44.06396484375,45.568115234375,"[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(5578.536999999999), 'mean_duration_us': np.float64(44.62829599999999), 'median_duration_us': np.float64(44.415), 'std_dev_duration_us': np.float64(0.4176719099771976), 'min_duration_us': np.float64(44.064), 'max_duration_us': np.float64(45.568)}]","[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(44.63)}]","{'op_shape': (4, 192, 128, 128), 'dtype_in_out': ('float', 'float'), 'stride_input': (3145728, 16384, 128, 1), 'stride_output': (16384, 65536, 128, 1)}",True,0.04816773148823458,98.10907849510669 aten::convolution,CONV_fwd,python3,CPU,thread 10586 (python3),"((1, 384, 3, 32, 32), (768, 384, 3, 1, 1), (768,), (), (), (), (), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'ScalarList', 'ScalarList', 'ScalarList', 'Scalar', 'ScalarList', 'Scalar')","((1179648, 3072, 1024, 32, 1), (1152, 3, 1, 1, 1), (1,), (), (), (), (), (), ())","('', '', '', '[1, 1, 1]', '[0, 0, 0]', '[1, 1, 1]', 'False', '[0, 0, 0]', '1')",136113,125,10065.179,80.521432,8.44938261090237,1.811939328,5.4375,317.7931034482759,matrix_bf16,0.13147732023217665,0.0005511163491296946,41.782585629646206,0.1751409749510105,43.366654296875,0.1819913322180833,5420.831787109375,78.921,72.531,142.112,0.13150228146379642,0.129866455385642,0.1326706660379824,41.79051813690855,41.27066389083023,42.1618226967602,43.357666015625,42.975830078125,43.90380859375,"[{'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16, __nv_bfloat16, float, true, false, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(355.5439999999999), 'mean_duration_us': np.float64(2.8443519999999993), 'median_duration_us': np.float64(2.848), 'std_dev_duration_us': np.float64(0.0836824718564169), 'min_duration_us': np.float64(2.656), 'max_duration_us': np.float64(3.104)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(417.598), 'mean_duration_us': np.float64(3.340784), 'median_duration_us': np.float64(3.328), 'std_dev_duration_us': np.float64(0.03836719098396442), 'min_duration_us': np.float64(3.232), 'max_duration_us': np.float64(3.425)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(517.5329999999999), 'mean_duration_us': np.float64(4.140263999999999), 'median_duration_us': np.float64(4.159), 'std_dev_duration_us': np.float64(0.05790352583392486), 'min_duration_us': np.float64(3.968), 'max_duration_us': np.float64(4.288)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(1974.3799999999999), 'mean_duration_us': np.float64(15.795039999999998), 'median_duration_us': np.float64(15.776), 'std_dev_duration_us': np.float64(0.09840547952222992), 'min_duration_us': np.float64(15.584), 'max_duration_us': np.float64(16.127)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x128x64_warpgroupsize2x1x1_g1_execute_segment_k_off_kernel__5x_cudnn', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(2155.7850000000008), 'mean_duration_us': np.float64(17.246280000000006), 'median_duration_us': np.float64(17.248), 'std_dev_duration_us': np.float64(0.07720762656629208), 'min_duration_us': np.float64(17.088), 'max_duration_us': np.float64(17.696)}]","[{'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(2.84)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(3.34)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(4.14)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(15.8)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc...', 'stream': 7, 'mean_duration_us': np.float64(17.25)}]","{'convNd': 'conv3d', 'input_shape': (1, 384, 3, 32, 32), 'filter_shape': (768, 384, 3, 1, 1), 'dtype_input_weight': ('c10::BFloat16', 'c10::BFloat16'), 'input_stride': (1179648, 3072, 1024, 32, 1), 'weight_stride': (1152, 3, 1, 1, 1), 'bias': False, 'stride': (1, 1, 1), 'padding': (0, 0, 0), 'dilation': (1, 1, 1), 'transposed_conv': False, 'output_padding': (0, 0, 0), 'groups': 1}",True,0.04680601675340004,98.15588451186008 @@ -94,7 +94,7 @@ aten::mul,elementwise,python3,CPU,thread 10586 (python3),"((1, 96, 4, 256, 256), aten::mul,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 1, 32, 32), (384, 1, 1, 1))","('c10::BFloat16', 'c10::BFloat16')","((393216, 1024, 1024, 32, 1), (1, 1, 1, 1))","('', '')",133495,1260,12213.59,9.693325396825397,41.26575915564248,0.000393216,1.500732421875,0.24987798926305516,vector_bf16,0.46235935079163537,0.0114868282093333,0.11553342489278541,0.0028703055359583455,3.405628410218254,0.08649731780561377,4291.091796875,8.16,7.49,1471.158,0.46391224067943,0.4167591278934437,0.486901093216498,0.11592145789549446,0.10413893288503814,0.1216658661429219,3.39208984375,3.23193359375,3.77587890625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 1260, 'total_duration_us': np.float64(4291.061), 'mean_duration_us': np.float64(3.405603968253968), 'median_duration_us': np.float64(3.392), 'std_dev_duration_us': np.float64(0.08644539234261618), 'min_duration_us': np.float64(3.232), 'max_duration_us': np.float64(3.776)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(3.41)}]","{'shape_in1': (1, 384, 1, 32, 32), 'shape_in2': (384, 1, 1, 1), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (393216, 1024, 1024, 32, 1), 'stride_input2': (1, 1, 1, 1), 'stride_output': None}",True,0.037051309175931864,98.56522582839584 aten::convolution,CONV_fwd,python3,CPU,thread 10586 (python3),"((1, 16, 126, 32, 32), (1536, 16, 1, 2, 2), (1536,), (), (), (), (), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'ScalarList', 'ScalarList', 'ScalarList', 'Scalar', 'ScalarList', 'Scalar')","((2064384, 129024, 1024, 32, 1), (64, 4, 4, 2, 1), (1,), (), (), (), (), (), ())","('', '', '', '[1, 2, 2]', '[0, 0, 0]', '[1, 1, 1]', 'False', '[0, 0, 0]', '1')",11770,10,2484.06,248.406,31.136808535522356,6.341787648,98.625,61.32319391634981,matrix_bf16,0.29432080846053194,0.002429381320169562,18.04869201084205,0.1489774217935162,351.3926025390625,2.9030591533399024,3513.926025390625,251.4725,199.903,281.503,0.29419852117792245,0.28942935553340227,0.29873995509497675,18.041192964097085,17.74873249445898,18.319688196850894,351.51708984375,346.17333984375,357.309326171875,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(7.296999999999999), 'mean_duration_us': np.float64(0.7296999999999999), 'median_duration_us': np.float64(0.736), 'std_dev_duration_us': np.float64(0.012853404218338436), 'min_duration_us': np.float64(0.704), 'max_duration_us': np.float64(0.737)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(43.648), 'mean_duration_us': np.float64(4.364800000000001), 'median_duration_us': np.float64(4.368), 'std_dev_duration_us': np.float64(0.07039999999999988), 'min_duration_us': np.float64(4.256), 'max_duration_us': np.float64(4.512)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(84.416), 'mean_duration_us': np.float64(8.4416), 'median_duration_us': np.float64(8.416), 'std_dev_duration_us': np.float64(0.16860083036568962), 'min_duration_us': np.float64(8.16), 'max_duration_us': np.float64(8.704)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize128x64x64_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(762.554), 'mean_duration_us': np.float64(76.2554), 'median_duration_us': np.float64(76.1115), 'std_dev_duration_us': np.float64(1.3488461142769423), 'min_duration_us': np.float64(73.663), 'max_duration_us': np.float64(78.912)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16, __nv_bfloat16, float, true, false, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(1104.886), 'mean_duration_us': np.float64(110.48859999999999), 'median_duration_us': np.float64(110.431), 'std_dev_duration_us': np.float64(0.2850695353768956), 'min_duration_us': np.float64(110.111), 'max_duration_us': np.float64(110.975)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(1511.123), 'mean_duration_us': np.float64(151.1123), 'median_duration_us': np.float64(151.135), 'std_dev_duration_us': np.float64(1.1013839521256856), 'min_duration_us': np.float64(148.959), 'max_duration_us': np.float64(153.534)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.73)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(4.36)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(8.44)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc...', 'stream': 7, 'mean_duration_us': np.float64(76.26)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(110.49)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(151.11)}]","{'convNd': 'conv3d', 'input_shape': (1, 16, 126, 32, 32), 'filter_shape': (1536, 16, 1, 2, 2), 'dtype_input_weight': ('c10::BFloat16', 'c10::BFloat16'), 'input_stride': (2064384, 129024, 1024, 32, 1), 'weight_stride': (64, 4, 4, 2, 1), 'bias': False, 'stride': (1, 2, 2), 'padding': (0, 0, 0), 'dilation': (1, 1, 1), 'transposed_conv': False, 'output_padding': (0, 0, 0), 'groups': 1}",True,0.030340893588647246,98.59556672198448 aten::silu,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 2, 64, 64),)","('c10::BFloat16',)","((3145728, 8192, 4096, 64, 1),)","('',)",136270,625,9579.717,15.327547200000001,22.25407710617379,0.003145728,12.0,0.25,vector_bf16,2.263510598763327,0.025250001876547308,0.5658776496908318,0.006312500469136827,5.55972578125,0.06293178273687887,3474.82861328125,8.57,7.96,99.641,2.272970564586549,2.160536891720813,2.29995124958722,0.5682426411466373,0.5401342229302033,0.574987812396805,5.535888671875,5.470947265625,5.823974609375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(3474.8329999999996), 'mean_duration_us': np.float64(5.559732799999999), 'median_duration_us': np.float64(5.536), 'std_dev_duration_us': np.float64(0.06282354179891485), 'min_duration_us': np.float64(5.471), 'max_duration_us': np.float64(5.824)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(a...', 'stream': 7, 'mean_duration_us': np.float64(5.56)}]","{'op_shape': (1, 384, 2, 64, 64), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (3145728, 8192, 4096, 64, 1), 'stride_output': None}",True,0.03000330810396984,98.62557003008845 -aten::mm,GEMM,python3,CPU,thread 10586 (python3),"((512, 10240), (10240, 4096))","('c10::BFloat16', 'c10::BFloat16')","((10240, 1), (1, 10240))","('', '')",289,48,1597.278,33.276625,2.6353552437267873,42.94967296,94.0,435.74468085106383,matrix_bf16,1.3724764570786148,0.01613094400980978,598.0493157653199,7.028973049380965,71.82601420084636,0.846523518853255,3447.648681640625,33.0355,29.011,39.51,1.3711251186526003,1.3328587891965415,1.4019999854982375,597.4604772341544,580.7861277179825,610.9140362341257,71.8870849609375,70.303955078125,73.950927734375,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(3447.648), 'mean_duration_us': np.float64(71.82600000000001), 'median_duration_us': np.float64(71.887), 'std_dev_duration_us': np.float64(0.8376598354940977), 'min_duration_us': np.float64(70.304), 'max_duration_us': np.float64(73.951)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(71.83)}]","{'M': 512, 'N': 4096, 'K': 10240, 'bias': False, 'stride_A': (10240, 1), 'stride_B': (1, 10240), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.029768623762951807,98.6553386538514 +aten::mm,GEMM,python3,CPU,thread 10586 (python3),"((512, 10240), (10240, 4096))","('c10::BFloat16', 'c10::BFloat16')","((10240, 1), (1, 10240))","('', '')",289,48,1597.278,33.276625,2.6353552437267873,42.94967296,94.0,435.74468085106383,matrix_bf16,1.3724764570786148,0.01613094400980978,598.0493157653199,7.028973049380965,71.82601420084636,0.846523518853255,3447.648681640625,33.0355,29.011,39.51,1.3711251186526003,1.3328587891965415,1.4019999854982375,597.4604772341544,580.7861277179825,610.9140362341257,71.8870849609375,70.303955078125,73.950927734375,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(3447.648), 'mean_duration_us': np.float64(71.82600000000001), 'median_duration_us': np.float64(71.887), 'std_dev_duration_us': np.float64(0.8376598354940977), 'min_duration_us': np.float64(70.304), 'max_duration_us': np.float64(73.951)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(71.83)}]","{'M': 512, 'N': 4096, 'K': 10240, 'bias': False, 'stride_A': (10240, 1), 'stride_B': (1, 10240), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.029768623762951807,98.6553386538514 aten::fill_,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 4, 66, 66), ())","('c10::BFloat16', 'Scalar')","((6690816, 17424, 4356, 66, 1), ())","('', '0.')",136296,625,12954.463,20.7271408,46.15295399603662,,,,,,,,,5.31633359375,0.06651587998524273,3322.70849609375,6.46,5.95,186.783,,,,,,,5.31201171875,5.215087890625,5.568115234375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(3322.7019999999998), 'mean_duration_us': np.float64(5.316323199999999), 'median_duration_us': np.float64(5.312), 'std_dev_duration_us': np.float64(0.06649757545775631), 'min_duration_us': np.float64(5.215), 'max_duration_us': np.float64(5.568)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Fi...', 'stream': 7, 'mean_duration_us': np.float64(5.32)}]",,False,0.0286898313105119,98.68402848516192 aten::mul,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 2, 64, 64), ())","('c10::BFloat16', 'double')","((3145728, 8192, 4096, 64, 1), ())","('', '')",136267,625,8004.826,12.8077216,7.520131897280622,0.003145728,12.000007629394531,0.24999984105438167,vector_bf16,2.3908406192747966,0.026366114053941625,0.5977097748050586,0.0065915243227070985,5.2636234375,0.059448436124279705,3289.7646484375,9.2,8.48,35.301,2.3976386453293634,2.259816736966721,2.4272223942733353,0.5994092802381839,0.5649538250537116,0.6068052127719695,5.248046875,5.18408203125,5.568115234375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(3289.74), 'mean_duration_us': np.float64(5.263584), 'median_duration_us': np.float64(5.248), 'std_dev_duration_us': np.float64(0.05943141041570529), 'min_duration_us': np.float64(5.184), 'max_duration_us': np.float64(5.568)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(5.26)}]","{'shape_in1': (1, 384, 2, 64, 64), 'shape_in2': (), 'dtype_in1_in2_out': ('c10::BFloat16', 'double', None), 'stride_input1': (3145728, 8192, 4096, 64, 1), 'stride_input2': (), 'stride_output': None}",True,0.028405378601799064,98.71243386376372 aten::add,elementwise,python3,CPU,thread 10586 (python3),"((1, 384, 2, 64, 64), (), ())","('c10::BFloat16', 'double', 'Scalar')","((3145728, 8192, 4096, 64, 1), (), ())","('', '', '1')",136269,625,9268.452,14.829523199999999,16.24265923345745,0.003145728,12.000007629394531,0.24999984105438167,vector_bf16,2.395279259382274,0.02884453593639032,0.5988194341264254,0.007211129399384996,5.253988671875,0.06425591865551121,3283.742919921875,8.34,7.49,68.081,2.3976386453293634,2.2861799290276794,2.4428685335102855,0.5994092802381839,0.5715446188786373,0.6107167450943218,5.248046875,5.15087890625,5.50390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctorOnSelf_add, std::array >(int, at::native::CUDAFunctorOnSelf_add, std::array)', 'stream': 7, 'count': 625, 'total_duration_us': np.float64(3283.716), 'mean_duration_us': np.float64(5.2539456), 'median_duration_us': np.float64(5.248), 'std_dev_duration_us': np.float64(0.06423713755017425), 'min_duration_us': np.float64(5.151), 'max_duration_us': np.float64(5.504)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(5.25)}]","{'shape_in1': (1, 384, 2, 64, 64), 'shape_in2': (), 'dtype_in1_in2_out': ('c10::BFloat16', 'double', None), 'stride_input1': (3145728, 8192, 4096, 64, 1), 'stride_input2': (), 'stride_output': None}",True,0.028353384159459604,98.74078724792318 @@ -149,11 +149,11 @@ aten::mul,elementwise,python3,CPU,thread 10586 (python3),"((1, 512, 4096), (1, 5 "aten::_upsample_nearest_exact2d->void at::native::(anonymous namespace)::upsample_nearest2d_out_frame(float const*, float*, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, float, float) (Synthetic Op)",other,python3,CPU,thread 10586 (python3),"((2, 384, 32, 32), (), (), ())","('float', 'ScalarList', 'Scalar', 'Scalar')","((1024, 2048, 32, 1), (), (), ())","('', '[64, 64]', '2.', '2.')",1675601,125,4128.878,33.031023999999995,8.010636777942752,,,,,,,,,7.954091796875,0.061588769495308324,994.261474609375,31.161,27.751,94.031,,,,,,,7.967041015625,7.80810546875,8.159912109375,"[{'name': 'void at::native::(anonymous namespace)::upsample_nearest2d_out_frame(float const*, float*, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, float, float)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(994.258), 'mean_duration_us': np.float64(7.954064000000001), 'median_duration_us': np.float64(7.967), 'std_dev_duration_us': np.float64(0.06137435868504052), 'min_duration_us': np.float64(7.808), 'max_duration_us': np.float64(8.16)}]","[{'name': 'void at::native::(anonymous namespace)::upsample_nearest2d_out_f...', 'stream': 7, 'mean_duration_us': np.float64(7.95)}]",,False,0.008584922215902669,99.55263534909784 aten::mean,reduce,python3,CPU,thread 10586 (python3),"((1, 512, 4096), (), (), ())","('float', 'ScalarList', 'Scalar', '')","((2097152, 4096, 1), (), (), ())","('', '[-1]', 'True', '')",61,98,1275.081,13.011030612244896,1.6957931784884226,0.002097152,8.000003814697266,0.2499998807907673,vector_fp32,0.8294320467827867,0.005987909755465986,0.20735791281973875,0.0014969767250523713,10.114205496651786,0.0732592187330293,991.192138671875,12.524999999999999,11.65,22.39,0.8295650486974578,0.8090930546542022,0.8429152602114663,0.2073911632825515,0.2022731672121883,0.21072871456958517,10.112060546875,9.951904296875,10.367919921875,"[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, float, 4> >(at::native::ReduceOp, unsigned int, float, 4>)', 'stream': 7, 'count': 98, 'total_duration_us': np.float64(991.187), 'mean_duration_us': np.float64(10.11415306122449), 'median_duration_us': np.float64(10.112), 'std_dev_duration_us': np.float64(0.07290703502361782), 'min_duration_us': np.float64(9.952), 'max_duration_us': np.float64(10.368)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 6, 'total_duration_us': np.float64(20.96), 'mean_duration_us': np.float64(3.4933333333333336), 'median_duration_us': np.float64(3.536), 'std_dev_duration_us': np.float64(0.11599233691163502), 'min_duration_us': np.float64(3.296), 'max_duration_us': np.float64(3.648)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16, __nv_bfloat16, float, true, false, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 6, 'total_duration_us': np.float64(69.888), 'mean_duration_us': np.float64(11.648000000000001), 'median_duration_us': np.float64(11.6), 'std_dev_duration_us': np.float64(0.15567487487281534), 'min_duration_us': np.float64(11.52), 'max_duration_us': np.float64(11.968)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 6, 'total_duration_us': np.float64(95.2), 'mean_duration_us': np.float64(15.866666666666667), 'median_duration_us': np.float64(15.888), 'std_dev_duration_us': np.float64(0.062425066190504724), 'min_duration_us': np.float64(15.744), 'max_duration_us': np.float64(15.936)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 6, 'total_duration_us': np.float64(253.95), 'mean_duration_us': np.float64(42.324999999999996), 'median_duration_us': np.float64(42.352000000000004), 'std_dev_duration_us': np.float64(0.20668010708983695), 'min_duration_us': np.float64(42.015), 'max_duration_us': np.float64(42.656)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn', 'stream': 7, 'count': 6, 'total_duration_us': np.float64(480.379), 'mean_duration_us': np.float64(80.06316666666667), 'median_duration_us': np.float64(79.983), 'std_dev_duration_us': np.float64(0.3969082247685089), 'min_duration_us': np.float64(79.584), 'max_duration_us': np.float64(80.799)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.77)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(3.49)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(11.65)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(15.87)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(42.32)}, {'name': 'sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc...', 'stream': 7, 'mean_duration_us': np.float64(80.06)}]","{'convNd': 'conv3d', 'input_shape': (1, 96, 3, 258, 258), 'filter_shape': (96, 96, 3, 3, 3), 'dtype_input_weight': ('c10::BFloat16', 'c10::BFloat16'), 'input_stride': (19170432, 199692, 66564, 258, 1), 'weight_stride': (2592, 27, 9, 3, 1), 'bias': False, 'stride': (1, 1, 1), 'padding': (0, 0, 0), 'dilation': (1, 1, 1), 'transposed_conv': False, 'output_padding': (0, 0, 0), 'groups': 1}",True,0.007986757408261241,99.5691805266288 -aten::bmm,GEMM,python3,CPU,thread 10586 (python3),"((64, 512, 512), (64, 512, 64))","('c10::BFloat16', 'c10::BFloat16')","((262144, 512, 1), (64, 4096, 1))","('', '')",215,48,1339.549,27.90727083333333,3.2634101126561648,2.147483648,40.0,51.2,matrix_bf16,2.291045759224885,0.062142092073413965,117.30154287231413,3.181675114158794,18.321060180664062,0.5155002027466918,879.410888671875,27.561,24.95,47.791,2.309636210368003,2.114055150925983,2.3874524637640877,118.25337397084175,108.23962372741032,122.2375661447213,18.1600341796875,17.568115234375,19.840087890625,"[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_32x6_nn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(879.4110000000001), 'mean_duration_us': np.float64(18.3210625), 'median_duration_us': np.float64(18.159999999999997), 'std_dev_duration_us': np.float64(0.5100926143950885), 'min_duration_us': np.float64(17.568), 'max_duration_us': np.float64(19.84)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_32x6_nn_align8::Params)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(879.4110000000001), 'mean_duration_us': np.float64(18.3210625), 'median_duration_us': np.float64(18.159999999999997), 'std_dev_duration_us': np.float64(0.5100926143950885), 'min_duration_us': np.float64(17.568), 'max_duration_us': np.float64(19.84)}]","[{'name': 'void cutlass::Kernel2 > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(865.0190000000001), 'mean_duration_us': np.float64(6.920152000000001), 'median_duration_us': np.float64(6.912), 'std_dev_duration_us': np.float64(0.09543945146531393), 'min_duration_us': np.float64(6.688), 'max_duration_us': np.float64(7.2)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(6.92)}]","{'shape_in1': (1, 192, 2, 64, 64), 'shape_in2': (1, 192, 2, 64, 64), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (1572864, 4096, 786432, 64, 1), 'stride_input2': (8192, 0, 4096, 64, 1), 'stride_output': None}",True,0.007468984250855273,99.58424275900187 aten::mul,elementwise,python3,CPU,thread 10586 (python3),"((1, 512, 10240), (1, 512, 10240))","('c10::BFloat16', 'c10::BFloat16')","((5242880, 10240, 1), (5242880, 10240, 1))","('', '')",270,96,932.324,9.711708333333332,1.9445434962406805,0.00524288,30.0,0.16666666666666666,vector_bf16,3.543690217481817,0.21215718719947602,0.5906150362469695,0.03535953119991269,8.908638000488281,0.5345586607333144,855.229248046875,10.195,6.95,17.28,3.4373894855636298,3.1609307185437774,3.8250020447663715,0.5728982475939384,0.5268217864239629,0.6375003407943953,9.151611328125,8.22412109375,9.951904296875,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(855.2280000000001), 'mean_duration_us': np.float64(8.908625), 'median_duration_us': np.float64(9.151499999999999), 'std_dev_duration_us': np.float64(0.5317951761486744), 'min_duration_us': np.float64(8.224), 'max_duration_us': np.float64(9.952)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Bi...', 'stream': 7, 'mean_duration_us': np.float64(8.91)}]","{'shape_in1': (1, 512, 10240), 'shape_in2': (1, 512, 10240), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (5242880, 10240, 1), 'stride_input2': (5242880, 10240, 1), 'stride_output': None}",True,0.007384452439064787,99.59162721144094 aten::mul,elementwise,python3,CPU,thread 10586 (python3),"((1, 192, 2, 64, 64), (192, 1, 1, 1))","('c10::BFloat16', 'c10::BFloat16')","((1572864, 4096, 786432, 64, 1), (1, 1, 1, 1))","('', '')",136210,125,1221.163,9.769304,2.096697850092055,0.001572864,6.0003662109375,0.24998474214220323,vector_bf16,0.9483483391091755,0.013677064698882607,0.23707261501319396,0.003419057492012392,6.635900390625,0.09621578052530944,829.487548828125,8.76,8.01,19.51,0.9498517116320212,0.9060709714165173,0.9831150011444266,0.23744843520566122,0.22650391815209367,0.24576375005722134,6.6240234375,6.39990234375,6.944091796875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(829.485), 'mean_duration_us': np.float64(6.63588), 'median_duration_us': np.float64(6.624), 'std_dev_duration_us': np.float64(0.09586303562896387), 'min_duration_us': np.float64(6.4), 'max_duration_us': np.float64(6.944)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(6.64)}]","{'shape_in1': (1, 192, 2, 64, 64), 'shape_in2': (192, 1, 1, 1), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (1572864, 4096, 786432, 64, 1), 'stride_input2': (1, 1, 1, 1), 'stride_output': None}",True,0.007162186474686599,99.59878939791562 -aten::bmm,GEMM,python3,CPU,thread 10586 (python3),"((64, 512, 64), (64, 64, 512))","('c10::BFloat16', 'c10::BFloat16')","((64, 4096, 1), (64, 1, 4096))","('', '')",117,48,1724.109,35.9189375,3.929348830283864,2.147483648,40.0,51.2,matrix_bf16,2.4355642974451306,0.01757383136199424,124.7008920291907,0.89978016573411,17.221954345703125,0.12420286115001129,826.65380859375,34.811,32.08,54.66,2.44080771517063,2.4005630025431066,2.4777347136449515,124.96935501673627,122.90882573020707,126.86001733862152,17.18408203125,16.927978515625,17.47216796875,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(35.235), 'mean_duration_us': np.float64(0.7340625), 'median_duration_us': np.float64(0.736), 'std_dev_duration_us': np.float64(0.012128489618112675), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(0.768)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(791.415), 'mean_duration_us': np.float64(16.4878125), 'median_duration_us': np.float64(16.448), 'std_dev_duration_us': np.float64(0.12022958181641505), 'min_duration_us': np.float64(16.191), 'max_duration_us': np.float64(16.736)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.73)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(16.49)}]","{'B': 64, 'M': 512, 'N': 512, 'K': 64, 'bias': False, 'stride_A': (64, 4096, 1), 'stride_B': (64, 1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16')}",True,0.007137718625822456,99.60592711654145 +aten::bmm,GEMM,python3,CPU,thread 10586 (python3),"((64, 512, 64), (64, 64, 512))","('c10::BFloat16', 'c10::BFloat16')","((64, 4096, 1), (64, 1, 4096))","('', '')",117,48,1724.109,35.9189375,3.929348830283864,2.147483648,40.0,51.2,matrix_bf16,2.4355642974451306,0.01757383136199424,124.7008920291907,0.89978016573411,17.221954345703125,0.12420286115001129,826.65380859375,34.811,32.08,54.66,2.44080771517063,2.4005630025431066,2.4777347136449515,124.96935501673627,122.90882573020707,126.86001733862152,17.18408203125,16.927978515625,17.47216796875,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(35.235), 'mean_duration_us': np.float64(0.7340625), 'median_duration_us': np.float64(0.736), 'std_dev_duration_us': np.float64(0.012128489618112675), 'min_duration_us': np.float64(0.703), 'max_duration_us': np.float64(0.768)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 48, 'total_duration_us': np.float64(791.415), 'mean_duration_us': np.float64(16.4878125), 'median_duration_us': np.float64(16.448), 'std_dev_duration_us': np.float64(0.12022958181641505), 'min_duration_us': np.float64(16.191), 'max_duration_us': np.float64(16.736)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.73)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(16.49)}]","{'B': 64, 'M': 512, 'N': 512, 'K': 64, 'bias': False, 'stride_A': (64, 4096, 1), 'stride_B': (64, 1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'transpose': (True, False)}",True,0.007137718625822456,99.60592711654145 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 16, 1, 32, 32), (1, 16, 1, 32, 32)), ())","('TensorList', 'Scalar')","(((32768, 2048, 1024, 32, 1), (16384, 1024, 1024, 32, 1)), ())","('', '2')",137503,124,3674.907,29.636346774193548,8.864738983904363,,,,,,,,,6.580936554939516,0.11247082224149711,816.0361328125,27.775,24.77,97.291,,,,,,,6.56103515625,6.337890625,7.072021484375,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 124, 'total_duration_us': np.float64(335.27000000000004), 'mean_duration_us': np.float64(2.7037903225806454), 'median_duration_us': np.float64(2.688), 'std_dev_duration_us': np.float64(0.030540632425151173), 'min_duration_us': np.float64(2.624), 'max_duration_us': np.float64(2.753)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 124, 'total_duration_us': np.float64(480.76599999999996), 'mean_duration_us': np.float64(3.8771451612903225), 'median_duration_us': np.float64(3.872), 'std_dev_duration_us': np.float64(0.10933469578497643), 'min_duration_us': np.float64(3.679), 'max_duration_us': np.float64(4.384)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.7)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(3.88)}]",,False,0.007046040608496564,99.61297315714995 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 384, 1, 32, 32), (1, 384, 1, 32, 32)), ())","('TensorList', 'Scalar')","(((786432, 1024, 1024, 32, 1), (786432, 1024, 1024, 32, 1)), ())","('', '3')",136146,125,5926.513,47.412104,129.10687908066237,,,,,,,,,6.5173359375,0.05036303960161556,814.6669921875,34.751,30.231,1477.888,,,,,,,6.52685546875,6.39990234375,6.65576171875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(404.3470000000001), 'mean_duration_us': np.float64(3.2347760000000005), 'median_duration_us': np.float64(3.232), 'std_dev_duration_us': np.float64(0.0388331536705428), 'min_duration_us': np.float64(3.167), 'max_duration_us': np.float64(3.328)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(410.338), 'mean_duration_us': np.float64(3.2827040000000003), 'median_duration_us': np.float64(3.296), 'std_dev_duration_us': np.float64(0.03089285328356698), 'min_duration_us': np.float64(3.2), 'max_duration_us': np.float64(3.36)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(3.23)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(3.28)}]",,False,0.007034218802997286,99.62000737595295 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((1, 192, 2, 128, 128), (1, 192, 2, 128, 128), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((12582912, 16384, 3145728, 128, 1), (12582912, 16384, 3145728, 128, 1), ())","('', '', 'False')",136674,125,1647.858,13.182864,2.203431108682474,0.006291456,24.0,0.25,vector_bf16,3.8727517812605514,0.07683976457039038,0.9681879453151379,0.019209941142597595,6.500748046875,0.1307143278303918,812.593505859375,12.66,11.79,33.3,3.893160671677305,3.6074478583327503,4.053768094384144,0.9732901679193262,0.9018619645831876,1.013442023596036,6.464111328125,6.2080078125,6.97607421875,"[{'name': 'Memcpy DtoD (Device -> Device)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(812.594), 'mean_duration_us': np.float64(6.500752), 'median_duration_us': np.float64(6.464), 'std_dev_duration_us': np.float64(0.13016805482144977), 'min_duration_us': np.float64(6.208), 'max_duration_us': np.float64(6.976)}]","[{'name': 'Memcpy DtoD (Device -> Device)', 'stream': 7, 'mean_duration_us': np.float64(6.5)}]","{'op_shape': (1, 192, 2, 128, 128), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (12582912, 16384, 3145728, 128, 1), 'stride_output': (12582912, 16384, 3145728, 128, 1)}",True,0.007016315344704603,99.62702369129765 @@ -170,7 +170,7 @@ aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((2, 384, 64, 64), ( aten::add,elementwise,python3,CPU,thread 10586 (python3),"((1, 512, 1), (), ())","('float', 'double', 'Scalar')","((512, 1, 1), (), ())","('', '', '1')",62,398,28535.096,71.69622110552764,72.24882090067554,5.12e-07,0.00391387939453125,0.12475633528265107,vector_fp32,0.0025275849950335337,0.00016920998884523925,0.00031533224109580145,2.1110018101550315e-05,1.6312439884971734,0.11396083874683917,649.235107421875,12.075,7.811,594.917,0.002421141293389025,0.0018063597678916828,0.00298260894251242,0.0003020527149647126,0.00022535482484418654,0.0003720993612491128,1.695068359375,1.3759765625,2.27197265625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctorOnSelf_add, std::array >(int, at::native::CUDAFunctorOnSelf_add, std::array)', 'stream': 7, 'count': 398, 'total_duration_us': np.float64(649.243), 'mean_duration_us': np.float64(1.6312638190954776), 'median_duration_us': np.float64(1.695), 'std_dev_duration_us': np.float64(0.11377535941290237), 'min_duration_us': np.float64(1.376), 'max_duration_us': np.float64(2.272)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(1.63)}]","{'shape_in1': (1, 512, 1), 'shape_in2': (), 'dtype_in1_in2_out': ('float', 'double', None), 'stride_input1': (512, 1, 1), 'stride_input2': (), 'stride_output': None}",True,0.005605801933781832,99.69553331818506 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((1, 192, 2, 64, 64), (1, 192, 2, 64, 64), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((1572864, 4096, 786432, 64, 1), (1572864, 4096, 786432, 64, 1), ())","('', '', 'False')",136190,250,3299.214,13.196856,3.1298698491076933,0.001572864,6.0,0.25,vector_bf16,2.4615487442628354,0.09569376083450526,0.6153871860657089,0.023923440208626315,2.559751953125,0.09976258957132868,639.93798828125,12.8,11.23,56.3,2.4575437512874307,2.234246902722386,2.731877851796883,0.6143859378218577,0.5585617256805965,0.6829694629492208,2.56005859375,2.302978515625,2.81591796875,"[{'name': 'Memcpy DtoD (Device -> Device)', 'stream': 7, 'count': 250, 'total_duration_us': np.float64(639.931), 'mean_duration_us': np.float64(2.559724), 'median_duration_us': np.float64(2.56), 'std_dev_duration_us': np.float64(0.0995714006329127), 'min_duration_us': np.float64(2.303), 'max_duration_us': np.float64(2.816)}]","[{'name': 'Memcpy DtoD (Device -> Device)', 'stream': 7, 'mean_duration_us': np.float64(2.56)}]","{'op_shape': (1, 192, 2, 64, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (1572864, 4096, 786432, 64, 1), 'stride_output': (1572864, 4096, 786432, 64, 1)}",True,0.005525526224934113,99.70105884440999 aten::rsqrt,elementwise,python3,CPU,thread 10586 (python3),"((1, 512, 1),)","('float',)","((512, 1, 1),)","('',)",63,398,3597.005,9.037701005025125,1.9780873018784815,5.12e-07,0.00390625,0.125,vector_fp32,0.0025937150266682725,0.0001762301149914232,0.00032421437833353407,2.20287643739279e-05,1.5865680943781406,0.10896669586695705,631.4541015625,8.835,6.77,19.141,0.0025096807778608827,0.002064634014275166,0.003199926759488842,0.00031371009723261034,0.0002580792517843958,0.00039999084493610526,1.632080078125,1.280029296875,1.98388671875,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array >(int, at::native::rsqrt_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#2}::operator()() const::{lambda(float)#1}, std::array)', 'stream': 7, 'count': 398, 'total_duration_us': np.float64(631.458), 'mean_duration_us': np.float64(1.5865778894472362), 'median_duration_us': np.float64(1.632), 'std_dev_duration_us': np.float64(0.10877768908088056), 'min_duration_us': np.float64(1.28), 'max_duration_us': np.float64(1.984)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::rs...', 'stream': 7, 'mean_duration_us': np.float64(1.59)}]","{'op_shape': (1, 512, 1), 'dtype_in_out': ('float', None), 'stride_input': (512, 1, 1), 'stride_output': None}",True,0.00545227234813313,99.70651111675812 -aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((64,), (32256, 1536), (1536, 64), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1536, 1), (1, 1536), (), ())","('', '', '', '1', '1')",23755,10,1128.173,112.8173,6.547578908777401,6.343852032,98.6251220703125,61.34307996786878,matrix_bf16,2.0125274971643616,0.03236526403270227,123.45463519608825,1.985384979739244,51.3980224609375,0.8239198280239874,513.980224609375,114.76150000000001,99.741,121.322,2.0173457220871027,1.9753941718943824,2.0690259065891663,123.75019995282713,121.17676265457904,126.92042164349144,51.2635498046875,49.98291015625,52.35205078125,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(513.98), 'mean_duration_us': np.float64(51.398), 'median_duration_us': np.float64(51.2635), 'std_dev_duration_us': np.float64(0.7816431410816572), 'min_duration_us': np.float64(49.983), 'max_duration_us': np.float64(52.352)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(51.4)}]","{'M': 32256, 'N': 64, 'K': 1536, 'bias': True, 'stride_A': (1536, 1), 'stride_B': (1, 1536), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.004437947523328548,99.71094906428145 +aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((64,), (32256, 1536), (1536, 64), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1536, 1), (1, 1536), (), ())","('', '', '', '1', '1')",23755,10,1128.173,112.8173,6.547578908777401,6.343852032,98.6251220703125,61.34307996786878,matrix_bf16,2.0125274971643616,0.03236526403270227,123.45463519608825,1.985384979739244,51.3980224609375,0.8239198280239874,513.980224609375,114.76150000000001,99.741,121.322,2.0173457220871027,1.9753941718943824,2.0690259065891663,123.75019995282713,121.17676265457904,126.92042164349144,51.2635498046875,49.98291015625,52.35205078125,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(513.98), 'mean_duration_us': np.float64(51.398), 'median_duration_us': np.float64(51.2635), 'std_dev_duration_us': np.float64(0.7816431410816572), 'min_duration_us': np.float64(49.983), 'max_duration_us': np.float64(52.352)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(51.4)}]","{'M': 32256, 'N': 64, 'K': 1536, 'bias': True, 'stride_A': (1536, 1), 'stride_B': (1, 1536), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.004437947523328548,99.71094906428145 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 96, 1, 256, 256), (1, 96, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((6291456, 65536, 65536, 256, 1), (25165824, 262144, 65536, 256, 1)), ())","('', '2')",137125,6,209.561,34.926833333333335,2.9566278370242434,,,,,,,,,85.1993408203125,0.13079502174487145,511.196044921875,34.565,31.92,40.38,,,,,,,85.1834716796875,85.054931640625,85.406982421875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 6, 'total_duration_us': np.float64(116.35099999999998), 'mean_duration_us': np.float64(19.39183333333333), 'median_duration_us': np.float64(19.3435), 'std_dev_duration_us': np.float64(0.28385936463130645), 'min_duration_us': np.float64(19.04), 'max_duration_us': np.float64(19.776)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 6, 'total_duration_us': np.float64(394.8449999999999), 'mean_duration_us': np.float64(65.80749999999999), 'median_duration_us': np.float64(65.9195), 'std_dev_duration_us': np.float64(0.32912852099648376), 'min_duration_us': np.float64(65.311), 'max_duration_us': np.float64(66.176)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(19.39)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(65.81)}]",,False,0.004413907603586437,99.71536297188504 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((2, 384, 32, 32), (2, 384, 32, 32), ())","('float', 'float', 'Scalar')","((393216, 1024, 32, 1), (1024, 2048, 32, 1), ())","('', '', 'False')",136170,125,1053.396,8.427168,0.7355205172024829,0.000786432,6.0,0.125,vector_fp32,1.5514938582227376,0.018800109167587304,0.1939367322778422,0.002350013645948413,4.055693359375,0.049734497798205886,506.961669921875,8.28,7.531,11.3,1.5484799769258504,1.4786437787468443,1.5855413631944872,0.1935599971157313,0.18483047234335553,0.1981926703993109,4.06298828125,3.968017578125,4.2548828125,"[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(506.9650000000001), 'mean_duration_us': np.float64(4.055720000000001), 'median_duration_us': np.float64(4.063), 'std_dev_duration_us': np.float64(0.04955572217211653), 'min_duration_us': np.float64(3.968), 'max_duration_us': np.float64(4.255)}]","[{'name': 'void at::native::elementwise_kernel<128, 2, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(4.06)}]","{'op_shape': (2, 384, 32, 32), 'dtype_in_out': ('float', 'float'), 'stride_input': (393216, 1024, 32, 1), 'stride_output': (1024, 2048, 32, 1)}",True,0.004377346013968128,99.71974031789901 aten::copy_,elementwise,python3,CPU,thread 10586 (python3),"((2, 384, 32, 32), (2, 384, 32, 32), ())","('float', 'c10::BFloat16', 'Scalar')","((1024, 2048, 32, 1), (1024, 2048, 32, 1), ())","('', '', 'False')",136163,125,1251.578,10.012624,5.4428702678499645,0.000786432,4.5,0.16666666666666666,vector_fp32,1.2071152697345138,0.0253209275256658,0.20118587828908566,0.004220154587610973,3.91069921875,0.08253535567456256,488.83740234375,9.42,8.57,69.65,1.208639411669064,1.1520148317339214,1.2496671946204576,0.201439901944844,0.1920024719556536,0.2082778657700763,3.904052734375,3.77587890625,4.095947265625,"[{'name': 'void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(488.83199999999994), 'mean_duration_us': np.float64(3.9106559999999995), 'median_duration_us': np.float64(3.904), 'std_dev_duration_us': np.float64(0.08223975719808521), 'min_duration_us': np.float64(3.776), 'max_duration_us': np.float64(4.096)}]","[{'name': 'void at::native::unrolled_elementwise_kernel, std::array >(int, at::native::CUDAFunctorOnSelf_add, std::array)', 'stream': 7, 'count': 96, 'total_duration_us': np.float64(205.347), 'mean_duration_us': np.float64(2.13903125), 'median_duration_us': np.float64(2.144), 'std_dev_duration_us': np.float64(0.03587201007430212), 'min_duration_us': np.float64(2.08), 'max_duration_us': np.float64(2.336)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(2.14)}]","{'shape_in1': (512, 512), 'shape_in2': (), 'dtype_in1_in2_out': ('long int', 'long int', None), 'stride_input1': (512, 1), 'stride_input2': (), 'stride_output': None}",True,0.001773100074847934,99.84099583698828 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 409, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((80412672, 26804224, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",368183,1,32.18,32.18,,,,,,,,,,204.031005859375,,204.031005859375,32.18,32.18,32.18,,,,,,,204.031005859375,204.031005859375,204.031005859375,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.152), 'mean_duration_us': np.float64(5.152), 'median_duration_us': np.float64(5.152), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.152), 'max_duration_us': np.float64(5.152)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(198.879), 'mean_duration_us': np.float64(198.879), 'median_duration_us': np.float64(198.879), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(198.879), 'max_duration_us': np.float64(198.879)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.15)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(198.88)}]",,False,0.0017616998743949924,99.84275753686268 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 405, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((79626240, 26542080, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",365921,1,32.06,32.06,,,,,,,,,,202.110107421875,,202.110107421875,32.06,32.06,32.06,,,,,,,202.110107421875,202.110107421875,202.110107421875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.216), 'mean_duration_us': np.float64(5.216), 'median_duration_us': np.float64(5.216), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.216), 'max_duration_us': np.float64(5.216)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(196.894), 'mean_duration_us': np.float64(196.894), 'median_duration_us': np.float64(196.894), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(196.894), 'max_duration_us': np.float64(196.894)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.22)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(196.89)}]",,False,0.001745113931872111,99.84450265079455 -aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (512, 4096), (4096, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (4096, 1), (1, 4096), (), ())","('', '', '', '1', '1')",11845,10,815.439,81.5439,7.319777143693196,6.443237376,17.5029296875,351.06979858282654,matrix_bf16,0.9087376362124789,0.015042430661435026,319.0303389097488,5.280943102506133,20.2012451171875,0.330968715991367,202.012451171875,82.58600000000001,62.13,90.651,0.9038934008585031,0.8851245198101989,0.940269050556598,317.3296741797408,310.74048689048755,330.1000661925704,20.3045654296875,19.51904296875,20.735107421875,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(202.012), 'mean_duration_us': np.float64(20.2012), 'median_duration_us': np.float64(20.3045), 'std_dev_duration_us': np.float64(0.31397190957154153), 'min_duration_us': np.float64(19.519), 'max_duration_us': np.float64(20.735)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpg...', 'stream': 7, 'mean_duration_us': np.float64(20.2)}]","{'M': 512, 'N': 1536, 'K': 4096, 'bias': True, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.0017442707217794377,99.84624692151633 +aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (512, 4096), (4096, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (4096, 1), (1, 