[clang] [Headers][X86] Allow _mm_max_pu8 / _mm_min_pu8 to be used in constexpr (PR #160489)

[via cfe-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-clang

Author: Simon Pilgrim (RKSimon)

<details>
<summary>Changes</summary>

These were missed in an earlier patch

---
Full diff: https://github.com/llvm/llvm-project/pull/160489.diff


2 Files Affected:

- (modified) clang/lib/Headers/xmmintrin.h (+4-6) 
- (modified) clang/test/CodeGen/X86/mmx-builtins.c (+2-2) 


``````````diff
diff --git a/clang/lib/Headers/xmmintrin.h b/clang/lib/Headers/xmmintrin.h
index 4891e3ce077b5..d876b4735a7d2 100644
--- a/clang/lib/Headers/xmmintrin.h
+++ b/clang/lib/Headers/xmmintrin.h
@@ -2363,9 +2363,8 @@ _mm_max_pi16(__m64 __a, __m64 __b) {
 /// \param __b
 ///    A 64-bit integer vector containing one of the source operands.
 /// \returns A 64-bit integer vector containing the comparison results.
-static __inline__ __m64 __DEFAULT_FN_ATTRS_SSE2
-_mm_max_pu8(__m64 __a, __m64 __b)
-{
+static __inline__ __m64 __DEFAULT_FN_ATTRS_SSE2_CONSTEXPR
+_mm_max_pu8(__m64 __a, __m64 __b) {
   return (__m64)__builtin_elementwise_max((__v8qu)__a, (__v8qu)__b);
 }
 
@@ -2400,9 +2399,8 @@ _mm_min_pi16(__m64 __a, __m64 __b) {
 /// \param __b
 ///    A 64-bit integer vector containing one of the source operands.
 /// \returns A 64-bit integer vector containing the comparison results.
-static __inline__ __m64 __DEFAULT_FN_ATTRS_SSE2
-_mm_min_pu8(__m64 __a, __m64 __b)
-{
+static __inline__ __m64 __DEFAULT_FN_ATTRS_SSE2_CONSTEXPR
+_mm_min_pu8(__m64 __a, __m64 __b) {
   return (__m64)__builtin_elementwise_min((__v8qu)__a, (__v8qu)__b);
 }
 
diff --git a/clang/test/CodeGen/X86/mmx-builtins.c b/clang/test/CodeGen/X86/mmx-builtins.c
index 43d9ec5e6cc8b..266c78ebe7a3f 100644
--- a/clang/test/CodeGen/X86/mmx-builtins.c
+++ b/clang/test/CodeGen/X86/mmx-builtins.c
@@ -371,7 +371,6 @@ __m64 test_mm_max_pi16(__m64 a, __m64 b) {
   // CHECK: call <4 x i16> @llvm.smax.v4i16(
   return _mm_max_pi16(a, b);
 }
-
 TEST_CONSTEXPR(match_v4hi(_mm_max_pi16((__m64)(__v4hi){+1, -2, +3, -4}, (__m64)(__v4hi){-1, 2, -3, 4}), 1, 2, 3, 4));
 
 __m64 test_mm_max_pu8(__m64 a, __m64 b) {
@@ -379,13 +378,13 @@ __m64 test_mm_max_pu8(__m64 a, __m64 b) {
   // CHECK: call <8 x i8> @llvm.umax.v8i8(
   return _mm_max_pu8(a, b);
 }
+TEST_CONSTEXPR(match_v8qi(_mm_max_pu8((__m64)(__v8qs){ 16, 17, 18, -19, -20, 21, -22, -23}, (__m64)(__v8qs){ 1, -2, -3, 4, 5, 0, 7, -8}),  16, -2, -3, -19, -20, 21, -22, -8));
 
 __m64 test_mm_min_pi16(__m64 a, __m64 b) {
   // CHECK-LABEL: test_mm_min_pi16
   // CHECK: call <4 x i16> @llvm.smin.v4i16(
   return _mm_min_pi16(a, b);
 }
-
 TEST_CONSTEXPR(match_v4hi(_mm_min_pi16((__m64)(__v4hi){+1, -2, +3, -4}, (__m64)(__v4hi){-1, 2, -3, 4}), -1, -2, -3, -4));
 
 __m64 test_mm_min_pu8(__m64 a, __m64 b) {
@@ -393,6 +392,7 @@ __m64 test_mm_min_pu8(__m64 a, __m64 b) {
   // CHECK: call <8 x i8> @llvm.umin.v8i8(
   return _mm_min_pu8(a, b);
 }
+TEST_CONSTEXPR(match_v8qi(_mm_min_pu8((__m64)(__v8qs){ 16, 17, 18, -19, -20, 21, -22, -23}, (__m64)(__v8qs){ 1, -2, -3, 4, 5, 0, 7, -8}),  1, 17, 18, 4, 5, 0, 7, -23));
 
 int test_mm_movemask_pi8(__m64 a) {
   // CHECK-LABEL: test_mm_movemask_pi8

``````````

</details>


https://github.com/llvm/llvm-project/pull/160489