[Windows] Fix '__builtin_clz' on windows (#3312)

dchigarev · web-flow · commit 36fece6765c0 · 2025-01-31T14:43:04.000+01:00
Closes #3273 This recent PR in upstream (triton-lang/triton#5621) brought a new faster logic for `pext_i32` that is used in `ReduceOpToLLVM` pattern. The new logic of `pext_i32` uses `__builtin_clz` intrinsic, that is natively available in GCC and Clang, but is missing in MSVC. It seems that the Windows version of this intrinsic was incorrectly copied from [the given source](https://gist.github.com/pps83/3210a2f980fd02bb2ba2e5a1fc4a2ef0#file-ctz_clz-cpp-L44-L55), so that it misses `r ^ 31` at the end of it, causing `tt.reduce(...)` lowering to produce incorrect llvm IR in some scenarious. Signed-off-by: dchigarev <dmitry.chigarev@intel.com>
diff --git a/lib/Conversion/TritonGPUToLLVM/Utility.cpp b/lib/Conversion/TritonGPUToLLVM/Utility.cpp
@@ -15,7 +15,7 @@
 static int __builtin_clz(unsigned x) {
   unsigned long r;
   _BitScanReverse(&r, x);
-  return static_cast<int>(r);
+  return static_cast<int>(r ^ 31);
 }
 
 static int __builtin_ctz(unsigned x) {

Original file line number	Diff line number	Diff line change
`@@ -15,7 +15,7 @@`
`15`	`15`	`static int __builtin_clz(unsigned x) {`
`16`	`16`	`unsigned long r;`
`17`	`17`	`_BitScanReverse(&r, x);`
`18`		`- return static_cast<int>(r);`
	`18`	`+ return static_cast<int>(r ^ 31);`
`19`	`19`	`}`
`20`	`20`
`21`	`21`	`static int __builtin_ctz(unsigned x) {`