4096), (), ())","('', '', '', '1', '1')",11845,10,815.439,81.5439,7.319777143693196,6.443237376,17.5029296875,351.06979858282654,matrix_bf16,0.9087376362124789,0.015042430661435026,319.0303389097488,5.280943102506133,20.2012451171875,0.330968715991367,202.012451171875,82.58600000000001,62.13,90.651,0.9038934008585031,0.8851245198101989,0.940269050556598,317.3296741797408,310.74048689048755,330.1000661925704,20.3045654296875,19.51904296875,20.735107421875,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(202.012), 'mean_duration_us': np.float64(20.2012), 'median_duration_us': np.float64(20.3045), 'std_dev_duration_us': np.float64(0.31397190957154153), 'min_duration_us': np.float64(19.519), 'max_duration_us': np.float64(20.735)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x64x64_warpg...', 'stream': 7, 'mean_duration_us': np.float64(20.2)}]","{'M': 512, 'N': 1536, 'K': 4096, 'bias': True, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.0017442707217794377,99.84624692151633 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 401, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((78839808, 26279936, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",363659,1,31.57,31.57,,,,,,,,,,200.5419921875,,200.5419921875,31.57,31.57,31.57,,,,,,,200.5419921875,200.5419921875,200.5419921875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.376), 'mean_duration_us': np.float64(5.376), 'median_duration_us': np.float64(5.376), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.376), 'max_duration_us': np.float64(5.376)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(195.166), 'mean_duration_us': np.float64(195.166), 'median_duration_us': np.float64(195.166), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(195.166), 'max_duration_us': np.float64(195.166)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.38)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(195.17)}]",,False,0.0017315740858090112,99.84797849560213 aten::sub,elementwise,python3,CPU,thread 10586 (python3),"((1, 16, 126, 32, 32), (1, 16, 126, 32, 32), ())","('float', 'float', 'Scalar')","((2064384, 129024, 1024, 32, 1), (2064384, 129024, 1024, 32, 1), ())","('', '', '1')",35903,26,3047.456,117.20984615384616,300.2458250024726,,,,,,,,,7.646597055288462,1.8691362373908282,198.8115234375,9.4205,6.891,961.381,,,,,,,6.719970703125,5.43896484375,10.847900390625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 26, 'total_duration_us': np.float64(198.81199999999998), 'mean_duration_us': np.float64(7.646615384615384), 'median_duration_us': np.float64(6.72), 'std_dev_duration_us': np.float64(1.83283238305599), 'min_duration_us': np.float64(5.439), 'max_duration_us': np.float64(10.848)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(7.65)}]",,False,0.0017166324029668425,99.8496951280051 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 397, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((78053376, 26017792, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",361397,1,30.48,30.48,,,,,,,,,,198.43017578125,,198.43017578125,30.48,30.48,30.48,,,,,,,198.43017578125,198.43017578125,198.43017578125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.184), 'mean_duration_us': np.float64(5.184), 'median_duration_us': np.float64(5.184), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.184), 'max_duration_us': np.float64(5.184)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(193.246), 'mean_duration_us': np.float64(193.246), 'median_duration_us': np.float64(193.246), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(193.246), 'max_duration_us': np.float64(193.246)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.18)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(193.25)}]",,False,0.0017133396675549537,99.85140846767266 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 393, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((77266944, 25755648, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",359135,1,30.79,30.79,,,,,,,,,,196.670166015625,,196.670166015625,30.79,30.79,30.79,,,,,,,196.670166015625,196.670166015625,196.670166015625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.184), 'mean_duration_us': np.float64(5.184), 'median_duration_us': np.float64(5.184), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.184), 'max_duration_us': np.float64(5.184)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(191.486), 'mean_duration_us': np.float64(191.486), 'median_duration_us': np.float64(191.486), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(191.486), 'max_duration_us': np.float64(191.486)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.18)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(191.49)}]",,False,0.0016981429136597513,99.85310661058631 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 389, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((76480512, 25493504, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",356873,1,33.721,33.721,,,,,,,,,,194.973876953125,,194.973876953125,33.721,33.721,33.721,,,,,,,194.973876953125,194.973876953125,194.973876953125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.472), 'mean_duration_us': np.float64(5.472), 'median_duration_us': np.float64(5.472), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.472), 'max_duration_us': np.float64(5.472)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(189.502), 'mean_duration_us': np.float64(189.502), 'median_duration_us': np.float64(189.502), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(189.502), 'max_duration_us': np.float64(189.502)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.47)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(189.5)}]",,False,0.0016834963543500183,99.85479010694065 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 385, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((75694080, 25231360, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",354611,1,30.91,30.91,,,,,,,,,,192.9580078125,,192.9580078125,30.91,30.91,30.91,,,,,,,192.9580078125,192.9580078125,192.9580078125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.504), 'mean_duration_us': np.float64(5.504), 'median_duration_us': np.float64(5.504), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.504), 'max_duration_us': np.float64(5.504)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(187.454), 'mean_duration_us': np.float64(187.454), 'median_duration_us': np.float64(187.454), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(187.454), 'max_duration_us': np.float64(187.454)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.5)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(187.45)}]",,False,0.0016660903900120121,99.85645619733067 -aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((9216,), (1, 1536), (1536, 9216), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1536, 1), (1, 1536), (), ())","('', '', '', '1', '1')",11837,10,478.645,47.8645,8.267741479186844,0.028320768,27.0380859375,0.9989164589879727,matrix_bf16,1.4872589982806343,0.018189369309396873,1.4856474921604907,0.01816966038176725,19.065478515625,0.23314521563266258,190.65478515625,45.8305,35.19,64.891,1.4902983977761906,1.4596057722753608,1.5145049343088541,1.4886835983420417,1.4580242295597088,1.5128639060996127,19.0240478515625,18.719970703125,19.424072265625,"[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x4_tn_align8::Params)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(190.655), 'mean_duration_us': np.float64(19.0655), 'median_duration_us': np.float64(19.024), 'std_dev_duration_us': np.float64(0.22117380043757456), 'min_duration_us': np.float64(18.72), 'max_duration_us': np.float64(19.424)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x4_tn_align8::Params)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(190.655), 'mean_duration_us': np.float64(19.0655), 'median_duration_us': np.float64(19.024), 'std_dev_duration_us': np.float64(0.22117380043757456), 'min_duration_us': np.float64(18.72), 'max_duration_us': np.float64(19.424)}]","[{'name': 'void cutlass::Kernel2(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.471), 'mean_duration_us': np.float64(5.471), 'median_duration_us': np.float64(5.471), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.471), 'max_duration_us': np.float64(5.471)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(185.182), 'mean_duration_us': np.float64(185.182), 'median_duration_us': np.float64(185.182), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(185.182), 'max_duration_us': np.float64(185.182)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.47)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(185.18)}]",,False,0.0016461864157744617,99.85974858702642 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 377, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((74121216, 24707072, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",350087,1,31.541,31.541,,,,,,,,,,188.927001953125,,188.927001953125,31.541,31.541,31.541,,,,,,,188.927001953125,188.927001953125,188.927001953125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.12), 'mean_duration_us': np.float64(5.12), 'median_duration_us': np.float64(5.12), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.12), 'max_duration_us': np.float64(5.12)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(183.807), 'mean_duration_us': np.float64(183.807), 'median_duration_us': np.float64(183.807), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(183.807), 'max_duration_us': np.float64(183.807)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.12)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(183.81)}]",,False,0.0016312847854116947,99.86137987181183 aten::clamp_min,elementwise,python3,CPU,thread 10586 (python3),"((1, 1, 32, 32), ())","('c10::BFloat16', 'Scalar')","((1024, 1024, 32, 1), ())","('', '9.9999999999999998e-13')",133595,126,1213.961,9.634611111111111,1.4658080964285733,1.024e-06,0.00390625,0.25,vector_bf16,0.0027503370285301356,0.00021421160684349512,0.0006875842571325339,5.355290171087378e-05,1.4983084542410714,0.11733577524842156,188.786865234375,9.295,8.75,20.43,0.0026685567043104818,0.0023272598141212374,0.0030498483912015997,0.0006671391760776204,0.0005818149535303093,0.0007624620978003999,1.534912109375,1.343017578125,1.760009765625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::launch_clamp_scalar(at::TensorIteratorBase&, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::{lambda()#1}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::(anonymous namespace)::launch_clamp_scalar(at::TensorIteratorBase&, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::{lambda()#1}::operator()() const::{lambda()#9}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 126, 'total_duration_us': np.float64(188.79000000000002), 'mean_duration_us': np.float64(1.4983333333333335), 'median_duration_us': np.float64(1.535), 'std_dev_duration_us': np.float64(0.11685658012050631), 'min_duration_us': np.float64(1.343), 'max_duration_us': np.float64(1.76)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(a...', 'stream': 7, 'mean_duration_us': np.float64(1.5)}]","{'op_shape': (1, 1, 32, 32), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (1024, 1024, 32, 1), 'stride_output': None}",True,0.0016300747789287089,99.86300994659076 @@ -334,7 +334,7 @@ aten::arange,other,python3,CPU,thread 10586 (python3),"((), (), (), (0,))","('Sc aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 129, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((25362432, 8454144, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",209843,1,39.59,39.59,,,,,,,,,,68.864013671875,,68.864013671875,39.59,39.59,39.59,,,,,,,68.864013671875,68.864013671875,68.864013671875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.216), 'mean_duration_us': np.float64(5.216), 'median_duration_us': np.float64(5.216), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.216), 'max_duration_us': np.float64(5.216)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(63.648), 'mean_duration_us': np.float64(63.648), 'median_duration_us': np.float64(63.648), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(63.648), 'max_duration_us': np.float64(63.648)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.22)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(63.65)}]",,False,0.0005946043530251155,99.96871443074026 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 125, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((24576000, 8192000, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",207581,1,40.25,40.25,,,,,,,,,,66.719970703125,,66.719970703125,40.25,40.25,40.25,,,,,,,66.719970703125,66.719970703125,66.719970703125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.024), 'mean_duration_us': np.float64(5.024), 'median_duration_us': np.float64(5.024), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.024), 'max_duration_us': np.float64(5.024)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(61.696), 'mean_duration_us': np.float64(61.696), 'median_duration_us': np.float64(61.696), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(61.696), 'max_duration_us': np.float64(61.696)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.02)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(61.7)}]",,False,0.0005760916754404758,99.9692905224157 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 121, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((23789568, 7929856, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",205319,1,35.57,35.57,,,,,,,,,,64.63916015625,,64.63916015625,35.57,35.57,35.57,,,,,,,64.63916015625,64.63916015625,64.63916015625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.248), 'mean_duration_us': np.float64(5.248), 'median_duration_us': np.float64(5.248), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.248), 'max_duration_us': np.float64(5.248)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(59.391), 'mean_duration_us': np.float64(59.391), 'median_duration_us': np.float64(59.391), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(59.391), 'max_duration_us': np.float64(59.391)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.25)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(59.39)}]",,False,0.000558124976390842,99.9698486473921 -aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (1, 1536), (1536, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1536, 1), (1, 1536), (), ())","('', '', '', '1', '1')",11828,10,275.092,27.5092,2.1801160927294165,0.004720128,4.5087890625,0.9983755685510072,matrix_bf16,0.7337507692909802,0.01088589454230261,0.7325588414656212,0.010868211152857691,6.4446044921875,0.0944805958845497,64.446044921875,27.6605,24.32,31.73,0.7314624163130894,0.7206960017863788,0.7576627242067374,0.7302742057602741,0.7195252805359136,0.7564319530498064,6.4635009765625,6.239990234375,6.56005859375,"[{'name': 'std::enable_if::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(64.446), 'mean_duration_us': np.float64(6.444599999999999), 'median_duration_us': np.float64(6.4635), 'std_dev_duration_us': np.float64(0.08959374978200195), 'min_duration_us': np.float64(6.24), 'max_duration_us': np.float64(6.56)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(64.446), 'mean_duration_us': np.float64(6.444599999999999), 'median_duration_us': np.float64(6.4635), 'std_dev_duration_us': np.float64(0.08959374978200195), 'min_duration_us': np.float64(6.24), 'max_duration_us': np.float64(6.56)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.215), 'mean_duration_us': np.float64(5.215), 'median_duration_us': np.float64(5.215), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.215), 'max_duration_us': np.float64(5.215)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(57.28), 'mean_duration_us': np.float64(57.28), 'median_duration_us': np.float64(57.28), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(57.28), 'max_duration_us': np.float64(57.28)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.22)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(57.28)}]",,False,0.0005396122988062024,99.97094471721934 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 113, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((22216704, 7405568, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",200795,1,38.96,38.96,,,,,,,,,,60.447998046875,,60.447998046875,38.96,38.96,38.96,,,,,,,60.447998046875,60.447998046875,60.447998046875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.216), 'mean_duration_us': np.float64(5.216), 'median_duration_us': np.float64(5.216), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.216), 'max_duration_us': np.float64(5.216)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(55.232), 'mean_duration_us': np.float64(55.232), 'median_duration_us': np.float64(55.232), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(55.232), 'max_duration_us': np.float64(55.232)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.22)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(55.23)}]",,False,0.0005219365072385407,99.97146665372658 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 109, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((21430272, 7143424, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",198533,1,34.63,34.63,,,,,,,,,,58.623046875,,58.623046875,34.63,34.63,34.63,,,,,,,58.623046875,58.623046875,58.623046875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.408), 'mean_duration_us': np.float64(5.408), 'median_duration_us': np.float64(5.408), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.408), 'max_duration_us': np.float64(5.408)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(53.215), 'mean_duration_us': np.float64(53.215), 'median_duration_us': np.float64(53.215), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(53.215), 'max_duration_us': np.float64(53.215)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.41)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(53.22)}]",,False,0.0005061790186317107,99.97197283274521 @@ -366,7 +366,7 @@ aten::fill_,elementwise,python3,CPU,thread 10586 (python3),"((1, 192, 3, 130, 13 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 73, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((14352384, 4784128, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",178175,1,95.131,95.131,,,,,,,,,,41.886962890625,,41.886962890625,95.131,95.131,95.131,,,,,,,41.886962890625,41.886962890625,41.886962890625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.312), 'mean_duration_us': np.float64(5.312), 'median_duration_us': np.float64(5.312), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.312), 'max_duration_us': np.float64(5.312)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(36.575), 'mean_duration_us': np.float64(36.575), 'median_duration_us': np.float64(36.575), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(36.575), 'max_duration_us': np.float64(36.575)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.31)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(36.58)}]",,False,0.0003616717809746126,99.98300773772182 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 69, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((13565952, 4521984, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",175913,1,32.611,32.611,,,,,,,,,,39.2001953125,,39.2001953125,32.611,32.611,32.611,,,,,,,39.2001953125,39.2001953125,39.2001953125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.152), 'mean_duration_us': np.float64(5.152), 'median_duration_us': np.float64(5.152), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.152), 'max_duration_us': np.float64(5.152)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(34.048), 'mean_duration_us': np.float64(34.048), 'median_duration_us': np.float64(34.048), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(34.048), 'max_duration_us': np.float64(34.048)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.15)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(34.05)}]",,False,0.0003384729632999417,99.98334621068513 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 65, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((12779520, 4259840, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",173651,1,32.74,32.74,,,,,,,,,,37.72802734375,,37.72802734375,32.74,32.74,32.74,,,,,,,37.72802734375,37.72802734375,37.72802734375,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.184), 'mean_duration_us': np.float64(5.184), 'median_duration_us': np.float64(5.184), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.184), 'max_duration_us': np.float64(5.184)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(32.544), 'mean_duration_us': np.float64(32.544), 'median_duration_us': np.float64(32.544), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(32.544), 'max_duration_us': np.float64(32.544)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.18)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(32.54)}]",,False,0.00032576157115289347,99.98367197225627 -aten::addmm,GEMM,python3,CPU,thread 10586 (python3),"((1536,), (1, 256), (256, 1536), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (256, 1), (1, 256), (), ())","('', '', '', '1', '1')",11821,10,817.02,81.702,12.331542617747921,0.000787968,0.75634765625,0.9935442220787605,matrix_bf16,0.21448386595136726,0.00806436425374349,0.2130992057450963,0.008012302509045338,3.702392578125,0.13998001440387378,37.02392578125,85.49600000000001,51.941,93.481,0.21458858291056837,0.2031447969482834,0.22537036547800748,0.21320324667486423,0.20183333925333002,0.2239154244484529,3.6959228515625,3.51904296875,3.904052734375,"[{'name': 'std::enable_if::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(37.024), 'mean_duration_us': np.float64(3.7024), 'median_duration_us': np.float64(3.696), 'std_dev_duration_us': np.float64(0.13277665457451465), 'min_duration_us': np.float64(3.519), 'max_duration_us': np.float64(3.904)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(37.024), 'mean_duration_us': np.float64(3.7024), 'median_duration_us': np.float64(3.696), 'std_dev_duration_us': np.float64(0.13277665457451465), 'min_duration_us': np.float64(3.519), 'max_duration_us': np.float64(3.904)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.152), 'mean_duration_us': np.float64(5.152), 'median_duration_us': np.float64(5.152), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.152), 'max_duration_us': np.float64(5.152)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(30.464), 'mean_duration_us': np.float64(30.464), 'median_duration_us': np.float64(30.464), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(30.464), 'max_duration_us': np.float64(30.464)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.15)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(30.46)}]",,False,0.00030752715289883594,99.98429918143556 aten::silu,elementwise,python3,CPU,thread 10586 (python3),"((1, 1536),)","('c10::BFloat16',)","((1536, 1),)","('',)",11823,20,288.493,14.42465,5.051917877348531,1.536e-06,0.005859375,0.25,vector_bf16,0.003644687309592985,0.00021436644321489302,0.0009111718273982463,5.3591610803723256e-05,1.69124755859375,0.0985900969293492,33.824951171875,14.059999999999999,9.04,22.301,0.0035915293676331817,0.003368467942711819,0.0039180794021485285,0.0008978823419082954,0.0008421169856779548,0.0009795198505371321,1.7120361328125,1.568115234375,1.823974609375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::(anonymous namespace)::silu_kernel(at::TensorIteratorBase&)::{lambda()#1}::operator()() const::{lambda()#6}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 20, 'total_duration_us': np.float64(33.824000000000005), 'mean_duration_us': np.float64(1.6912000000000003), 'median_duration_us': np.float64(1.712), 'std_dev_duration_us': np.float64(0.09614655480047114), 'min_duration_us': np.float64(1.568), 'max_duration_us': np.float64(1.824)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::(a...', 'stream': 7, 'mean_duration_us': np.float64(1.69)}]","{'op_shape': (1, 1536), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (1536, 1), 'stride_output': None}",True,0.0002920605717739781,99.98459124200734 aten::cat,elementwise,python3,CPU,thread 10586 (python3),"(((1, 3, 57, 256, 256), (1, 3, 4, 256, 256)), ())","('TensorList', 'Scalar')","(((11206656, 3735552, 65536, 256, 1), (786432, 262144, 65536, 256, 1)), ())","('', '2')",169127,1,38.75,38.75,,,,,,,,,,33.5361328125,,33.5361328125,38.75,38.75,38.75,,,,,,,33.5361328125,33.5361328125,33.5361328125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.216), 'mean_duration_us': np.float64(5.216), 'median_duration_us': np.float64(5.216), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.216), 'max_duration_us': np.float64(5.216)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(28.32), 'mean_duration_us': np.float64(28.32), 'median_duration_us': np.float64(28.32), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(28.32), 'max_duration_us': np.float64(28.32)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(5.22)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(28.32)}]",,False,0.00028956677792489713,99.98488080878526 diff --git a/tests/traces/h100/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv b/tests/traces/h100/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv index 8561eeec9..a60423b76 100644 --- a/tests/traces/h100/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv +++ b/tests/traces/h100/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv @@ -1,12 +1,12 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::bmm,3137,3137,64,False,"(200768, 64, 1)","(200768, 3137, 1)","('c10::BFloat16', 'c10::BFloat16')",96,120.923369472,1875.4220581054688,61.49096477794793,1.034168107132991,1.0365710460416007,0.006868565124284747,1.0245090238002632,1.0407138419080384,63.59199465019184,63.739753681984716,0.4223546961324301,62.99804829719181,63.99449819669006,python3,CPU,thread 17751 (python3),matrix_bf16,"[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(22819.528), 'mean_duration_us': np.float64(1901.6273333333331), 'median_duration_us': np.float64(1897.1419999999998), 'std_dev_duration_us': np.float64(12.13592587687027), 'min_duration_us': np.float64(1889.59), 'max_duration_us': np.float64(1919.478)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(20388.114), 'mean_duration_us': np.float64(1699.0095000000001), 'median_duration_us': np.float64(1693.687), 'std_dev_duration_us': np.float64(11.565890623582206), 'min_duration_us': np.float64(1682.295), 'max_duration_us': np.float64(1718.295)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_32x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1300.504), 'mean_duration_us': np.float64(108.37533333333333), 'median_duration_us': np.float64(107.8715), 'std_dev_duration_us': np.float64(0.8493324175819239), 'min_duration_us': np.float64(107.583), 'max_duration_us': np.float64(109.599)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_64x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(490.207), 'mean_duration_us': np.float64(40.85058333333333), 'median_duration_us': np.float64(40.7355), 'std_dev_duration_us': np.float64(0.4567669825949428), 'min_duration_us': np.float64(40.288), 'max_duration_us': np.float64(41.632)}]","[{'name': 'void cutlass::Kernel2::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.056), 'mean_duration_us': np.float64(5.056), 'median_duration_us': np.float64(5.056), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.056), 'max_duration_us': np.float64(5.056)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(22819.528), 'mean_duration_us': np.float64(1901.6273333333331), 'median_duration_us': np.float64(1897.1419999999998), 'std_dev_duration_us': np.float64(12.13592587687027), 'min_duration_us': np.float64(1889.59), 'max_duration_us': np.float64(1919.478)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(20388.114), 'mean_duration_us': np.float64(1699.0095000000001), 'median_duration_us': np.float64(1693.687), 'std_dev_duration_us': np.float64(11.565890623582206), 'min_duration_us': np.float64(1682.295), 'max_duration_us': np.float64(1718.295)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_32x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1300.504), 'mean_duration_us': np.float64(108.37533333333333), 'median_duration_us': np.float64(107.8715), 'std_dev_duration_us': np.float64(0.8493324175819239), 'min_duration_us': np.float64(107.583), 'max_duration_us': np.float64(109.599)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_64x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(490.207), 'mean_duration_us': np.float64(40.85058333333333), 'median_duration_us': np.float64(40.7355), 'std_dev_duration_us': np.float64(0.4567669825949428), 'min_duration_us': np.float64(40.288), 'max_duration_us': np.float64(41.632)}]","[{'name': 'void cutlass::Kernel2::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.056), 'mean_duration_us': np.float64(5.056), 'median_duration_us': np.float64(5.056), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.056), 'max_duration_us': np.float64(5.056)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(c10::BFloat16*, c10::BFloat16 const*, int)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(68688.625), 'mean_duration_us': np.float64(5724.052083333333), 'median_duration_us': np.float64(5706.610000000001), 'std_dev_duration_us': np.float64(26.445577605900787), 'min_duration_us': np.float64(5701.537), 'max_duration_us': np.float64(5767.809)}]","[{'name': 'void at::native::(anonymous namespace)::cunn_SoftMaxForward<8, c...', 'stream': 7, 'mean_duration_us': np.float64(5724.05)}]",,False,43.35495142967854,43.35495142967854 -aten::bmm,GEMM,python3,CPU,thread 17751 (python3),"((96, 3137, 64), (96, 64, 3137))","('c10::BFloat16', 'c10::BFloat16')","((200768, 64, 1), (200768, 3137, 1))","('', '')",258,12,292.997,24.416416666666667,4.301036375283885,120.923369472,1875.4220581054688,61.49096477794793,matrix_bf16,1.034168107132991,0.006868565124284747,63.59199465019184,0.4223546961324301,1901.6273600260417,12.675563152100723,22819.5283203125,22.1555,21.361,33.891,1.0365710460416007,1.0245090238002632,1.0407138419080384,63.739753681984716,62.99804829719181,63.99449819669006,1897.14208984375,1889.590087890625,1919.47802734375,"[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(22819.528), 'mean_duration_us': np.float64(1901.6273333333331), 'median_duration_us': np.float64(1897.1419999999998), 'std_dev_duration_us': np.float64(12.13592587687027), 'min_duration_us': np.float64(1889.59), 'max_duration_us': np.float64(1919.478)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(20388.114), 'mean_duration_us': np.float64(1699.0095000000001), 'median_duration_us': np.float64(1693.687), 'std_dev_duration_us': np.float64(11.565890623582206), 'min_duration_us': np.float64(1682.295), 'max_duration_us': np.float64(1718.295)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(22819.528), 'mean_duration_us': np.float64(1901.6273333333331), 'median_duration_us': np.float64(1897.1419999999998), 'std_dev_duration_us': np.float64(12.13592587687027), 'min_duration_us': np.float64(1889.59), 'max_duration_us': np.float64(1919.478)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(20388.114), 'mean_duration_us': np.float64(1699.0095000000001), 'median_duration_us': np.float64(1693.687), 'std_dev_duration_us': np.float64(11.565890623582206), 'min_duration_us': np.float64(1682.295), 'max_duration_us': np.float64(1718.295)}]","[{'name': 'void cutlass::Kernel2 >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(15495.243999999999), 'mean_duration_us': np.float64(1291.2703333333332), 'median_duration_us': np.float64(1291.129), 'std_dev_duration_us': np.float64(0.8817396945181291), 'min_duration_us': np.float64(1289.657), 'max_duration_us': np.float64(1292.505)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(1291.27)}]","{'shape_in1': (8, 12, 3137, 3137), 'shape_in2': (), 'dtype_in1_in2_out': ('c10::BFloat16', 'double', None), 'stride_input1': (118089228, 9840769, 3137, 1), 'stride_input2': (), 'stride_output': None}",True,9.780302539912961,80.40709451266034 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((1, 128, 3, 224, 224), (1, 128, 3, 224, 224), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((19267584, 150528, 50176, 224, 1), (19267584, 150528, 50176, 224, 1), ())","('', '', 'False')",9,1,3650.483,3650.483,,0.019267584,73.5,0.25,vector_bf16,0.022265925108137966,,0.0055664812770344916,,3461.35791015625,,3461.35791015625,3650.483,3650.483,3650.483,0.022265925108137966,0.022265925108137966,0.022265925108137966,0.0055664812770344916,0.0055664812770344916,0.0055664812770344916,3461.35791015625,3461.35791015625,3461.35791015625,"[{'name': 'Memcpy HtoD (Pageable -> Device)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3461.358), 'mean_duration_us': np.float64(3461.358), 'median_duration_us': np.float64(3461.358), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3461.358), 'max_duration_us': np.float64(3461.358)}]","[{'name': 'Memcpy HtoD (Pageable -> Device)', 'stream': 7, 'mean_duration_us': np.float64(3461.36)}]","{'op_shape': (1, 128, 3, 224, 224), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (19267584, 150528, 50176, 224, 1), 'stride_output': (19267584, 150528, 50176, 224, 1)}",True,2.1847431585968318,82.59183767125717 -aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((3072,), (25089, 768), (768, 3072), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",338,12,743.901,61.991749999999996,25.512201070806743,118.461828096,188.26318359375,600.0854801938983,matrix_bf16,1.1977992273146254,0.00558864229869653,718.7819244989773,3.353663097445264,164.81243896484375,0.7722403076739952,1977.749267578125,51.841,50.471,139.492,1.197756432545682,1.1843049675225192,1.205370482867186,718.7562439795063,710.6842151317701,723.3253250229063,164.8150634765625,163.77392578125,166.68701171875,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(10.270000000000001), 'mean_duration_us': np.float64(0.8558333333333334), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.10186333436958013), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.087)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1967.4789999999998), 'mean_duration_us': np.float64(163.95658333333333), 'median_duration_us': np.float64(164.04700000000003), 'std_dev_duration_us': np.float64(0.7758030740608224), 'min_duration_us': np.float64(162.815), 'max_duration_us': np.float64(165.887)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.86)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(163.96)}]","{'M': 25089, 'N': 3072, 'K': 768, 'bias': True, 'stride_A': (768, 1), 'stride_B': (1, 768), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,1.2483176527578892,83.84015532401506 -aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((768,), (25089, 3072), (3072, 768), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (3072, 1), (1, 3072), (), ())","('', '', '', '1', '1')",349,12,678.273,56.52275,12.923427719561372,118.40402304,188.2587890625,599.8066613237056,matrix_bf16,1.2440432234830532,0.01934222001063396,746.1854124197507,11.601592407166898,158.7135213216146,2.416313978777062,1904.562255859375,50.001,47.841,93.352,1.2421041593862796,1.2205975873943669,1.2951800166077196,745.0223488577723,732.1225637147852,776.8576015746579,158.927001953125,152.4140625,161.72705078125,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(13.661999999999999), 'mean_duration_us': np.float64(1.1384999999999998), 'median_duration_us': np.float64(1.168), 'std_dev_duration_us': np.float64(0.18668846598901961), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.376)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1890.9), 'mean_duration_us': np.float64(157.57500000000002), 'median_duration_us': np.float64(158.031), 'std_dev_duration_us': np.float64(2.2915636582909897), 'min_duration_us': np.float64(151.199), 'max_duration_us': np.float64(160.415)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(1.14)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(157.58)}]","{'M': 25089, 'N': 768, 'K': 3072, 'bias': True, 'stride_A': (3072, 1), 'stride_B': (1, 3072), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,1.2021234054997467,85.0422787295148 +aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((3072,), (25089, 768), (768, 3072), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",338,12,743.901,61.991749999999996,25.512201070806743,118.461828096,188.26318359375,600.0854801938983,matrix_bf16,1.1977992273146254,0.00558864229869653,718.7819244989773,3.353663097445264,164.81243896484375,0.7722403076739952,1977.749267578125,51.841,50.471,139.492,1.197756432545682,1.1843049675225192,1.205370482867186,718.7562439795063,710.6842151317701,723.3253250229063,164.8150634765625,163.77392578125,166.68701171875,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(10.270000000000001), 'mean_duration_us': np.float64(0.8558333333333334), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.10186333436958013), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.087)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1967.4789999999998), 'mean_duration_us': np.float64(163.95658333333333), 'median_duration_us': np.float64(164.04700000000003), 'std_dev_duration_us': np.float64(0.7758030740608224), 'min_duration_us': np.float64(162.815), 'max_duration_us': np.float64(165.887)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.86)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(163.96)}]","{'M': 25089, 'N': 3072, 'K': 768, 'bias': True, 'stride_A': (768, 1), 'stride_B': (1, 768), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,1.2483176527578892,83.84015532401506 +aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((768,), (25089, 3072), (3072, 768), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (3072, 1), (1, 3072), (), ())","('', '', '', '1', '1')",349,12,678.273,56.52275,12.923427719561372,118.40402304,188.2587890625,599.8066613237056,matrix_bf16,1.2440432234830532,0.01934222001063396,746.1854124197507,11.601592407166898,158.7135213216146,2.416313978777062,1904.562255859375,50.001,47.841,93.352,1.2421041593862796,1.2205975873943669,1.2951800166077196,745.0223488577723,732.1225637147852,776.8576015746579,158.927001953125,152.4140625,161.72705078125,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(13.661999999999999), 'mean_duration_us': np.float64(1.1384999999999998), 'median_duration_us': np.float64(1.168), 'std_dev_duration_us': np.float64(0.18668846598901961), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.376)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1890.9), 'mean_duration_us': np.float64(157.57500000000002), 'median_duration_us': np.float64(158.031), 'std_dev_duration_us': np.float64(2.2915636582909897), 'min_duration_us': np.float64(151.199), 'max_duration_us': np.float64(160.415)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(1.14)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(157.58)}]","{'M': 25089, 'N': 768, 'K': 3072, 'bias': True, 'stride_A': (3072, 1), 'stride_B': (1, 3072), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,1.2021234054997467,85.0422787295148 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((8, 12, 64, 3137), (8, 12, 64, 3137), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((2409216, 200768, 3137, 1), (7227648, 64, 1, 2304), ())","('', '', 'False')",256,12,89.85000000000001,7.487500000000001,1.5630513572671072,0.019273728,73.5234375,0.25,vector_bf16,0.5011573421889868,0.0005911931389301384,0.1252893355472467,0.0001477982847325346,153.83394368489584,0.18149232699190526,1846.00732421875,6.835,6.64,11.98,0.5012429309440907,0.5003593038782586,0.501919680888061,0.12531073273602267,0.12508982596956464,0.12547992022201526,153.8074951171875,153.60009765625,154.0791015625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1846.007), 'mean_duration_us': np.float64(153.83391666666668), 'median_duration_us': np.float64(153.8075), 'std_dev_duration_us': np.float64(0.17379991289475016), 'min_duration_us': np.float64(153.6), 'max_duration_us': np.float64(154.079)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(153.83)}]","{'op_shape': (8, 12, 64, 3137), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (2409216, 200768, 3137, 1), 'stride_output': (7227648, 64, 1, 2304)}",True,1.1651646483805833,86.20744337789539 aten::gelu,elementwise,python3,CPU,thread 17751 (python3),"((1, 25089, 3072), ())","('c10::BFloat16', '')","((77073408, 3072, 1), ())","('', '')",341,12,189.321,15.77675,3.88976879717221,,,,,,,,,127.84999593098958,0.5000303915750315,1534.199951171875,13.945,13.67,24.36,,,,,,,127.79150390625,127.10302734375,128.702880859375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::GeluCUDAKernelImpl(at::TensorIteratorBase&, at::native::GeluType)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::GeluCUDAKernelImpl(at::TensorIteratorBase&, at::native::GeluType)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1534.2), 'mean_duration_us': np.float64(127.85000000000001), 'median_duration_us': np.float64(127.7915), 'std_dev_duration_us': np.float64(0.47875602694204833), 'min_duration_us': np.float64(127.103), 'max_duration_us': np.float64(128.703)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Ge...', 'stream': 7, 'mean_duration_us': np.float64(127.85)}]",,False,0.9683577758334274,87.17580115372881 -aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((2304,), (25096, 768), (768, 2304), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",226,12,646.32,53.86000000000001,4.4953152382128945,88.871159808,150.42626953125,563.4264735955439,matrix_bf16,1.2520954903075088,0.01029725378044037,705.4637467088432,5.801745385231927,125.98337809244792,1.042859664778292,1511.800537109375,51.576,49.961,63.221,1.256009414663169,1.2344605374354898,1.2635748083268794,707.6689553064726,695.527747400138,711.931498379779,125.5830078125,124.8310546875,127.775146484375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(10.656000000000002), 'mean_duration_us': np.float64(0.8880000000000002), 'median_duration_us': np.float64(0.8640000000000001), 'std_dev_duration_us': np.float64(0.10939835464941874), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.12)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1501.144), 'mean_duration_us': np.float64(125.09533333333333), 'median_duration_us': np.float64(124.703), 'std_dev_duration_us': np.float64(1.0343101834341377), 'min_duration_us': np.float64(123.967), 'max_duration_us': np.float64(127.007)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.89)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(125.1)}]","{'M': 25096, 'N': 2304, 'K': 768, 'bias': True, 'stride_A': (768, 1), 'stride_B': (1, 768), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.9542196924858387,88.13002084621465 -aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((2304,), (25088, 768), (768, 2304), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",90,12,769.69,64.14083333333333,30.10791482418877,88.842829824,150.37939453125,563.422437389805,matrix_bf16,1.2798326110548646,0.009074582161354192,721.0864091714902,5.11282319964417,123.21262613932292,0.8791740453911183,1478.551513671875,51.1755,49.451,151.542,1.282579072306334,1.258985328811022,1.293013730436367,722.6338270639899,709.3405825967112,728.5129475809422,122.943115234375,121.950927734375,125.2470703125,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(10.401000000000003), 'mean_duration_us': np.float64(0.8667500000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.11950113318849044), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.089)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1468.15), 'mean_duration_us': np.float64(122.34583333333335), 'median_duration_us': np.float64(122.079), 'std_dev_duration_us': np.float64(0.8646479855345122), 'min_duration_us': np.float64(121.151), 'max_duration_us': np.float64(124.479)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.87)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(122.35)}]","{'M': 25088, 'N': 2304, 'K': 768, 'bias': True, 'stride_A': (768, 1), 'stride_B': (1, 768), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.93323354243416,89.06325438864882 +aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((2304,), (25096, 768), (768, 2304), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",226,12,646.32,53.86000000000001,4.4953152382128945,88.871159808,150.42626953125,563.4264735955439,matrix_bf16,1.2520954903075088,0.01029725378044037,705.4637467088432,5.801745385231927,125.98337809244792,1.042859664778292,1511.800537109375,51.576,49.961,63.221,1.256009414663169,1.2344605374354898,1.2635748083268794,707.6689553064726,695.527747400138,711.931498379779,125.5830078125,124.8310546875,127.775146484375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(10.656000000000002), 