From 173a1f12bebfd652664f1ae76dcdd096d5b1fe08 Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Fri, 8 Aug 2025 16:32:44 +0100 Subject: [PATCH] [Headers][X86] Allow _mm512_set1_epi8/16/pd/ps intrinsics to be used in constexpr Pulled out of #152288 as I need this to proceed with several other patches --- clang/lib/Headers/avx512fintrin.h | 8 ++++---- clang/test/CodeGen/X86/avx512f-builtins.c | 18 ++++++++++++++++++ 2 files changed, 22 insertions(+), 4 deletions(-) diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h index e3bc71ed8e918..95b80cc59bb02 100644 --- a/clang/lib/Headers/avx512fintrin.h +++ b/clang/lib/Headers/avx512fintrin.h @@ -277,20 +277,20 @@ _mm512_setzero_pd(void) { return __extension__(__m512d){0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0}; } -static __inline __m512 __DEFAULT_FN_ATTRS512 +static __inline __m512 __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_set1_ps(float __w) { return __extension__ (__m512){ __w, __w, __w, __w, __w, __w, __w, __w, __w, __w, __w, __w, __w, __w, __w, __w }; } -static __inline __m512d __DEFAULT_FN_ATTRS512 +static __inline __m512d __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_set1_pd(double __w) { return __extension__ (__m512d){ __w, __w, __w, __w, __w, __w, __w, __w }; } -static __inline __m512i __DEFAULT_FN_ATTRS512 +static __inline __m512i __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_set1_epi8(char __w) { return __extension__ (__m512i)(__v64qi){ @@ -304,7 +304,7 @@ _mm512_set1_epi8(char __w) __w, __w, __w, __w, __w, __w, __w, __w }; } -static __inline __m512i __DEFAULT_FN_ATTRS512 +static __inline __m512i __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_set1_epi16(short __w) { return __extension__ (__m512i)(__v32hi){ diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c index d59799e2c24e5..8c446f5715c00 100644 --- a/clang/test/CodeGen/X86/avx512f-builtins.c +++ b/clang/test/CodeGen/X86/avx512f-builtins.c @@ -439,6 +439,22 @@ __m512d test_mm512_set1_pd(double d) // CHECK: insertelement <8 x double> {{.*}}, i32 7 return _mm512_set1_pd(d); } +TEST_CONSTEXPR(match_m512d(_mm512_set1_pd(-100.0), -100.0, -100.0, -100.0, -100.0, -100.0, -100.0, -100.0, -100.0)); + +__m512 test_mm512_set1_ps(float d) +{ + // CHECK-LABEL: test_mm512_set1_ps + // CHECK: insertelement <16 x float> {{.*}}, i32 0 + // CHECK: insertelement <16 x float> {{.*}}, i32 1 + // CHECK: insertelement <16 x float> {{.*}}, i32 2 + // CHECK: insertelement <16 x float> {{.*}}, i32 3 + // CHECK: insertelement <16 x float> {{.*}}, i32 4 + // CHECK: insertelement <16 x float> {{.*}}, i32 5 + // CHECK: insertelement <16 x float> {{.*}}, i32 6 + // CHECK: insertelement <16 x float> {{.*}}, i32 15 + return _mm512_set1_ps(d); +} +TEST_CONSTEXPR(match_m512(_mm512_set1_ps(-55.0f), -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f, -55.0f)); __mmask16 test_mm512_knot(__mmask16 a) { @@ -9046,6 +9062,7 @@ __m512i test_mm512_set1_epi8(char d) // CHECK: insertelement <64 x i8> {{.*}}, i32 63 return _mm512_set1_epi8(d); } +TEST_CONSTEXPR(match_v64qi(_mm512_set1_epi8(127), 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127)); __m512i test_mm512_set1_epi16(short d) { @@ -9061,6 +9078,7 @@ __m512i test_mm512_set1_epi16(short d) // CHECK: insertelement <32 x i16> {{.*}}, i32 31 return _mm512_set1_epi16(d); } +TEST_CONSTEXPR(match_v32hi(_mm512_set1_epi16(-511), -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511, -511)); __m512i test_mm512_set1_epi32(int d) {