third_party/pto-isa: Update reference to recent state - #663
Open
ChristosMatzoros wants to merge 1 commit into
Open
third_party/pto-isa: Update reference to recent state#663ChristosMatzoros wants to merge 1 commit into
ChristosMatzoros wants to merge 1 commit into
Conversation
Signed-off-by: Christos Konstantinos Matzoros <christos.konstantinos.matzoros@h-partners.com>
EdwardXuy
pushed a commit
to EdwardXuy/sgl-kernel-npu
that referenced
this pull request
Aug 10, 2026
1. Always install triton-ascend 3.2.1 (per dev advice) — fixes triton.language.extra.cann missing on CANN 8.5.0 2. sglang stub: uninstall partial sglang first, then create clean stub providing only is_npu (real sglang __init__ needs orjson etc.) 3. Merge PR sgl-project#663: bump pto-isa to 6081a2d3 (tests pass per developer) 4. F injection kept (fixes NameError in test_swiglu_quant; test now fails on precision assertion, not missing import)
EdwardXuy
pushed a commit
to EdwardXuy/sgl-kernel-npu
that referenced
this pull request
Aug 10, 2026
1. Revert pto-isa to af66b6a (pre-sgl-project#663) — sgl-project#663's 6081a2d3 causes SYNC_AIV_ONLY_ALL ambiguity in mega_kernel.cpp, breaking Build/Daily/DeepEP 2. Fix sglang stub: uninstall partial sglang (--no-deps installs real pkg whose __init__ needs orjson) then create clean stub providing only is_npu 3. F injection kept (works, fixes NameError in test_swiglu_quant) No triton version change, no pto-isa bump — safe incremental fix on 9cb432b base
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Update third_party/pto-isa: bump
af66b6ad→6081a2d3One-line change: the
third_party/pto-isasubmodule reference.The current pin is from 2026-05-06 — three months old — and predates the whole fused
multiply-add instruction family, including
TMULADDDST, whichPR #601 needs for the A5
causal_conv1dpath.
af66b6ada43c40ad9583731883fb9bef65ac840a(2026-05-06)6081a2d3—mastertip (2026-08-06)https://gitcode.com/cann/pto-isa.gitNotable additions
New instructions, for both
npu/a5andnpu/a2a3:TMulAddDst,TFusedMulAdd,TFusedMulAddReluTSubRelu,TSubReluConv,TAddDeqRelu,TAddReluConvTPrefetchAsyncTPairReduceSum,TTernOp, and (a5)TInterleave,TDeInterleavePlus dtype coverage extensions (
TRowMax/TRowMinuint8/int8,TABSint8/16/32,TCmpbf16),A5
TMOVDN→ZZ, fixpipe support inTPUSH, and precision fixes to the new fused ops.TMULADDDSTis the one PR #601 depends on: an element-wise fused multiply-accumulate,dst = src0 * src1 + dst, lowering to a single instruction on both families (vmulaon dav-c310,vmlaon dav-c220).Validation
mega_chunk_gdn— 910B2.sobuilt at6081a2d3is byte-identical to one built atb963dcb6, 143 commits earlier — the intervening range does not touch the a2a3 codegen these kernels use