'mean_duration_us': np.float64(0.8880000000000002), 'median_duration_us': np.float64(0.8640000000000001), 'std_dev_duration_us': np.float64(0.10939835464941874), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.12)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1501.144), 'mean_duration_us': np.float64(125.09533333333333), 'median_duration_us': np.float64(124.703), 'std_dev_duration_us': np.float64(1.0343101834341377), 'min_duration_us': np.float64(123.967), 'max_duration_us': np.float64(127.007)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.89)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(125.1)}]","{'M': 25096, 'N': 2304, 'K': 768, 'bias': True, 'stride_A': (768, 1), 'stride_B': (1, 768), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.9542196924858387,88.13002084621465 +aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((2304,), (25088, 768), (768, 2304), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",90,12,769.69,64.14083333333333,30.10791482418877,88.842829824,150.37939453125,563.422437389805,matrix_bf16,1.2798326110548646,0.009074582161354192,721.0864091714902,5.11282319964417,123.21262613932292,0.8791740453911183,1478.551513671875,51.1755,49.451,151.542,1.282579072306334,1.258985328811022,1.293013730436367,722.6338270639899,709.3405825967112,728.5129475809422,122.943115234375,121.950927734375,125.2470703125,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(10.401000000000003), 'mean_duration_us': np.float64(0.8667500000000002), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.11950113318849044), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.089)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1468.15), 'mean_duration_us': np.float64(122.34583333333335), 'median_duration_us': np.float64(122.079), 'std_dev_duration_us': np.float64(0.8646479855345122), 'min_duration_us': np.float64(121.151), 'max_duration_us': np.float64(124.479)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.87)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(122.35)}]","{'M': 25088, 'N': 2304, 'K': 768, 'bias': True, 'stride_A': (768, 1), 'stride_B': (1, 768), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.93323354243416,89.06325438864882 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((8, 12, 3137, 64), (8, 12, 3137, 64), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((2409216, 200768, 64, 1), (7227648, 64, 2304, 1), ())","('', '', 'False')",248,24,344.025,14.334375,5.342183097252023,0.019273728,73.5234375,0.25,vector_bf16,1.3045592230689447,0.017159701006116225,0.32613980576723617,0.004289925251529056,59.10631306966146,0.7768103961300407,1418.551513671875,13.67,11.12,37.18,1.3030007203173533,1.2801593994964995,1.3310603589276682,0.3257501800793383,0.32003984987412487,0.33276508973191704,59.16748046875,57.919921875,60.222900390625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(1418.552), 'mean_duration_us': np.float64(59.10633333333333), 'median_duration_us': np.float64(59.167500000000004), 'std_dev_duration_us': np.float64(0.7604596563190504), 'min_duration_us': np.float64(57.92), 'max_duration_us': np.float64(60.223)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(59.11)}]","{'op_shape': (8, 12, 3137, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (2409216, 200768, 64, 1), 'stride_output': (7227648, 64, 2304, 1)}",True,0.8953626857015512,89.95861707435037 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((3136, 12, 8, 64), (3136, 12, 8, 64), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((6144, 512, 64, 1), (18432, 64, 2304, 1), ())","('', '', 'False')",112,24,330.516,13.771500000000001,2.6301797921682906,0.019267584,73.5,0.25,vector_bf16,1.321548671940372,0.03358092858653137,0.330387167985093,0.008395232146632842,58.35430908203125,1.4830178369992342,1400.50341796875,13.415500000000002,10.99,20.901,1.3177170168568133,1.2810912333096607,1.3622695981806492,0.3294292542142033,0.32027280832741517,0.3405673995451623,58.4954833984375,56.574951171875,60.159912109375,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(1400.504), 'mean_duration_us': np.float64(58.35433333333333), 'median_duration_us': np.float64(58.4955), 'std_dev_duration_us': np.float64(1.4517576320523424), 'min_duration_us': np.float64(56.575), 'max_duration_us': np.float64(60.16)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(58.35)}]","{'op_shape': (3136, 12, 8, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (6144, 512, 64, 1), 'stride_output': (18432, 64, 2304, 1)}",True,0.8839710715903936,90.84258814594077 -aten::bmm,GEMM,python3,CPU,thread 17751 (python3),"((37632, 8, 8), (37632, 8, 64))","('c10::BFloat16', 'c10::BFloat16')","((64, 8, 1), (512, 64, 1))","('', '')",141,12,263.402,21.950166666666664,3.913974284147725,0.308281344,78.09375,3.764705882352941,matrix_bf16,0.75563543226023,0.006153496711181706,2.844745156744395,0.023166105265625295,108.37532552083333,0.8871153674083552,1300.50390625,19.936,19.75,32.12,0.7591178128449387,0.747152268948895,0.7611539560477466,2.8578552954162397,2.8128085419252518,2.8655207757091636,107.87158203125,107.5830078125,109.59912109375,"[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_32x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1300.504), 'mean_duration_us': np.float64(108.37533333333333), 'median_duration_us': np.float64(107.8715), 'std_dev_duration_us': np.float64(0.8493324175819239), 'min_duration_us': np.float64(107.583), 'max_duration_us': np.float64(109.599)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_32x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1300.504), 'mean_duration_us': np.float64(108.37533333333333), 'median_duration_us': np.float64(107.8715), 'std_dev_duration_us': np.float64(0.8493324175819239), 'min_duration_us': np.float64(107.583), 'max_duration_us': np.float64(109.599)}]","[{'name': 'void cutlass::Kernel2(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(10.336), 'mean_duration_us': np.float64(10.336), 'median_duration_us': np.float64(10.336), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(10.336), 'max_duration_us': np.float64(10.336)}, {'name': 'Memset (Device)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(32.448), 'mean_duration_us': np.float64(32.448), 'median_duration_us': np.float64(32.448), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(32.448), 'max_duration_us': np.float64(32.448)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16, __nv_bfloat16, float, true, false, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(43.488), 'mean_duration_us': np.float64(43.488), 'median_duration_us': np.float64(43.488), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(43.488), 'max_duration_us': np.float64(43.488)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(64.864), 'mean_duration_us': np.float64(64.864), 'median_duration_us': np.float64(64.864), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(64.864), 'max_duration_us': np.float64(64.864)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(281.343), 'mean_duration_us': np.float64(281.343), 'median_duration_us': np.float64(281.343), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(281.343), 'max_duration_us': np.float64(281.343)}, {'name': 'void cutlass__5x_cudnn::Kernel(cutlass_tensorop_bf16_s16816fprop_optimized_bf16_256x64_32x4_nhwc_align8::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(863.099), 'mean_duration_us': np.float64(863.099), 'median_duration_us': np.float64(863.099), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(863.099), 'max_duration_us': np.float64(863.099)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(10.34)}, {'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(32.45)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(43.49)}, {'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(64.86)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(281.34)}, {'name': 'void cutlass__5x_cudnn::Kernel, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(1140.8580000000002), 'mean_duration_us': np.float64(95.07150000000001), 'median_duration_us': np.float64(95.007), 'std_dev_duration_us': np.float64(0.5674224616632664), 'min_duration_us': np.float64(93.823), 'max_duration_us': np.float64(96.0)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 24, 'total_duration_us': np.float64(891.774), 'mean_duration_us': np.float64(37.15725), 'median_duration_us': np.float64(37.471999999999994), 'std_dev_duration_us': np.float64(1.5359198723132224), 'min_duration_us': np.float64(35.04), 'max_duration_us': np.float64(39.296)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(37.16)}]","{'shape_in1': (1, 25089, 768), 'shape_in2': (1, 25089, 768), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (19268352, 768, 1), 'stride_input2': (19268352, 768, 1), 'stride_output': None}",True,0.5628705555887215,94.4754075561639 aten::cat,other,python3,CPU,thread 17751 (python3),"(((1, 1, 768), (1, 25088, 768)), ())","('TensorList', 'Scalar')","(((768, 768, 1), (19267584, 768, 1)), ())","('', '1')",323,12,109.401,9.11675,0.8928409330996102,,,,,,,,,71.43155924479167,0.42238302928029525,857.1787109375,8.86,8.52,11.791,,,,,,,71.52001953125,70.55908203125,72.222900390625,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy_contig, unsigned int, 3, 128, 1>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 128, 1>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(857.179), 'mean_duration_us': np.float64(71.43158333333334), 'median_duration_us': np.float64(71.52), 'std_dev_duration_us': np.float64(0.40444395127411975), 'min_duration_us': np.float64(70.559), 'max_duration_us': np.float64(72.223)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy_cont...', 'stream': 7, 'mean_duration_us': np.float64(71.43)}]",,False,0.541034869269274,95.01644242543317 aten::cat,other,python3,CPU,thread 17751 (python3),"(((1, 1, 768), (1, 25088, 768)), ())","('TensorList', 'Scalar')","(((19268352, 768, 1), (19267584, 768, 1)), ())","('', '1')",322,12,161.371,13.447583333333334,1.2300299669914432,,,,,,,,,71.00215657552083,0.4918832522059854,852.02587890625,12.835,12.591,16.01,,,,,,,70.991455078125,69.98291015625,71.583984375,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy_contig, unsigned int, 3, 128, 1>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 128, 1>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(852.026), 'mean_duration_us': np.float64(71.00216666666667), 'median_duration_us': np.float64(70.9915), 'std_dev_duration_us': np.float64(0.4709237081405958), 'min_duration_us': np.float64(69.983), 'max_duration_us': np.float64(71.584)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy_cont...', 'stream': 7, 'mean_duration_us': np.float64(71.0)}]",,False,0.5377824998755629,95.55422492530873 @@ -26,8 +26,8 @@ aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((3136, 8, 12, 64), aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((8, 3137, 12, 64), (8, 3137, 12, 64), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((2409216, 768, 64, 1), (2409216, 64, 200768, 1), ())","('', '', 'False')",285,12,107.042,8.920166666666667,0.5312781899829084,0.019273728,73.5234375,0.25,vector_bf16,1.3642432961565705,0.004485938567823921,0.3410608240391426,0.0011214846419559802,56.511678059895836,0.18626793258170687,678.14013671875,8.76,8.45,10.14,1.3650071737511964,1.3542534621853786,1.3712368352461475,0.3412517934377991,0.33856336554634464,0.34280920881153687,56.4794921875,56.222900390625,56.927978515625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(678.1399999999999), 'mean_duration_us': np.float64(56.511666666666656), 'median_duration_us': np.float64(56.4795), 'std_dev_duration_us': np.float64(0.17831869098766825), 'min_duration_us': np.float64(56.223), 'max_duration_us': np.float64(56.928)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(56.51)}]","{'op_shape': (8, 3137, 12, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (2409216, 768, 64, 1), 'stride_output': (2409216, 64, 200768, 1)}",True,0.42802913270512655,97.8060464690606 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((1, 224, 14, 8, 768), (1, 224, 14, 8, 768), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((19267584, 86016, 6144, 768, 1), (19273728, 10752, 768, 2409216, 1), ())","('', '', 'False')",320,12,185.35399999999998,15.446166666666665,12.469253851215822,0.019267584,73.5,0.25,vector_bf16,1.449765337831095,0.008992930670082923,0.36244133445777377,0.0022482326675207307,53.162434895833336,0.3298869750019682,637.94921875,11.54,11.0,54.921,1.4491246426956061,1.4361629069733586,1.4641051154429465,0.36228116067390154,0.35904072674333964,0.36602627886073663,53.18408203125,52.639892578125,53.6640625,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(637.9490000000001), 'mean_duration_us': np.float64(53.16241666666667), 'median_duration_us': np.float64(53.184), 'std_dev_duration_us': np.float64(0.31582021740576044), 'min_duration_us': np.float64(52.64), 'max_duration_us': np.float64(53.664)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(53.16)}]","{'op_shape': (1, 224, 14, 8, 768), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (19267584, 86016, 6144, 768, 1), 'stride_output': (19273728, 10752, 768, 2409216, 1)}",True,0.40266139109935045,98.20870786015996 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((3136, 12, 64, 8), (3136, 12, 64, 8), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((6144, 512, 8, 1), (18432, 64, 1, 2304), ())","('', '', 'False')",120,12,94.36,7.863333333333333,2.5734141924806053,0.019267584,73.5,0.25,vector_bf16,1.509537166523619,0.011166934986269333,0.37738429163090476,0.0027917337465673334,51.05816650390625,0.37748484724352016,612.697998046875,6.8100000000000005,6.59,15.27,1.5095392213808339,1.493149131610688,1.5262928435995127,0.37738480534520846,0.373287282902672,0.38157321089987817,51.0555419921875,50.4951171875,51.615966796875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(612.698), 'mean_duration_us': np.float64(51.058166666666665), 'median_duration_us': np.float64(51.055499999999995), 'std_dev_duration_us': np.float64(0.36143575577902604), 'min_duration_us': np.float64(50.495), 'max_duration_us': np.float64(51.616)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(51.06)}]","{'op_shape': (3136, 12, 64, 8), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (6144, 512, 8, 1), 'stride_output': (18432, 64, 1, 2304)}",True,0.38672330173982483,98.59543116189978 -aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((768,), (25096, 768), (768, 768), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",293,12,678.969,56.58075,14.591958003858277,29.623719936,74.64990234375,378.4516787347187,matrix_bf16,1.6549605575811857,0.014483722221198377,626.3226012563458,5.481388988939864,47.301188151041664,0.41462497011250526,567.6142578125,49.13549999999999,48.321,92.931,1.6516730714337187,1.6286148406327146,1.672023619806524,625.0784466050197,616.3520204497272,632.78014579988,47.39208984375,46.815185546875,48.06298828125,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(10.337000000000003), 'mean_duration_us': np.float64(0.8614166666666669), 'median_duration_us': np.float64(0.7995000000000001), 'std_dev_duration_us': np.float64(0.12347837754935968), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.152)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(557.277), 'mean_duration_us': np.float64(46.439750000000004), 'median_duration_us': np.float64(46.432), 'std_dev_duration_us': np.float64(0.4246459554734984), 'min_duration_us': np.float64(45.951), 'max_duration_us': np.float64(47.295)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.86)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(46.44)}]","{'M': 25096, 'N': 768, 'K': 768, 'bias': True, 'stride_A': (768, 1), 'stride_B': (1, 768), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.3582673039500553,98.95369846584984 -aten::bmm,GEMM,python3,CPU,thread 17751 (python3),"((37632, 8, 64), (37632, 64, 8))","('c10::BFloat16', 'c10::BFloat16')","((512, 64, 1), (512, 8, 1))","('', '')",122,12,309.624,25.802000000000003,7.451258197598086,0.308281344,78.09375,3.764705882352941,matrix_bf16,2.004803684989608,0.023356309562369927,7.54749622584323,0.0879296359995103,40.850606282552086,0.4770833730161301,490.207275390625,22.356,22.03,47.061,2.01023028655931,1.9669262704706054,2.032542129875167,7.567925784693873,7.404898900595221,7.651923312471216,40.7354736328125,40.2880859375,41.632080078125,"[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_64x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(490.207), 'mean_duration_us': np.float64(40.85058333333333), 'median_duration_us': np.float64(40.7355), 'std_dev_duration_us': np.float64(0.4567669825949428), 'min_duration_us': np.float64(40.288), 'max_duration_us': np.float64(41.632)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_16x16_64x1_nn_align8::Params)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(490.207), 'mean_duration_us': np.float64(40.85058333333333), 'median_duration_us': np.float64(40.7355), 'std_dev_duration_us': np.float64(0.4567669825949428), 'min_duration_us': np.float64(40.288), 'max_duration_us': np.float64(41.632)}]","[{'name': 'void cutlass::Kernel2, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(468.6050000000001), 'mean_duration_us': np.float64(39.05041666666667), 'median_duration_us': np.float64(38.976), 'std_dev_duration_us': np.float64(0.46841407222195597), 'min_duration_us': np.float64(38.432), 'max_duration_us': np.float64(39.936)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(39.05)}]","{'shape_in1': (1, 25088, 768), 'shape_in2': (1, 25088, 768), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (19268352, 768, 1), 'stride_input2': (19267584, 768, 1), 'stride_output': None}",True,0.2957746887555609,99.55888264619874 aten::cat,other,python3,CPU,thread 17751 (python3),"(((128, 1, 768), (128, 196, 768)), ())","('TensorList', 'Scalar')","(((0, 768, 1), (150528, 1, 196)), ())","('', '1')",25,1,184.903,184.903,,,,,,,,,,161.31103515625,,161.31103515625,184.903,184.903,184.903,,,,,,,161.31103515625,161.31103515625,161.31103515625,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(161.311), 'mean_duration_us': np.float64(161.311), 'median_duration_us': np.float64(161.311), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(161.311), 'max_duration_us': np.float64(161.311)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy(c10::BFloat16*, c10::BFloat16 const*, int, int, int, bool const*, int, bool)', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(92.672), 'mean_duration_us': np.float64(7.722666666666666), 'median_duration_us': np.float64(7.712), 'std_dev_duration_us': np.float64(0.04617057745168701), 'min_duration_us': np.float64(7.679), 'max_duration_us': np.float64(7.841)}]","[{'name': 'void (anonymous namespace)::softmax_warp_forward >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(53.983), 'mean_duration_us': np.float64(53.983), 'median_duration_us': np.float64(53.983), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(53.983), 'max_duration_us': np.float64(53.983)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(53.98)}]","{'shape_in1': (128, 197, 768), 'shape_in2': (1, 197, 768), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (151296, 768, 1), 'stride_input2': (151296, 768, 1), 'stride_output': None}",True,0.034072984275610695,99.93211522571212 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((1, 196, 128, 768), (1, 196, 128, 768), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((19267584, 98304, 768, 1), (150528, 768, 150528, 1), ())","('', '', 'False')",64,1,11.43,11.43,,0.019267584,73.5,0.25,vector_bf16,1.489457525164785,,0.37236438129119626,,51.743896484375,,51.743896484375,11.43,11.43,11.43,1.489457525164785,1.489457525164785,1.489457525164785,0.37236438129119626,0.37236438129119626,0.37236438129119626,51.743896484375,51.743896484375,51.743896484375,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(51.744), 'mean_duration_us': np.float64(51.744), 'median_duration_us': np.float64(51.744), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(51.744), 'max_duration_us': np.float64(51.744)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(51.74)}]","{'op_shape': (1, 196, 128, 768), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (19267584, 98304, 768, 1), 'stride_output': (150528, 768, 150528, 1)}",True,0.032659761509111845,99.96477498722123 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((1, 8, 1, 1, 1, 768), (1, 8, 1, 1, 1, 768), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((6144, 768, 768, 6144, 768, 1), (0, 0, 19268352, 19268352, 768, 1), ())","('', '', 'False')",203,12,148.974,12.414499999999999,0.8435089588358649,6.144e-06,0.0234375,0.25,vector_bf16,0.007097606117564984,0.0004125868896499309,0.001774401529391246,0.00010314672241248273,3.4720255533854165,0.17761358856305998,41.664306640625,12.1605,11.71,14.69,0.006981779442363712,0.006736935885423638,0.008347565801476076,0.001745444860590928,0.0016842339713559096,0.002086891450369019,3.52001953125,2.944091796875,3.64794921875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 12, 'total_duration_us': np.float64(41.66400000000001), 'mean_duration_us': np.float64(3.472000000000001), 'median_duration_us': np.float64(3.52), 'std_dev_duration_us': np.float64(0.17008233300375442), 'min_duration_us': np.float64(2.944), 'max_duration_us': np.float64(3.648)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(3.47)}]","{'op_shape': (1, 8, 1, 1, 1, 768), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (6144, 768, 768, 6144, 768, 1), 'stride_output': (0, 0, 19268352, 19268352, 768, 1)}",True,0.026297716460843248,99.99107270368208 -aten::addmm,GEMM,python3,CPU,thread 17751 (python3),"((400,), (1, 768), (768, 400), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (19268352, 1), (1, 768), (), ())","('', '', '', '1', '1')",3526,1,42.541,42.541,,0.0006148,0.58892822265625,0.9955694890662244,matrix_bf16,0.12214145811000046,,0.12160030904437684,,5.055908203125,,5.055908203125,42.541,42.541,42.541,0.12214145811000046,0.12214145811000046,0.12214145811000046,0.12160030904437684,0.12160030904437684,0.12160030904437684,5.055908203125,5.055908203125,5.055908203125,"[{'name': 'std::enable_if::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.056), 'mean_duration_us': np.float64(5.056), 'median_duration_us': np.float64(5.056), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.056), 'max_duration_us': np.float64(5.056)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(5.056), 'mean_duration_us': np.float64(5.056), 'median_duration_us': np.float64(5.056), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(5.056), 'max_duration_us': np.float64(5.056)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.999), 'mean_duration_us': np.float64(3.999), 'median_duration_us': np.float64(3.999), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.999), 'max_duration_us': np.float64(3.999)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(4.0)}]","{'op_shape': (1, 768, 8), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (6144, 8, 1), 'stride_output': (6144, 1, 768)}",True,0.0025241073945317946,99.99678800399684 aten::copy_,elementwise,python3,CPU,thread 17751 (python3),"((1, 400), (1, 400), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((400, 1), (400, 1), ())","('', '', 'False')",3531,1,125986.084,125986.084,,4e-07,0.00152587890625,0.25,vector_bf16,0.0005747259493115847,,0.00014368148732789616,,2.783935546875,,2.783935546875,125986.084,125986.084,125986.084,0.0005747259493115847,0.0005747259493115847,0.0005747259493115847,0.00014368148732789616,0.00014368148732789616,0.00014368148732789616,2.783935546875,2.783935546875,2.783935546875,"[{'name': 'Memcpy DtoH (Device -> Pageable)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(2.784), 'mean_duration_us': np.float64(2.784), 'median_duration_us': np.float64(2.784), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(2.784), 'max_duration_us': np.float64(2.784)}]","[{'name': 'Memcpy DtoH (Device -> Pageable)', 'stream': 7, 'mean_duration_us': np.float64(2.78)}]","{'op_shape': (1, 400), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (400, 1), 'stride_output': (400, 1)}",True,0.001757167070808672,99.99854517106765 "aten::upsample_nearest1d->void at::native::(anonymous namespace)::upsample_nearest1d_out_frame(c10::BFloat16 const*, unsigned long, unsigned long, unsigned long, unsigned long, c10::BFloat16*, float) (Synthetic Op)",other,python3,CPU,thread 17751 (python3),"((1, 768, 8), (), ())","('c10::BFloat16', 'ScalarList', '')","((6144, 1, 768), (), ())","('', '[128]', '')",50125,1,77.211,77.211,,,,,,,,,,2.304931640625,,2.304931640625,77.211,77.211,77.211,,,,,,,2.304931640625,2.304931640625,2.304931640625,"[{'name': 'void at::native::(anonymous namespace)::upsample_nearest1d_out_frame(c10::BFloat16 const*, unsigned long, unsigned long, unsigned long, unsigned long, c10::BFloat16*, float)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(2.305), 'mean_duration_us': np.float64(2.305), 'median_duration_us': np.float64(2.305), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(2.305), 'max_duration_us': np.float64(2.305)}]","[{'name': 'void at::native::(anonymous namespace)::upsample_nearest1d_out_f...', 'stream': 7, 'mean_duration_us': np.float64(2.31)}]",,False,0.001454828932342776,99.99999999999999 diff --git a/tests/traces/h100/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv b/tests/traces/h100/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv index 7a68a42d8..15ce8ba71 100644 --- a/tests/traces/h100/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv +++ b/tests/traces/h100/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv @@ -1,5 +1,5 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::addmm,141,512,512,True,"(512, 1)","(1, 512)","('c10::BFloat16', 'c10::BFloat16')",1,0.0739968,0.7763671875,90.89622641509433,0.14024191474990233,0.1405526757713708,0.00274580737810573,0.1338930163829104,0.14451834091795607,12.747460835993481,12.775707840161862,0.24958352913253534,12.170369932540957,13.136171837212327,python3,CPU,thread 19392 (python3),matrix_bf16,"[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align8::Params)', 'stream': 7, 'count': 17, 'total_duration_us': np.float64(98.718), 'mean_duration_us': np.float64(5.8069411764705885), 'median_duration_us': np.float64(5.792), 'std_dev_duration_us': np.float64(0.11144002374700734), 'min_duration_us': np.float64(5.633), 'max_duration_us': np.float64(6.08)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_64x3_tn_align2::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(41.28), 'mean_duration_us': np.float64(41.28), 'median_duration_us': np.float64(41.28), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(41.28), 'max_duration_us': np.float64(41.28)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 4, 'total_duration_us': np.float64(24.510999999999996), 'mean_duration_us': np.float64(6.127749999999999), 'median_duration_us': np.float64(6.0954999999999995), 'std_dev_duration_us': np.float64(0.13305708361451504), 'min_duration_us': np.float64(5.984), 'max_duration_us': np.float64(6.336)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align8::Params)', 'stream': 7, 'count': 17, 'total_duration_us': np.float64(98.718), 'mean_duration_us': np.float64(5.8069411764705885), 'median_duration_us': np.float64(5.792), 'std_dev_duration_us': np.float64(0.11144002374700734), 'min_duration_us': np.float64(5.633), 'max_duration_us': np.float64(6.08)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_64x3_tn_align2::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(41.28), 'mean_duration_us': np.float64(41.28), 'median_duration_us': np.float64(41.28), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(41.28), 'max_duration_us': np.float64(41.28)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 4, 'total_duration_us': np.float64(24.510999999999996), 'mean_duration_us': np.float64(6.127749999999999), 'median_duration_us': np.float64(6.0954999999999995), 'std_dev_duration_us': np.float64(0.13305708361451504), 'min_duration_us': np.float64(5.984), 'max_duration_us': np.float64(6.336)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align8::Params)', 'stream': 7, 'count': 17, 'total_duration_us': np.float64(98.718), 'mean_duration_us': np.float64(5.8069411764705885), 'median_duration_us': np.float64(5.792), 'std_dev_duration_us': np.float64(0.11144002374700734), 'min_duration_us': np.float64(5.633), 'max_duration_us': np.float64(6.08)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_64x3_tn_align2::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(41.28), 'mean_duration_us': np.float64(41.28), 'median_duration_us': np.float64(41.28), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(41.28), 'max_duration_us': np.float64(41.28)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align8::Params)', 'stream': 7, 'count': 17, 'total_duration_us': np.float64(98.718), 'mean_duration_us': np.float64(5.8069411764705885), 'median_duration_us': np.float64(5.792), 'std_dev_duration_us': np.float64(0.11144002374700734), 'min_duration_us': np.float64(5.633), 'max_duration_us': np.float64(6.08)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_64x3_tn_align2::Params)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(41.28), 'mean_duration_us': np.float64(41.28), 'median_duration_us': np.float64(41.28), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(41.28), 'max_duration_us': np.float64(41.28)}]","[{'name': 'void cutlass::Kernel2 >, false, false, false, false, false, true, false>(pytorch_flash::Flash_fwd_params)', 'stream': 7, 'count': 4, 'total_duration_us': np.float64(37.408), 'mean_duration_us': np.float64(9.352), 'median_duration_us': np.float64(9.12), 'std_dev_duration_us': np.float64(0.4205139712304459), 'min_duration_us': np.float64(9.088), 'max_duration_us': np.float64(10.08)}]","[{'name': 'void pytorch_flash::flash_fwd_kernel(int, float, c10::BFloat16 const*, c10::BFloat16 const*, c10::BFloat16 const*, float*, float*, c10::BFloat16*)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(30.911999999999995), 'mean_duration_us': np.float64(3.0911999999999997), 'median_duration_us': np.float64(3.056), 'std_dev_duration_us': np.float64(0.08591833331716812), 'min_duration_us': np.float64(3.008), 'max_duration_us': np.float64(3.263)}]","[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_ke...', 'stream': 7, 'mean_duration_us': np.float64(3.09)}]","{'op_shape': (1, 141, 512), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (72192, 512, 1), 'stride_output': None, 'num_channels': 512, 'has_bias': True, 'is_affine': True, 'is_training': True}",True,9.989766911224601,78.01487545534088 -aten::addmm,GEMM,python3,CPU,thread 19392 (python3),"((2048,), (141, 512), (512, 2048), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (512, 1), (1, 512), (), ())","('', '', '', '1', '1')",123,4,140.042,35.0105,2.3895007149332814,0.2959872,2.6923828125,104.84221980413493,matrix_bf16,0.4609334034383986,0.011414977294292354,48.32528119835659,1.196771558547411,6.12774658203125,0.1536288548372083,24.510986328125,35.1905,32.38,37.281,0.4631792165968792,0.44558015289765723,0.4717950276621787,48.56073723515704,46.71561233045623,49.46403799265606,6.0955810546875,5.98388671875,6.3359375,"[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 4, 'total_duration_us': np.float64(24.510999999999996), 'mean_duration_us': np.float64(6.127749999999999), 'median_duration_us': np.float64(6.0954999999999995), 'std_dev_duration_us': np.float64(0.13305708361451504), 'min_duration_us': np.float64(5.984), 'max_duration_us': np.float64(6.336)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 4, 'total_duration_us': np.float64(24.510999999999996), 'mean_duration_us': np.float64(6.127749999999999), 'median_duration_us': np.float64(6.0954999999999995), 'std_dev_duration_us': np.float64(0.13305708361451504), 'min_duration_us': np.float64(5.984), 'max_duration_us': np.float64(6.336)}]","[{'name': 'void cutlass::Kernel2, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 9, 'total_duration_us': np.float64(14.752), 'mean_duration_us': np.float64(1.6391111111111112), 'median_duration_us': np.float64(1.632), 'std_dev_duration_us': np.float64(0.03927239472777688), 'min_duration_us': np.float64(1.6), 'max_duration_us': np.float64(1.728)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(1.64)}]","{'shape_in1': (1, 141, 512), 'shape_in2': (1, 141, 512), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (72192, 512, 1), 'stride_input2': (72192, 512, 1), 'stride_output': None}",True,4.767577913885836,90.70358994427423 aten::gelu,elementwise,python3,CPU,thread 19392 (python3),"((1, 141, 2048), ())","('c10::BFloat16', '')","((288768, 2048, 1), ())","('', '')",126,4,61.120000000000005,15.280000000000001,6.821089844494549,,,,,,,,,1.9676513671875,0.018187520729568748,7.87060546875,11.925,11.76,25.51,,,,,,,1.9674072265625,1.951904296875,1.98388671875,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::GeluCUDAKernelImpl(at::TensorIteratorBase&, at::native::GeluType)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::GeluCUDAKernelImpl(at::TensorIteratorBase&, at::native::GeluType)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 4, 'total_duration_us': np.float64(7.871), 'mean_duration_us': np.float64(1.96775), 'median_duration_us': np.float64(1.9675), 'std_dev_duration_us': np.float64(0.015753967754188188), 'min_duration_us': np.float64(1.952), 'max_duration_us': np.float64(1.984)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Ge...', 'stream': 7, 'mean_duration_us': np.float64(1.97)}]",,False,2.5435182416444895,93.24710818591872 aten::all,reduce,python3,CPU,thread 19392 (python3),"((1, 141),)","('bool',)","((141, 1),)","('',)",37,1,16.5,16.5,,,,,,,,,,3.64794921875,,3.64794921875,16.5,16.5,16.5,,,,,,,3.64794921875,3.64794921875,3.64794921875,"[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, bool, 4> >(at::native::ReduceOp, unsigned int, bool, 4>)', 'stream': 7, 'count': 1, 'total_duration_us': np.float64(3.648), 'mean_duration_us': np.float64(3.648), 'median_duration_us': np.float64(3.648), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(3.648), 'max_duration_us': np.float64(3.648)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(cutlass_75_tensorop_bf16_s1688gemm_bf16_256x128_tn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(130015.7), 'mean_duration_us': np.float64(1083.4641666666666), 'median_duration_us': np.float64(1082.965), 'std_dev_duration_us': np.float64(3.932529614919927), 'min_duration_us': np.float64(1075.574), 'max_duration_us': np.float64(1096.694)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(99311.41), 'mean_duration_us': np.float64(827.5950833333334), 'median_duration_us': np.float64(826.9205), 'std_dev_duration_us': np.float64(2.9474507419783706), 'min_duration_us': np.float64(825.432), 'max_duration_us': np.float64(851.8)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 240, 'total_duration_us': np.float64(1579.851), 'mean_duration_us': np.float64(6.5827125), 'median_duration_us': np.float64(6.592), 'std_dev_duration_us': np.float64(0.13116311667951205), 'min_duration_us': np.float64(6.303), 'max_duration_us': np.float64(6.975)}]","[{'name': 'void cutlass::Kernel2(cublasLt::cublasSplitKParams, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(140.35000000000002), 'mean_duration_us': np.float64(2.339166666666667), 'median_duration_us': np.float64(2.336), 'std_dev_duration_us': np.float64(0.037292164801142315), 'min_duration_us': np.float64(2.272), 'max_duration_us': np.float64(2.496)}, {'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(385.4959999999999), 'mean_duration_us': np.float64(6.424933333333332), 'median_duration_us': np.float64(6.336), 'std_dev_duration_us': np.float64(0.19279919317489072), 'min_duration_us': np.float64(6.239), 'max_duration_us': np.float64(6.912)}]","[{'name': 'void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, _...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(422.845), 'mean_duration_us': np.float64(7.047416666666667), 'median_duration_us': np.float64(6.976), 'std_dev_duration_us': np.float64(0.17790056133194818), 'min_duration_us': np.float64(6.752), 'max_duration_us': np.float64(7.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_tn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(185.69699999999997), 'mean_duration_us': np.float64(3.0949499999999994), 'median_duration_us': np.float64(3.104), 'std_dev_duration_us': np.float64(0.04453478977159322), 'min_duration_us': np.float64(3.008), 'max_duration_us': np.float64(3.168)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_nn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(167.428), 'mean_duration_us': np.float64(2.7904666666666667), 'median_duration_us': np.float64(2.816), 'std_dev_duration_us': np.float64(0.037250264726874344), 'min_duration_us': np.float64(2.72), 'max_duration_us': np.float64(2.88)}]","[{'name': 'void cutlass::Kernel2(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(35.168), 'mean_duration_us': np.float64(2.344533333333333), 'median_duration_us': np.float64(2.368), 'std_dev_duration_us': np.float64(0.10418372660300108), 'min_duration_us': np.float64(2.176), 'max_duration_us': np.float64(2.56)}, {'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(83.104), 'mean_duration_us': np.float64(5.540266666666667), 'median_duration_us': np.float64(5.376), 'std_dev_duration_us': np.float64(0.32818683025916123), 'min_duration_us': np.float64(5.152), 'max_duration_us': np.float64(6.176)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void gemv2T_kernel_val(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align2::Params)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(56.54299999999999), 'mean_duration_us': np.float64(11.308599999999998), 'median_duration_us': np.float64(11.296), 'std_dev_duration_us': np.float64(0.10403768547983014), 'min_duration_us': np.float64(11.168), 'max_duration_us': np.float64(11.487)}]","[{'name': 'void cutlass::Kernel2, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(23.615000000000002), 'mean_duration_us': np.float64(4.723000000000001), 'median_duration_us': np.float64(4.704), 'std_dev_duration_us': np.float64(0.06861486719363401), 'min_duration_us': np.float64(4.64), 'max_duration_us': np.float64(4.831)}]","[{'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched > >(cublasDotParams, cublasGemvTensorStridedBatched >)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(10.143), 'mean_duration_us': np.float64(2.0286), 'median_duration_us': np.float64(2.047), 'std_dev_duration_us': np.float64(0.02545270123189289), 'min_duration_us': np.float64(1.984), 'max_duration_us': np.float64(2.048)}, {'name': 'void reduce_1Block_kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16> >(float const*, float, cublasGemvTensorStridedBatched, int, float const*, float, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, cublasPointerMode_t, cublasLtEpilogue_t, cublasGemvTensorStridedBatched::value_type, float>::type const>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(13.183000000000002), 'mean_duration_us': np.float64(2.6366000000000005), 'median_duration_us': np.float64(2.688), 'std_dev_duration_us': np.float64(0.08984341934721768), 'min_duration_us': np.float64(2.527), 'max_duration_us': np.float64(2.72)}]","[{'name': 'void dot_kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(18.784), 'mean_duration_us': np.float64(3.7567999999999997), 'median_duration_us': np.float64(3.744), 'std_dev_duration_us': np.float64(0.03263372488699366), 'min_duration_us': np.float64(3.712), 'max_duration_us': np.float64(3.808)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(17.918), 'mean_duration_us': np.float64(3.5835999999999997), 'median_duration_us': np.float64(3.552), 'std_dev_duration_us': np.float64(0.049834124854360604), 'min_duration_us': np.float64(3.551), 'max_duration_us': np.float64(3.68)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(cutlass_75_tensorop_bf16_s1688gemm_bf16_256x128_tn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(130015.7), 'mean_duration_us': np.float64(1083.4641666666666), 'median_duration_us': np.float64(1082.965), 'std_dev_duration_us': np.float64(3.932529614919927), 'min_duration_us': np.float64(1075.574), 'max_duration_us': np.float64(1096.694)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(99311.41), 'mean_duration_us': np.float64(827.5950833333334), 'median_duration_us': np.float64(826.9205), 'std_dev_duration_us': np.float64(2.9474507419783706), 'min_duration_us': np.float64(825.432), 'max_duration_us': np.float64(851.8)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 240, 'total_duration_us': np.float64(1579.851), 'mean_duration_us': np.float64(6.5827125), 'median_duration_us': np.float64(6.592), 'std_dev_duration_us': np.float64(0.13116311667951205), 'min_duration_us': np.float64(6.303), 'max_duration_us': np.float64(6.975)}]","[{'name': 'void cutlass::Kernel2(cublasLt::cublasSplitKParams, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(140.35000000000002), 'mean_duration_us': np.float64(2.339166666666667), 'median_duration_us': np.float64(2.336), 'std_dev_duration_us': np.float64(0.037292164801142315), 'min_duration_us': np.float64(2.272), 'max_duration_us': np.float64(2.496)}, {'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(385.4959999999999), 'mean_duration_us': np.float64(6.424933333333332), 'median_duration_us': np.float64(6.336), 'std_dev_duration_us': np.float64(0.19279919317489072), 'min_duration_us': np.float64(6.239), 'max_duration_us': np.float64(6.912)}]","[{'name': 'void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, _...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(422.845), 'mean_duration_us': np.float64(7.047416666666667), 'median_duration_us': np.float64(6.976), 'std_dev_duration_us': np.float64(0.17790056133194818), 'min_duration_us': np.float64(6.752), 'max_duration_us': np.float64(7.