Fix BF16 detection in x86_64 ukernel cmake (iree-org#17983)

`_mm512_cvtneps_pbh` was used incorrectly in IREE_UK_BUILD_X86_64_AVX512_BF16 cmake test, so it always failed. Signed-off-by: vient <[email protected]>
nod-ai · Aug 8, 2024 · e57ccaa · e57ccaa
1 parent 7ab66ff
commit e57ccaa
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/runtime/src/iree/builtins/ukernel/arch/x86_64/CMakeLists.txt b/runtime/src/iree/builtins/ukernel/arch/x86_64/CMakeLists.txt
@@ -316,8 +316,8 @@ if(IREE_UK_TRY_X86_64_AVX512_BF16)
   string(JOIN "\n" IREE_UK_BUILD_X86_64_AVX512_BF16_TEST
     "#include <immintrin.h>"
     "int main() {"
-    "  __m256bh a;"
-    "  _mm512_cvtneps_pbh(a);"
+    "  __m512 a;"
+    "  __m256bh b = _mm512_cvtneps_pbh(a);"
     "  return 0;"
     "}"
   )