jit/a64: avoid redundant icache flush on macOS after generateProgram

src/jit_compiler_a64.cpp

@@ -116,7 +116,7 @@ void JitCompilerA64::enableWriting()
 
 void JitCompilerA64::enableExecution()
 {
-	setPagesRX(code, CodeSize + CalcDatasetItemSize);
+	setPagesRXKeepIcache(code, CodeSize + CalcDatasetItemSize);
 }
 
 void JitCompilerA64::enableAll()

src/virtual_memory.c

@@ -198,6 +198,22 @@ void setPagesRX(void* ptr, size_t bytes) {
 #endif
 }
 
+void setPagesRXKeepIcache(void* ptr, size_t bytes) {
+#if defined(USE_PTHREAD_JIT_WP) && defined(MAC_OS_VERSION_11_0) \
+	&& MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_VERSION_11_0
+	if (__builtin_available(macOS 11.0, *)) {
+		/* The A64 JIT already called __builtin___clear_cache on the exact
+		   modified range inside generateProgram. Only toggle write-protect
+		   here; re-flushing the full buffer is redundant and measurably
+		   expensive on Apple Silicon. */
+		pthread_jit_write_protect_np(1);
+		return;
+	}
+#endif
+	char *errfunc;
+	pageProtect(ptr, bytes, PAGE_EXECUTE_READ, &errfunc);
+}
+
 void setPagesRWX(void* ptr, size_t bytes) {
 	char *errfunc;
 	pageProtect(ptr, bytes, PAGE_EXECUTE_READWRITE, &errfunc);

src/virtual_memory.h

@@ -39,6 +39,7 @@ extern "C" {
 void* allocMemoryPages(size_t);
 void setPagesRW(void*, size_t);
 void setPagesRX(void*, size_t);
+void setPagesRXKeepIcache(void*, size_t);
 void setPagesRWX(void*, size_t);
 void* allocLargePagesMemory(size_t);
 void freePagedMemory(void*, size_t);