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_tn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(185.69699999999997), 'mean_duration_us': np.float64(3.0949499999999994), 'median_duration_us': np.float64(3.104), 'std_dev_duration_us': np.float64(0.04453478977159322), 'min_duration_us': np.float64(3.008), 'max_duration_us': np.float64(3.168)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_nn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(167.428), 'mean_duration_us': np.float64(2.7904666666666667), 'median_duration_us': np.float64(2.816), 'std_dev_duration_us': np.float64(0.037250264726874344), 'min_duration_us': np.float64(2.72), 'max_duration_us': np.float64(2.88)}]","[{'name': 'void cutlass::Kernel2(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(35.168), 'mean_duration_us': np.float64(2.344533333333333), 'median_duration_us': np.float64(2.368), 'std_dev_duration_us': np.float64(0.10418372660300108), 'min_duration_us': np.float64(2.176), 'max_duration_us': np.float64(2.56)}, {'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(83.104), 'mean_duration_us': np.float64(5.540266666666667), 'median_duration_us': np.float64(5.376), 'std_dev_duration_us': np.float64(0.32818683025916123), 'min_duration_us': np.float64(5.152), 'max_duration_us': np.float64(6.176)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void gemv2T_kernel_val(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align2::Params)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(56.54299999999999), 'mean_duration_us': np.float64(11.308599999999998), 'median_duration_us': np.float64(11.296), 'std_dev_duration_us': np.float64(0.10403768547983014), 'min_duration_us': np.float64(11.168), 'max_duration_us': np.float64(11.487)}]","[{'name': 'void cutlass::Kernel2, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(23.615000000000002), 'mean_duration_us': np.float64(4.723000000000001), 'median_duration_us': np.float64(4.704), 'std_dev_duration_us': np.float64(0.06861486719363401), 'min_duration_us': np.float64(4.64), 'max_duration_us': np.float64(4.831)}]","[{'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched > >(cublasDotParams, cublasGemvTensorStridedBatched >)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(10.143), 'mean_duration_us': np.float64(2.0286), 'median_duration_us': np.float64(2.047), 'std_dev_duration_us': np.float64(0.02545270123189289), 'min_duration_us': np.float64(1.984), 'max_duration_us': np.float64(2.048)}, {'name': 'void reduce_1Block_kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16> >(float const*, float, cublasGemvTensorStridedBatched, int, float const*, float, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, cublasPointerMode_t, cublasLtEpilogue_t, cublasGemvTensorStridedBatched::value_type, float>::type const>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(13.183000000000002), 'mean_duration_us': np.float64(2.6366000000000005), 'median_duration_us': np.float64(2.688), 'std_dev_duration_us': np.float64(0.08984341934721768), 'min_duration_us': np.float64(2.527), 'max_duration_us': np.float64(2.72)}]","[{'name': 'void dot_kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(18.784), 'mean_duration_us': np.float64(3.7567999999999997), 'median_duration_us': np.float64(3.744), 'std_dev_duration_us': np.float64(0.03263372488699366), 'min_duration_us': np.float64(3.712), 'max_duration_us': np.float64(3.808)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(17.918), 'mean_duration_us': np.float64(3.5835999999999997), 'median_duration_us': np.float64(3.552), 'std_dev_duration_us': np.float64(0.049834124854360604), 'min_duration_us': np.float64(3.551), 'max_duration_us': np.float64(3.68)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel(c10::BFloat16*, c10::BFloat16 const*, int)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(262422.467), 'mean_duration_us': np.float64(2186.8538916666666), 'median_duration_us': np.float64(2170.955), 'std_dev_duration_us': np.float64(52.45910857528236), 'min_duration_us': np.float64(2146.764), 'max_duration_us': np.float64(2395.817)}]","[{'name': 'void at::native::(anonymous namespace)::cunn_SoftMaxForward<8, c...', 'stream': 7, 'mean_duration_us': np.float64(2186.85)}]",,False,46.30226398196923,46.30226398196923 -aten::bmm,GEMM,python3,CPU,thread 21587 (python3),"((16, 5185, 64), (16, 64, 5185))","('c10::BFloat16', 'c10::BFloat16')","((331840, 64, 1), (331840, 1, 64))","('', '')",130,120,2904.9230000000002,24.20769166666667,6.205454380123629,55.0588928,840.6953430175781,62.458121588556374,matrix_bf16,0.81363516256602,0.0029586788585072117,50.8181239122733,0.18479352388613465,1083.464170328776,3.9490247035939565,130015.70043945312,22.675,21.861,54.201,0.8139994260729402,0.8038093453714655,0.8195930552523396,50.8408751266788,50.20442182722889,51.190242698087026,1082.965087890625,1075.573974609375,1096.694091796875,"[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_256x128_tn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(130015.7), 'mean_duration_us': np.float64(1083.4641666666666), 'median_duration_us': np.float64(1082.965), 'std_dev_duration_us': np.float64(3.932529614919927), 'min_duration_us': np.float64(1075.574), 'max_duration_us': np.float64(1096.694)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(99311.41), 'mean_duration_us': np.float64(827.5950833333334), 'median_duration_us': np.float64(826.9205), 'std_dev_duration_us': np.float64(2.9474507419783706), 'min_duration_us': np.float64(825.432), 'max_duration_us': np.float64(851.8)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_256x128_tn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(130015.7), 'mean_duration_us': np.float64(1083.4641666666666), 'median_duration_us': np.float64(1082.965), 'std_dev_duration_us': np.float64(3.932529614919927), 'min_duration_us': np.float64(1075.574), 'max_duration_us': np.float64(1096.694)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_tensorop_bf16_s1688gemm_bf16_128x128_nn_align1::Params)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(99311.41), 'mean_duration_us': np.float64(827.5950833333334), 'median_duration_us': np.float64(826.9205), 'std_dev_duration_us': np.float64(2.9474507419783706), 'min_duration_us': np.float64(825.432), 'max_duration_us': np.float64(851.8)}]","[{'name': 'void cutlass::Kernel2 Pageable)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(8133.554), 'mean_duration_us': np.float64(1626.7108), 'median_duration_us': np.float64(347.196), 'std_dev_duration_us': np.float64(1579.6924965375256), 'min_duration_us': np.float64(325.565), 'max_duration_us': np.float64(3620.445)}]","[{'name': 'Memcpy DtoH (Device -> Pageable)', 'stream': 7, 'mean_duration_us': np.float64(1626.71)}]","{'op_shape': (1, 5184, 768), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (3981312, 768, 1), 'stride_output': (3981312, 768, 1)}",True,1.435098059781039,90.05025103739173 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((1024,), (5185, 4096), (4096, 1024), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (4096, 1), (1, 4096), (), ())","('', '', '', '1', '1')",180,120,5342.39,44.51991666666667,3.9975768597113572,43.50024192,58.63671875,707.4929218573046,matrix_bf16,0.9361141033731365,0.008279770802943752,662.2941021872913,5.857879237683533,65.6862569173177,0.5816449146059576,7882.350830078125,43.34,41.401,64.581,0.9368314933785674,0.9167019964401024,0.9516789078184636,662.8016505383448,648.5601739338325,673.3060911624533,65.630859375,64.60693359375,67.072021484375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(115.61200000000002), 'mean_duration_us': np.float64(0.9634333333333335), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.18370559478566667), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.6)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(7766.737), 'mean_duration_us': np.float64(64.72280833333333), 'median_duration_us': np.float64(64.639), 'std_dev_duration_us': np.float64(0.5570301951096441), 'min_duration_us': np.float64(63.711), 'max_duration_us': np.float64(65.951)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.96)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(64.72)}]","{'M': 5185, 'N': 1024, 'K': 4096, 'bias': True, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,1.3907753124513664,91.4410263498431 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((4096,), (5185, 1024), (1024, 4096), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",168,120,5464.575,45.538125,5.356131963095423,43.51617024,58.642578125,707.68126561199,matrix_bf16,1.0540770662571635,0.006644526312569384,745.9505923014428,4.70220679027121,58.338828531901044,0.3674647953274667,7000.659423828125,43.894999999999996,42.42,87.671,1.0538078218406994,1.0409530341917432,1.069334988558061,745.7600530720407,736.6629606794539,756.7483380659514,58.3514404296875,57.504150390625,59.072021484375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(116.09), 'mean_duration_us': np.float64(0.9674166666666667), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.15527371441711854), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.568)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(6884.5689999999995), 'mean_duration_us': np.float64(57.37140833333333), 'median_duration_us': np.float64(57.344), 'std_dev_duration_us': np.float64(0.31250691981227396), 'min_duration_us': np.float64(56.639), 'max_duration_us': np.float64(58.048)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.97)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(57.37)}]","{'M': 5185, 'N': 4096, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,1.235208189463912,92.67623453930702 +aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((1024,), (5185, 4096), (4096, 1024), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (4096, 1), (1, 4096), (), ())","('', '', '', '1', '1')",180,120,5342.39,44.51991666666667,3.9975768597113572,43.50024192,58.63671875,707.4929218573046,matrix_bf16,0.9361141033731365,0.008279770802943752,662.2941021872913,5.857879237683533,65.6862569173177,0.5816449146059576,7882.350830078125,43.34,41.401,64.581,0.9368314933785674,0.9167019964401024,0.9516789078184636,662.8016505383448,648.5601739338325,673.3060911624533,65.630859375,64.60693359375,67.072021484375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(115.61200000000002), 'mean_duration_us': np.float64(0.9634333333333335), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.18370559478566667), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.6)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(7766.737), 'mean_duration_us': np.float64(64.72280833333333), 'median_duration_us': np.float64(64.639), 'std_dev_duration_us': np.float64(0.5570301951096441), 'min_duration_us': np.float64(63.711), 'max_duration_us': np.float64(65.951)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.96)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(64.72)}]","{'M': 5185, 'N': 1024, 'K': 4096, 'bias': True, 'stride_A': (4096, 1), 'stride_B': (1, 4096), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,1.3907753124513664,91.4410263498431 +aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((4096,), (5185, 1024), (1024, 4096), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",168,120,5464.575,45.538125,5.356131963095423,43.51617024,58.642578125,707.68126561199,matrix_bf16,1.0540770662571635,0.006644526312569384,745.9505923014428,4.70220679027121,58.338828531901044,0.3674647953274667,7000.659423828125,43.894999999999996,42.42,87.671,1.0538078218406994,1.0409530341917432,1.069334988558061,745.7600530720407,736.6629606794539,756.7483380659514,58.3514404296875,57.504150390625,59.072021484375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(116.09), 'mean_duration_us': np.float64(0.9674166666666667), 'median_duration_us': np.float64(0.96), 'std_dev_duration_us': np.float64(0.15527371441711854), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.568)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(6884.5689999999995), 'mean_duration_us': np.float64(57.37140833333333), 'median_duration_us': np.float64(57.344), 'std_dev_duration_us': np.float64(0.31250691981227396), 'min_duration_us': np.float64(56.639), 'max_duration_us': np.float64(58.048)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.97)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(57.37)}]","{'M': 5185, 'N': 4096, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,1.235208189463912,92.67623453930702 aten::mul,elementwise,python3,CPU,thread 21587 (python3),"((1, 5185, 4096), (1, 5185, 4096))","('c10::BFloat16', 'c10::BFloat16')","((21237760, 4096, 1), (21237760, 4096, 1))","('', '')",173,120,1010.96,8.424666666666667,1.5790246608434646,0.02123776,121.5234375,0.16666666666666666,vector_bf16,2.927888439760909,0.04325446081552555,0.4879814066268182,0.007209076802587591,43.531089274088544,0.6437695588697182,5223.730712890625,8.105,7.73,23.72,2.938878306392254,2.850444490464644,2.987306416359983,0.48981305106537576,0.47507408174410737,0.49788440272666395,43.3594970703125,42.656005859375,44.7041015625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(5223.73), 'mean_duration_us': np.float64(43.53108333333333), 'median_duration_us': np.float64(43.3595), 'std_dev_duration_us': np.float64(0.6410638629566389), 'min_duration_us': np.float64(42.656), 'max_duration_us': np.float64(44.704)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Bi...', 'stream': 7, 'mean_duration_us': np.float64(43.53)}]","{'shape_in1': (1, 5185, 4096), 'shape_in2': (1, 5185, 4096), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (21237760, 4096, 1), 'stride_input2': (21237760, 4096, 1), 'stride_output': None}",True,0.9216838822575284,93.59791842156454 aten::copy_,elementwise,python3,CPU,thread 21587 (python3),"((1, 16, 5185, 64), (1, 16, 5185, 64), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((5309440, 331840, 64, 1), (5309440, 64, 1024, 1), ())","('', '', 'False')",99,360,3863.671,10.732419444444444,2.8546070292674024,0.00530944,20.25390625,0.25,vector_bf16,1.5309547885496972,0.13893879350192556,0.3827386971374244,0.03473469837548139,13.994538709852431,1.3518866999959953,5038.033935546875,11.614999999999998,6.63,20.97,1.6147809574724805,1.30388310090533,1.6675266923534036,0.4036952393681202,0.32597077522633255,0.41688167308835095,13.152099609375,12.736083984375,16.2880859375,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 360, 'total_duration_us': np.float64(5038.033), 'mean_duration_us': np.float64(13.994536111111112), 'median_duration_us': np.float64(13.152), 'std_dev_duration_us': np.float64(1.3500063184322875), 'min_duration_us': np.float64(12.736), 'max_duration_us': np.float64(16.288)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(13.99)}]","{'op_shape': (1, 16, 5185, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (5309440, 331840, 64, 1), 'stride_output': (5309440, 64, 1024, 1)}",True,0.8889192287805522,94.4868376503451 aten::sigmoid,elementwise,python3,CPU,thread 21587 (python3),"((1, 5185, 4096),)","('c10::BFloat16',)","((21237760, 4096, 1),)","('',)",172,120,1101.429,9.178575,1.5475797187622904,0.02123776,81.015625,0.25,vector_bf16,2.5562344827415298,0.02997232525707459,0.6390586206853824,0.007493081314268645,33.237467447916664,0.39434098956693103,3988.49609375,8.82,8.32,16.73,2.5699959365698373,2.4787321363746453,2.589966876121147,0.6424989841424593,0.6196830340936614,0.6474917190302869,33.054931640625,32.800048828125,34.27197265625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::sigmoid_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array >(int, at::native::sigmoid_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(3988.498), 'mean_duration_us': np.float64(33.23748333333334), 'median_duration_us': np.float64(33.055), 'std_dev_duration_us': np.float64(0.39268874407375365), 'min_duration_us': np.float64(32.8), 'max_duration_us': np.float64(34.272)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::si...', 'stream': 7, 'mean_duration_us': np.float64(33.24)}]","{'op_shape': (1, 5185, 4096), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (21237760, 4096, 1), 'stride_output': None}",True,0.7037369968143413,95.19057464715944 @@ -16,47 +16,47 @@ aten::copy_,elementwise,python3,CPU,thread 21587 (python3),"((1, 5185, 1024), (1 aten::add,elementwise,python3,CPU,thread 21587 (python3),"((1, 5185, 1024), (1, 5185, 1024), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((5309440, 1024, 1), (5309440, 1024, 1), ())","('', '', '1')",58,245,3153.262,12.870457142857143,2.072352795697699,0.00530944,30.380859375,0.16666666666666666,vector_bf16,3.37217742955636,0.11828050989378604,0.5620295715927267,0.019713418315631014,9.458608697385204,0.33553775984771844,2317.359130859375,12.61,11.41,25.44,3.363270289970874,2.9363337107880643,3.714658167221795,0.5605450483284791,0.4893889517980108,0.619109694536966,9.471923828125,8.575927734375,10.84912109375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 245, 'total_duration_us': np.float64(2317.353), 'mean_duration_us': np.float64(9.458583673469388), 'median_duration_us': np.float64(9.472), 'std_dev_duration_us': np.float64(0.3348456698101192), 'min_duration_us': np.float64(8.576), 'max_duration_us': np.float64(10.849)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(9.46)}]","{'shape_in1': (1, 5185, 1024), 'shape_in2': (1, 5185, 1024), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (5309440, 1024, 1), 'stride_input2': (5309440, 1024, 1), 'stride_output': None}",True,0.4088787645666147,97.72585118137282 aten::copy_,elementwise,python3,CPU,thread 21587 (python3),"((1, 5185, 16, 64), (1, 5185, 16, 64), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((5309440, 1024, 64, 1), (5309440, 64, 331840, 1), ())","('', '', 'False')",143,120,1354.8029999999999,11.290024999999998,1.3857915037287256,0.00530944,20.25390625,0.25,vector_bf16,1.360252899597523,0.016036729759886573,0.3400632248993808,0.004009182439971644,15.615273030598958,0.18619000389162943,1873.832763671875,11.096,10.51,21.59,1.3627939741822284,1.2962473730796167,1.391370338926121,0.34069849354555715,0.3240618432699042,0.3478425847315303,15.583984375,15.263916015625,16.384033203125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(1873.8329999999999), 'mean_duration_us': np.float64(15.615274999999999), 'median_duration_us': np.float64(15.584), 'std_dev_duration_us': np.float64(0.1854254999768552), 'min_duration_us': np.float64(15.264), 'max_duration_us': np.float64(16.384)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(15.62)}]","{'op_shape': (1, 5185, 16, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (5309440, 1024, 64, 1), 'stride_output': (5309440, 64, 331840, 1)}",True,0.3306222221716982,98.05647340354452 aten::copy_,elementwise,python3,CPU,thread 21587 (python3),"((1, 3, 1008, 1008), (1, 3, 1008, 1008), ())","('c10::BFloat16', 'c10::BFloat16', 'Scalar')","((3048192, 1016064, 1008, 1), (3048192, 1016064, 1008, 1), ())","('', '', 'False')",38,5,2087.843,417.56859999999995,19.572307153220333,0.003048192,11.6279296875,0.25,vector_bf16,0.037495513635406354,0.0015161349549736524,0.009373878408851588,0.0003790337387434131,325.5904296875,12.710085512152338,1627.9521484375,422.057,384.266,431.807,0.03655991693240216,0.036482892684950884,0.03997772855227399,0.00913997923310054,0.009120723171237721,0.009994432138068497,333.5009765625,304.989013671875,334.205078125,"[{'name': 'Memcpy HtoD (Pageable -> Device)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(1627.952), 'mean_duration_us': np.float64(325.5904), 'median_duration_us': np.float64(333.501), 'std_dev_duration_us': np.float64(11.368233417730304), 'min_duration_us': np.float64(304.989), 'max_duration_us': np.float64(334.205)}]","[{'name': 'Memcpy HtoD (Pageable -> Device)', 'stream': 7, 'mean_duration_us': np.float64(325.59)}]","{'op_shape': (1, 3, 1008, 1008), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (3048192, 1016064, 1008, 1), 'stride_output': (3048192, 1016064, 1008, 1)}",True,0.28723863054400445,98.34371203408853 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((768,), (5, 768), (768, 768), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",2995,240,6472.147,26.967279166666668,3.633627911660878,0.00590208,1.14111328125,4.932605905006419,matrix_bf16,0.1818424780903004,0.00362785334275486,0.896957281209216,0.017894770820969894,6.582722981770833,0.13143096002245483,1579.853515625,26.6205,22.661,43.221,0.18151343372467688,0.1715451250962548,0.18983786745167913,0.8953342350283321,0.8461644970248511,0.9363953859859782,6.592041015625,6.302978515625,6.97509765625,"[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 240, 'total_duration_us': np.float64(1579.851), 'mean_duration_us': np.float64(6.5827125), 'median_duration_us': np.float64(6.592), 'std_dev_duration_us': np.float64(0.13116311667951205), 'min_duration_us': np.float64(6.303), 'max_duration_us': np.float64(6.975)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 240, 'total_duration_us': np.float64(1579.851), 'mean_duration_us': np.float64(6.5827125), 'median_duration_us': np.float64(6.592), 'std_dev_duration_us': np.float64(0.13116311667951205), 'min_duration_us': np.float64(6.303), 'max_duration_us': np.float64(6.975)}]","[{'name': 'void cutlass::Kernel2(cudnn::engines_precompiled::nhwc2nchw_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(48.35), 'mean_duration_us': np.float64(9.67), 'median_duration_us': np.float64(9.663), 'std_dev_duration_us': np.float64(0.0547539952880155), 'min_duration_us': np.float64(9.6), 'max_duration_us': np.float64(9.759)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(58.303000000000004), 'mean_duration_us': np.float64(11.6606), 'median_duration_us': np.float64(11.679), 'std_dev_duration_us': np.float64(0.15088485676170435), 'min_duration_us': np.float64(11.424), 'max_duration_us': np.float64(11.872)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, false, true, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t, __nv_bfloat16 const*, __nv_bfloat16*)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(222.207), 'mean_duration_us': np.float64(44.4414), 'median_duration_us': np.float64(44.384), 'std_dev_duration_us': np.float64(0.1792647204555328), 'min_duration_us': np.float64(44.224), 'max_duration_us': np.float64(44.768)}, {'name': 'void cutlass__5x_cudnn::Kernel(cutlass_tensorop_bf16_s16816fprop_optimized_bf16_256x64_32x4_nhwc_align8::Params)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(982.9660000000001), 'mean_duration_us': np.float64(196.59320000000002), 'median_duration_us': np.float64(196.542), 'std_dev_duration_us': np.float64(0.4292092263686784), 'min_duration_us': np.float64(195.966), 'max_duration_us': np.float64(197.245)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(2.17)}, {'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(5.63)}, {'name': 'void cudnn::engines_precompiled::nhwcToNchwKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(9.67)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(11.66)}, {'name': 'void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16,...', 'stream': 7, 'mean_duration_us': np.float64(44.44)}, {'name': 'void cutlass__5x_cudnn::Kernel >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(934.809), 'mean_duration_us': np.float64(7.790075), 'median_duration_us': np.float64(7.776), 'std_dev_duration_us': np.float64(0.06863528763204346), 'min_duration_us': np.float64(7.711), 'max_duration_us': np.float64(8.032)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(7.79)}]","{'shape_in1': (1, 5185, 1024), 'shape_in2': (), 'dtype_in1_in2_out': ('c10::BFloat16', 'double', None), 'stride_input1': (5309440, 1024, 1), 'stride_input2': (), 'stride_output': None}",True,0.16493869247189352,99.02574088752509 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((768,), (5, 3072), (3072, 768), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (3072, 1), (1, 3072), (), ())","('', '', '', '1', '1')",3102,60,1877.59,31.293166666666664,3.028820457138705,0.0235968,4.5380859375,4.9588444157520986,matrix_bf16,0.543219529864628,0.011771522206847568,2.6937411321966906,0.05837314716032782,8.76400146484375,0.19430162370386625,525.840087890625,30.235500000000002,28.151,39.811,0.5487002632111675,0.514544104751848,0.5569473850725798,2.720919236146404,2.551544160506864,2.7618154303348956,8.67236328125,8.5439453125,9.248046875,"[{'name': 'void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, true, true, false>(cublasLt::cublasSplitKParams, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(140.35000000000002), 'mean_duration_us': np.float64(2.339166666666667), 'median_duration_us': np.float64(2.336), 'std_dev_duration_us': np.float64(0.037292164801142315), 'min_duration_us': np.float64(2.272), 'max_duration_us': np.float64(2.496)}, {'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(385.4959999999999), 'mean_duration_us': np.float64(6.424933333333332), 'median_duration_us': np.float64(6.336), 'std_dev_duration_us': np.float64(0.19279919317489072), 'min_duration_us': np.float64(6.239), 'max_duration_us': np.float64(6.912)}]","[{'name': 'void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, _...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void cutlass::Kernel2(cublasLt::cublasSplitKParams, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(140.35000000000002), 'mean_duration_us': np.float64(2.339166666666667), 'median_duration_us': np.float64(2.336), 'std_dev_duration_us': np.float64(0.037292164801142315), 'min_duration_us': np.float64(2.272), 'max_duration_us': np.float64(2.496)}, {'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(385.4959999999999), 'mean_duration_us': np.float64(6.424933333333332), 'median_duration_us': np.float64(6.336), 'std_dev_duration_us': np.float64(0.19279919317489072), 'min_duration_us': np.float64(6.239), 'max_duration_us': np.float64(6.912)}]","[{'name': 'void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, _...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void cutlass::Kernel2(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 180, 'total_duration_us': np.float64(457.9199999999999), 'mean_duration_us': np.float64(2.5439999999999996), 'median_duration_us': np.float64(2.4645), 'std_dev_duration_us': np.float64(0.22104903930525865), 'min_duration_us': np.float64(2.24), 'max_duration_us': np.float64(3.232)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.54)}]","{'op_shape': (1, 12, 5, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (3840, 320, 64, 1), 'stride_output': (3840, 64, 768, 1)}",True,0.08079478707812955,99.19931578910217 aten::layer_norm,NORM_fwd,python3,CPU,thread 21587 (python3),"((1, 5, 768), (), (768,), (768,), (), ())","('c10::BFloat16', 'ScalarList', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((3840, 768, 1), (), (1,), (1,), (), ())","('', '[768]', '', '', '1.0000000000000001e-05', 'True')",2982,125,3175.6730000000002,25.405384,2.3197281882763496,2.3808e-05,0.0234375,0.96875,vector_bf16,0.006822840126415809,0.00030302975427877585,0.006609626372465316,0.00029356007445756414,3.60912109375,0.1614401127924182,451.14013671875,24.67,22.72,35.53,0.006857172752043597,0.006000077248614175,0.00738487975937202,0.0066428861035422344,0.005812574834594981,0.007154102266891644,3.583984375,3.327880859375,4.095947265625,"[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_kernel(int, float, c10::BFloat16 const*, c10::BFloat16 const*, c10::BFloat16 const*, float*, float*, c10::BFloat16*)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(451.14099999999996), 'mean_duration_us': np.float64(3.6091279999999997), 'median_duration_us': np.float64(3.584), 'std_dev_duration_us': np.float64(0.16081094370719926), 'min_duration_us': np.float64(3.328), 'max_duration_us': np.float64(4.096)}]","[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_ke...', 'stream': 7, 'mean_duration_us': np.float64(3.61)}]","{'op_shape': (1, 5, 768), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (3840, 768, 1), 'stride_output': None, 'num_channels': 768, 'has_bias': True, 'is_affine': True, 'is_training': True}",True,0.07959992876873168,99.2789157178709 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((1024,), (5184, 1024), (1024, 1024), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",4653,20,943.795,47.18975,8.063849262865133,10.876944384,22.251953125,466.1641358729044,matrix_bf16,1.0808561306902909,0.02489510390079241,503.85636416617047,11.605204597379057,21.5982177734375,0.4948575067787944,431.96435546875,44.596000000000004,38.921,66.401,1.0730756837665036,1.0491400290246446,1.1269681965945002,500.22939884923824,489.0714550399473,525.3521555217206,21.7440185546875,20.7041015625,22.239990234375,"[{'name': 'Memset (Device)', 'stream': 7, 'count': 20, 'total_duration_us': np.float64(17.248), 'mean_duration_us': np.float64(0.8624), 'median_duration_us': np.float64(0.8), 'std_dev_duration_us': np.float64(0.13365305832639968), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(1.248)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 20, 'total_duration_us': np.float64(414.7149999999999), 'mean_duration_us': np.float64(20.735749999999996), 'median_duration_us': np.float64(20.768), 'std_dev_duration_us': np.float64(0.42032010123238245), 'min_duration_us': np.float64(19.936), 'max_duration_us': np.float64(21.279)}]","[{'name': 'Memset (Device)', 'stream': 7, 'mean_duration_us': np.float64(0.86)}, {'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize128x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(20.74)}]","{'M': 5184, 'N': 1024, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.07621652149159028,99.35513223936249 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((3072,), (5, 768), (768, 3072), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (768, 1), (1, 768), (), ())","('', '', '', '1', '1')",3090,60,1736.734,28.945566666666664,3.9998917034774846,0.02360832,4.54248046875,4.9564656562399225,matrix_bf16,0.676291858187444,0.016913222300935213,3.3520173687007455,0.08382980547093656,7.047428385416667,0.17938446301713049,422.845703125,27.045499999999997,25.38,42.511,0.6827817266046057,0.6333940996039219,0.7054456557708997,3.3841841786239235,3.139396101551847,3.496517165172114,6.97607421875,6.751953125,7.52001953125,"[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(422.845), 'mean_duration_us': np.float64(7.047416666666667), 'median_duration_us': np.float64(6.976), 'std_dev_duration_us': np.float64(0.17790056133194818), 'min_duration_us': np.float64(6.752), 'max_duration_us': np.float64(7.52)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_64x64_64x6_tn_align8::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(422.845), 'mean_duration_us': np.float64(7.047416666666667), 'median_duration_us': np.float64(6.976), 'std_dev_duration_us': np.float64(0.17790056133194818), 'min_duration_us': np.float64(6.752), 'max_duration_us': np.float64(7.52)}]","[{'name': 'void cutlass::Kernel2 >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})', 'stream': 7, 'count': 120, 'total_duration_us': np.float64(287.13699999999994), 'mean_duration_us': np.float64(2.392808333333333), 'median_duration_us': np.float64(2.368), 'std_dev_duration_us': np.float64(0.24387456802740948), 'min_duration_us': np.float64(1.984), 'max_duration_us': np.float64(2.913)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.39)}]","{'shape_in1': (1, 12, 5, 5), 'shape_in2': (1, 1, 5, 5), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (300, 25, 5, 1), 'stride_input2': (25, 25, 5, 1), 'stride_output': None}",True,0.050662586775340884,99.48040243718685 aten::cat,other,python3,CPU,thread 21587 (python3),"(((1, 1, 1024), (1, 5184, 1024)), ())","('TensorList', 'Scalar')","(((1024, 1024, 1), (5308416, 1, 5184)), ())","('', '1')",50,5,195.19400000000002,39.0388,1.6605418392801794,,,,,,,,,43.961181640625,0.536907726929032,219.805908203125,38.851,37.771,41.831,,,,,,,43.7119140625,43.551025390625,44.799072265625,"[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy, unsigned int, 3, 64, 64>(at::native::(anonymous namespace)::OpaqueType<2u>*, at::native::(anonymous namespace)::CatArrInputTensorMetadata, unsigned int, 64, 64>, at::native::(anonymous namespace)::TensorSizeStride, int, unsigned int)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(219.806), 'mean_duration_us': np.float64(43.961200000000005), 'median_duration_us': np.float64(43.712), 'std_dev_duration_us': np.float64(0.48020553932665044), 'min_duration_us': np.float64(43.551), 'max_duration_us': np.float64(44.799)}]","[{'name': 'void at::native::(anonymous namespace)::CatArrayBatchedCopy(at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, int, int, unsigned int, unsigned int, long)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(195.071), 'mean_duration_us': np.float64(39.0142), 'median_duration_us': np.float64(39.072), 'std_dev_duration_us': np.float64(0.2134070289376613), 'min_duration_us': np.float64(38.624), 'max_duration_us': np.float64(39.231)}]","[{'name': 'void at::native::(anonymous namespace)::indexSelectLargeIndex, std::array >(int, at::native::CUDAFunctor_add, std::array)', 'stream': 7, 'count': 125, 'total_duration_us': np.float64(187.747), 'mean_duration_us': np.float64(1.5019760000000002), 'median_duration_us': np.float64(1.44), 'std_dev_duration_us': np.float64(0.09206636423797784), 'min_duration_us': np.float64(1.344), 'max_duration_us': np.float64(1.633)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::CU...', 'stream': 7, 'mean_duration_us': np.float64(1.5)}]","{'shape_in1': (1, 5, 768), 'shape_in2': (1, 5, 768), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (3840, 768, 1), 'stride_input2': (3840, 768, 1), 'stride_output': None}",True,0.03312632448302226,99.58673035299077 -aten::bmm,GEMM,python3,CPU,thread 21587 (python3),"((12, 5, 64), (12, 64, 5))","('c10::BFloat16', 'c10::BFloat16')","((320, 64, 1), (320, 1, 64))","('', '')",3046,60,1214.72,20.245333333333335,2.022170981946381,3.84e-05,0.01522064208984375,2.406015037593985,matrix_bf16,0.005157845853882534,7.520311885488376e-05,0.012409854686033166,0.0001809398348388166,3.09495849609375,0.04488577520163302,185.697509765625,19.445999999999998,18.391,25.8,0.005141746106654082,0.0050379284833538835,0.0053057511565619666,0.012371118452100047,0.012121331689272503,0.012765717068419771,3.10400390625,3.008056640625,3.16796875,"[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_tn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(185.69699999999997), 'mean_duration_us': np.float64(3.0949499999999994), 'median_duration_us': np.float64(3.104), 'std_dev_duration_us': np.float64(0.04453478977159322), 'min_duration_us': np.float64(3.008), 'max_duration_us': np.float64(3.168)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_tn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(185.69699999999997), 'mean_duration_us': np.float64(3.0949499999999994), 'median_duration_us': np.float64(3.104), 'std_dev_duration_us': np.float64(0.04453478977159322), 'min_duration_us': np.float64(3.008), 'max_duration_us': np.float64(3.168)}]","[{'name': 'void cutlass::Kernel2 >(int, at::native::GeluCUDAKernelImpl(at::TensorIteratorBase&, at::native::GeluType)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 20, 'total_duration_us': np.float64(176.76800000000006), 'mean_duration_us': np.float64(8.838400000000004), 'median_duration_us': np.float64(8.832), 'std_dev_duration_us': np.float64(0.06681796165702758), 'min_duration_us': np.float64(8.704), 'max_duration_us': np.float64(8.992)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Ge...', 'stream': 7, 'mean_duration_us': np.float64(8.84)}]",,False,0.031189342770556128,99.65068447846807 -aten::bmm,GEMM,python3,CPU,thread 21587 (python3),"((12, 5, 5), (12, 5, 64))","('c10::BFloat16', 'c10::BFloat16')","((25, 5, 1), (320, 64, 1))","('', '')",3057,60,1222.686,20.3781,1.7533072835383063,3.84e-05,0.01522064208984375,2.406015037593985,matrix_bf16,0.005720637383200418,7.720630551923328e-05,0.01376393956860251,0.0001857595320763497,2.790397135416667,0.037545531768877524,167.423828125,19.674999999999997,18.88,27.731,0.00566777874111323,0.0055418921668362155,0.005867710259402208,0.013636760880873938,0.01333387589013225,0.014117799120366216,2.81591796875,2.719970703125,2.8798828125,"[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_nn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(167.428), 'mean_duration_us': np.float64(2.7904666666666667), 'median_duration_us': np.float64(2.816), 'std_dev_duration_us': np.float64(0.037250264726874344), 'min_duration_us': np.float64(2.72), 'max_duration_us': np.float64(2.88)}]","[{'name': 'void cutlass::Kernel2(cutlass_75_wmma_tensorop_bf16_s161616gemm_bf16_32x32_32x1_nn_align1::Params)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(167.428), 'mean_duration_us': np.float64(2.7904666666666667), 'median_duration_us': np.float64(2.816), 'std_dev_duration_us': np.float64(0.037250264726874344), 'min_duration_us': np.float64(2.72), 'max_duration_us': np.float64(2.88)}]","[{'name': 'void cutlass::Kernel2(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(163.17200000000005), 'mean_duration_us': np.float64(2.719533333333334), 'median_duration_us': np.float64(2.784), 'std_dev_duration_us': np.float64(0.14011381881250043), 'min_duration_us': np.float64(2.527), 'max_duration_us': np.float64(3.008)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.72)}]","{'op_shape': (1, 5, 12, 64), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (3840, 768, 64, 1), 'stride_output': (3840, 64, 320, 1)}",True,0.02879014930310264,99.70901517118162 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((1,), (5184, 1024), (1024, 1), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",4613,15,485.997,32.3998,4.716143751595606,0.010622016,10.136842727661133,0.9993194251711081,matrix_bf16,1.3512750024987534,0.06778795277493481,1.3503553587451418,0.06774181800057408,7.884765625,0.3991367711291624,118.271484375,33.77,25.26,40.071,1.3669515855572998,1.2534521794207405,1.4441704978936545,1.3660212727158554,1.252599111418207,1.4431876318041597,7.77587890625,7.360107421875,8.47998046875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(35.168), 'mean_duration_us': np.float64(2.344533333333333), 'median_duration_us': np.float64(2.368), 'std_dev_duration_us': np.float64(0.10418372660300108), 'min_duration_us': np.float64(2.176), 'max_duration_us': np.float64(2.56)}, {'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(83.104), 'mean_duration_us': np.float64(5.540266666666667), 'median_duration_us': np.float64(5.376), 'std_dev_duration_us': np.float64(0.32818683025916123), 'min_duration_us': np.float64(5.152), 'max_duration_us': np.float64(6.176)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void gemv2T_kernel_val(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(35.168), 'mean_duration_us': np.float64(2.344533333333333), 'median_duration_us': np.float64(2.368), 'std_dev_duration_us': np.float64(0.10418372660300108), 'min_duration_us': np.float64(2.176), 'max_duration_us': np.float64(2.56)}, {'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 15, 'total_duration_us': np.float64(83.104), 'mean_duration_us': np.float64(5.540266666666667), 'median_duration_us': np.float64(5.376), 'std_dev_duration_us': np.float64(0.32818683025916123), 'min_duration_us': np.float64(5.152), 'max_duration_us': np.float64(6.176)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.34)}, {'name': 'void gemv2T_kernel_val(c10::BFloat16*, c10::BFloat16 const*, int, int, int, bool const*, int, bool)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(115.74799999999999), 'mean_duration_us': np.float64(1.9291333333333331), 'median_duration_us': np.float64(1.857), 'std_dev_duration_us': np.float64(0.09073284349610612), 'min_duration_us': np.float64(1.824), 'max_duration_us': np.float64(2.049)}]","[{'name': 'void (anonymous namespace)::softmax_warp_forward >(int, at::native::sigmoid_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#4}::operator()() const::{lambda(c10::BFloat16)#1}, std::array)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(105.02100000000002), 'mean_duration_us': np.float64(1.7503500000000003), 'median_duration_us': np.float64(1.76), 'std_dev_duration_us': np.float64(0.021235838418422137), 'min_duration_us': np.float64(1.696), 'max_duration_us': np.float64(1.792)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::si...', 'stream': 7, 'mean_duration_us': np.float64(1.75)}]","{'op_shape': (1, 5, 3072), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (15360, 3072, 1), 'stride_output': None}",True,0.01853025448675789,99.76883635680612 aten::mul,elementwise,python3,CPU,thread 21587 (python3),"((1, 5, 3072), ())","('c10::BFloat16', 'double')","((15360, 3072, 1), ())","('', '')",3093,60,830.406,13.8401,1.4801328124100492,1.536e-05,0.05860137939453125,0.24996745215466734,vector_bf16,0.03920225464616464,0.0007341958893775424,0.009799287712620246,0.00018352507585013398,1.56800537109375,0.029571126288607766,94.080322265625,13.504999999999999,12.501,23.49,0.0391858956873735,0.03762759874420691,0.040033562589470335,0.009795198505371323,0.009405674988787562,0.010007087641164307,1.568115234375,1.534912109375,1.633056640625,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(94.07899999999998), 'mean_duration_us': np.float64(1.567983333333333), 'median_duration_us': np.float64(1.568), 'std_dev_duration_us': np.float64(0.02925092116308288), 'min_duration_us': np.float64(1.535), 'max_duration_us': np.float64(1.633)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(1.57)}]","{'shape_in1': (1, 5, 3072), 'shape_in2': (), 'dtype_in1_in2_out': ('c10::BFloat16', 'double', None), 'stride_input1': (15360, 3072, 1), 'stride_input2': (), 'stride_output': None}",True,0.016599691185428117,99.78543604799155 aten::mul,elementwise,python3,CPU,thread 21587 (python3),"((1, 5, 3072), (1, 5, 3072))","('c10::BFloat16', 'c10::BFloat16')","((15360, 3072, 1), (15360, 3072, 1))","('', '')",3095,60,482.448,8.040799999999999,0.7044778039592068,1.536e-05,0.087890625,0.16666666666666666,vector_bf16,0.06203925053109231,0.0007057964384430683,0.010339875088515386,0.00011763273974051123,1.4857014973958333,0.01701343762587246,89.14208984375,7.8455,7.5,10.26,0.06261193564438547,0.06004252584698584,0.06265350373443983,0.010435322607397579,0.010007087641164307,0.010442250622406639,1.471923828125,1.470947265625,1.534912109375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(89.14699999999999), 'mean_duration_us': np.float64(1.4857833333333332), 'median_duration_us': np.float64(1.472), 'std_dev_duration_us': np.float64(0.016880058912482752), 'min_duration_us': np.float64(1.471), 'max_duration_us': np.float64(1.535)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::Bi...', 'stream': 7, 'mean_duration_us': np.float64(1.49)}]","{'shape_in1': (1, 5, 3072), 'shape_in2': (1, 5, 3072), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (15360, 3072, 1), 'stride_input2': (15360, 3072, 1), 'stride_output': None}",True,0.015728381104521377,99.80116442909608 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((768,), (5184, 1024), (1024, 768), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",4574,5,309.125,61.825,1.7314820241631146,8.157708288,19.22021484375,404.7715875314382,matrix_bf16,1.1611975590495807,0.005419567689335076,470.0197794141297,2.1936870173462433,17.356396484375,0.08054890760489654,86.781982421875,61.251,60.521,64.821,1.1598689677963243,1.1577380219065116,1.170708864691617,469.4820034233688,468.6194570726056,473.86968569835346,17.3759765625,17.215087890625,17.407958984375,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(86.78200000000001), 'mean_duration_us': np.float64(17.3564), 'median_duration_us': np.float64(17.376), 'std_dev_duration_us': np.float64(0.07209049868047837), 'min_duration_us': np.float64(17.215), 'max_duration_us': np.float64(17.408)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(17.36)}]","{'M': 5184, 'N': 768, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1}",True,0.015311959759184679,99.81647638885526 +aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((768,), (5184, 1024), (1024, 768), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",4574,5,309.125,61.825,1.7314820241631146,8.157708288,19.22021484375,404.7715875314382,matrix_bf16,1.1611975590495807,0.005419567689335076,470.0197794141297,2.1936870173462433,17.356396484375,0.08054890760489654,86.781982421875,61.251,60.521,64.821,1.1598689677963243,1.1577380219065116,1.170708864691617,469.4820034233688,468.6194570726056,473.86968569835346,17.3759765625,17.215087890625,17.407958984375,"[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__5x_cublas', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(86.78200000000001), 'mean_duration_us': np.float64(17.3564), 'median_duration_us': np.float64(17.376), 'std_dev_duration_us': np.float64(0.07209049868047837), 'min_duration_us': np.float64(17.215), 'max_duration_us': np.float64(17.408)}]","[{'name': 'sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warp...', 'stream': 7, 'mean_duration_us': np.float64(17.36)}]","{'M': 5184, 'N': 768, 'K': 1024, 'bias': True, 'stride_A': (1024, 1), 'stride_B': (1, 1024), 'dtype_A_B': ('c10::BFloat16', 'c10::BFloat16'), 'B': 1, 'transpose': (True, False)}",True,0.015311959759184679,99.81647638885526 aten::mul,elementwise,python3,CPU,thread 21587 (python3),"((1, 5, 768), ())","('c10::BFloat16', 'double')","((3840, 768, 1), ())","('', '')",2998,60,771.562,12.859366666666666,0.6683400628033774,3.84e-06,0.01465606689453125,0.24986985944820403,vector_bf16,0.01071372856755311,0.0007718860894261398,0.0026770378513407043,0.00019287106867493307,1.4415568033854167,0.10085870516947602,86.493408203125,12.690000000000001,11.981,15.2,0.010440757671255598,0.009800300171259536,0.012005975205035284,0.0026088306518493947,0.0024487996263428307,0.0029999313370207893,1.471923828125,1.280029296875,1.568115234375,"[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor >, std::array >(int, at::native::AUnaryFunctor >, std::array)', 'stream': 7, 'count': 60, 'total_duration_us': np.float64(86.49500000000002), 'mean_duration_us': np.float64(1.4415833333333337), 'median_duration_us': np.float64(1.472), 'std_dev_duration_us': np.float64(0.10003121040733015), 'min_duration_us': np.float64(1.28), 'max_duration_us': np.float64(1.568)}]","[{'name': 'void at::native::vectorized_elementwise_kernel<4, at::native::AU...', 'stream': 7, 'mean_duration_us': np.float64(1.44)}]","{'shape_in1': (1, 5, 768), 'shape_in2': (), 'dtype_in1_in2_out': ('c10::BFloat16', 'double', None), 'stride_input1': (3840, 768, 1), 'stride_input2': (), 'stride_output': None}",True,0.015261043235941896,99.8317374320912 aten::layer_norm,NORM_fwd,python3,CPU,thread 21587 (python3),"((1, 5184, 1024), (), (1024,), (1024,), (), ())","('c10::BFloat16', 'ScalarList', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((5308416, 1024, 1), (), (1,), (1,), (), ())","('', '[1024]', '', '', '1.0000000000000001e-05', 'True')",4550,5,133.12,26.624000000000002,4.62429778452902,0.026548224,20.26171875,1.2495662232504339,vector_bf16,1.5077311487443041,0.015796372462077515,1.884009917213458,0.01973861347849532,14.092578125,0.1477430461918904,70.462890625,24.69,23.72,34.83,1.5090417457168623,1.4886485920147798,1.5262982214115337,1.8856475949226608,1.8601649988709843,1.9072107040830646,14.0791015625,13.919921875,14.27197265625,"[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_kernel(int, float, c10::BFloat16 