[clang] 267f592 - [Headers][X86] Allow _mm_cmov_si128/_mm256_cmov_si256 intrinsics to be used in constexpr (#153236)

[via cfe-commits]

Author: Simon Pilgrim
Date: 2025-08-13T08:53:26+01:00
New Revision: 267f592ca0cc251e8b26372546ea8545bf2b4a27

URL: https://github.com/llvm/llvm-project/commit/267f592ca0cc251e8b26372546ea8545bf2b4a27
DIFF: https://github.com/llvm/llvm-project/commit/267f592ca0cc251e8b26372546ea8545bf2b4a27.diff

LOG: [Headers][X86] Allow _mm_cmov_si128/_mm256_cmov_si256 intrinsics to be used in constexpr (#153236)

Added: 
    

Modified: 
    clang/lib/Headers/xopintrin.h
    clang/test/CodeGen/X86/xop-builtins.c

Removed: 
    


################################################################################
diff  --git a/clang/lib/Headers/xopintrin.h b/clang/lib/Headers/xopintrin.h
index 976cdf4902a40..fb88a9060574e 100644
--- a/clang/lib/Headers/xopintrin.h
+++ b/clang/lib/Headers/xopintrin.h
@@ -20,6 +20,14 @@
 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("xop"), __min_vector_width__(128)))
 #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("xop"), __min_vector_width__(256)))
 
+#if defined(__cplusplus) && (__cplusplus >= 201103L)
+#define __DEFAULT_FN_ATTRS_CONSTEXPR __DEFAULT_FN_ATTRS constexpr
+#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256 constexpr
+#else
+#define __DEFAULT_FN_ATTRS_CONSTEXPR __DEFAULT_FN_ATTRS
+#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256
+#endif
+
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_maccs_epi16(__m128i __A, __m128i __B, __m128i __C)
 {
@@ -182,13 +190,13 @@ _mm_hsubq_epi32(__m128i __A)
   return (__m128i)__builtin_ia32_vphsubdq((__v4si)__A);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS
+static __inline__ __m128i __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm_cmov_si128(__m128i __A, __m128i __B, __m128i __C)
 {
   return (__m128i)(((__v2du)__A & (__v2du)__C) | ((__v2du)__B & ~(__v2du)__C));
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
+static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR
 _mm256_cmov_si256(__m256i __A, __m256i __B, __m256i __C)
 {
   return (__m256i)(((__v4du)__A & (__v4du)__C) | ((__v4du)__B & ~(__v4du)__C));
@@ -766,5 +774,7 @@ _mm256_frcz_pd(__m256d __A)
 
 #undef __DEFAULT_FN_ATTRS
 #undef __DEFAULT_FN_ATTRS256
+#undef __DEFAULT_FN_ATTRS_CONSTEXPR
+#undef __DEFAULT_FN_ATTRS256_CONSTEXPR
 
 #endif /* __XOPINTRIN_H */

diff  --git a/clang/test/CodeGen/X86/xop-builtins.c b/clang/test/CodeGen/X86/xop-builtins.c
index e05ed92d9fa14..eb9f832ac9b0b 100644
--- a/clang/test/CodeGen/X86/xop-builtins.c
+++ b/clang/test/CodeGen/X86/xop-builtins.c
@@ -9,6 +9,7 @@
 
 
 #include <x86intrin.h>
+#include "builtin_test_helpers.h"
 
 // NOTE: This should match the tests in llvm/test/CodeGen/X86/xop-intrinsics-fast-isel.ll
 
@@ -182,6 +183,7 @@ __m128i test_mm_cmov_si128(__m128i a, __m128i b, __m128i c) {
   // CHECK-NEXT: %{{.*}} = or <2 x i64> [[AND]], [[ANDN]]
   return _mm_cmov_si128(a, b, c);
 }
+TEST_CONSTEXPR(match_v4si(_mm_cmov_si128((__m128i)(__v4si){+1,+2,+3,+4}, (__m128i)(__v4si){-4,-3,-2,-1}, (__m128i)(__v4si){-1,0,0,-1}), +1, -3, -2, +4));
 
 __m256i test_mm256_cmov_si256(__m256i a, __m256i b, __m256i c) {
   // CHECK-LABEL: test_mm256_cmov_si256
@@ -191,6 +193,7 @@ __m256i test_mm256_cmov_si256(__m256i a, __m256i b, __m256i c) {
   // CHECK-NEXT: %{{.*}} = or <4 x i64> [[AND]], [[ANDN]]
   return _mm256_cmov_si256(a, b, c);
 }
+TEST_CONSTEXPR(match_v4di(_mm256_cmov_si256((__m256i)(__v4di){+1,+2,+3,+4}, (__m256i)(__v4di){-4,-3,-2,-1}, (__m256i)(__v4di){0,-1,0,-1}), -4, +2, -2, +4));
 
 __m128i test_mm_perm_epi8(__m128i a, __m128i b, __m128i c) {
   // CHECK-LABEL: test_mm_perm_epi8