const*, c10::BFloat16 const*, c10::BFloat16 const*, float*, float*, c10::BFloat16*)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(70.463), 'mean_duration_us': np.float64(14.0926), 'median_duration_us': np.float64(14.079), 'std_dev_duration_us': np.float64(0.1321144957981524), 'min_duration_us': np.float64(13.92), 'max_duration_us': np.float64(14.272)}]","[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_ke...', 'stream': 7, 'mean_duration_us': np.float64(14.09)}]","{'op_shape': (1, 5184, 1024), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (5308416, 1024, 1), 'stride_output': None, 'num_channels': 1024, 'has_bias': True, 'is_affine': True, 'is_training': True}",True,0.01243259160087899,99.84417002369209 aten::mul,elementwise,python3,CPU,thread 21587 (python3),"((1, 5184, 1024), (1, 5184, 1024))","('c10::BFloat16', 'c10::BFloat16')","((5309440, 1024, 1), (5309440, 0, 1))","('', '')",4549,5,81.06,16.212,0.8936274391489991,0.005308416,30.375,0.16666666666666666,vector_bf16,2.4348515656969374,0.012220510682827207,0.4058085942828229,0.002036751780471206,13.08134765625,0.06585518566848884,65.40673828125,16.08,15.25,17.38,2.433585129383674,2.4158296288285617,2.445720663192231,0.40559752156394563,0.40263827147142695,0.4076201105320385,13.087890625,13.02294921875,13.18408203125,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(65.407), 'mean_duration_us': np.float64(13.081399999999999), 'median_duration_us': np.float64(13.088), 'std_dev_duration_us': np.float64(0.058854396607220374), 'min_duration_us': np.float64(13.023), 'max_duration_us': np.float64(13.184)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(13.08)}]","{'shape_in1': (1, 5184, 1024), 'shape_in2': (1, 5184, 1024), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (5309440, 1024, 1), 'stride_input2': (5309440, 0, 1), 'stride_output': None}",True,0.011540475529510099,99.8557104992216 aten::div,elementwise,python3,CPU,thread 21587 (python3),"((1, 5184, 768), (1, 5184, 1))","('c10::BFloat16', 'c10::BFloat16')","((3981312, 768, 1), (5184, 1, 1))","('', '')",4580,5,51.49,10.298,0.28630403420140665,0.003981312,15.1973876953125,0.2498373454782043,vector_bf16,1.3826359183838934,0.01195742787988091,0.345434087611851,0.002987412040256516,11.526220703125,0.0993899153066534,57.631103515625,10.2,9.99,10.63,1.379467910814295,1.3718716899472456,1.3988316646521795,0.34464260101020777,0.3427447813531179,0.34948038986755814,11.552001953125,11.39208984375,11.615966796875,"[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(57.631), 'mean_duration_us': np.float64(11.5262), 'median_duration_us': np.float64(11.552), 'std_dev_duration_us': np.float64(0.0889413289759043), 'min_duration_us': np.float64(11.392), 'max_duration_us': np.float64(11.616)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(11.53)}]","{'shape_in1': (1, 5184, 768), 'shape_in2': (1, 5184, 1), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (3981312, 768, 1), 'stride_input2': (5184, 1, 1), 'stride_output': None}",True,0.010168529379967471,99.86587902860157 -aten::addmm,GEMM,python3,CPU,thread 21587 (python3),"((4,), (5184, 1024), (1024, 4), (), ())","('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((1,), (1024, 1), (1, 1024), (), ())","('', '', '', '1', '1')",4705,5,189.091,37.818200000000004,6.091385408919716,0.042488064,10.172370910644531,3.98331674558037,matrix_bf16,0.9433046851197732,0.009662687965852772,3.7574813484220106,0.03848954678169942,11.308544921875,0.11635617708924226,56.542724609375,34.69,32.26,45.5,0.944281152935074,0.9285668824041996,0.9550979447359216,3.7613709290222186,3.6987760120720066,3.8044576369359917,11.2958984375,11.16796875,11.487060546875,"[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align2::Params)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(56.54299999999999), 'mean_duration_us': np.float64(11.308599999999998), 'median_duration_us': np.float64(11.296), 'std_dev_duration_us': np.float64(0.10403768547983014), 'min_duration_us': np.float64(11.168), 'max_duration_us': np.float64(11.487)}]","[{'name': 'void cutlass::Kernel2(cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_tn_align2::Params)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(56.54299999999999), 'mean_duration_us': np.float64(11.308599999999998), 'median_duration_us': np.float64(11.296), 'std_dev_duration_us': np.float64(0.10403768547983014), 'min_duration_us': np.float64(11.168), 'max_duration_us': np.float64(11.487)}]","[{'name': 'void cutlass::Kernel2, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp, unsigned int, c10::BFloat16, 4>)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(31.743), 'mean_duration_us': np.float64(3.1742999999999997), 'median_duration_us': np.float64(3.216), 'std_dev_duration_us': np.float64(0.22374005005809758), 'min_duration_us': np.float64(2.848), 'max_duration_us': np.float64(3.551)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp(at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, int, int, unsigned int, long)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(28.159999999999997), 'mean_duration_us': np.float64(5.632), 'median_duration_us': np.float64(5.792), 'std_dev_duration_us': np.float64(0.2638787600395301), 'min_duration_us': np.float64(5.28), 'max_duration_us': np.float64(5.888)}]","[{'name': 'void at::native::(anonymous namespace)::indexSelectSmallIndex > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast > >(at::TensorIteratorBase&, at::native::BinaryFunctor > const&)::{lambda(int)#1})', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(26.686999999999998), 'mean_duration_us': np.float64(2.6687), 'median_duration_us': np.float64(2.656), 'std_dev_duration_us': np.float64(0.20862696374150683), 'min_duration_us': np.float64(2.335), 'max_duration_us': np.float64(3.008)}]","[{'name': 'void at::native::elementwise_kernel<128, 4, at::native::gpu_kern...', 'stream': 7, 'mean_duration_us': np.float64(2.67)}]","{'shape_in1': (1, 768), 'shape_in2': (1, 1), 'dtype_in1_in2_out': ('c10::BFloat16', 'c10::BFloat16', None), 'stride_input1': (768, 1), 'stride_input2': (1, 1), 'stride_output': None}",True,0.004708701485337204,99.89113353797683 aten::index_select,other,python3,CPU,thread 21587 (python3),"((16, 768), (), (5,))","('c10::BFloat16', 'Scalar', 'long int')","((768, 1), (), (1,))","('', '0', '')",2926,5,70.461,14.0922,0.47537795489484025,,,,,,,,,5.292626953125,0.21976044118159957,26.463134765625,13.95,13.721,14.92,,,,,,,5.216064453125,5.087890625,5.59912109375,"[{'name': 'void at::native::(anonymous namespace)::indexSelectSmallIndex(at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, at::cuda::detail::TensorInfo, int, int, unsigned int, long)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(26.463), 'mean_duration_us': np.float64(5.2926), 'median_duration_us': np.float64(5.216), 'std_dev_duration_us': np.float64(0.19653355947522047), 'min_duration_us': np.float64(5.088), 'max_duration_us': np.float64(5.599)}]","[{'name': 'void at::native::(anonymous namespace)::indexSelectSmallIndex, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp, unsigned int, c10::BFloat16, 4>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(24.704), 'mean_duration_us': np.float64(4.9408), 'median_duration_us': np.float64(4.96), 'std_dev_duration_us': np.float64(0.05935115837117255), 'min_duration_us': np.float64(4.832), 'max_duration_us': np.float64(4.992)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(23.615000000000002), 'mean_duration_us': np.float64(4.723000000000001), 'median_duration_us': np.float64(4.704), 'std_dev_duration_us': np.float64(0.06861486719363401), 'min_duration_us': np.float64(4.64), 'max_duration_us': np.float64(4.831)}]","[{'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched > >(cublasDotParams, cublasGemvTensorStridedBatched >)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(10.143), 'mean_duration_us': np.float64(2.0286), 'median_duration_us': np.float64(2.047), 'std_dev_duration_us': np.float64(0.02545270123189289), 'min_duration_us': np.float64(1.984), 'max_duration_us': np.float64(2.048)}, {'name': 'void reduce_1Block_kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16> >(float const*, float, cublasGemvTensorStridedBatched, int, float const*, float, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, cublasPointerMode_t, cublasLtEpilogue_t, cublasGemvTensorStridedBatched::value_type, float>::type const>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(13.183000000000002), 'mean_duration_us': np.float64(2.6366000000000005), 'median_duration_us': np.float64(2.688), 'std_dev_duration_us': np.float64(0.08984341934721768), 'min_duration_us': np.float64(2.527), 'max_duration_us': np.float64(2.72)}]","[{'name': 'void dot_kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>, float, float)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(23.615000000000002), 'mean_duration_us': np.float64(4.723000000000001), 'median_duration_us': np.float64(4.704), 'std_dev_duration_us': np.float64(0.06861486719363401), 'min_duration_us': np.float64(4.64), 'max_duration_us': np.float64(4.831)}]","[{'name': 'void gemv2T_kernel_val, cublasGemvTensorStridedBatched > >(cublasDotParams, cublasGemvTensorStridedBatched >)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(10.143), 'mean_duration_us': np.float64(2.0286), 'median_duration_us': np.float64(2.047), 'std_dev_duration_us': np.float64(0.02545270123189289), 'min_duration_us': np.float64(1.984), 'max_duration_us': np.float64(2.048)}, {'name': 'void reduce_1Block_kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16> >(float const*, float, cublasGemvTensorStridedBatched, int, float const*, float, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, cublasPointerMode_t, cublasLtEpilogue_t, cublasGemvTensorStridedBatched::value_type, float>::type const>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(13.183000000000002), 'mean_duration_us': np.float64(2.6366000000000005), 'median_duration_us': np.float64(2.688), 'std_dev_duration_us': np.float64(0.08984341934721768), 'min_duration_us': np.float64(2.527), 'max_duration_us': np.float64(2.72)}]","[{'name': 'void dot_kernel >(at::TensorIteratorBase&, c10::ArrayRef, c10::ArrayRef)::{lambda(char*, char const*, long)#1}>(at::TensorIteratorBase&, c10::ArrayRef, c10::ArrayRef, at::native::index_kernel_impl >(at::TensorIteratorBase&, c10::ArrayRef, c10::ArrayRef)::{lambda(char*, char const*, long)#1} const&)::{lambda(int)#1}>(long, at::native::gpu_index_kernel >(at::TensorIteratorBase&, c10::ArrayRef, c10::ArrayRef)::{lambda(char*, char const*, long)#1}>(at::TensorIteratorBase&, c10::ArrayRef, c10::ArrayRef, at::native::index_kernel_impl >(at::TensorIteratorBase&, c10::ArrayRef, c10::ArrayRef)::{lambda(char*, char const*, long)#1} const&)::{lambda(int)#1})', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(22.176000000000002), 'mean_duration_us': np.float64(4.4352), 'median_duration_us': np.float64(4.256), 'std_dev_duration_us': np.float64(0.26400560600108475), 'min_duration_us': np.float64(4.192), 'max_duration_us': np.float64(4.832)}]","[{'name': 'void at::native::index_elementwise_kernel<128, 4, at::native::gp...', 'stream': 7, 'mean_duration_us': np.float64(4.44)}]",,False,0.00391277542784406,99.91235676823943 aten::add_,elementwise,python3,CPU,thread 21587 (python3),"((1, 5184, 4), (5184, 4), ())","('c10::BFloat16', 'float', 'Scalar')","((20736, 4, 1), (4, 1), ())","('', '', '1')",4712,5,421.52700000000004,84.3054,9.921064776524748,2.0736e-05,0.19775390625,0.1,vector_bf16,0.04684234817911755,0.0011139861040140278,0.0046842348179117555,0.00011139861040140246,4.42880859375,0.10751042271143953,22.14404296875,81.631,74.171,95.092,0.04695635559486953,0.04500087739747802,0.04799924046340775,0.0046956355594869525,0.004500087739747801,0.004799924046340774,4.416015625,4.320068359375,4.60791015625,"[{'name': 'void at::native::unrolled_elementwise_kernel, std::array, 4, TrivialOffsetCalculator<2, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<2>, at::native::memory::StoreWithCast<1> >(int, at::native::CUDAFunctor_add, std::array, TrivialOffsetCalculator<2, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<2>, at::native::memory::StoreWithCast<1>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(22.144000000000002), 'mean_duration_us': np.float64(4.428800000000001), 'median_duration_us': np.float64(4.416), 'std_dev_duration_us': np.float64(0.09621309682158638), 'min_duration_us': np.float64(4.32), 'max_duration_us': np.float64(4.608)}]","[{'name': 'void at::native::unrolled_elementwise_kernel Device)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(20.255), 'mean_duration_us': np.float64(4.051), 'median_duration_us': np.float64(4.032), 'std_dev_duration_us': np.float64(0.02550294100687212), 'min_duration_us': np.float64(4.032), 'max_duration_us': np.float64(4.096)}]","[{'name': 'Memcpy HtoD (Pageable -> Device)', 'stream': 7, 'mean_duration_us': np.float64(4.05)}]","{'op_shape': (5184, 4), 'dtype_in_out': ('float', 'float'), 'stride_input': (4, 1), 'stride_output': (4, 1)}",True,0.003573805781991728,99.91983770641667 aten::layer_norm,NORM_fwd,python3,CPU,thread 21587 (python3),"((1, 1024), (), (1024,), (1024,), (), ())","('c10::BFloat16', 'ScalarList', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar')","((5309440, 1), (), (1,), (1,), (), ())","('', '[1024]', '', '', '1.0000000000000001e-05', 'True')",2904,5,152.483,30.4966,1.8555972892845027,1.1264e-05,0.015625,0.6875,vector_bf16,0.004074626006048601,0.0001555257392488542,0.002801305379158413,0.00010692394573358708,4.025634765625,0.15245814796610938,20.128173828125,29.71,28.911,33.66,0.004096,0.003878676684776326,0.004302401846390563,0.002816,0.002666590220783724,0.0029579012693935116,4.0,3.80810546875,4.22412109375,"[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_kernel(int, float, c10::BFloat16 const*, c10::BFloat16 const*, c10::BFloat16 const*, float*, float*, c10::BFloat16*)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(20.128), 'mean_duration_us': np.float64(4.0256), 'median_duration_us': np.float64(4.0), 'std_dev_duration_us': np.float64(0.13636656481704024), 'min_duration_us': np.float64(3.808), 'max_duration_us': np.float64(4.224)}]","[{'name': 'void at::native::(anonymous namespace)::vectorized_layer_norm_ke...', 'stream': 7, 'mean_duration_us': np.float64(4.03)}]","{'op_shape': (1, 1024), 'dtype_in_out': ('c10::BFloat16', None), 'stride_input': (5309440, 1), 'stride_output': None, 'num_channels': 1024, 'has_bias': True, 'is_affine': True, 'is_training': True}",True,0.0035514490344764962,99.92338915545115 -aten::mm,GEMM,python3,CPU,thread 21587 (python3),"((1, 768), (768, 768))","('c10::BFloat16', 'c10::BFloat16')","((768, 1), (1, 768))","('', '')",4499,5,170.004,34.0008,0.48391135551875536,0.001179648,1.1279296875,0.9974025974025974,matrix_bf16,0.31485008481014465,0.003056303467113609,0.3140322923820663,0.0030483650165496886,3.75673828125,0.03655128437400035,18.78369140625,34.071,33.491,34.471,0.31590617019889145,0.31057963328631877,0.31862806629834256,0.31508563469188133,0.3097729329401205,0.31780046093133385,3.743896484375,3.7119140625,3.80810546875,"[{'name': 'std::enable_if::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(18.784), 'mean_duration_us': np.float64(3.7567999999999997), 'median_duration_us': np.float64(3.744), 'std_dev_duration_us': np.float64(0.03263372488699366), 'min_duration_us': np.float64(3.712), 'max_duration_us': np.float64(3.808)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(18.784), 'mean_duration_us': np.float64(3.7567999999999997), 'median_duration_us': np.float64(3.744), 'std_dev_duration_us': np.float64(0.03263372488699366), 'min_duration_us': np.float64(3.712), 'max_duration_us': np.float64(3.808)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel, unsigned int, long, 4> >(at::native::ReduceOp, unsigned int, long, 4>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(18.527), 'mean_duration_us': np.float64(3.7054), 'median_duration_us': np.float64(3.712), 'std_dev_duration_us': np.float64(0.13307982566865645), 'min_duration_us': np.float64(3.488), 'max_duration_us': np.float64(3.871)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(17.918), 'mean_duration_us': np.float64(3.5835999999999997), 'median_duration_us': np.float64(3.552), 'std_dev_duration_us': np.float64(0.049834124854360604), 'min_duration_us': np.float64(3.551), 'max_duration_us': np.float64(3.68)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel::type internal::gemvx::kernel, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float> >(cublasGemvParamsEx, cublasGemvTensorStridedBatched<__nv_bfloat16 const>, cublasGemvTensorStridedBatched<__nv_bfloat16>, float>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(17.918), 'mean_duration_us': np.float64(3.5835999999999997), 'median_duration_us': np.float64(3.552), 'std_dev_duration_us': np.float64(0.049834124854360604), 'min_duration_us': np.float64(3.551), 'max_duration_us': np.float64(3.68)}]","[{'name': 'std::enable_if::type internal::gemvx::kernel Device)', 'stream': 7, 'count': 10, 'total_duration_us': np.float64(17.44), 'mean_duration_us': np.float64(1.7440000000000002), 'median_duration_us': np.float64(1.7440000000000002), 'std_dev_duration_us': np.float64(0.9569284194755635), 'min_duration_us': np.float64(0.768), 'max_duration_us': np.float64(2.72)}]","[{'name': 'Memcpy HtoD (Pageable -> Device)', 'stream': 7, 'mean_duration_us': np.float64(1.74)}]","{'op_shape': (1, 2), 'dtype_in_out': ('int', 'int'), 'stride_input': (2, 1), 'stride_output': (2, 1)}",True,0.0030771758357429615,99.93621098707234 aten::linalg_norm,reduce,python3,CPU,thread 21587 (python3),"((1, 1, 768), (), (), (), ())","('c10::BFloat16', '', 'ScalarList', 'Scalar', '')","((768, 768, 1), (), (), (), ())","('', '', '[-1]', 'True', '')",4581,5,56.662,11.3324,0.30352644036393284,7.68e-07,0.0014667510986328125,0.49934980494148246,vector_bf16,0.00048495757643147807,3.789683268713756e-05,0.00024216347119595264,1.892377601022212e-05,3.187255859375,0.25357873774649625,15.936279296875,11.211,11.071,11.85,0.0004954890671700488,0.000440935675789179,0.0005224021892362551,0.00024742236904200097,0.00022018114369706728,0.0002608614312961274,3.10400390625,2.944091796875,3.488037109375,"[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp, unsigned int, c10::BFloat16, 4>)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(15.936), 'mean_duration_us': np.float64(3.1872), 'median_duration_us': np.float64(3.104), 'std_dev_duration_us': np.float64(0.2268165778773677), 'min_duration_us': np.float64(2.944), 'max_duration_us': np.float64(3.488)}]","[{'name': 'void at::native::reduce_kernel<512, 1, at::native::ReduceOp Pageable)', 'stream': 7, 'count': 5, 'total_duration_us': np.float64(14.717999999999998), 'mean_duration_us': np.float64(2.9435999999999996), 'median_duration_us': np.float64(2.944), 'std_dev_duration_us': np.float64(0.0004898979485565817), 'min_duration_us': np.float64(2.943), 'max_duration_us': np.float64(2.944)}]","[{'name': 'Memcpy DtoH (Device -> Pageable)', 'stream': 7, 'mean_duration_us': np.float64(2.94)}]","{'op_shape': (1, 5184, 4), 'dtype_in_out': ('c10::BFloat16', 'c10::BFloat16'), 'stride_input': (20736, 4, 1), 'stride_output': (20736, 4, 1)}",True,0.0025969580683059554,99.94161976921399 diff --git a/tests/traces/inference/sglang_decode/perf_csvs/GEMM.csv b/tests/traces/inference/sglang_decode/perf_csvs/GEMM.csv index 5f769439f..c3d2c8a93 100644 --- a/tests/traces/inference/sglang_decode/perf_csvs/GEMM.csv +++ b/tests/traces/inference/sglang_decode/perf_csvs/GEMM.csv @@ -1,10 +1,10 @@ -name,param: B,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: transpose,param: stride_A,param: stride_B,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aiter::gemm_a16w16,1,32,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,6.442450944,193.75,31.710967741935484,3.0281005673551444,3.0523747073479246,0.09384870781134186,2.759264130723115,3.1382935344697542,96.02399941073553,96.7937558810098,2.9760333460278057,87.49893584084036,99.51832503629507,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(2417.659), 'mean_duration_us': np.float64(67.15719444444444), 'median_duration_us': np.float64(66.55850000000001), 'std_dev_duration_us': np.float64(2.1321042529681073), 'min_duration_us': np.float64(64.736), 'max_duration_us': np.float64(73.629)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(67.16)}]","[[32, 4096], [24576, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",67.15719943576389,66.55859375,2.162312900832966,64.736328125,73.62890625,2417.6591796875,36,3603 -aten::mm,1,32,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,6.442450944,193.75,31.710967741935484,3.0281005673551444,3.0523747073479246,0.09384870781134186,2.759264130723115,3.1382935344697542,96.02399941073553,96.7937558810098,2.9760333460278057,87.49893584084036,99.51832503629507,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(2417.659), 'mean_duration_us': np.float64(67.15719444444444), 'median_duration_us': np.float64(66.55850000000001), 'std_dev_duration_us': np.float64(2.1321042529681073), 'min_duration_us': np.float64(64.736), 'max_duration_us': np.float64(73.629)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(67.16)}]","[[32, 4096], [4096, 24576]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",67.15719943576389,66.55859375,2.162312900832966,64.736328125,73.62890625,2417.6591796875,36,3618 -aiter::gemm_a16w16,1,32,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,3.221225472,97.0,31.670103092783506,2.5179118357004993,2.48711682149981,0.18476128484204382,2.1481037191766696,2.7963528144767222,79.74252741517458,78.76724614069502,5.851408938502671,68.03066624031679,88.56078191827311,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO1_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(1461.9070000000002), 'mean_duration_us': np.float64(40.60852777777778), 'median_duration_us': np.float64(40.8995), 'std_dev_duration_us': np.float64(2.950849194969241), 'min_duration_us': np.float64(36.373), 'max_duration_us': np.float64(47.35)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(40.61)}]","[[32, 12288], [4096, 12288], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[12288, 1], [12288, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",40.608425564236114,40.8994140625,2.9926972240857443,36.373046875,47.349609375,1461.9033203125,36,3698 -aten::mm,1,32,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(12288, 1)","(1, 12288)",1,3.221225472,97.0,31.670103092783506,2.5179118357004993,2.48711682149981,0.18476128484204382,2.1481037191766696,2.7963528144767222,79.74252741517458,78.76724614069502,5.851408938502671,68.03066624031679,88.56078191827311,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO1_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(1461.9070000000002), 'mean_duration_us': np.float64(40.60852777777778), 'median_duration_us': np.float64(40.8995), 'std_dev_duration_us': np.float64(2.950849194969241), 'min_duration_us': np.float64(36.373), 'max_duration_us': np.float64(47.35)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(40.61)}]","[[32, 12288], [12288, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[12288, 1], [1, 12288]]","['', '']",40.608425564236114,40.8994140625,2.9926972240857443,36.373046875,47.349609375,1461.9033203125,36,3713 -aiter::gemm_a16w16,1,32,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,1.610612736,48.625,31.588688946015424,2.3118616016609934,2.401485497079251,0.3492055932142782,0.34269780634320524,2.461955778375065,73.02867702110615,75.8597783756037,11.030946862254629,10.825374407057343,77.76995528193521,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA2048_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(903.515), 'mean_duration_us': np.float64(25.097638888888888), 'median_duration_us': np.float64(21.231), 'std_dev_duration_us': np.float64(20.9245116153716), 'min_duration_us': np.float64(20.71), 'max_duration_us': np.float64(148.781)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(25.1)}]","[[32, 4096], [6144, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",25.097737630208332,21.2314453125,21.22134930970088,20.7099609375,148.78125,903.5185546875,36,2306 -aten::mm,1,32,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,1.610612736,48.625,31.588688946015424,2.3118616016609934,2.401485497079251,0.3492055932142782,0.34269780634320524,2.461955778375065,73.02867702110615,75.8597783756037,11.030946862254629,10.825374407057343,77.76995528193521,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA2048_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(903.515), 'mean_duration_us': np.float64(25.097638888888888), 'median_duration_us': np.float64(21.231), 'std_dev_duration_us': np.float64(20.9245116153716), 'min_duration_us': np.float64(20.71), 'max_duration_us': np.float64(148.781)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(25.1)}]","[[32, 4096], [4096, 6144]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",25.097737630208332,21.2314453125,21.22134930970088,20.7099609375,148.78125,903.5185546875,36,2321 -aiter::gemm_a16w16,1,32,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,1.073741824,32.5,31.50769230769231,2.045055758676163,2.0135627422757225,0.13064427853186372,1.7048516918266645,2.2566353647180546,64.4349875964428,63.44271532585662,4.116299729742416,53.71594253632322,71.10137272219347,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB2048_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(602.3440000000002), 'mean_duration_us': np.float64(16.731777777777783), 'median_duration_us': np.float64(16.924500000000002), 'std_dev_duration_us': np.float64(1.079763299964681), 'min_duration_us': np.float64(15.102), 'max_duration_us': np.float64(19.989)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(16.73)}]","[[32, 4096], [4096, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",16.731689453125,16.9248046875,1.0951818785939924,15.1015625,19.9892578125,602.3408203125,36,3402 -aten::mm,1,32,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,1.073741824,32.5,31.50769230769231,2.045055758676163,2.0135627422757225,0.13064427853186372,1.7048516918266645,2.2566353647180546,64.4349875964428,63.44271532585662,4.116299729742416,53.71594253632322,71.10137272219347,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB2048_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(602.3440000000002), 'mean_duration_us': np.float64(16.731777777777783), 'median_duration_us': np.float64(16.924500000000002), 'std_dev_duration_us': np.float64(1.079763299964681), 'min_duration_us': np.float64(15.102), 'max_duration_us': np.float64(19.989)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(16.73)}]","[[32, 4096], [4096, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",16.731689453125,16.9248046875,1.0951818785939924,15.1015625,19.9892578125,602.3408203125,36,3417 -aten::mm,1,32,151936,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,39.829110784,1196.5234375,31.745303777219156,3.161654456289559,3.161654456289559,,3.161654456289559,3.161654456289559,100.36768115351072,100.36768115351072,,100.36768115351072,100.36768115351072,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_3_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 1, 'total_duration_us': np.float64(396.832), 'mean_duration_us': np.float64(396.832), 'median_duration_us': np.float64(396.832), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(396.832), 'max_duration_us': np.float64(396.832)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(396.83)}]","[[32, 4096], [4096, 151936]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",396.83203125,396.83203125,,396.83203125,396.83203125,396.83203125,1,60441 +name,param: B,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: transpose,param: stride_A,param: stride_B,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aiter::gemm_a16w16,1,32,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,256,32,64,1,6.442450944,193.75,31.710967741935484,3.0281005673551444,3.0523747073479246,0.09384870781134186,2.759264130723115,3.1382935344697542,96.02399941073553,96.7937558810098,2.9760333460278057,87.49893584084036,99.51832503629507,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(2417.659), 'mean_duration_us': np.float64(67.15719444444444), 'median_duration_us': np.float64(66.55850000000001), 'std_dev_duration_us': np.float64(2.1321042529681073), 'min_duration_us': np.float64(64.736), 'max_duration_us': np.float64(73.629)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(67.16)}]","[[32, 4096], [24576, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",67.15719943576389,66.55859375,2.162312900832966,64.736328125,73.62890625,2417.6591796875,36,3603 +aten::mm,1,32,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",256,32,64,1,6.442450944,193.75,31.710967741935484,3.0281005673551444,3.0523747073479246,0.09384870781134186,2.759264130723115,3.1382935344697542,96.02399941073553,96.7937558810098,2.9760333460278057,87.49893584084036,99.51832503629507,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(2417.659), 'mean_duration_us': np.float64(67.15719444444444), 'median_duration_us': np.float64(66.55850000000001), 'std_dev_duration_us': np.float64(2.1321042529681073), 'min_duration_us': np.float64(64.736), 'max_duration_us': np.float64(73.629)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x32x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(67.16)}]","[[32, 4096], [4096, 24576]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",67.15719943576389,66.55859375,2.162312900832966,64.736328125,73.62890625,2417.6591796875,36,3618 +aiter::gemm_a16w16,1,32,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,32,32,256,1,3.221225472,97.0,31.670103092783506,2.5179118357004993,2.48711682149981,0.18476128484204382,2.1481037191766696,2.7963528144767222,79.74252741517458,78.76724614069502,5.851408938502671,68.03066624031679,88.56078191827311,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO1_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(1461.9070000000002), 'mean_duration_us': np.float64(40.60852777777778), 'median_duration_us': np.float64(40.8995), 'std_dev_duration_us': np.float64(2.950849194969241), 'min_duration_us': np.float64(36.373), 'max_duration_us': np.float64(47.35)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(40.61)}]","[[32, 12288], [4096, 12288], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[12288, 1], [12288, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",40.608425564236114,40.8994140625,2.9926972240857443,36.373046875,47.349609375,1461.9033203125,36,3698 +aten::mm,1,32,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(12288, 1)","(1, 12288)",32,32,256,1,3.221225472,97.0,31.670103092783506,2.5179118357004993,2.48711682149981,0.18476128484204382,2.1481037191766696,2.7963528144767222,79.74252741517458,78.76724614069502,5.851408938502671,68.03066624031679,88.56078191827311,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO1_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(1461.9070000000002), 'mean_duration_us': np.float64(40.60852777777778), 'median_duration_us': np.float64(40.8995), 'std_dev_duration_us': np.float64(2.950849194969241), 'min_duration_us': np.float64(36.373), 'max_duration_us': np.float64(47.35)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(40.61)}]","[[32, 12288], [12288, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[12288, 1], [1, 12288]]","['', '']",40.608425564236114,40.8994140625,2.9926972240857443,36.373046875,47.349609375,1461.9033203125,36,3713 +aiter::gemm_a16w16,1,32,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,64,32,256,1,1.610612736,48.625,31.588688946015424,2.3118616016609934,2.401485497079251,0.3492055932142782,0.34269780634320524,2.461955778375065,73.02867702110615,75.8597783756037,11.030946862254629,10.825374407057343,77.76995528193521,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA2048_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(903.515), 'mean_duration_us': np.float64(25.097638888888888), 'median_duration_us': np.float64(21.231), 'std_dev_duration_us': np.float64(20.9245116153716), 'min_duration_us': np.float64(20.71), 'max_duration_us': np.float64(148.781)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(25.1)}]","[[32, 4096], [6144, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",25.097737630208332,21.2314453125,21.22134930970088,20.7099609375,148.78125,903.5185546875,36,2306 +aten::mm,1,32,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",64,32,256,1,1.610612736,48.625,31.588688946015424,2.3118616016609934,2.401485497079251,0.3492055932142782,0.34269780634320524,2.461955778375065,73.02867702110615,75.8597783756037,11.030946862254629,10.825374407057343,77.76995528193521,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA2048_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(903.515), 'mean_duration_us': np.float64(25.097638888888888), 'median_duration_us': np.float64(21.231), 'std_dev_duration_us': np.float64(20.9245116153716), 'min_duration_us': np.float64(20.71), 'max_duration_us': np.float64(148.781)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(25.1)}]","[[32, 4096], [4096, 6144]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",25.097737630208332,21.2314453125,21.22134930970088,20.7099609375,148.78125,903.5185546875,36,2321 +aiter::gemm_a16w16,1,32,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,16,32,512,1,1.073741824,32.5,31.50769230769231,2.045055758676163,2.0135627422757225,0.13064427853186372,1.7048516918266645,2.2566353647180546,64.4349875964428,63.44271532585662,4.116299729742416,53.71594253632322,71.10137272219347,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB2048_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(602.3440000000002), 'mean_duration_us': np.float64(16.731777777777783), 'median_duration_us': np.float64(16.924500000000002), 'std_dev_duration_us': np.float64(1.079763299964681), 'min_duration_us': np.float64(15.102), 'max_duration_us': np.float64(19.989)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(16.73)}]","[[32, 4096], [4096, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",16.731689453125,16.9248046875,1.0951818785939924,15.1015625,19.9892578125,602.3408203125,36,3402 +aten::mm,1,32,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",16,32,512,1,1.073741824,32.5,31.50769230769231,2.045055758676163,2.0135627422757225,0.13064427853186372,1.7048516918266645,2.2566353647180546,64.4349875964428,63.44271532585662,4.116299729742416,53.71594253632322,71.10137272219347,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB2048_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(602.3440000000002), 'mean_duration_us': np.float64(16.731777777777783), 'median_duration_us': np.float64(16.924500000000002), 'std_dev_duration_us': np.float64(1.079763299964681), 'min_duration_us': np.float64(15.102), 'max_duration_us': np.float64(19.989)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x32x512_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(16.73)}]","[[32, 4096], [4096, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",16.731689453125,16.9248046875,1.0951818785939924,15.1015625,19.9892578125,602.3408203125,36,3417 +aten::mm,1,32,151936,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",256,48,64,1,39.829110784,1196.5234375,31.745303777219156,3.161654456289559,3.161654456289559,,3.161654456289559,3.161654456289559,100.36768115351072,100.36768115351072,,100.36768115351072,100.36768115351072,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_3_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 1, 'total_duration_us': np.float64(396.832), 'mean_duration_us': np.float64(396.832), 'median_duration_us': np.float64(396.832), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(396.832), 'max_duration_us': np.float64(396.832)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(396.83)}]","[[32, 4096], [4096, 151936]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",396.83203125,396.83203125,,396.83203125,396.83203125,396.83203125,1,60441 diff --git a/tests/traces/inference/sglang_prefilldecode/perf_csvs/GEMM.csv b/tests/traces/inference/sglang_prefilldecode/perf_csvs/GEMM.csv index 08311e347..c054922b1 100644 --- a/tests/traces/inference/sglang_prefilldecode/perf_csvs/GEMM.csv +++ b/tests/traces/inference/sglang_prefilldecode/perf_csvs/GEMM.csv @@ -1,10 +1,10 @@ -name,param: B,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: transpose,param: stride_A,param: stride_B,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aiter::gemm_a16w16,1,15360,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,3092.37645312,1032.0,2857.6744186046512,0.15989855459699648,0.1598142700762255,0.0009942644080381477,0.15676364830950568,0.1614992228734229,456.93800904369596,456.69715132480445,2.841283964179698,447.9794675412107,461.5121978299117,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA4_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(243643.021), 'mean_duration_us': np.float64(6767.861694444445), 'median_duration_us': np.float64(6771.1755), 'std_dev_duration_us': np.float64(41.71130077943137), 'min_duration_us': np.float64(6700.53), 'max_duration_us': np.float64(6902.942)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(6767.86)}]","[[15360, 4096], [24576, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",6767.86181640625,6771.17529296875,42.30302141272883,6700.5302734375,6902.9423828125,243643.025390625,36,163 -aten::mm,1,15360,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,3092.37645312,1032.0,2857.6744186046512,0.15989855459699648,0.1598142700762255,0.0009942644080381477,0.15676364830950568,0.1614992228734229,456.93800904369596,456.69715132480445,2.841283964179698,447.9794675412107,461.5121978299117,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA4_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(243643.021), 'mean_duration_us': np.float64(6767.861694444445), 'median_duration_us': np.float64(6771.1755), 'std_dev_duration_us': np.float64(41.71130077943137), 'min_duration_us': np.float64(6700.53), 'max_duration_us': np.float64(6902.942)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(6767.86)}]","[[15360, 4096], [4096, 24576]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",6767.86181640625,6771.17529296875,42.30302141272883,6700.5302734375,6902.9423828125,243643.025390625,36,169 -aiter::gemm_a16w16,1,15360,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,1546.18822656,576.0,2560.0,0.22784430668329525,0.22773631836216326,0.0021934151326576576,0.222524755068712,0.23237480025639273,583.281425109236,583.004975007138,5.615142739603605,569.6633729759027,594.8794886563654,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(95439.041), 'mean_duration_us': np.float64(2651.0844722222223), 'median_duration_us': np.float64(2652.1014999999998), 'std_dev_duration_us': np.float64(25.310139447666806), 'min_duration_us': np.float64(2599.162), 'max_duration_us': np.float64(2714.214)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(2651.08)}]","[[15360, 12288], [4096, 12288], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[12288, 1], [12288, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",2651.0844997829863,2652.1015625,25.669117118336754,2599.162109375,2714.2138671875,95439.0419921875,36,172 -aten::mm,1,15360,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(12288, 1)","(1, 12288)",1,1546.18822656,576.0,2560.0,0.22784430668329525,0.22773631836216326,0.0021934151326576576,0.222524755068712,0.23237480025639273,583.281425109236,583.004975007138,5.615142739603605,569.6633729759027,594.8794886563654,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(95439.041), 'mean_duration_us': np.float64(2651.0844722222223), 'median_duration_us': np.float64(2652.1014999999998), 'std_dev_duration_us': np.float64(25.310139447666806), 'min_duration_us': np.float64(2599.162), 'max_duration_us': np.float64(2714.214)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(2651.08)}]","[[15360, 12288], [12288, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[12288, 1], [1, 12288]]","['', '']",2651.0844997829863,2652.1015625,25.669117118336754,2599.162109375,2714.2138671875,95439.0419921875,36,178 -aiter::gemm_a16w16,1,15360,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,773.09411328,348.0,2118.6206896551726,0.25629236994956,0.257305160693752,0.006763183331447345,0.22010358625052645,0.26236321765011933,542.9863175758953,545.1320370008318,14.328620133935333,466.3160116976671,555.8481411180459,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(51295.484000000004), 'mean_duration_us': np.float64(1424.8745555555556), 'median_duration_us': np.float64(1418.1775), 'std_dev_duration_us': np.float64(42.04861841200046), 'min_duration_us': np.float64(1390.837), 'max_duration_us': np.float64(1657.876)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(1424.87)}]","[[15360, 4096], [6144, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",1424.874538845486,1418.177734375,42.645073981571386,1390.8369140625,1657.8759765625,51295.4833984375,36,84 -aten::mm,1,15360,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,773.09411328,348.0,2118.6206896551726,0.25629236994956,0.257305160693752,0.006763183331447345,0.22010358625052645,0.26236321765011933,542.9863175758953,545.1320370008318,14.328620133935333,466.3160116976671,555.8481411180459,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(51295.484000000004), 'mean_duration_us': np.float64(1424.8745555555556), 'median_duration_us': np.float64(1418.1775), 'std_dev_duration_us': np.float64(42.04861841200046), 'min_duration_us': np.float64(1390.837), 'max_duration_us': np.float64(1657.876)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(1424.87)}]","[[15360, 4096], [4096, 6144]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",1424.874538845486,1418.177734375,42.645073981571386,1390.8369140625,1657.8759765625,51295.4833984375,36,90 -aiter::gemm_a16w16,1,15360,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",nan,nan,1,515.39607552,272.0,1807.0588235294115,0.34829642598889776,0.34917679608781427,0.007117676621272211,0.32493575013378695,0.3596699771038994,629.3921297869965,630.9830103422149,12.862060341498934,587.1780143594079,649.944805684223,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(29491.870999999996), 'mean_duration_us': np.float64(819.2186388888888), 'median_duration_us': np.float64(816.819), 'std_dev_duration_us': np.float64(16.860455720993983), 'min_duration_us': np.float64(792.984), 'max_duration_us': np.float64(877.751)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(819.22)}]","[[15360, 4096], [4096, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",819.2186686197916,816.81884765625,17.09959181546983,792.984375,877.7509765625,29491.8720703125,36,149 -aten::mm,1,15360,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,515.39607552,272.0,1807.0588235294115,0.34829642598889776,0.34917679608781427,0.007117676621272211,0.32493575013378695,0.3596699771038994,629.3921297869965,630.9830103422149,12.862060341498934,587.1780143594079,649.944805684223,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(29491.870999999996), 'mean_duration_us': np.float64(819.2186388888888), 'median_duration_us': np.float64(816.819), 'std_dev_duration_us': np.float64(16.860455720993983), 'min_duration_us': np.float64(792.984), 'max_duration_us': np.float64(877.751)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(819.22)}]","[[15360, 4096], [4096, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",819.2186686197916,816.81884765625,17.09959181546983,792.984375,877.7509765625,29491.8720703125,36,155 -aten::mm,1,15,151936,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",1,18.66989568,1191.464111328125,14.943798835999154,2.8632861644348853,2.8632861644348853,,2.8632861644348853,2.8632861644348853,42.788372451214514,42.788372451214514,,42.788372451214514,42.788372451214514,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 1, 'total_duration_us': np.float64(436.331), 'mean_duration_us': np.float64(436.331), 'median_duration_us': np.float64(436.331), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(436.331), 'max_duration_us': np.float64(436.331)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(436.33)}]","[[15, 4096], [4096, 151936]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",436.3310546875,436.3310546875,,436.3310546875,436.3310546875,436.3310546875,1,3771 +name,param: B,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: transpose,param: stride_A,param: stride_B,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aiter::gemm_a16w16,1,15360,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,128,256,32,1,3092.37645312,1032.0,2857.6744186046512,0.15989855459699648,0.1598142700762255,0.0009942644080381477,0.15676364830950568,0.1614992228734229,456.93800904369596,456.69715132480445,2.841283964179698,447.9794675412107,461.5121978299117,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA4_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(243643.021), 'mean_duration_us': np.float64(6767.861694444445), 'median_duration_us': np.float64(6771.1755), 'std_dev_duration_us': np.float64(41.71130077943137), 'min_duration_us': np.float64(6700.53), 'max_duration_us': np.float64(6902.942)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(6767.86)}]","[[15360, 4096], [24576, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",6767.86181640625,6771.17529296875,42.30302141272883,6700.5302734375,6902.9423828125,243643.025390625,36,163 +aten::mm,1,15360,24576,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",128,256,32,1,3092.37645312,1032.0,2857.6744186046512,0.15989855459699648,0.1598142700762255,0.0009942644080381477,0.15676364830950568,0.1614992228734229,456.93800904369596,456.69715132480445,2.841283964179698,447.9794675412107,461.5121978299117,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA4_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(243643.021), 'mean_duration_us': np.float64(6767.861694444445), 'median_duration_us': np.float64(6771.1755), 'std_dev_duration_us': np.float64(41.71130077943137), 'min_duration_us': np.float64(6700.53), 'max_duration_us': np.float64(6902.942)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(6767.86)}]","[[15360, 4096], [4096, 24576]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",6767.86181640625,6771.17529296875,42.30302141272883,6700.5302734375,6902.9423828125,243643.025390625,36,169 +aiter::gemm_a16w16,1,15360,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,320,224,64,1,1546.18822656,576.0,2560.0,0.22784430668329525,0.22773631836216326,0.0021934151326576576,0.222524755068712,0.23237480025639273,583.281425109236,583.004975007138,5.615142739603605,569.6633729759027,594.8794886563654,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(95439.041), 'mean_duration_us': np.float64(2651.0844722222223), 'median_duration_us': np.float64(2652.1014999999998), 'std_dev_duration_us': np.float64(25.310139447666806), 'min_duration_us': np.float64(2599.162), 'max_duration_us': np.float64(2714.214)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(2651.08)}]","[[15360, 12288], [4096, 12288], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[12288, 1], [12288, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",2651.0844997829863,2652.1015625,25.669117118336754,2599.162109375,2714.2138671875,95439.0419921875,36,172 +aten::mm,1,15360,4096,12288,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(12288, 1)","(1, 12288)",320,224,64,1,1546.18822656,576.0,2560.0,0.22784430668329525,0.22773631836216326,0.0021934151326576576,0.222524755068712,0.23237480025639273,583.281425109236,583.004975007138,5.615142739603605,569.6633729759027,594.8794886563654,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(95439.041), 'mean_duration_us': np.float64(2651.0844722222223), 'median_duration_us': np.float64(2652.1014999999998), 'std_dev_duration_us': np.float64(25.310139447666806), 'min_duration_us': np.float64(2599.162), 'max_duration_us': np.float64(2714.214)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(2651.08)}]","[[15360, 12288], [12288, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[12288, 1], [1, 12288]]","['', '']",2651.0844997829863,2652.1015625,25.669117118336754,2599.162109375,2714.2138671875,95439.0419921875,36,178 +aiter::gemm_a16w16,1,15360,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,256,192,64,1,773.09411328,348.0,2118.6206896551726,0.25629236994956,0.257305160693752,0.006763183331447345,0.22010358625052645,0.26236321765011933,542.9863175758953,545.1320370008318,14.328620133935333,466.3160116976671,555.8481411180459,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(51295.484000000004), 'mean_duration_us': np.float64(1424.8745555555556), 'median_duration_us': np.float64(1418.1775), 'std_dev_duration_us': np.float64(42.04861841200046), 'min_duration_us': np.float64(1390.837), 'max_duration_us': np.float64(1657.876)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(1424.87)}]","[[15360, 4096], [6144, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",1424.874538845486,1418.177734375,42.645073981571386,1390.8369140625,1657.8759765625,51295.4833984375,36,84 +aten::mm,1,15360,6144,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",256,192,64,1,773.09411328,348.0,2118.6206896551726,0.25629236994956,0.257305160693752,0.006763183331447345,0.22010358625052645,0.26236321765011933,542.9863175758953,545.1320370008318,14.328620133935333,466.3160116976671,555.8481411180459,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(51295.484000000004), 'mean_duration_us': np.float64(1424.8745555555556), 'median_duration_us': np.float64(1418.1775), 'std_dev_duration_us': np.float64(42.04861841200046), 'min_duration_us': np.float64(1390.837), 'max_duration_us': np.float64(1657.876)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(1424.87)}]","[[15360, 4096], [4096, 6144]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",1424.874538845486,1418.177734375,42.645073981571386,1390.8369140625,1657.8759765625,51295.4833984375,36,90 +aiter::gemm_a16w16,1,15360,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')","(True, False)",,,320,224,64,1,515.39607552,272.0,1807.0588235294115,0.34829642598889776,0.34917679608781427,0.007117676621272211,0.32493575013378695,0.3596699771038994,629.3921297869965,630.9830103422149,12.862060341498934,587.1780143594079,649.944805684223,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(29491.870999999996), 'mean_duration_us': np.float64(819.2186388888888), 'median_duration_us': np.float64(816.819), 'std_dev_duration_us': np.float64(16.860455720993983), 'min_duration_us': np.float64(792.984), 'max_duration_us': np.float64(877.751)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(819.22)}]","[[15360, 4096], [4096, 4096], [], [], [], [], []]","['c10::BFloat16', 'c10::BFloat16', '', 'Scalar', '', '', '']","[[4096, 1], [4096, 1], [], [], [], [], []]","['', '', '', '15', '', '', '']",819.2186686197916,816.81884765625,17.09959181546983,792.984375,877.7509765625,29491.8720703125,36,149 +aten::mm,1,15360,4096,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",320,224,64,1,515.39607552,272.0,1807.0588235294115,0.34829642598889776,0.34917679608781427,0.007117676621272211,0.32493575013378695,0.3596699771038994,629.3921297869965,630.9830103422149,12.862060341498934,587.1780143594079,649.944805684223,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA1_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA0_LBSPPB256_LBSPPM0_LPA0_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_14_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA2_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 36, 'total_duration_us': np.float64(29491.870999999996), 'mean_duration_us': np.float64(819.2186388888888), 'median_duration_us': np.float64(816.819), 'std_dev_duration_us': np.float64(16.860455720993983), 'min_duration_us': np.float64(792.984), 'max_duration_us': np.float64(877.751)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT320x224x64_MI16x1...', 'stream': 3, 'mean_duration_us': np.float64(819.22)}]","[[15360, 4096], [4096, 4096]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",819.2186686197916,816.81884765625,17.09959181546983,792.984375,877.7509765625,29491.8720703125,36,155 +aten::mm,1,15,151936,4096,False,"('c10::BFloat16', 'c10::BFloat16')","(True, False)","(4096, 1)","(1, 4096)",256,16,64,1,18.66989568,1191.464111328125,14.943798835999154,2.8632861644348853,2.8632861644348853,,2.8632861644348853,2.8632861644348853,42.788372451214514,42.788372451214514,,42.788372451214514,42.788372451214514,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 3, 'count': 1, 'total_duration_us': np.float64(436.331), 'mean_duration_us': np.float64(436.331), 'median_duration_us': np.float64(436.331), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(436.331), 'max_duration_us': np.float64(436.331)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x64_MI16x16...', 'stream': 3, 'mean_duration_us': np.float64(436.33)}]","[[15, 4096], [4096, 151936]]","['c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [1, 4096]]","['', '']",436.3310546875,436.3310546875,,436.3310546875,436.3310546875,436.3310546875,1,3771 diff --git a/tests/traces/inference/vllm_decode_full/perf_csvs/GEMM.csv b/tests/traces/inference/vllm_decode_full/perf_csvs/GEMM.csv index 9e8de306e..7be113e67 100644 --- a/tests/traces/inference/vllm_decode_full/perf_csvs/GEMM.csv +++ b/tests/traces/inference/vllm_decode_full/perf_csvs/GEMM.csv @@ -1,9 +1,9 @@ -name,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: B,param: transpose,param: stride_A,param: stride_B,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -vllm::rocm_unquantized_gemm,32,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,1,0.94388224,28.623046875,31.44865233708632,1.8642714721006335,1.8950323840240895,0.07657548307134485,1.757319604322717,1.9694817404677987,58.628825388040944,59.596214612693444,2.4081957446551607,55.265333281491216,61.937546540211464,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS8_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(387.01300000000003), 'mean_duration_us': np.float64(16.125541666666667), 'median_duration_us': np.float64(15.838999999999999), 'std_dev_duration_us': np.float64(0.6518056956874158), 'min_duration_us': np.float64(15.239), 'max_duration_us': np.float64(17.079)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(16.13)}]","[[32, 2880], [5120, 2880], [5120]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",16.12548828125,15.8388671875,0.6657737750991694,15.2392578125,17.0791015625,387.01171875,24,2361 -aten::addmm,32,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",1,0.94388224,28.623046875,31.44865233708632,1.8642714721006335,1.8950323840240895,0.07657548307134485,1.757319604322717,1.9694817404677987,58.628825388040944,59.596214612693444,2.4081957446551607,55.265333281491216,61.937546540211464,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS8_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(387.01300000000003), 'mean_duration_us': np.float64(16.125541666666667), 'median_duration_us': np.float64(15.838999999999999), 'std_dev_duration_us': np.float64(0.6518056956874158), 'min_duration_us': np.float64(15.239), 'max_duration_us': np.float64(17.079)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(16.13)}]","[[5120], [32, 2880], [2880, 5120], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",16.12548828125,15.8388671875,0.6657737750991694,15.2392578125,17.0791015625,387.01171875,24,2403 -vllm::rocm_unquantized_gemm,32,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,1,0.75506688,22.9312744140625,31.402000500391264,1.5803851220890681,1.5924917828534704,0.08821376381905677,1.466222141130233,1.7786800849526838,49.62725439465183,50.00742776203366,2.770088655587423,46.042308409456325,55.85411291772015,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(366.21299999999997), 'mean_duration_us': np.float64(15.258874999999998), 'median_duration_us': np.float64(15.099), 'std_dev_duration_us': np.float64(0.8094309375779587), 'min_duration_us': np.float64(13.519), 'max_duration_us': np.float64(16.399)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(15.26)}]","[[32, 4096], [2880, 4096], [2880]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [4096, 1], [1]]","['', '', '']",15.2589111328125,15.09912109375,0.8270028908902994,13.5185546875,16.3994140625,366.2138671875,24,3273 -aten::addmm,32,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(4096, 1)","(1, 4096)",1,0.75506688,22.9312744140625,31.402000500391264,1.5803851220890681,1.5924917828534704,0.08821376381905677,1.466222141130233,1.7786800849526838,49.62725439465183,50.00742776203366,2.770088655587423,46.042308409456325,55.85411291772015,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(366.21299999999997), 'mean_duration_us': np.float64(15.258874999999998), 'median_duration_us': np.float64(15.099), 'std_dev_duration_us': np.float64(0.8094309375779587), 'min_duration_us': np.float64(13.519), 'max_duration_us': np.float64(16.399)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(15.26)}]","[[2880], [32, 4096], [4096, 2880], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [4096, 1], [1, 4096], [], []]","['', '', '', '1', '1']",15.2589111328125,15.09912109375,0.8270028908902994,13.5185546875,16.3994140625,366.2138671875,24,3315 -vllm::rocm_unquantized_gemm,32,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16', '')",1,"(True, False)",,,1,37.06454016,1117.05859375,31.643371437963122,4.233520397461509,4.233520397461509,,4.233520397461509,4.233520397461509,133.9628584270678,133.9628584270678,,133.9628584270678,133.9628584270678,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT7_2_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(276.678), 'mean_duration_us': np.float64(276.678), 'median_duration_us': np.float64(276.678), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(276.678), 'max_duration_us': np.float64(276.678)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(276.68)}]","[[32, 2880], [201088, 2880], []]","['c10::BFloat16', 'c10::BFloat16', '']","[[2880, 1], [2880, 1], []]","['', '', '']",276.677734375,276.677734375,,276.677734375,276.677734375,276.677734375,1,91 -aten::mm,32,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",1,37.06454016,1117.05859375,31.643371437963122,4.233520397461509,4.233520397461509,,4.233520397461509,4.233520397461509,133.9628584270678,133.9628584270678,,133.9628584270678,133.9628584270678,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT7_2_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(276.678), 'mean_duration_us': np.float64(276.678), 'median_duration_us': np.float64(276.678), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(276.678), 'max_duration_us': np.float64(276.678)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(276.68)}]","[[32, 2880], [2880, 201088]]","['c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [1, 2880]]","['', '']",276.677734375,276.677734375,,276.677734375,276.677734375,276.677734375,1,99 -vllm::rocm_unquantized_gemm,32,32,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,1,0.005899264,0.35357666015625,15.911617469359571,0.04457977743861225,0.04281870834038781,0.0033816962515098373,0.04138326226291694,0.05010558901940082,0.7093363654723844,0.681314907644327,0.05380825715159169,0.658474638761718,0.7972609655536492,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(200.656), 'mean_duration_us': np.float64(8.360666666666667), 'median_duration_us': np.float64(8.658999999999999), 'std_dev_duration_us': np.float64(0.5917746380356477), 'min_duration_us': np.float64(7.399), 'max_duration_us': np.float64(8.959)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(8.36)}]","[[32, 2880], [32, 2880], [32]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",8.360514322916666,8.65869140625,0.6044340976345506,7.3994140625,8.958984375,200.65234375,24,3363 -aten::addmm,32,32,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",1,0.005899264,0.35357666015625,15.911617469359571,0.04457977743861225,0.04281870834038781,0.0033816962515098373,0.04138326226291694,0.05010558901940082,0.7093363654723844,0.681314907644327,0.05380825715159169,0.658474638761718,0.7972609655536492,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(200.656), 'mean_duration_us': np.float64(8.360666666666667), 'median_duration_us': np.float64(8.658999999999999), 'std_dev_duration_us': np.float64(0.5917746380356477), 'min_duration_us': np.float64(7.399), 'max_duration_us': np.float64(8.959)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(8.36)}]","[[32], [32, 2880], [2880, 32], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",8.360514322916666,8.65869140625,0.6044340976345506,7.3994140625,8.958984375,200.65234375,24,3405 +name,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: B,param: transpose,param: stride_A,param: stride_B,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +vllm::rocm_unquantized_gemm,32,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,32,32,256,1,0.94388224,28.623046875,31.44865233708632,1.8642714721006335,1.8950323840240895,0.07657548307134485,1.757319604322717,1.9694817404677987,58.628825388040944,59.596214612693444,2.4081957446551607,55.265333281491216,61.937546540211464,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS8_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(387.01300000000003), 'mean_duration_us': np.float64(16.125541666666667), 'median_duration_us': np.float64(15.838999999999999), 'std_dev_duration_us': np.float64(0.6518056956874158), 'min_duration_us': np.float64(15.239), 'max_duration_us': np.float64(17.079)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(16.13)}]","[[32, 2880], [5120, 2880], [5120]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",16.12548828125,15.8388671875,0.6657737750991694,15.2392578125,17.0791015625,387.01171875,24,2361 +aten::addmm,32,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",32,32,256,1,0.94388224,28.623046875,31.44865233708632,1.8642714721006335,1.8950323840240895,0.07657548307134485,1.757319604322717,1.9694817404677987,58.628825388040944,59.596214612693444,2.4081957446551607,55.265333281491216,61.937546540211464,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS8_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(387.01300000000003), 'mean_duration_us': np.float64(16.125541666666667), 'median_duration_us': np.float64(15.838999999999999), 'std_dev_duration_us': np.float64(0.6518056956874158), 'min_duration_us': np.float64(15.239), 'max_duration_us': np.float64(17.079)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(16.13)}]","[[5120], [32, 2880], [2880, 5120], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",16.12548828125,15.8388671875,0.6657737750991694,15.2392578125,17.0791015625,387.01171875,24,2403 +vllm::rocm_unquantized_gemm,32,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,32,32,256,1,0.75506688,22.9312744140625,31.402000500391264,1.5803851220890681,1.5924917828534704,0.08821376381905677,1.466222141130233,1.7786800849526838,49.62725439465183,50.00742776203366,2.770088655587423,46.042308409456325,55.85411291772015,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(366.21299999999997), 'mean_duration_us': np.float64(15.258874999999998), 'median_duration_us': np.float64(15.099), 'std_dev_duration_us': np.float64(0.8094309375779587), 'min_duration_us': np.float64(13.519), 'max_duration_us': np.float64(16.399)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(15.26)}]","[[32, 4096], [2880, 4096], [2880]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [4096, 1], [1]]","['', '', '']",15.2589111328125,15.09912109375,0.8270028908902994,13.5185546875,16.3994140625,366.2138671875,24,3273 +aten::addmm,32,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(4096, 1)","(1, 4096)",32,32,256,1,0.75506688,22.9312744140625,31.402000500391264,1.5803851220890681,1.5924917828534704,0.08821376381905677,1.466222141130233,1.7786800849526838,49.62725439465183,50.00742776203366,2.770088655587423,46.042308409456325,55.85411291772015,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(366.21299999999997), 'mean_duration_us': np.float64(15.258874999999998), 'median_duration_us': np.float64(15.099), 'std_dev_duration_us': np.float64(0.8094309375779587), 'min_duration_us': np.float64(13.519), 'max_duration_us': np.float64(16.399)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x32x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(15.26)}]","[[2880], [32, 4096], [4096, 2880], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [4096, 1], [1, 4096], [], []]","['', '', '', '1', '1']",15.2589111328125,15.09912109375,0.8270028908902994,13.5185546875,16.3994140625,366.2138671875,24,3315 +vllm::rocm_unquantized_gemm,32,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16', '')",1,"(True, False)",,,448,32,64,1,37.06454016,1117.05859375,31.643371437963122,4.233520397461509,4.233520397461509,,4.233520397461509,4.233520397461509,133.9628584270678,133.9628584270678,,133.9628584270678,133.9628584270678,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT7_2_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(276.678), 'mean_duration_us': np.float64(276.678), 'median_duration_us': np.float64(276.678), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(276.678), 'max_duration_us': np.float64(276.678)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(276.68)}]","[[32, 2880], [201088, 2880], []]","['c10::BFloat16', 'c10::BFloat16', '']","[[2880, 1], [2880, 1], []]","['', '', '']",276.677734375,276.677734375,,276.677734375,276.677734375,276.677734375,1,91 +aten::mm,32,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",448,32,64,1,37.06454016,1117.05859375,31.643371437963122,4.233520397461509,4.233520397461509,,4.233520397461509,4.233520397461509,133.9628584270678,133.9628584270678,,133.9628584270678,133.9628584270678,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT7_2_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(276.678), 'mean_duration_us': np.float64(276.678), 'median_duration_us': np.float64(276.678), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(276.678), 'max_duration_us': np.float64(276.678)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT448x32x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(276.68)}]","[[32, 2880], [2880, 201088]]","['c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [1, 2880]]","['', '']",276.677734375,276.677734375,,276.677734375,276.677734375,276.677734375,1,99 +vllm::rocm_unquantized_gemm,32,32,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,16,16,256,1,0.005899264,0.35357666015625,15.911617469359571,0.04457977743861225,0.04281870834038781,0.0033816962515098373,0.04138326226291694,0.05010558901940082,0.7093363654723844,0.681314907644327,0.05380825715159169,0.658474638761718,0.7972609655536492,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(200.656), 'mean_duration_us': np.float64(8.360666666666667), 'median_duration_us': np.float64(8.658999999999999), 'std_dev_duration_us': np.float64(0.5917746380356477), 'min_duration_us': np.float64(7.399), 'max_duration_us': np.float64(8.959)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(8.36)}]","[[32, 2880], [32, 2880], [32]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",8.360514322916666,8.65869140625,0.6044340976345506,7.3994140625,8.958984375,200.65234375,24,3363 +aten::addmm,32,32,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",16,16,256,1,0.005899264,0.35357666015625,15.911617469359571,0.04457977743861225,0.04281870834038781,0.0033816962515098373,0.04138326226291694,0.05010558901940082,0.7093363654723844,0.681314907644327,0.05380825715159169,0.658474638761718,0.7972609655536492,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(200.656), 'mean_duration_us': np.float64(8.360666666666667), 'median_duration_us': np.float64(8.658999999999999), 'std_dev_duration_us': np.float64(0.5917746380356477), 'min_duration_us': np.float64(7.399), 'max_duration_us': np.float64(8.959)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(8.36)}]","[[32], [32, 2880], [2880, 32], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",8.360514322916666,8.65869140625,0.6044340976345506,7.3994140625,8.958984375,200.65234375,24,3405 diff --git a/tests/traces/inference/vllm_prefilldecode_piecewise/perf_csvs/GEMM.csv b/tests/traces/inference/vllm_prefilldecode_piecewise/perf_csvs/GEMM.csv index 9666889e3..46be8c177 100644 --- a/tests/traces/inference/vllm_prefilldecode_piecewise/perf_csvs/GEMM.csv +++ b/tests/traces/inference/vllm_prefilldecode_piecewise/perf_csvs/GEMM.csv @@ -1,9 +1,9 @@ -name,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: B,param: transpose,param: stride_A,param: stride_B,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -vllm::rocm_unquantized_gemm,8192,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,1,241.63385344,153.134765625,1504.818315158472,0.4211595086883962,0.42079397165643295,0.008487170663371079,0.4066829475183831,0.4376299566168604,633.7685422774422,633.2184754568752,12.771649858116453,611.9839478882944,658.5535739790589,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(9153.921), 'mean_duration_us': np.float64(381.41337500000003), 'median_duration_us': np.float64(381.5965), 'std_dev_duration_us': np.float64(7.5112827677906235), 'min_duration_us': np.float64(366.916), 'max_duration_us': np.float64(394.837)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(381.41)}]","[[8192, 2880], [5120, 2880], [5120]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",381.4132893880208,381.5966796875,7.672717566300248,366.916015625,394.8369140625,9153.9189453125,24,19447 -aten::addmm,8192,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",1,241.63385344,153.134765625,1504.818315158472,0.4211595086883962,0.42079397165643295,0.008487170663371079,0.4066829475183831,0.4376299566168604,633.7685422774422,633.2184754568752,12.771649858116453,611.9839478882944,658.5535739790589,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(9153.921), 'mean_duration_us': np.float64(381.41337500000003), 'median_duration_us': np.float64(381.5965), 'std_dev_duration_us': np.float64(7.5112827677906235), 'min_duration_us': np.float64(366.916), 'max_duration_us': np.float64(394.837)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(381.41)}]","[[5120], [8192, 2880], [2880, 5120], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",381.4132893880208,381.5966796875,7.672717566300248,366.916015625,394.8369140625,9153.9189453125,24,19489 -vllm::rocm_unquantized_gemm,8192,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,1,193.29712128,131.5054931640625,1401.7855495208823,0.43805050836717513,0.43837419125105387,0.005620968034274064,0.42766093161505264,0.448351574265411,614.0528725893824,614.5066065786309,7.879391764764194,599.4889140326192,628.4927579101919,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT10_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(7556.130999999999), 'mean_duration_us': np.float64(314.8387916666666), 'median_duration_us': np.float64(314.557), 'std_dev_duration_us': np.float64(3.953142241508301), 'min_duration_us': np.float64(307.557), 'max_duration_us': np.float64(322.437)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(314.84)}]","[[8192, 4096], [2880, 4096], [2880]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [4096, 1], [1]]","['', '', '']",314.8387044270833,314.556640625,4.038205716642757,307.556640625,322.4365234375,7556.12890625,24,19704 -aten::addmm,8192,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(4096, 1)","(1, 4096)",1,193.29712128,131.5054931640625,1401.7855495208823,0.43805050836717513,0.43837419125105387,0.005620968034274064,0.42766093161505264,0.448351574265411,614.0528725893824,614.5066065786309,7.879391764764194,599.4889140326192,628.4927579101919,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT10_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(7556.130999999999), 'mean_duration_us': np.float64(314.8387916666666), 'median_duration_us': np.float64(314.557), 'std_dev_duration_us': np.float64(3.953142241508301), 'min_duration_us': np.float64(307.557), 'max_duration_us': np.float64(322.437)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(314.84)}]","[[2880], [8192, 4096], [4096, 2880], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [4096, 1], [1, 4096], [], []]","['', '', '', '1', '1']",314.8387044270833,314.556640625,4.038205716642757,307.556640625,322.4365234375,7556.12890625,24,19746 -vllm::rocm_unquantized_gemm,8192,32,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,1,1.510211584,45.67584228515625,31.531985573653074,1.945676571984176,1.9282116063691763,0.08192372666792049,1.8006411208282849,2.1614835702071398,61.35104559879981,60.80034055498329,2.5832177674327674,56.77778984528399,68.15586875345969,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(591.769), 'mean_duration_us': np.float64(24.657041666666668), 'median_duration_us': np.float64(24.839), 'std_dev_duration_us': np.float64(1.0004206398296778), 'min_duration_us': np.float64(22.158), 'max_duration_us': np.float64(26.599)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(24.66)}]","[[8192, 2880], [32, 2880], [32]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",24.657063802083332,24.8388671875,1.0218617974069066,22.158203125,26.5986328125,591.76953125,24,19798 -aten::addmm,8192,32,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",1,1.510211584,45.67584228515625,31.531985573653074,1.945676571984176,1.9282116063691763,0.08192372666792049,1.8006411208282849,2.1614835702071398,61.35104559879981,60.80034055498329,2.5832177674327674,56.77778984528399,68.15586875345969,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(591.769), 'mean_duration_us': np.float64(24.657041666666668), 'median_duration_us': np.float64(24.839), 'std_dev_duration_us': np.float64(1.0004206398296778), 'min_duration_us': np.float64(22.158), 'max_duration_us': np.float64(26.599)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(24.66)}]","[[32], [8192, 2880], [2880, 32], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",24.657063802083332,24.8388671875,1.0218617974069066,22.158203125,26.5986328125,591.76953125,24,19840 -vllm::rocm_unquantized_gemm,46,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16', '')",1,"(True, False)",,,1,53.28027648,1122.505126953125,45.266636231695244,3.3955777686924575,3.3955777686924575,,3.3955777686924575,3.3955777686924575,153.7063836518329,153.7063836518329,,153.7063836518329,153.7063836518329,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_3_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(346.637), 'mean_duration_us': np.float64(346.637), 'median_duration_us': np.float64(346.637), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(346.637), 'max_duration_us': np.float64(346.637)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(346.64)}]","[[46, 2880], [201088, 2880], []]","['c10::BFloat16', 'c10::BFloat16', '']","[[2880, 1], [2880, 1], []]","['', '', '']",346.63671875,346.63671875,,346.63671875,346.63671875,346.63671875,1,633 -aten::mm,46,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",1,53.28027648,1122.505126953125,45.266636231695244,3.3955777686924575,3.3955777686924575,,3.3955777686924575,3.3955777686924575,153.7063836518329,153.7063836518329,,153.7063836518329,153.7063836518329,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_3_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(346.637), 'mean_duration_us': np.float64(346.637), 'median_duration_us': np.float64(346.637), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(346.637), 'max_duration_us': np.float64(346.637)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(346.64)}]","[[46, 2880], [2880, 201088]]","['c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [1, 2880]]","['', '']",346.63671875,346.63671875,,346.63671875,346.63671875,346.63671875,1,641 +name,param: M,param: N,param: K,param: bias,param: dtype_A_B,param: B,param: transpose,param: stride_A,param: stride_B,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,num_kernels,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +vllm::rocm_unquantized_gemm,8192,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,256,192,64,1,241.63385344,153.134765625,1504.818315158472,0.4211595086883962,0.42079397165643295,0.008487170663371079,0.4066829475183831,0.4376299566168604,633.7685422774422,633.2184754568752,12.771649858116453,611.9839478882944,658.5535739790589,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(9153.921), 'mean_duration_us': np.float64(381.41337500000003), 'median_duration_us': np.float64(381.5965), 'std_dev_duration_us': np.float64(7.5112827677906235), 'min_duration_us': np.float64(366.916), 'max_duration_us': np.float64(394.837)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(381.41)}]","[[8192, 2880], [5120, 2880], [5120]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",381.4132893880208,381.5966796875,7.672717566300248,366.916015625,394.8369140625,9153.9189453125,24,19447 +aten::addmm,8192,5120,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",256,192,64,1,241.63385344,153.134765625,1504.818315158472,0.4211595086883962,0.42079397165643295,0.008487170663371079,0.4066829475183831,0.4376299566168604,633.7685422774422,633.2184754568752,12.771649858116453,611.9839478882944,658.5535739790589,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(9153.921), 'mean_duration_us': np.float64(381.41337500000003), 'median_duration_us': np.float64(381.5965), 'std_dev_duration_us': np.float64(7.5112827677906235), 'min_duration_us': np.float64(366.916), 'max_duration_us': np.float64(394.837)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(381.41)}]","[[5120], [8192, 2880], [2880, 5120], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",381.4132893880208,381.5966796875,7.672717566300248,366.916015625,394.8369140625,9153.9189453125,24,19489 +vllm::rocm_unquantized_gemm,8192,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,160,256,64,1,193.29712128,131.5054931640625,1401.7855495208823,0.43805050836717513,0.43837419125105387,0.005620968034274064,0.42766093161505264,0.448351574265411,614.0528725893824,614.5066065786309,7.879391764764194,599.4889140326192,628.4927579101919,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT10_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(7556.130999999999), 'mean_duration_us': np.float64(314.8387916666666), 'median_duration_us': np.float64(314.557), 'std_dev_duration_us': np.float64(3.953142241508301), 'min_duration_us': np.float64(307.557), 'max_duration_us': np.float64(322.437)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(314.84)}]","[[8192, 4096], [2880, 4096], [2880]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[4096, 1], [4096, 1], [1]]","['', '', '']",314.8387044270833,314.556640625,4.038205716642757,307.556640625,322.4365234375,7556.12890625,24,19704 +aten::addmm,8192,2880,4096,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(4096, 1)","(1, 4096)",160,256,64,1,193.29712128,131.5054931640625,1401.7855495208823,0.43805050836717513,0.43837419125105387,0.005620968034274064,0.42766093161505264,0.448351574265411,614.0528725893824,614.5066065786309,7.879391764764194,599.4889140326192,628.4927579101919,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT10_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(7556.130999999999), 'mean_duration_us': np.float64(314.8387916666666), 'median_duration_us': np.float64(314.557), 'std_dev_duration_us': np.float64(3.953142241508301), 'min_duration_us': np.float64(307.557), 'max_duration_us': np.float64(322.437)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT160x256x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(314.84)}]","[[2880], [8192, 4096], [4096, 2880], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [4096, 1], [1, 4096], [], []]","['', '', '', '1', '1']",314.8387044270833,314.556640625,4.038205716642757,307.556640625,322.4365234375,7556.12890625,24,19746 +vllm::rocm_unquantized_gemm,8192,32,2880,True,"('c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",,,32,64,128,1,1.510211584,45.67584228515625,31.531985573653074,1.945676571984176,1.9282116063691763,0.08192372666792049,1.8006411208282849,2.1614835702071398,61.35104559879981,60.80034055498329,2.5832177674327674,56.77778984528399,68.15586875345969,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(591.769), 'mean_duration_us': np.float64(24.657041666666668), 'median_duration_us': np.float64(24.839), 'std_dev_duration_us': np.float64(1.0004206398296778), 'min_duration_us': np.float64(22.158), 'max_duration_us': np.float64(26.599)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(24.66)}]","[[8192, 2880], [32, 2880], [32]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [2880, 1], [1]]","['', '', '']",24.657063802083332,24.8388671875,1.0218617974069066,22.158203125,26.5986328125,591.76953125,24,19798 +aten::addmm,8192,32,2880,True,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",32,64,128,1,1.510211584,45.67584228515625,31.531985573653074,1.945676571984176,1.9282116063691763,0.08192372666792049,1.8006411208282849,2.1614835702071398,61.35104559879981,60.80034055498329,2.5832177674327674,56.77778984528399,68.15586875345969,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(591.769), 'mean_duration_us': np.float64(24.657041666666668), 'median_duration_us': np.float64(24.839), 'std_dev_duration_us': np.float64(1.0004206398296778), 'min_duration_us': np.float64(22.158), 'max_duration_us': np.float64(26.599)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(24.66)}]","[[32], [8192, 2880], [2880, 32], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2880, 1], [1, 2880], [], []]","['', '', '', '1', '1']",24.657063802083332,24.8388671875,1.0218617974069066,22.158203125,26.5986328125,591.76953125,24,19840 +vllm::rocm_unquantized_gemm,46,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16', '')",1,"(True, False)",,,256,48,64,1,53.28027648,1122.505126953125,45.266636231695244,3.3955777686924575,3.3955777686924575,,3.3955777686924575,3.3955777686924575,153.7063836518329,153.7063836518329,,153.7063836518329,153.7063836518329,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_3_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(346.637), 'mean_duration_us': np.float64(346.637), 'median_duration_us': np.float64(346.637), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(346.637), 'max_duration_us': np.float64(346.637)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(346.64)}]","[[46, 2880], [201088, 2880], []]","['c10::BFloat16', 'c10::BFloat16', '']","[[2880, 1], [2880, 1], []]","['', '', '']",346.63671875,346.63671875,,346.63671875,346.63671875,346.63671875,1,633 +aten::mm,46,201088,2880,False,"('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)","(2880, 1)","(1, 2880)",256,48,64,1,53.28027648,1122.505126953125,45.266636231695244,3.3955777686924575,3.3955777686924575,,3.3955777686924575,3.3955777686924575,153.7063836518329,153.7063836518329,,153.7063836518329,153.7063836518329,Unknown,Unknown,Unknown,matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_3_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(346.637), 'mean_duration_us': np.float64(346.637), 'median_duration_us': np.float64(346.637), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(346.637), 'max_duration_us': np.float64(346.637)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x48x64_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(346.64)}]","[[46, 2880], [2880, 201088]]","['c10::BFloat16', 'c10::BFloat16']","[[2880, 1], [1, 2880]]","['', '']",346.63671875,346.63671875,,346.63671875,346.63671875,346.63671875,1,641 diff --git a/tests/traces/mi300/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv b/tests/traces/mi300/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv index c06ff7626..e1b9e24ef 100644 --- a/tests/traces/mi300/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv +++ b/tests/traces/mi300/Falconsai_nsfw_image_detection__1016002_perf_report_csvs/GEMM.csv @@ -1,5 +1,5 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::addmm,788,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.930167808,3.43505859375,258.2422174840085,0.4241487805689779,0.4218650440352281,0.026973370696042442,0.3697610105263158,0.49369108285370095,109.53312163727101,108.94336445064621,6.965663061564173,95.48790329744361,127.49187998822111,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(409.207), 'mean_duration_us': np.float64(8.525145833333333), 'median_duration_us': np.float64(8.538), 'std_dev_duration_us': np.float64(0.526665611401237), 'min_duration_us': np.float64(7.296), 'max_duration_us': np.float64(9.741)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(8.53)}]","[[768], [788, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",8.525146484375,8.5380859375,0.532247235181306,7.2958984375,9.7412109375,409.20703125,48,33 -aten::addmm,788,768,3072,True,"(3072, 1)","(1, 3072)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",3.71885568,10.27294921875,345.2345643804363,0.47239675564457034,0.4723091971155956,0.026823575886993672,0.42852684415609643,0.5069627370162699,163.08768814968468,163.05745991907628,9.260425536471832,147.9422783675531,175.02105967092564,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(274.459), 'mean_duration_us': np.float64(22.871583333333334), 'median_duration_us': np.float64(22.811500000000002), 'std_dev_duration_us': np.float64(1.2678273711572705), 'min_duration_us': np.float64(21.248), 'max_duration_us': np.float64(25.137)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(22.87)}]","[[768], [788, 3072], [3072, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [3072, 1], [1, 3072], [], []]","['', '', '', '1', '1']",22.871622721354168,22.8115234375,1.324204222903877,21.248046875,25.13720703125,274.45947265625,12,127 -aten::addmm,788,3072,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",3.720671232,10.27734375,345.2554161915621,0.5159044962901375,0.5139950705199006,0.024138253495302988,0.48002148088216107,0.5553845755554997,178.11882158174964,177.45958199275964,8.333862756658265,165.7300161628605,191.74953277978813,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x160x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_5_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(251.16200000000003), 'mean_duration_us': np.float64(20.93016666666667), 'median_duration_us': np.float64(20.9675), 'std_dev_duration_us': np.float64(0.9275455813896989), 'min_duration_us': np.float64(19.404), 'max_duration_us': np.float64(22.45)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x160x128_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(20.93)}]","[[3072], [788, 768], [768, 3072], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",20.930216471354168,20.967529296875,0.9688740862432101,19.40380859375,22.4501953125,251.16259765625,12,117 -aten::addmm,4,768,768,True,"(151296, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.004721664,1.13818359375,3.956241956241956,0.15919178429073857,0.15919178429073857,,0.15919178429073857,0.15919178429073857,0.6298012161000391,0.6298012161000391,,0.6298012161000391,0.6298012161000391,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(7.497), 'mean_duration_us': np.float64(7.497), 'median_duration_us': np.float64(7.497), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(7.497), 'max_duration_us': np.float64(7.497)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(7.5)}]","[[768], [4, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [151296, 1], [1, 768], [], []]","['', '', '', '1', '1']",7.4970703125,7.4970703125,,7.4970703125,7.4970703125,7.4970703125,1,1379 +name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aten::addmm,788,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",64,32,128,0.930167808,3.43505859375,258.2422174840085,0.4241487805689779,0.4218650440352281,0.026973370696042442,0.3697610105263158,0.49369108285370095,109.53312163727101,108.94336445064621,6.965663061564173,95.48790329744361,127.49187998822111,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(409.207), 'mean_duration_us': np.float64(8.525145833333333), 'median_duration_us': np.float64(8.538), 'std_dev_duration_us': np.float64(0.526665611401237), 'min_duration_us': np.float64(7.296), 'max_duration_us': np.float64(9.741)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(8.53)}]","[[768], [788, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",8.525146484375,8.5380859375,0.532247235181306,7.2958984375,9.7412109375,409.20703125,48,33 +aten::addmm,788,768,3072,True,"(3072, 1)","(1, 3072)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",64,32,256,3.71885568,10.27294921875,345.2345643804363,0.47239675564457034,0.4723091971155956,0.026823575886993672,0.42852684415609643,0.5069627370162699,163.08768814968468,163.05745991907628,9.260425536471832,147.9422783675531,175.02105967092564,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_2_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB2_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(274.459), 'mean_duration_us': np.float64(22.871583333333334), 'median_duration_us': np.float64(22.811500000000002), 'std_dev_duration_us': np.float64(1.2678273711572705), 'min_duration_us': np.float64(21.248), 'max_duration_us': np.float64(25.137)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x32x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(22.87)}]","[[768], [788, 3072], [3072, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [3072, 1], [1, 3072], [], []]","['', '', '', '1', '1']",22.871622721354168,22.8115234375,1.324204222903877,21.248046875,25.13720703125,274.45947265625,12,127 +aten::addmm,788,3072,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",64,160,128,3.720671232,10.27734375,345.2554161915621,0.5159044962901375,0.5139950705199006,0.024138253495302988,0.48002148088216107,0.5553845755554997,178.11882158174964,177.45958199275964,8.333862756658265,165.7300161628605,191.74953277978813,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x160x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_5_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(251.16200000000003), 'mean_duration_us': np.float64(20.93016666666667), 'median_duration_us': np.float64(20.9675), 'std_dev_duration_us': np.float64(0.9275455813896989), 'min_duration_us': np.float64(19.404), 'max_duration_us': np.float64(22.45)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x160x128_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(20.93)}]","[[3072], [788, 768], [768, 3072], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",20.930216471354168,20.967529296875,0.9688740862432101,19.40380859375,22.4501953125,251.16259765625,12,117 +aten::addmm,4,768,768,True,"(151296, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.004721664,1.13818359375,3.956241956241956,0.15919178429073857,0.15919178429073857,,0.15919178429073857,0.15919178429073857,0.6298012161000391,0.6298012161000391,,0.6298012161000391,0.6298012161000391,python3,CPU,thread 950 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(7.497), 'mean_duration_us': np.float64(7.497), 'median_duration_us': np.float64(7.497), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(7.497), 'max_duration_us': np.float64(7.497)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(7.5)}]","[[768], [4, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [151296, 1], [1, 768], [], []]","['', '', '', '1', '1']",7.4970703125,7.4970703125,,7.4970703125,7.4970703125,7.4970703125,1,1379 diff --git a/tests/traces/mi300/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv b/tests/traces/mi300/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv index 3d41c3d7e..190b00cdf 100644 --- a/tests/traces/mi300/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv +++ b/tests/traces/mi300/Qwen_Qwen1.5-0.5B-Chat__1016005_perf_report_csvs/GEMM.csv @@ -1,6 +1,6 @@ -name,param: B,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: transpose,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::mm,1,8944,2816,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",51.581550592,71.0078125,692.7688414567059,0.6738529762661254,0.674108665064226,0.019829532765433946,0.62806071132492,0.7091151238095238,466.8243456800369,467.0014789124705,13.737282440537477,435.1008913490395,491.2528627809524,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(5308.273), 'mean_duration_us': np.float64(110.58902083333334), 'median_duration_us': np.float64(110.453), 'std_dev_duration_us': np.float64(3.2462516941953186), 'min_duration_us': np.float64(105.0), 'max_duration_us': np.float64(118.551)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(110.59)}]","[[8944, 1024], [1024, 2816]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",110.58902994791667,110.452880859375,3.280588398059727,105.0,118.55078125,5308.2734375,48,231 -aten::addmm,1,8944,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",18.766086144,36.939453125,484.4883413525089,0.9445121748402515,0.951401332232734,0.024813139318591027,0.8603128998015335,0.9788727841162896,457.6051369756043,460.94285341400456,12.02167671221286,416.81156986901215,474.2524515716137,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(2954.763), 'mean_duration_us': np.float64(41.038375), 'median_duration_us': np.float64(40.7125), 'std_dev_duration_us': np.float64(1.1135532272552386), 'min_duration_us': np.float64(39.57), 'max_duration_us': np.float64(45.023)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(41.04)}]","[[1024], [8944, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",41.038370768229164,40.71240234375,1.1213617423390887,39.56982421875,45.02294921875,2954.7626953125,72,90 -aten::mm,1,8944,1024,2816,False,"(2816, 1)","(1, 2816)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",51.581550592,71.0078125,692.7688414567059,0.753305737501413,0.7572560566039392,0.025395412136689215,0.703202779015716,0.7950246671011402,521.8667430315435,524.6034010195848,17.593150244249767,487.1569745278536,550.7683175571602,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(2374.795), 'mean_duration_us': np.float64(98.94979166666667), 'median_duration_us': np.float64(98.32499999999999), 'std_dev_duration_us': np.float64(3.3136632546066833), 'min_duration_us': np.float64(93.654), 'max_duration_us': np.float64(105.883)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(98.95)}]","[[8944, 2816], [2816, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[2816, 1], [1, 2816]]","['', '']",98.94974772135417,98.324951171875,3.3849417047301227,93.65380859375,105.8828125,2374.7939453125,24,253 -aten::mm,1,8944,1024,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",18.756927488,36.9375,484.27749576988157,0.9490013078576646,0.9555973894680865,0.030823547515567696,0.861800215638343,0.9878291064508096,459.57997685165213,462.774310735841,14.927150401583084,417.35045028328057,478.38340592059774,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(980.549), 'mean_duration_us': np.float64(40.856208333333335), 'median_duration_us': np.float64(40.5325), 'std_dev_duration_us': np.float64(1.3524004269928913), 'min_duration_us': np.float64(39.209), 'max_duration_us': np.float64(44.943)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(40.86)}]","[[8944, 1024], [1024, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",40.856160481770836,40.532470703125,1.3815012260133064,39.208984375,44.94287109375,980.5478515625,24,204 -aten::bmm,1,32,559,1,False,"(32, 1, 1)","(559, 559, 1)","('float', 'float')","(False, False)",3.5776e-05,0.07049179077148438,0.4840088749391201,0.0027034068147724836,0.0027034068147724836,,0.0027034068147724836,0.0027034068147724836,0.00130847289092078,0.00130847289092078,,0.00130847289092078,0.00130847289092078,python3,CPU,thread 23232 (python3),matrix_fp32,"[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB8_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(27.342), 'mean_duration_us': np.float64(27.342), 'median_duration_us': np.float64(27.342), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(27.342), 'max_duration_us': np.float64(27.342)}]","[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_...', 'stream': 0, 'mean_duration_us': np.float64(27.34)}]","[[1, 32, 1], [1, 1, 559]]","['float', 'float']","[[32, 1, 1], [559, 559, 1]]","['', '']",27.341796875,27.341796875,,27.341796875,27.341796875,27.341796875,1,50 +name,param: B,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: transpose,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aten::mm,1,8944,2816,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",128,224,64,51.581550592,71.0078125,692.7688414567059,0.6738529762661254,0.674108665064226,0.019829532765433946,0.62806071132492,0.7091151238095238,466.8243456800369,467.0014789124705,13.737282440537477,435.1008913490395,491.2528627809524,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 48, 'total_duration_us': np.float64(5308.273), 'mean_duration_us': np.float64(110.58902083333334), 'median_duration_us': np.float64(110.453), 'std_dev_duration_us': np.float64(3.2462516941953186), 'min_duration_us': np.float64(105.0), 'max_duration_us': np.float64(118.551)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x224x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(110.59)}]","[[8944, 1024], [1024, 2816]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",110.58902994791667,110.452880859375,3.280588398059727,105.0,118.55078125,5308.2734375,48,231 +aten::addmm,1,8944,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",256,128,64,18.766086144,36.939453125,484.4883413525089,0.9445121748402515,0.951401332232734,0.024813139318591027,0.8603128998015335,0.9788727841162896,457.6051369756043,460.94285341400456,12.02167671221286,416.81156986901215,474.2524515716137,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 72, 'total_duration_us': np.float64(2954.763), 'mean_duration_us': np.float64(41.038375), 'median_duration_us': np.float64(40.7125), 'std_dev_duration_us': np.float64(1.1135532272552386), 'min_duration_us': np.float64(39.57), 'max_duration_us': np.float64(45.023)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(41.04)}]","[[1024], [8944, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",41.038370768229164,40.71240234375,1.1213617423390887,39.56982421875,45.02294921875,2954.7626953125,72,90 +aten::mm,1,8944,1024,2816,False,"(2816, 1)","(1, 2816)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",256,128,64,51.581550592,71.0078125,692.7688414567059,0.753305737501413,0.7572560566039392,0.025395412136689215,0.703202779015716,0.7950246671011402,521.8667430315435,524.6034010195848,17.593150244249767,487.1569745278536,550.7683175571602,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(2374.795), 'mean_duration_us': np.float64(98.94979166666667), 'median_duration_us': np.float64(98.32499999999999), 'std_dev_duration_us': np.float64(3.3136632546066833), 'min_duration_us': np.float64(93.654), 'max_duration_us': np.float64(105.883)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(98.95)}]","[[8944, 2816], [2816, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[2816, 1], [1, 2816]]","['', '']",98.94974772135417,98.324951171875,3.3849417047301227,93.65380859375,105.8828125,2374.7939453125,24,253 +aten::mm,1,8944,1024,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')","(True, False)",256,128,64,18.756927488,36.9375,484.27749576988157,0.9490013078576646,0.9555973894680865,0.030823547515567696,0.861800215638343,0.9878291064508096,459.57997685165213,462.774310735841,14.927150401583084,417.35045028328057,478.38340592059774,python3,CPU,thread 23232 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(980.549), 'mean_duration_us': np.float64(40.856208333333335), 'median_duration_us': np.float64(40.5325), 'std_dev_duration_us': np.float64(1.3524004269928913), 'min_duration_us': np.float64(39.209), 'max_duration_us': np.float64(44.943)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(40.86)}]","[[8944, 1024], [1024, 1024]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",40.856160481770836,40.532470703125,1.3815012260133064,39.208984375,44.94287109375,980.5478515625,24,204 +aten::bmm,1,32,559,1,False,"(32, 1, 1)","(559, 559, 1)","('float', 'float')","(False, False)",16,16,128,3.5776e-05,0.07049179077148438,0.4840088749391201,0.0027034068147724836,0.0027034068147724836,,0.0027034068147724836,0.0027034068147724836,0.00130847289092078,0.00130847289092078,,0.00130847289092078,0.00130847289092078,python3,CPU,thread 23232 (python3),matrix_fp32,"[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB8_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(27.342), 'mean_duration_us': np.float64(27.342), 'median_duration_us': np.float64(27.342), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(27.342), 'max_duration_us': np.float64(27.342)}]","[{'name': 'Cijk_Ailk_Bljk_S_B_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_...', 'stream': 0, 'mean_duration_us': np.float64(27.34)}]","[[1, 32, 1], [1, 1, 559]]","['float', 'float']","[[32, 1, 1], [559, 559, 1]]","['', '']",27.341796875,27.341796875,,27.341796875,27.341796875,27.341796875,1,50 diff --git a/tests/traces/mi300/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv b/tests/traces/mi300/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv index a8b1568bf..868fd496d 100644 --- a/tests/traces/mi300/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv +++ b/tests/traces/mi300/facebook_timesformer-base-finetuned-k400__1016002_perf_report_csvs/GEMM.csv @@ -1,12 +1,12 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::bmm,3137,3137,64,False,"(200768, 64, 1)","(200768, 3137, 1)","('c10::BFloat16', 'c10::BFloat16')",96,"(False, False)",120.923369472,1875.4220581054688,61.49096477794793,1.622501115772592,1.6235078720754257,0.015640743540936836,1.5955931251879383,1.650130231333166,99.76915896215367,99.8310653785112,0.9617644101766607,98.11456066086738,101.4680999339348,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA0_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(14545.617999999999), 'mean_duration_us': np.float64(1212.1348333333333), 'median_duration_us': np.float64(1211.283), 'std_dev_duration_us': np.float64(11.187026748525364), 'min_duration_us': np.float64(1191.738), 'max_duration_us': np.float64(1232.471)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(1212.13)}]","[[96, 3137, 64], [96, 64, 3137]]","['c10::BFloat16', 'c10::BFloat16']","[[200768, 64, 1], [200768, 3137, 1]]","['', '']",1212.1348063151042,1211.282958984375,11.68448034000017,1191.73779296875,1232.47119140625,14545.61767578125,12,263 -aten::bmm,8,8,64,False,"(512, 64, 1)","(512, 8, 1)","('c10::BFloat16', 'c10::BFloat16')",37632,"(False, False)",0.308281344,78.09375,3.764705882352941,0.07216108359587865,0.07079892969470838,0.006905450073665989,0.0670236026823021,0.09344305334841453,0.2716652558903667,0.266537147085961,0.025996988512624886,0.2523241512745491,0.3517856126057959,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x448x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB128_LBSPPM0_LPA16_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_7_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(13710.097999999998), 'mean_duration_us': np.float64(1142.5081666666665), 'median_duration_us': np.float64(1156.6174999999998), 'std_dev_duration_us': np.float64(84.56194532494443), 'min_duration_us': np.float64(876.333), 'max_duration_us': np.float64(1221.767)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x448x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(1142.51)}]","[[37632, 8, 64], [37632, 64, 8]]","['c10::BFloat16', 'c10::BFloat16']","[[512, 64, 1], [512, 8, 1]]","['', '']",1142.5081787109375,1156.61767578125,88.32208238575072,876.3330078125,1221.76708984375,13710.09814453125,12,125 -aten::bmm,3137,64,3137,False,"(9840769, 3137, 1)","(200768, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",96,"(False, False)",120.923369472,1875.4220581054688,61.49096477794793,2.4286535365647803,2.427563364803113,0.012475139298374256,2.4066591549942156,2.443586240277836,149.3402490747436,149.27321336134503,0.7671083511963279,147.98779333227526,150.25847543280264,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB128_LBSPPM0_LPA32_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_7_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(9716.842), 'mean_duration_us': np.float64(809.7368333333334), 'median_duration_us': np.float64(810.0809999999999), 'std_dev_duration_us': np.float64(3.9871760230630486), 'min_duration_us': np.float64(804.769), 'max_duration_us': np.float64(817.117)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x224x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(809.74)}]","[[96, 3137, 3137], [96, 3137, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[9840769, 3137, 1], [200768, 64, 1]]","['', '']",809.73681640625,810.0810546875,4.1644939138616754,804.76904296875,817.1171875,9716.841796875,12,283 -aten::bmm,8,64,8,False,"(64, 8, 1)","(512, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",37632,"(False, False)",0.308281344,78.09375,3.764705882352941,0.2938623046710218,0.2916583826887007,0.018777507522173983,0.263886762057468,0.33216485167117266,1.106305146996788,1.0980080289456968,0.07069179302465489,0.9934560453928208,1.2505029709973559,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA2048_LBSPPB128_LBSPPM0_LPA0_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT4_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(3356.223), 'mean_duration_us': np.float64(279.68525), 'median_duration_us': np.float64(280.7645), 'std_dev_duration_us': np.float64(16.826475344552502), 'min_duration_us': np.float64(246.526), 'max_duration_us': np.float64(310.312)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x32_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(279.69)}]","[[37632, 8, 8], [37632, 8, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[64, 8, 1], [512, 64, 1]]","['', '']",279.6852213541667,280.764404296875,17.57473286070068,246.52587890625,310.31201171875,3356.22265625,12,145 -aten::addmm,25089,3072,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",118.461828096,188.26318359375,600.0854801938983,0.7750207865568397,0.7801452150870591,0.01212663828834957,0.7530007157427342,0.7876945569263141,465.078720861214,468.15381601649005,7.277019560401961,451.8647960928278,472.68406643924715,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB128_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(3057.256), 'mean_duration_us': np.float64(254.77133333333333), 'median_duration_us': np.float64(253.0405), 'std_dev_duration_us': np.float64(3.8628236074434144), 'min_duration_us': np.float64(250.615), 'max_duration_us': np.float64(262.162)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x224x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(254.77)}]","[[3072], [25089, 768], [768, 3072], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",254.7713623046875,253.04052734375,4.034526324921192,250.615234375,262.162109375,3057.25634765625,12,345 -aten::addmm,25089,768,3072,True,"(3072, 1)","(1, 3072)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",118.40402304,188.2587890625,599.8066613237056,0.9113350589701215,0.9031991903131726,0.02340186508485172,0.8857262721309608,0.9628047351732908,546.6248390681109,541.7448908520182,14.036594565292738,531.2645181335635,577.4966937109461,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(2600.847), 'mean_duration_us': np.float64(216.73725000000002), 'median_duration_us': np.float64(218.561), 'std_dev_duration_us': np.float64(5.201450440742469), 'min_duration_us': np.float64(205.03), 'max_duration_us': np.float64(222.872)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(216.74)}]","[[768], [25089, 3072], [3072, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [3072, 1], [1, 3072], [], []]","['', '', '', '1', '1']",216.73714192708334,218.560791015625,5.432787018968682,205.02978515625,222.8720703125,2600.845703125,12,356 -aten::addmm,25096,2304,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",88.871159808,150.42626953125,563.4264735955439,0.8201891975798056,0.8173786300280079,0.028393483256752255,0.7829438965375163,0.8809421265789645,462.1163072735487,460.5327591090373,15.99764014444606,441.1313186492872,496.3461158201453,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(2310.257), 'mean_duration_us': np.float64(192.52141666666668), 'median_duration_us': np.float64(192.98250000000002), 'std_dev_duration_us': np.float64(6.2790327075956185), 'min_duration_us': np.float64(179.051), 'max_duration_us': np.float64(201.462)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(192.52)}]","[[2304], [25096, 768], [768, 2304], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",192.5213623046875,192.982421875,6.558283927465597,179.05078125,201.4619140625,2310.25634765625,12,231 -aten::addmm,25088,2304,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",88.842829824,150.37939453125,563.422437389805,0.8451509012425363,0.8482090610697497,0.026524564657361804,0.7771340822670733,0.8732377472905446,476.17698074026026,477.9000166040364,14.944534869954294,437.8547788096037,492.0017399992213,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(2241.0190000000002), 'mean_duration_us': np.float64(186.75158333333334), 'median_duration_us': np.float64(185.906), 'std_dev_duration_us': np.float64(5.873673700764756), 'min_duration_us': np.float64(180.574), 'max_duration_us': np.float64(202.905)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(186.75)}]","[[2304], [25088, 768], [768, 2304], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",186.7515869140625,185.906005859375,6.134801196057336,180.57421875,202.90478515625,2241.01904296875,12,93 -aten::addmm,25088,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",29.614276608,74.62646484375,378.44985768966535,1.1023945690056698,1.1055407737805627,0.03816811242336863,0.992285767994799,1.1764986232261758,417.2010677580557,418.3917485073764,14.444716714907022,375.53040768511187,445.24573653203345,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_8_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB8_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(1705.616), 'mean_duration_us': np.float64(71.06733333333334), 'median_duration_us': np.float64(70.7815), 'std_dev_duration_us': np.float64(2.4860613204200916), 'min_duration_us': np.float64(66.512), 'max_duration_us': np.float64(78.86)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(71.07)}]","[[768], [25088, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",71.06734212239583,70.781494140625,2.5394755658784947,66.51220703125,78.85986328125,1705.6162109375,24,162 -aten::addmm,25096,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",29.623719936,74.64990234375,378.4516787347187,1.0549091731169484,1.053384165887925,0.01133784738837991,1.0407406456236934,1.0775004846651746,399.2321474787631,398.6550059328566,4.290827377370407,393.87004446374175,407.7818672590083,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_8_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB8_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(890.515), 'mean_duration_us': np.float64(74.20958333333333), 'median_duration_us': np.float64(74.31), 'std_dev_duration_us': np.float64(0.7594008008877068), 'min_duration_us': np.float64(72.646), 'max_duration_us': np.float64(75.212)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(74.21)}]","[[768], [25096, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",74.20955403645833,74.309814453125,0.7931931743684959,72.64599609375,75.2119140625,890.5146484375,12,300 -aten::addmm,1,400,768,True,"(19268352, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.0006148,0.58892822265625,0.9955694890662244,0.0666796925185849,0.0666796925185849,,0.0666796925185849,0.0666796925185849,0.06638426741182053,0.06638426741182053,,0.06638426741182053,0.06638426741182053,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(9.261), 'mean_duration_us': np.float64(9.261), 'median_duration_us': np.float64(9.261), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(9.261), 'max_duration_us': np.float64(9.261)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(9.26)}]","[[400], [1, 768], [768, 400], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [19268352, 1], [1, 768], [], []]","['', '', '', '1', '1']",9.26123046875,9.26123046875,,9.26123046875,9.26123046875,9.26123046875,1,3577 +name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aten::bmm,3137,3137,64,False,"(200768, 64, 1)","(200768, 3137, 1)","('c10::BFloat16', 'c10::BFloat16')",96,"(False, False)",128,256,32,120.923369472,1875.4220581054688,61.49096477794793,1.622501115772592,1.6235078720754257,0.015640743540936836,1.5955931251879383,1.650130231333166,99.76915896215367,99.8310653785112,0.9617644101766607,98.11456066086738,101.4680999339348,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA0_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(14545.617999999999), 'mean_duration_us': np.float64(1212.1348333333333), 'median_duration_us': np.float64(1211.283), 'std_dev_duration_us': np.float64(11.187026748525364), 'min_duration_us': np.float64(1191.738), 'max_duration_us': np.float64(1232.471)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x256x32_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(1212.13)}]","[[96, 3137, 64], [96, 64, 3137]]","['c10::BFloat16', 'c10::BFloat16']","[[200768, 64, 1], [200768, 3137, 1]]","['', '']",1212.1348063151042,1211.282958984375,11.68448034000017,1191.73779296875,1232.47119140625,14545.61767578125,12,263 +aten::bmm,8,8,64,False,"(512, 64, 1)","(512, 8, 1)","('c10::BFloat16', 'c10::BFloat16')",37632,"(False, False)",16,448,64,0.308281344,78.09375,3.764705882352941,0.07216108359587865,0.07079892969470838,0.006905450073665989,0.0670236026823021,0.09344305334841453,0.2716652558903667,0.266537147085961,0.025996988512624886,0.2523241512745491,0.3517856126057959,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x448x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB128_LBSPPM0_LPA16_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_7_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(13710.097999999998), 'mean_duration_us': np.float64(1142.5081666666665), 'median_duration_us': np.float64(1156.6174999999998), 'std_dev_duration_us': np.float64(84.56194532494443), 'min_duration_us': np.float64(876.333), 'max_duration_us': np.float64(1221.767)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x448x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(1142.51)}]","[[37632, 8, 64], [37632, 64, 8]]","['c10::BFloat16', 'c10::BFloat16']","[[512, 64, 1], [512, 8, 1]]","['', '']",1142.5081787109375,1156.61767578125,88.32208238575072,876.3330078125,1221.76708984375,13710.09814453125,12,125 +aten::bmm,3137,64,3137,False,"(9840769, 3137, 1)","(200768, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",96,"(False, False)",64,224,64,120.923369472,1875.4220581054688,61.49096477794793,2.4286535365647803,2.427563364803113,0.012475139298374256,2.4066591549942156,2.443586240277836,149.3402490747436,149.27321336134503,0.7671083511963279,147.98779333227526,150.25847543280264,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB128_LBSPPM0_LPA32_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_7_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(9716.842), 'mean_duration_us': np.float64(809.7368333333334), 'median_duration_us': np.float64(810.0809999999999), 'std_dev_duration_us': np.float64(3.9871760230630486), 'min_duration_us': np.float64(804.769), 'max_duration_us': np.float64(817.117)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x224x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(809.74)}]","[[96, 3137, 3137], [96, 3137, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[9840769, 3137, 1], [200768, 64, 1]]","['', '']",809.73681640625,810.0810546875,4.1644939138616754,804.76904296875,817.1171875,9716.841796875,12,283 +aten::bmm,8,64,8,False,"(64, 8, 1)","(512, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",37632,"(False, False)",256,16,32,0.308281344,78.09375,3.764705882352941,0.2938623046710218,0.2916583826887007,0.018777507522173983,0.263886762057468,0.33216485167117266,1.106305146996788,1.0980080289456968,0.07069179302465489,0.9934560453928208,1.2505029709973559,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA2048_LBSPPB128_LBSPPM0_LPA0_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT4_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(3356.223), 'mean_duration_us': np.float64(279.68525), 'median_duration_us': np.float64(280.7645), 'std_dev_duration_us': np.float64(16.826475344552502), 'min_duration_us': np.float64(246.526), 'max_duration_us': np.float64(310.312)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x16x32_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(279.69)}]","[[37632, 8, 8], [37632, 8, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[64, 8, 1], [512, 64, 1]]","['', '']",279.6852213541667,280.764404296875,17.57473286070068,246.52587890625,310.31201171875,3356.22265625,12,145 +aten::addmm,25089,3072,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,224,64,118.461828096,188.26318359375,600.0854801938983,0.7750207865568397,0.7801452150870591,0.01212663828834957,0.7530007157427342,0.7876945569263141,465.078720861214,468.15381601649005,7.277019560401961,451.8647960928278,472.68406643924715,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x224x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB128_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(3057.256), 'mean_duration_us': np.float64(254.77133333333333), 'median_duration_us': np.float64(253.0405), 'std_dev_duration_us': np.float64(3.8628236074434144), 'min_duration_us': np.float64(250.615), 'max_duration_us': np.float64(262.162)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x224x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(254.77)}]","[[3072], [25089, 768], [768, 3072], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",254.7713623046875,253.04052734375,4.034526324921192,250.615234375,262.162109375,3057.25634765625,12,345 +aten::addmm,25089,768,3072,True,"(3072, 1)","(1, 3072)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,128,64,118.40402304,188.2587890625,599.8066613237056,0.9113350589701215,0.9031991903131726,0.02340186508485172,0.8857262721309608,0.9628047351732908,546.6248390681109,541.7448908520182,14.036594565292738,531.2645181335635,577.4966937109461,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(2600.847), 'mean_duration_us': np.float64(216.73725000000002), 'median_duration_us': np.float64(218.561), 'std_dev_duration_us': np.float64(5.201450440742469), 'min_duration_us': np.float64(205.03), 'max_duration_us': np.float64(222.872)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(216.74)}]","[[768], [25089, 3072], [3072, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [3072, 1], [1, 3072], [], []]","['', '', '', '1', '1']",216.73714192708334,218.560791015625,5.432787018968682,205.02978515625,222.8720703125,2600.845703125,12,356 +aten::addmm,25096,2304,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,192,64,88.871159808,150.42626953125,563.4264735955439,0.8201891975798056,0.8173786300280079,0.028393483256752255,0.7829438965375163,0.8809421265789645,462.1163072735487,460.5327591090373,15.99764014444606,441.1313186492872,496.3461158201453,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(2310.257), 'mean_duration_us': np.float64(192.52141666666668), 'median_duration_us': np.float64(192.98250000000002), 'std_dev_duration_us': np.float64(6.2790327075956185), 'min_duration_us': np.float64(179.051), 'max_duration_us': np.float64(201.462)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(192.52)}]","[[2304], [25096, 768], [768, 2304], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",192.5213623046875,192.982421875,6.558283927465597,179.05078125,201.4619140625,2310.25634765625,12,231 +aten::addmm,25088,2304,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,192,64,88.842829824,150.37939453125,563.422437389805,0.8451509012425363,0.8482090610697497,0.026524564657361804,0.7771340822670733,0.8732377472905446,476.17698074026026,477.9000166040364,14.944534869954294,437.8547788096037,492.0017399992213,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(2241.0190000000002), 'mean_duration_us': np.float64(186.75158333333334), 'median_duration_us': np.float64(185.906), 'std_dev_duration_us': np.float64(5.873673700764756), 'min_duration_us': np.float64(180.574), 'max_duration_us': np.float64(202.905)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(186.75)}]","[[2304], [25088, 768], [768, 2304], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",186.7515869140625,185.906005859375,6.134801196057336,180.57421875,202.90478515625,2241.01904296875,12,93 +aten::addmm,25088,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,128,64,29.614276608,74.62646484375,378.44985768966535,1.1023945690056698,1.1055407737805627,0.03816811242336863,0.992285767994799,1.1764986232261758,417.2010677580557,418.3917485073764,14.444716714907022,375.53040768511187,445.24573653203345,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_8_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB8_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 24, 'total_duration_us': np.float64(1705.616), 'mean_duration_us': np.float64(71.06733333333334), 'median_duration_us': np.float64(70.7815), 'std_dev_duration_us': np.float64(2.4860613204200916), 'min_duration_us': np.float64(66.512), 'max_duration_us': np.float64(78.86)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(71.07)}]","[[768], [25088, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",71.06734212239583,70.781494140625,2.5394755658784947,66.51220703125,78.85986328125,1705.6162109375,24,162 +aten::addmm,25096,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,128,64,29.623719936,74.64990234375,378.4516787347187,1.0549091731169484,1.053384165887925,0.01133784738837991,1.0407406456236934,1.0775004846651746,399.2321474787631,398.6550059328566,4.290827377370407,393.87004446374175,407.7818672590083,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB1024_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_8_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB8_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 12, 'total_duration_us': np.float64(890.515), 'mean_duration_us': np.float64(74.20958333333333), 'median_duration_us': np.float64(74.31), 'std_dev_duration_us': np.float64(0.7594008008877068), 'min_duration_us': np.float64(72.646), 'max_duration_us': np.float64(75.212)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(74.21)}]","[[768], [25096, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",74.20955403645833,74.309814453125,0.7931931743684959,72.64599609375,75.2119140625,890.5146484375,12,300 +aten::addmm,1,400,768,True,"(19268352, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.0006148,0.58892822265625,0.9955694890662244,0.0666796925185849,0.0666796925185849,,0.0666796925185849,0.0666796925185849,0.06638426741182053,0.06638426741182053,,0.06638426741182053,0.06638426741182053,python3,CPU,thread 13201 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(9.261), 'mean_duration_us': np.float64(9.261), 'median_duration_us': np.float64(9.261), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(9.261), 'max_duration_us': np.float64(9.261)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(9.26)}]","[[400], [1, 768], [768, 400], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [19268352, 1], [1, 768], [], []]","['', '', '', '1', '1']",9.26123046875,9.26123046875,,9.26123046875,9.26123046875,9.26123046875,1,3577 diff --git a/tests/traces/mi300/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv b/tests/traces/mi300/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv index d48769bff..e184c7ab4 100644 --- a/tests/traces/mi300/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv +++ b/tests/traces/mi300/gaunernst_bert-small-uncased__1016001_perf_report_csvs/GEMM.csv @@ -1,5 +1,5 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::addmm,141,512,512,True,"(512, 1)","(1, 512)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.0739968,0.7763671875,90.89622641509433,0.161716610441385,0.16249861988304096,0.007093321468841064,0.14825145296105283,0.17214618895198763,14.69942963776174,14.77051134502924,0.6447561542668266,13.475497634714564,15.64743896747548,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 17, 'total_duration_us': np.float64(85.736), 'mean_duration_us': np.float64(5.043294117647059), 'median_duration_us': np.float64(5.01), 'std_dev_duration_us': np.float64(0.2192558979569931), 'min_duration_us': np.float64(4.729), 'max_duration_us': np.float64(5.491)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(5.04)}]","[[512], [141, 512], [512, 512], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [512, 1], [1, 512], [], []]","['', '', '', '1', '1']",5.043313419117647,5.009765625,0.22600854465708484,4.72900390625,5.4912109375,85.736328125,17,48 -aten::addmm,141,2048,512,True,"(512, 1)","(1, 512)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.2959872,2.6923828125,104.84221980413493,0.35165267620425944,0.3881479117816242,0.07985573206817134,0.23241741624793985,0.39789746500584955,36.86804717331926,40.69428868352502,8.372252214111324,24.367157840575633,41.716453485651364,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 4, 'total_duration_us': np.float64(33.794), 'mean_duration_us': np.float64(8.4485), 'median_duration_us': np.float64(7.276), 'std_dev_duration_us': np.float64(2.1389002431156063), 'min_duration_us': np.float64(7.095), 'max_duration_us': np.float64(12.147)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(8.45)}]","[[2048], [141, 512], [512, 2048], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [512, 1], [1, 512], [], []]","['', '', '', '1', '1']",8.4486083984375,7.276123046875,2.469693365812908,7.09521484375,12.14697265625,33.79443359375,4,133 -aten::addmm,141,512,2048,True,"(2048, 1)","(1, 2048)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.295770624,2.689453125,104.87962962962963,0.38822161106300757,0.3865433379125297,0.0067010700297641225,0.38233527128293393,0.39746449714403687,40.71653878250636,40.540522116066896,0.7028057428438708,40.09918164649808,41.68592925139357,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 4, 'total_duration_us': np.float64(29.063000000000002), 'mean_duration_us': np.float64(7.265750000000001), 'median_duration_us': np.float64(7.296), 'std_dev_duration_us': np.float64(0.10763450887145837), 'min_duration_us': np.float64(7.095), 'max_duration_us': np.float64(7.376)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(7.27)}]","[[512], [141, 2048], [2048, 512], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2048, 1], [1, 2048], [], []]","['', '', '', '1', '1']",7.2657470703125,7.2958984375,0.12417225157361939,7.09521484375,7.3759765625,29.06298828125,4,143 -aten::addmm,141,30522,512,True,"(512, 1)","(1, 512)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",4.41119205,38.211021423339844,110.09495182588434,1.568977890630975,1.568977890630975,,1.568977890630975,1.568977890630975,172.73654528489485,172.73654528489485,,172.73654528489485,172.73654528489485,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x144x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_9_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(25.537), 'mean_duration_us': np.float64(25.537), 'median_duration_us': np.float64(25.537), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(25.537), 'max_duration_us': np.float64(25.537)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x144x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(25.54)}]","[[30522], [141, 512], [512, 30522], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [512, 1], [1, 512], [], []]","['', '', '', '1', '1']",25.537109375,25.537109375,,25.537109375,25.537109375,25.537109375,1,517 +name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aten::addmm,141,512,512,True,"(512, 1)","(1, 512)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,128,0.0739968,0.7763671875,90.89622641509433,0.161716610441385,0.16249861988304096,0.007093321468841064,0.14825145296105283,0.17214618895198763,14.69942963776174,14.77051134502924,0.6447561542668266,13.475497634714564,15.64743896747548,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 17, 'total_duration_us': np.float64(85.736), 'mean_duration_us': np.float64(5.043294117647059), 'median_duration_us': np.float64(5.01), 'std_dev_duration_us': np.float64(0.2192558979569931), 'min_duration_us': np.float64(4.729), 'max_duration_us': np.float64(5.491)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(5.04)}]","[[512], [141, 512], [512, 512], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [512, 1], [1, 512], [], []]","['', '', '', '1', '1']",5.043313419117647,5.009765625,0.22600854465708484,4.72900390625,5.4912109375,85.736328125,17,48 +aten::addmm,141,2048,512,True,"(512, 1)","(1, 512)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",32,64,128,0.2959872,2.6923828125,104.84221980413493,0.35165267620425944,0.3881479117816242,0.07985573206817134,0.23241741624793985,0.39789746500584955,36.86804717331926,40.69428868352502,8.372252214111324,24.367157840575633,41.716453485651364,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG16_16_1', 'stream': 0, 'count': 4, 'total_duration_us': np.float64(33.794), 'mean_duration_us': np.float64(8.4485), 'median_duration_us': np.float64(7.276), 'std_dev_duration_us': np.float64(2.1389002431156063), 'min_duration_us': np.float64(7.095), 'max_duration_us': np.float64(12.147)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT32x64x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(8.45)}]","[[2048], [141, 512], [512, 2048], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [512, 1], [1, 512], [], []]","['', '', '', '1', '1']",8.4486083984375,7.276123046875,2.469693365812908,7.09521484375,12.14697265625,33.79443359375,4,133 +aten::addmm,141,512,2048,True,"(2048, 1)","(1, 2048)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.295770624,2.689453125,104.87962962962963,0.38822161106300757,0.3865433379125297,0.0067010700297641225,0.38233527128293393,0.39746449714403687,40.71653878250636,40.540522116066896,0.7028057428438708,40.09918164649808,41.68592925139357,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 4, 'total_duration_us': np.float64(29.063000000000002), 'mean_duration_us': np.float64(7.265750000000001), 'median_duration_us': np.float64(7.296), 'std_dev_duration_us': np.float64(0.10763450887145837), 'min_duration_us': np.float64(7.095), 'max_duration_us': np.float64(7.376)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(7.27)}]","[[512], [141, 2048], [2048, 512], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [2048, 1], [1, 2048], [], []]","['', '', '', '1', '1']",7.2657470703125,7.2958984375,0.12417225157361939,7.09521484375,7.3759765625,29.06298828125,4,143 +aten::addmm,141,30522,512,True,"(512, 1)","(1, 512)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",128,144,64,4.41119205,38.211021423339844,110.09495182588434,1.568977890630975,1.568977890630975,,1.568977890630975,1.568977890630975,172.73654528489485,172.73654528489485,,172.73654528489485,172.73654528489485,python3,CPU,thread 6240 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x144x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_9_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(25.537), 'mean_duration_us': np.float64(25.537), 'median_duration_us': np.float64(25.537), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(25.537), 'max_duration_us': np.float64(25.537)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x144x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(25.54)}]","[[30522], [141, 512], [512, 30522], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [512, 1], [1, 512], [], []]","['', '', '', '1', '1']",25.537109375,25.537109375,,25.537109375,25.537109375,25.537109375,1,517 diff --git a/tests/traces/mi300/google_owlv2-large-patch14-ensemble__1016001_perf_report_csvs/GEMM.csv b/tests/traces/mi300/google_owlv2-large-patch14-ensemble__1016001_perf_report_csvs/GEMM.csv index 8fa0bbc2c..20512ecc4 100644 --- a/tests/traces/mi300/google_owlv2-large-patch14-ensemble__1016001_perf_report_csvs/GEMM.csv +++ b/tests/traces/mi300/google_owlv2-large-patch14-ensemble__1016001_perf_report_csvs/GEMM.csv @@ -1,19 +1,19 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::bmm,5185,64,5185,False,"(26884225, 5185, 1)","(331840, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",16,"(False, False)",55.0588928,840.6953430175781,62.458121588556374,1.9348781475913366,1.918903858994779,0.15942706277562066,1.5747165906047995,2.2076078621764026,120.84885460130042,119.85113054184595,9.957514871346126,98.35384028351152,137.88304027566676,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x288x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB128_LBSPPM0_LPA32_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_9_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(55056.329), 'mean_duration_us': np.float64(458.80274166666663), 'median_duration_us': np.float64(459.394), 'std_dev_duration_us': np.float64(39.331859811628505), 'min_duration_us': np.float64(399.316), 'max_duration_us': np.float64(559.804)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x288x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(458.8)}]","[[16, 5185, 5185], [16, 5185, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[26884225, 5185, 1], [331840, 64, 1]]","['', '']",458.80274658203126,459.39404296875,39.49678093830489,399.31591796875,559.80419921875,55056.32958984375,120,138 -aten::bmm,5185,5185,64,False,"(331840, 64, 1)","(331840, 1, 64)","('c10::BFloat16', 'c10::BFloat16')",16,"(True, False)",55.0588928,840.6953430175781,62.458121588556374,2.6174472214701487,2.656653715298328,0.0860623729797748,2.3349087987253174,2.7334355348837214,163.48083681021157,165.92960076879297,5.375294155770481,145.83401764897596,170.72524899224808,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(40460.247), 'mean_duration_us': np.float64(337.16872500000005), 'median_duration_us': np.float64(331.821), 'std_dev_duration_us': np.float64(11.541609001032239), 'min_duration_us': np.float64(322.5), 'max_duration_us': np.float64(377.545)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x32_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(337.17)}]","[[16, 5185, 64], [16, 64, 5185]]","['c10::BFloat16', 'c10::BFloat16']","[[331840, 64, 1], [331840, 1, 64]]","['', '']",337.1687459309896,331.82080078125,11.590012366575005,322.5,377.544921875,40460.24951171875,120,132 -aten::addmm,5185,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",10.87904256,22.255859375,466.1722246599386,0.7378558737948471,0.7472497512770964,0.02903544614369141,0.606347057077249,0.7866304697323809,343.9679141653467,348.3470789294297,13.535518522798442,282.6621565137079,366.70527606043646,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_5_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 480, 'total_duration_us': np.float64(15205.791000000001), 'mean_duration_us': np.float64(31.678731250000002), 'median_duration_us': np.float64(31.2305), 'std_dev_duration_us': np.float64(1.291014259160901), 'min_duration_us': np.float64(29.667), 'max_duration_us': np.float64(38.488)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(31.68)}]","[[1024], [5185, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",31.678751627604168,31.23046875,1.2923518699121088,29.6669921875,38.48779296875,15205.80078125,480,81 -aten::addmm,5185,1024,4096,True,"(4096, 1)","(1, 4096)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",43.50024192,58.63671875,707.4929218573046,0.6056462508366827,0.6083364881324489,0.015494993164156314,0.554854236436141,0.6339910213526537,428.49043561636665,430.39375946123783,10.962597987867907,392.5554449411091,448.54416012808576,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT160x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(12190.582999999999), 'mean_duration_us': np.float64(101.58819166666666), 'median_duration_us': np.float64(101.071), 'std_dev_duration_us': np.float64(2.68756934575908), 'min_duration_us': np.float64(96.981), 'max_duration_us': np.float64(110.813)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT160x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(101.59)}]","[[1024], [5185, 4096], [4096, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [4096, 1], [1, 4096], [], []]","['', '', '', '1', '1']",101.58817952473959,101.07080078125,2.698842558047398,96.98095703125,110.81298828125,12190.58154296875,120,184 -aten::addmm,5185,4096,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",43.51617024,58.642578125,707.68126561199,0.6371319783990542,0.6412084337656122,0.014159004993720388,0.5849558619152019,0.6554589191749379,450.88636483531377,453.7711959283303,10.020062573762525,413.96230468730255,463.85599747838717,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x288x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_9_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(11587.363), 'mean_duration_us': np.float64(96.56135833333333), 'median_duration_us': np.float64(95.899), 'std_dev_duration_us': np.float64(2.207858184590643), 'min_duration_us': np.float64(93.814), 'max_duration_us': np.float64(105.121)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x288x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(96.56)}]","[[4096], [5185, 1024], [1024, 4096], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",96.56133626302083,95.89892578125,2.2171139876584305,93.81396484375,105.12109375,11587.3603515625,120,172 -aten::addmm,5,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.00590208,1.14111328125,4.932605905006419,0.17462094776741063,0.1735374344593159,0.00965977537414959,0.1332602159986949,0.20587432680836765,0.8613363180953469,0.8559917739536859,0.04764786505156589,0.6573201283375931,1.0154969201041752,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 240, 'total_duration_us': np.float64(1649.788), 'mean_duration_us': np.float64(6.874116666666667), 'median_duration_us': np.float64(6.895), 'std_dev_duration_us': np.float64(0.40092175428075183), 'min_duration_us': np.float64(5.812), 'max_duration_us': np.float64(8.979)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.87)}]","[[768], [5, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",6.87412109375,6.89501953125,0.40177757505756456,5.81201171875,8.97900390625,1649.7890625,240,3045 -aten::addmm,5184,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",10.876944384,22.251953125,466.1641358729044,0.7510257004162115,0.7662881034053088,0.03379675993331028,0.696168550458181,0.8128479531877254,350.10124665286605,357.21603155362266,15.754837389615584,324.5288107462304,378.92056369381504,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_5_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 20, 'total_duration_us': np.float64(622.568), 'mean_duration_us': np.float64(31.1284), 'median_duration_us': np.float64(30.448999999999998), 'std_dev_duration_us': np.float64(1.3790227119231941), 'min_duration_us': np.float64(28.705), 'max_duration_us': np.float64(33.516)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(31.13)}]","[[1024], [5184, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",31.1284423828125,30.44921875,1.4148984334472765,28.705078125,33.51611328125,622.56884765625,20,4733 -aten::addmm,5,768,3072,True,"(3072, 1)","(1, 3072)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.0235968,4.5380859375,4.9588444157520986,0.5559981933749094,0.5508430454576891,0.024874190872702668,0.5183482444550822,0.6214427588317816,2.7571085363854246,2.731544959923741,0.12334724250545347,2.570408297430988,3.081637954342558,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(514.482), 'mean_duration_us': np.float64(8.5747), 'median_duration_us': np.float64(8.6385), 'std_dev_duration_us': np.float64(0.3650420660691038), 'min_duration_us': np.float64(7.657), 'max_duration_us': np.float64(9.18)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(8.57)}]","[[768], [5, 3072], [3072, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [3072, 1], [1, 3072], [], []]","['', '', '', '1', '1']",8.57470703125,8.638671875,0.3681263248527013,7.6572265625,9.18017578125,514.482421875,60,3154 -aten::addmm,5,3072,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.02360832,4.54248046875,4.9564656562399225,0.685134546466019,0.6908081954535799,0.04742426582345252,0.550129851567787,0.7974252046104798,3.3958458494623383,3.4239670958147435,0.23505674482633523,2.726699715768102,3.952410640071936,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(419.266), 'mean_duration_us': np.float64(6.987766666666667), 'median_duration_us': np.float64(6.895), 'std_dev_duration_us': np.float64(0.5233797654561062), 'min_duration_us': np.float64(5.973), 'max_duration_us': np.float64(8.658)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.99)}]","[[3072], [5, 768], [768, 3072], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",6.987784830729167,6.89501953125,0.5278124854875798,5.97314453125,8.658203125,419.26708984375,60,3142 -aten::bmm,5,64,5,False,"(25, 5, 1)","(320, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",12,"(False, False)",3.84e-05,0.01522064208984375,2.406015037593985,0.0025180843922148104,0.0025442837171007187,0.00025649061322683634,0.001843130709371828,0.0031350546710147706,0.006058548913599544,0.006121584883249849,0.0006171202724254714,0.004434600202999887,0.007542988682140802,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(384.538), 'mean_duration_us': np.float64(6.408966666666667), 'median_duration_us': np.float64(6.273), 'std_dev_duration_us': np.float64(0.7107715049312417), 'min_duration_us': np.float64(5.091), 'max_duration_us': np.float64(8.659)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.41)}]","[[12, 5, 5], [12, 5, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[25, 5, 1], [320, 64, 1]]","['', '']",6.408935546875,6.27294921875,0.7168049733241744,5.0908203125,8.6591796875,384.5361328125,60,3108 -aten::bmm,5,5,64,False,"(320, 64, 1)","(320, 1, 64)","('c10::BFloat16', 'c10::BFloat16')",12,"(True, False)",3.84e-05,0.01522064208984375,2.406015037593985,0.0027180057084622558,0.002746037133495757,0.0002617506725687591,0.0020313268286619845,0.0034628753045873504,0.006539562606826479,0.006607006636982273,0.000629776054300774,0.004887402896028835,0.008331730056150016,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(355.711), 'mean_duration_us': np.float64(5.928516666666667), 'median_duration_us': np.float64(5.812), 'std_dev_duration_us': np.float64(0.600517262912196), 'min_duration_us': np.float64(4.609), 'max_duration_us': np.float64(7.857)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(5.93)}]","[[12, 5, 64], [12, 64, 5]]","['c10::BFloat16', 'c10::BFloat16']","[[320, 64, 1], [320, 1, 64]]","['', '']",5.928507486979167,5.81201171875,0.6055981766940021,4.60888671875,7.85693359375,355.71044921875,60,3096 -aten::addmm,5184,1,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(False, False)",0.010622016,10.136842727661133,0.9993194251711081,0.47779096652487923,0.48919535270455516,0.08082671684087418,0.2017731886140128,0.5410927891427009,0.4774657940195904,0.48886241866109353,0.08077170821189028,0.20163586686069684,0.5407245350103154,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 15, 'total_duration_us': np.float64(69.66600000000001), 'mean_duration_us': np.float64(4.644400000000001), 'median_duration_us': np.float64(4.689), 'std_dev_duration_us': np.float64(0.2239195688932374), 'min_duration_us': np.float64(4.248), 'max_duration_us': np.float64(5.01)}, {'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA1_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB128_LBSPPM0_LPA16_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB4_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS0_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_1', 'stream': 0, 'count': 15, 'total_duration_us': np.float64(282.956), 'mean_duration_us': np.float64(18.863733333333336), 'median_duration_us': np.float64(17.158), 'std_dev_duration_us': np.float64(7.847680162244695), 'min_duration_us': np.float64(15.234), 'max_duration_us': np.float64(47.99)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::na...', 'stream': 0, 'mean_duration_us': np.float64(4.64)}, {'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(18.86)}]","[[1], [5184, 1024], [1024, 1], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",23.508138020833332,21.72802734375,8.146875535940318,19.64404296875,52.67919921875,352.6220703125,15,4691 -aten::addmm,5184,768,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",8.157708288,19.22021484375,404.7715875314382,0.7406942075203564,0.739259883724679,0.022667388614154498,0.7171168943482114,0.7710215584405881,299.8119702533551,299.23139673354467,9.17511487454336,290.26854377094014,312.08762023096034,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x64x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB4_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(136.148), 'mean_duration_us': np.float64(27.229599999999998), 'median_duration_us': np.float64(27.262), 'std_dev_duration_us': np.float64(0.7408710009171641), 'min_duration_us': np.float64(26.139), 'max_duration_us': np.float64(28.104)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x64x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(27.23)}]","[[768], [5184, 1024], [1024, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",27.2296875,27.26220703125,0.828210329661911,26.13916015625,28.10400390625,136.1484375,5,4651 -aten::mm,1,1,768,False,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(False, False)",1.536e-06,0.0029315948486328125,0.4996746909564086,0.00025551842828586206,0.0002581943157117664,1.1700299466573218e-05,0.0002434286598097595,0.00027095123735743494,0.00012767609168740538,0.00012901316490997828,5.846343520057406e-06,0.00012163514036037431,0.0001353874757908328,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA1_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB128_LBSPPM0_LPA16_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB4_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS0_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_1', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(60.253), 'mean_duration_us': np.float64(12.0506), 'median_duration_us': np.float64(11.906), 'std_dev_duration_us': np.float64(0.49213477828741153), 'min_duration_us': np.float64(11.345), 'max_duration_us': np.float64(12.628)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(12.05)}]","[[1, 768], [768, 1]]","['c10::BFloat16', 'c10::BFloat16']","[[768, 1], [1, 768]]","['', '']",12.0505859375,11.90576171875,0.5501090851891149,11.34521484375,12.6279296875,60.2529296875,5,4594 -aten::mm,1,768,768,False,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.001179648,1.1279296875,0.9974025974025974,0.17584613791289605,0.20490741561627612,0.08052748619397093,0.03231208142683724,0.21857160801299405,0.17538939469753786,0.20437518856272735,0.08031832389216842,0.032228153942611686,0.21800388955062264,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(58.97), 'mean_duration_us': np.float64(11.794), 'median_duration_us': np.float64(5.772), 'std_dev_duration_us': np.float64(12.405547646113815), 'min_duration_us': np.float64(5.411), 'max_duration_us': np.float64(36.603)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(11.79)}]","[[1, 768], [768, 768]]","['c10::BFloat16', 'c10::BFloat16']","[[768, 1], [1, 768]]","['', '']",11.79404296875,5.77197265625,13.86981421896285,5.4111328125,36.60302734375,58.97021484375,5,4573 -aten::bmm,5184,1,768,False,"(3981312, 768, 1)","(768, 1, 1)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, True)",0.007962624,7.6051025390625,0.9985072470746857,0.8477543358412456,0.8724736013675944,0.04534565219112569,0.7925381348085602,0.8920112154678027,0.8464888480764708,0.8711712138468936,0.045277962336167106,0.7913550711894016,0.8906796631165,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bjlk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x16x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB1_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(47.143), 'mean_duration_us': np.float64(9.4286), 'median_duration_us': np.float64(9.14), 'std_dev_duration_us': np.float64(0.45920696858823895), 'min_duration_us': np.float64(8.94), 'max_duration_us': np.float64(10.062)}]","[{'name': 'Cijk_Alik_Bjlk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x16x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(9.43)}]","[[1, 5184, 768], [1, 768, 1]]","['c10::BFloat16', 'c10::BFloat16']","[[3981312, 768, 1], [768, 1, 1]]","['', '']",9.42861328125,9.14013671875,0.5133638684162176,8.93994140625,10.06201171875,47.14306640625,5,4679 -aten::addmm,5184,4,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.042488064,10.172370910644531,3.98331674558037,1.7201543992644102,1.750681214297163,0.08366630382099807,1.574414428252252,1.7738530403572879,6.851919823573667,6.973517797082866,0.33326938905099623,6.271391356540541,7.065818519853837,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(31.067), 'mean_duration_us': np.float64(6.2134), 'median_duration_us': np.float64(6.093), 'std_dev_duration_us': np.float64(0.28699937282161453), 'min_duration_us': np.float64(6.013), 'max_duration_us': np.float64(6.775)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.21)}]","[[4], [5184, 1024], [1024, 4], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",6.21337890625,6.0927734375,0.3208009670495085,6.01318359375,6.77490234375,31.06689453125,5,4786 -aten::mm,1,768,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",0.001572864,1.50341796875,0.9977265345891523,0.2619856472467282,0.2587406238179195,0.012911050226540351,0.24732384740309482,0.2789257454859611,0.2613900319395742,0.25815238595928836,0.01288169740043261,0.24676156519074616,0.2782916174514039,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(30.145000000000003), 'mean_duration_us': np.float64(6.029000000000001), 'median_duration_us': np.float64(6.093), 'std_dev_duration_us': np.float64(0.2632732420888988), 'min_duration_us': np.float64(5.652), 'max_duration_us': np.float64(6.374)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.03)}]","[[1, 1024], [1024, 768]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",6.02890625,6.0927734375,0.2943662550613358,5.65185546875,6.3740234375,30.14453125,5,4580 +name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aten::bmm,5185,64,5185,False,"(26884225, 5185, 1)","(331840, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",16,"(False, False)",64,288,64,55.0588928,840.6953430175781,62.458121588556374,1.9348781475913366,1.918903858994779,0.15942706277562066,1.5747165906047995,2.2076078621764026,120.84885460130042,119.85113054184595,9.957514871346126,98.35384028351152,137.88304027566676,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x288x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA1024_LBSPPB128_LBSPPM0_LPA32_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT2_9_MO40_NTn1_NTA0_NTB4_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW2_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA2_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(55056.329), 'mean_duration_us': np.float64(458.80274166666663), 'median_duration_us': np.float64(459.394), 'std_dev_duration_us': np.float64(39.331859811628505), 'min_duration_us': np.float64(399.316), 'max_duration_us': np.float64(559.804)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x288x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(458.8)}]","[[16, 5185, 5185], [16, 5185, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[26884225, 5185, 1], [331840, 64, 1]]","['', '']",458.80274658203126,459.39404296875,39.49678093830489,399.31591796875,559.80419921875,55056.32958984375,120,138 +aten::bmm,5185,5185,64,False,"(331840, 64, 1)","(331840, 1, 64)","('c10::BFloat16', 'c10::BFloat16')",16,"(True, False)",256,128,32,55.0588928,840.6953430175781,62.458121588556374,2.6174472214701487,2.656653715298328,0.0860623729797748,2.3349087987253174,2.7334355348837214,163.48083681021157,165.92960076879297,5.375294155770481,145.83401764897596,170.72524899224808,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA2_GRVWB2_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB256_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT8_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(40460.247), 'mean_duration_us': np.float64(337.16872500000005), 'median_duration_us': np.float64(331.821), 'std_dev_duration_us': np.float64(11.541609001032239), 'min_duration_us': np.float64(322.5), 'max_duration_us': np.float64(377.545)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x128x32_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(337.17)}]","[[16, 5185, 64], [16, 64, 5185]]","['c10::BFloat16', 'c10::BFloat16']","[[331840, 64, 1], [331840, 1, 64]]","['', '']",337.1687459309896,331.82080078125,11.590012366575005,322.5,377.544921875,40460.24951171875,120,132 +aten::addmm,5185,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",128,160,64,10.87904256,22.255859375,466.1722246599386,0.7378558737948471,0.7472497512770964,0.02903544614369141,0.606347057077249,0.7866304697323809,343.9679141653467,348.3470789294297,13.535518522798442,282.6621565137079,366.70527606043646,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_5_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 480, 'total_duration_us': np.float64(15205.791000000001), 'mean_duration_us': np.float64(31.678731250000002), 'median_duration_us': np.float64(31.2305), 'std_dev_duration_us': np.float64(1.291014259160901), 'min_duration_us': np.float64(29.667), 'max_duration_us': np.float64(38.488)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(31.68)}]","[[1024], [5185, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",31.678751627604168,31.23046875,1.2923518699121088,29.6669921875,38.48779296875,15205.80078125,480,81 +aten::addmm,5185,1024,4096,True,"(4096, 1)","(1, 4096)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",160,128,64,43.50024192,58.63671875,707.4929218573046,0.6056462508366827,0.6083364881324489,0.015494993164156314,0.554854236436141,0.6339910213526537,428.49043561636665,430.39375946123783,10.962597987867907,392.5554449411091,448.54416012808576,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT160x128x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT5_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB4_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(12190.582999999999), 'mean_duration_us': np.float64(101.58819166666666), 'median_duration_us': np.float64(101.071), 'std_dev_duration_us': np.float64(2.68756934575908), 'min_duration_us': np.float64(96.981), 'max_duration_us': np.float64(110.813)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT160x128x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(101.59)}]","[[1024], [5185, 4096], [4096, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [4096, 1], [1, 4096], [], []]","['', '', '', '1', '1']",101.58817952473959,101.07080078125,2.698842558047398,96.98095703125,110.81298828125,12190.58154296875,120,184 +aten::addmm,5185,4096,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",128,288,64,43.51617024,58.642578125,707.68126561199,0.6371319783990542,0.6412084337656122,0.014159004993720388,0.5849558619152019,0.6554589191749379,450.88636483531377,453.7711959283303,10.020062573762525,413.96230468730255,463.85599747838717,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x288x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_9_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 120, 'total_duration_us': np.float64(11587.363), 'mean_duration_us': np.float64(96.56135833333333), 'median_duration_us': np.float64(95.899), 'std_dev_duration_us': np.float64(2.207858184590643), 'min_duration_us': np.float64(93.814), 'max_duration_us': np.float64(105.121)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x288x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(96.56)}]","[[4096], [5185, 1024], [1024, 4096], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",96.56133626302083,95.89892578125,2.2171139876584305,93.81396484375,105.12109375,11587.3603515625,120,172 +aten::addmm,5,768,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.00590208,1.14111328125,4.932605905006419,0.17462094776741063,0.1735374344593159,0.00965977537414959,0.1332602159986949,0.20587432680836765,0.8613363180953469,0.8559917739536859,0.04764786505156589,0.6573201283375931,1.0154969201041752,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 240, 'total_duration_us': np.float64(1649.788), 'mean_duration_us': np.float64(6.874116666666667), 'median_duration_us': np.float64(6.895), 'std_dev_duration_us': np.float64(0.40092175428075183), 'min_duration_us': np.float64(5.812), 'max_duration_us': np.float64(8.979)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.87)}]","[[768], [5, 768], [768, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",6.87412109375,6.89501953125,0.40177757505756456,5.81201171875,8.97900390625,1649.7890625,240,3045 +aten::addmm,5184,1024,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",128,160,64,10.876944384,22.251953125,466.1641358729044,0.7510257004162115,0.7662881034053088,0.03379675993331028,0.696168550458181,0.8128479531877254,350.10124665286605,357.21603155362266,15.754837389615584,324.5288107462304,378.92056369381504,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB128_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_5_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB1_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 20, 'total_duration_us': np.float64(622.568), 'mean_duration_us': np.float64(31.1284), 'median_duration_us': np.float64(30.448999999999998), 'std_dev_duration_us': np.float64(1.3790227119231941), 'min_duration_us': np.float64(28.705), 'max_duration_us': np.float64(33.516)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT128x160x64_MI16x...', 'stream': 0, 'mean_duration_us': np.float64(31.13)}]","[[1024], [5184, 1024], [1024, 1024], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",31.1284423828125,30.44921875,1.4148984334472765,28.705078125,33.51611328125,622.56884765625,20,4733 +aten::addmm,5,768,3072,True,"(3072, 1)","(1, 3072)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.0235968,4.5380859375,4.9588444157520986,0.5559981933749094,0.5508430454576891,0.024874190872702668,0.5183482444550822,0.6214427588317816,2.7571085363854246,2.731544959923741,0.12334724250545347,2.570408297430988,3.081637954342558,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(514.482), 'mean_duration_us': np.float64(8.5747), 'median_duration_us': np.float64(8.6385), 'std_dev_duration_us': np.float64(0.3650420660691038), 'min_duration_us': np.float64(7.657), 'max_duration_us': np.float64(9.18)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(8.57)}]","[[768], [5, 3072], [3072, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [3072, 1], [1, 3072], [], []]","['', '', '', '1', '1']",8.57470703125,8.638671875,0.3681263248527013,7.6572265625,9.18017578125,514.482421875,60,3154 +aten::addmm,5,3072,768,True,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.02360832,4.54248046875,4.9564656562399225,0.685134546466019,0.6908081954535799,0.04742426582345252,0.550129851567787,0.7974252046104798,3.3958458494623383,3.4239670958147435,0.23505674482633523,2.726699715768102,3.952410640071936,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(419.266), 'mean_duration_us': np.float64(6.987766666666667), 'median_duration_us': np.float64(6.895), 'std_dev_duration_us': np.float64(0.5233797654561062), 'min_duration_us': np.float64(5.973), 'max_duration_us': np.float64(8.658)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.99)}]","[[3072], [5, 768], [768, 3072], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [768, 1], [1, 768], [], []]","['', '', '', '1', '1']",6.987784830729167,6.89501953125,0.5278124854875798,5.97314453125,8.658203125,419.26708984375,60,3142 +aten::bmm,5,64,5,False,"(25, 5, 1)","(320, 64, 1)","('c10::BFloat16', 'c10::BFloat16')",12,"(False, False)",16,16,256,3.84e-05,0.01522064208984375,2.406015037593985,0.0025180843922148104,0.0025442837171007187,0.00025649061322683634,0.001843130709371828,0.0031350546710147706,0.006058548913599544,0.006121584883249849,0.0006171202724254714,0.004434600202999887,0.007542988682140802,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(384.538), 'mean_duration_us': np.float64(6.408966666666667), 'median_duration_us': np.float64(6.273), 'std_dev_duration_us': np.float64(0.7107715049312417), 'min_duration_us': np.float64(5.091), 'max_duration_us': np.float64(8.659)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.41)}]","[[12, 5, 5], [12, 5, 64]]","['c10::BFloat16', 'c10::BFloat16']","[[25, 5, 1], [320, 64, 1]]","['', '']",6.408935546875,6.27294921875,0.7168049733241744,5.0908203125,8.6591796875,384.5361328125,60,3108 +aten::bmm,5,5,64,False,"(320, 64, 1)","(320, 1, 64)","('c10::BFloat16', 'c10::BFloat16')",12,"(True, False)",16,16,256,3.84e-05,0.01522064208984375,2.406015037593985,0.0027180057084622558,0.002746037133495757,0.0002617506725687591,0.0020313268286619845,0.0034628753045873504,0.006539562606826479,0.006607006636982273,0.000629776054300774,0.004887402896028835,0.008331730056150016,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 60, 'total_duration_us': np.float64(355.711), 'mean_duration_us': np.float64(5.928516666666667), 'median_duration_us': np.float64(5.812), 'std_dev_duration_us': np.float64(0.600517262912196), 'min_duration_us': np.float64(4.609), 'max_duration_us': np.float64(7.857)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(5.93)}]","[[12, 5, 64], [12, 64, 5]]","['c10::BFloat16', 'c10::BFloat16']","[[320, 64, 1], [320, 1, 64]]","['', '']",5.928507486979167,5.81201171875,0.6055981766940021,4.60888671875,7.85693359375,355.71044921875,60,3096 +aten::addmm,5184,1,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(False, False)",16,16,32,0.010622016,10.136842727661133,0.9993194251711081,0.47779096652487923,0.48919535270455516,0.08082671684087418,0.2017731886140128,0.5410927891427009,0.4774657940195904,0.48886241866109353,0.08077170821189028,0.20163586686069684,0.5407245350103154,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1}>(int, at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int, bool)#1})', 'stream': 0, 'count': 15, 'total_duration_us': np.float64(69.66600000000001), 'mean_duration_us': np.float64(4.644400000000001), 'median_duration_us': np.float64(4.689), 'std_dev_duration_us': np.float64(0.2239195688932374), 'min_duration_us': np.float64(4.248), 'max_duration_us': np.float64(5.01)}, {'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA1_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB128_LBSPPM0_LPA16_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB4_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS0_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_1', 'stream': 0, 'count': 15, 'total_duration_us': np.float64(282.956), 'mean_duration_us': np.float64(18.863733333333336), 'median_duration_us': np.float64(17.158), 'std_dev_duration_us': np.float64(7.847680162244695), 'min_duration_us': np.float64(15.234), 'max_duration_us': np.float64(47.99)}]","[{'name': 'void at::native::elementwise_kernel_manual_unroll<128, 8, at::na...', 'stream': 0, 'mean_duration_us': np.float64(4.64)}, {'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(18.86)}]","[[1], [5184, 1024], [1024, 1], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",23.508138020833332,21.72802734375,8.146875535940318,19.64404296875,52.67919921875,352.6220703125,15,4691 +aten::addmm,5184,768,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,64,64,8.157708288,19.22021484375,404.7715875314382,0.7406942075203564,0.739259883724679,0.022667388614154498,0.7171168943482114,0.7710215584405881,299.8119702533551,299.23139673354467,9.17511487454336,290.26854377094014,312.08762023096034,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x64x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_4_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB4_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(136.148), 'mean_duration_us': np.float64(27.229599999999998), 'median_duration_us': np.float64(27.262), 'std_dev_duration_us': np.float64(0.7408710009171641), 'min_duration_us': np.float64(26.139), 'max_duration_us': np.float64(28.104)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT256x64x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(27.23)}]","[[768], [5184, 1024], [1024, 768], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",27.2296875,27.26220703125,0.828210329661911,26.13916015625,28.10400390625,136.1484375,5,4651 +aten::mm,1,1,768,False,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(False, False)",16,16,32,1.536e-06,0.0029315948486328125,0.4996746909564086,0.00025551842828586206,0.0002581943157117664,1.1700299466573218e-05,0.0002434286598097595,0.00027095123735743494,0.00012767609168740538,0.00012901316490997828,5.846343520057406e-06,0.00012163514036037431,0.0001353874757908328,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16x1_SN_LDSB0_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA1_GRVWB4_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA128_LBSPPB128_LBSPPM0_LPA16_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB4_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS0_SPO0_SRVW0_SSO0_SVW4_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_1', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(60.253), 'mean_duration_us': np.float64(12.0506), 'median_duration_us': np.float64(11.906), 'std_dev_duration_us': np.float64(0.49213477828741153), 'min_duration_us': np.float64(11.345), 'max_duration_us': np.float64(12.628)}]","[{'name': 'Cijk_Ailk_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x32_MI16x16...', 'stream': 0, 'mean_duration_us': np.float64(12.05)}]","[[1, 768], [768, 1]]","['c10::BFloat16', 'c10::BFloat16']","[[768, 1], [1, 768]]","['', '']",12.0505859375,11.90576171875,0.5501090851891149,11.34521484375,12.6279296875,60.2529296875,5,4594 +aten::mm,1,768,768,False,"(768, 1)","(1, 768)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.001179648,1.1279296875,0.9974025974025974,0.17584613791289605,0.20490741561627612,0.08052748619397093,0.03231208142683724,0.21857160801299405,0.17538939469753786,0.20437518856272735,0.08031832389216842,0.032228153942611686,0.21800388955062264,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(58.97), 'mean_duration_us': np.float64(11.794), 'median_duration_us': np.float64(5.772), 'std_dev_duration_us': np.float64(12.405547646113815), 'min_duration_us': np.float64(5.411), 'max_duration_us': np.float64(36.603)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(11.79)}]","[[1, 768], [768, 768]]","['c10::BFloat16', 'c10::BFloat16']","[[768, 1], [1, 768]]","['', '']",11.79404296875,5.77197265625,13.86981421896285,5.4111328125,36.60302734375,58.97021484375,5,4573 +aten::bmm,5184,1,768,False,"(3981312, 768, 1)","(768, 1, 1)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, True)",64,16,128,0.007962624,7.6051025390625,0.9985072470746857,0.8477543358412456,0.8724736013675944,0.04534565219112569,0.7925381348085602,0.8920112154678027,0.8464888480764708,0.8711712138468936,0.045277962336167106,0.7913550711894016,0.8906796631165,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bjlk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x16x128_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB1_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA256_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(47.143), 'mean_duration_us': np.float64(9.4286), 'median_duration_us': np.float64(9.14), 'std_dev_duration_us': np.float64(0.45920696858823895), 'min_duration_us': np.float64(8.94), 'max_duration_us': np.float64(10.062)}]","[{'name': 'Cijk_Alik_Bjlk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT64x16x128_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(9.43)}]","[[1, 5184, 768], [1, 768, 1]]","['c10::BFloat16', 'c10::BFloat16']","[[3981312, 768, 1], [768, 1, 1]]","['', '']",9.42861328125,9.14013671875,0.5133638684162176,8.93994140625,10.06201171875,47.14306640625,5,4679 +aten::addmm,5184,4,1024,True,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.042488064,10.172370910644531,3.98331674558037,1.7201543992644102,1.750681214297163,0.08366630382099807,1.574414428252252,1.7738530403572879,6.851919823573667,6.973517797082866,0.33326938905099623,6.271391356540541,7.065818519853837,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMB_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA0_NTB4_NTC0_NTD0_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(31.067), 'mean_duration_us': np.float64(6.2134), 'median_duration_us': np.float64(6.093), 'std_dev_duration_us': np.float64(0.28699937282161453), 'min_duration_us': np.float64(6.013), 'max_duration_us': np.float64(6.775)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.21)}]","[[4], [5184, 1024], [1024, 4], [], []]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16', 'Scalar', 'Scalar']","[[1], [1024, 1], [1, 1024], [], []]","['', '', '', '1', '1']",6.21337890625,6.0927734375,0.3208009670495085,6.01318359375,6.77490234375,31.06689453125,5,4786 +aten::mm,1,768,1024,False,"(1024, 1)","(1, 1024)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",16,16,256,0.001572864,1.50341796875,0.9977265345891523,0.2619856472467282,0.2587406238179195,0.012911050226540351,0.24732384740309482,0.2789257454859611,0.2613900319395742,0.25815238595928836,0.01288169740043261,0.24676156519074616,0.2782916174514039,python3,CPU,thread 14646 (python3),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT1_1_MO40_NTn1_NTA4_NTB0_NTC0_NTD0_NTM0_NEPBS0_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW1_SK0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA0_WSGRB0_WS64_WG16_4_4', 'stream': 0, 'count': 5, 'total_duration_us': np.float64(30.145000000000003), 'mean_duration_us': np.float64(6.029000000000001), 'median_duration_us': np.float64(6.093), 'std_dev_duration_us': np.float64(0.2632732420888988), 'min_duration_us': np.float64(5.652), 'max_duration_us': np.float64(6.374)}]","[{'name': 'Cijk_Alik_Bljk_B_BS_BH_Bias_HA_S_SAV_UserArgs_MT16x16x256_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(6.03)}]","[[1, 1024], [1024, 768]]","['c10::BFloat16', 'c10::BFloat16']","[[1024, 1], [1, 1024]]","['', '']",6.02890625,6.0927734375,0.2943662550613358,5.65185546875,6.3740234375,30.14453125,5,4580 diff --git a/tests/traces/torch_compile_triton/trace_perf_report_csvs/GEMM.csv b/tests/traces/torch_compile_triton/trace_perf_report_csvs/GEMM.csv index 5d00de206..e84d6320c 100644 --- a/tests/traces/torch_compile_triton/trace_perf_report_csvs/GEMM.csv +++ b/tests/traces/torch_compile_triton/trace_perf_report_csvs/GEMM.csv @@ -1,4 +1,4 @@ -name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first -aten::mm,32768,8192,2048,False,"(2048, 1)","(1, 2048)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",1099.511627776,672.0,1560.3809523809523,0.3769563534706426,0.3769563534706426,0.02830943904389355,0.3569385571511183,0.396974149790167,588.1955138345722,588.1955138345722,44.173509456681124,556.960125748945,619.4309019201996,python,CPU,thread 3182306 (python),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO1_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 2, 'total_duration_us': np.float64(3749.165), 'mean_duration_us': np.float64(1874.5825), 'median_duration_us': np.float64(1874.5825), 'std_dev_duration_us': np.float64(99.5475), 'min_duration_us': np.float64(1775.035), 'max_duration_us': np.float64(1974.13)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(1874.58)}]","[[32768, 2048], [2048, 8192], [32768, 8192]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2048, 1], [1, 2048], [8192, 1]]","['', '', '']",1874.58251953125,1874.58251953125,140.7812312508252,1775.03515625,1974.1298828125,3749.1650390625,2,37 -aten::mm,32768,2048,2048,False,"(2048, 1)","(1, 2048)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",274.877906944,264.0,992.969696969697,0.4938291597257226,0.4935452201585823,0.025304131748677804,0.46328380706575706,0.5249423915199687,490.3573910876508,490.0754477017098,25.126236034565874,460.0267815130523,521.2518874341313,python,CPU,thread 3182306 (python),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_12_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB4_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 4, 'total_duration_us': np.float64(2246.691), 'mean_duration_us': np.float64(561.67275), 'median_duration_us': np.float64(560.9114999999999), 'std_dev_duration_us': np.float64(24.95315869158651), 'min_duration_us': np.float64(527.342), 'max_duration_us': np.float64(597.526)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(561.67)}]","[[32768, 2048], [2048, 2048], [32768, 2048]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2048, 1], [1, 2048], [2048, 1]]","['', '', '']",561.6727294921875,560.91162109375,28.81344828937993,527.341796875,597.52587890625,2246.69091796875,4,6 -aten::mm,32768,2048,8192,False,"(8192, 1)","(1, 8192)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",1099.511627776,672.0,1560.3809523809523,0.34648342562333495,0.34648342562333495,,0.34648342562333495,0.34648342562333495,540.6461376583543,540.6461376583543,,540.6461376583543,540.6461376583543,python,CPU,thread 3182306 (python),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(2033.699), 'mean_duration_us': np.float64(2033.699), 'median_duration_us': np.float64(2033.699), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(2033.699), 'max_duration_us': np.float64(2033.699)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(2033.7)}]","[[32768, 8192], [8192, 2048], [32768, 2048]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[8192, 1], [1, 8192], [2048, 1]]","['', '', '']",2033.69921875,2033.69921875,,2033.69921875,2033.69921875,2033.69921875,1,40 +name,param: M,param: N,param: K,param: bias,param: stride_A,param: stride_B,param: dtype_A_B,param: B,param: transpose,impl_param: mt_m,impl_param: mt_n,impl_param: mt_k,GFLOPS_first,Data Moved (MB)_first,FLOPS/Byte_first,TB/s_mean,TB/s_median,TB/s_std,TB/s_min,TB/s_max,TFLOPS/s_mean,TFLOPS/s_median,TFLOPS/s_std,TFLOPS/s_min,TFLOPS/s_max,process_name_first,process_label_first,thread_name_first,Compute Spec,kernel_details__summarize_kernel_stats,trunc_kernel_details,Input Dims_first,Input type_first,Input Strides_first,Concrete Inputs_first,Kernel Time (µs)_mean,Kernel Time (µs)_median,Kernel Time (µs)_std,Kernel Time (µs)_min,Kernel Time (µs)_max,Kernel Time (µs)_sum,name_count,UID_first +aten::mm,32768,8192,2048,False,"(2048, 1)","(1, 2048)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,192,64,1099.511627776,672.0,1560.3809523809523,0.3769563534706426,0.3769563534706426,0.02830943904389355,0.3569385571511183,0.396974149790167,588.1955138345722,588.1955138345722,44.173509456681124,556.960125748945,619.4309019201996,python,CPU,thread 3182306 (python),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO1_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 2, 'total_duration_us': np.float64(3749.165), 'mean_duration_us': np.float64(1874.5825), 'median_duration_us': np.float64(1874.5825), 'std_dev_duration_us': np.float64(99.5475), 'min_duration_us': np.float64(1775.035), 'max_duration_us': np.float64(1974.13)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(1874.58)}]","[[32768, 2048], [2048, 8192], [32768, 8192]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2048, 1], [1, 2048], [8192, 1]]","['', '', '']",1874.58251953125,1874.58251953125,140.7812312508252,1775.03515625,1974.1298828125,3749.1650390625,2,37 +aten::mm,32768,2048,2048,False,"(2048, 1)","(1, 2048)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,192,64,274.877906944,264.0,992.969696969697,0.4938291597257226,0.4935452201585823,0.025304131748677804,0.46328380706575706,0.5249423915199687,490.3573910876508,490.0754477017098,25.126236034565874,460.0267815130523,521.2518874341313,python,CPU,thread 3182306 (python),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA512_LBSPPB512_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT4_12_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW4_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA4_VWB4_WSGRA0_WSGRB0_WS64_WG64_4_1', 'stream': 0, 'count': 4, 'total_duration_us': np.float64(2246.691), 'mean_duration_us': np.float64(561.67275), 'median_duration_us': np.float64(560.9114999999999), 'std_dev_duration_us': np.float64(24.95315869158651), 'min_duration_us': np.float64(527.342), 'max_duration_us': np.float64(597.526)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(561.67)}]","[[32768, 2048], [2048, 2048], [32768, 2048]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[2048, 1], [1, 2048], [2048, 1]]","['', '', '']",561.6727294921875,560.91162109375,28.81344828937993,527.341796875,597.52587890625,2246.69091796875,4,6 +aten::mm,32768,2048,8192,False,"(8192, 1)","(1, 8192)","('c10::BFloat16', 'c10::BFloat16')",1,"(True, False)",256,192,64,1099.511627776,672.0,1560.3809523809523,0.34648342562333495,0.34648342562333495,,0.34648342562333495,0.34648342562333495,540.6461376583543,540.6461376583543,,540.6461376583543,540.6461376583543,python,CPU,thread 3182306 (python),matrix_bf16,"[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTLA0_DTLB0_DTVA0_DTVB0_EPS0_FDSI0_GRPM1_GRVWA8_GRVWB8_GSUAMBSK_GLS0_ISA942_IU1_K1_LDSTI0_LBSPPA1024_LBSPPB256_LBSPPM0_LPA16_LPB16_LPM0_LRVW8_LWPMn1_MIAV0_MIWT8_6_MO40_NTn1_NTA0_NTB0_NTC0_NTD4_NTM0_NEPBS16_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO0_SRVW0_SSO0_SVW8_SK0_SKFTR0_SKXCCM0_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA8_VWB2_WSGRA0_WSGRB0_WS64_WG32_8_1', 'stream': 0, 'count': 1, 'total_duration_us': np.float64(2033.699), 'mean_duration_us': np.float64(2033.699), 'median_duration_us': np.float64(2033.699), 'std_dev_duration_us': np.float64(0.0), 'min_duration_us': np.float64(2033.699), 'max_duration_us': np.float64(2033.699)}]","[{'name': 'Cijk_Alik_Bljk_BBS_BH_Bias_HA_S_SAV_UserArgs_MT256x192x64_MI16x1...', 'stream': 0, 'mean_duration_us': np.float64(2033.7)}]","[[32768, 8192], [8192, 2048], [32768, 2048]]","['c10::BFloat16', 'c10::BFloat16', 'c10::BFloat16']","[[8192, 1], [1, 8192], [2048, 1]]","['', '', '']",2033.69921875,2033.69921875,,2033.69921875,2033.69921875,2033.69921875,1,40