Author: Pedro Lobo Date: 2025-08-07T10:46:22+01:00 New Revision: 5805e887458801f2756d0466b84b712472507f2f
URL: https://github.com/llvm/llvm-project/commit/5805e887458801f2756d0466b84b712472507f2f DIFF: https://github.com/llvm/llvm-project/commit/5805e887458801f2756d0466b84b712472507f2f.diff LOG: [Headers][X86] Allow AVX512 reduction intrinsics to be used in constexpr (#152363) Closes #152324. Part of #30794. This PR adds `constexpr` support for the following AVX512 integer reduction intrinsics: - `_mm512_reduce_add_epi32` - `_mm512_reduce_add_epi64` - `_mm512_reduce_mul_epi32` - `_mm512_reduce_mul_epi64` - `_mm512_reduce_and_epi32` - `_mm512_reduce_and_epi64` - `_mm512_reduce_or_epi32` - `_mm512_reduce_or_epi64` - `_mm512_reduce_max_epi32` - `_mm512_reduce_max_epi64` - `_mm512_reduce_min_epi32` - `_mm512_reduce_min_epi64` - `_mm512_reduce_max_epu32` - `_mm512_reduce_max_epu64` - `_mm512_reduce_min_epu32` - `_mm512_reduce_min_epu64` --------- Co-authored-by: Simon Pilgrim <llvm-...@redking.me.uk> Added: Modified: clang/lib/Headers/avx512fintrin.h clang/test/CodeGen/X86/avx512-reduceIntrin.c clang/test/CodeGen/X86/avx512-reduceMinMaxIntrin.c Removed: ################################################################################ diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h index 440552c457b3d..3ba21cd73888c 100644 --- a/clang/lib/Headers/avx512fintrin.h +++ b/clang/lib/Headers/avx512fintrin.h @@ -9337,19 +9337,23 @@ _mm512_mask_abs_pd(__m512d __W, __mmask8 __K, __m512d __A) * This takes log2(n) steps where n is the number of elements in the vector. */ -static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_add_epi64(__m512i __W) { +static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR +_mm512_reduce_add_epi64(__m512i __W) { return __builtin_reduce_add((__v8di)__W); } -static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_mul_epi64(__m512i __W) { +static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR +_mm512_reduce_mul_epi64(__m512i __W) { return __builtin_reduce_mul((__v8di)__W); } -static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_and_epi64(__m512i __W) { +static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR +_mm512_reduce_and_epi64(__m512i __W) { return __builtin_reduce_and((__v8di)__W); } -static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_or_epi64(__m512i __W) { +static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR +_mm512_reduce_or_epi64(__m512i __W) { return __builtin_reduce_or((__v8di)__W); } @@ -9400,22 +9404,22 @@ _mm512_mask_reduce_mul_pd(__mmask8 __M, __m512d __W) { return __builtin_ia32_reduce_fmul_pd512(1.0, __W); } -static __inline__ int __DEFAULT_FN_ATTRS512 +static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_add_epi32(__m512i __W) { return __builtin_reduce_add((__v16si)__W); } -static __inline__ int __DEFAULT_FN_ATTRS512 +static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_mul_epi32(__m512i __W) { return __builtin_reduce_mul((__v16si)__W); } -static __inline__ int __DEFAULT_FN_ATTRS512 +static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_and_epi32(__m512i __W) { return __builtin_reduce_and((__v16si)__W); } -static __inline__ int __DEFAULT_FN_ATTRS512 +static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_or_epi32(__m512i __W) { return __builtin_reduce_or((__v16si)__W); } @@ -9466,22 +9470,22 @@ _mm512_mask_reduce_mul_ps(__mmask16 __M, __m512 __W) { return __builtin_ia32_reduce_fmul_ps512(1.0f, __W); } -static __inline__ long long __DEFAULT_FN_ATTRS512 +static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_max_epi64(__m512i __V) { return __builtin_reduce_max((__v8di)__V); } -static __inline__ unsigned long long __DEFAULT_FN_ATTRS512 +static __inline__ unsigned long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_max_epu64(__m512i __V) { return __builtin_reduce_max((__v8du)__V); } -static __inline__ long long __DEFAULT_FN_ATTRS512 +static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_min_epi64(__m512i __V) { return __builtin_reduce_min((__v8di)__V); } -static __inline__ unsigned long long __DEFAULT_FN_ATTRS512 +static __inline__ unsigned long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_min_epu64(__m512i __V) { return __builtin_reduce_min((__v8du)__V); } @@ -9509,22 +9513,22 @@ _mm512_mask_reduce_min_epu64(__mmask8 __M, __m512i __V) { __V = _mm512_mask_mov_epi64(_mm512_set1_epi64(-1LL), __M, __V); return __builtin_reduce_min((__v8du)__V); } -static __inline__ int __DEFAULT_FN_ATTRS512 +static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_max_epi32(__m512i __V) { return __builtin_reduce_max((__v16si)__V); } -static __inline__ unsigned int __DEFAULT_FN_ATTRS512 +static __inline__ unsigned int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_max_epu32(__m512i __V) { return __builtin_reduce_max((__v16su)__V); } -static __inline__ int __DEFAULT_FN_ATTRS512 +static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_min_epi32(__m512i __V) { return __builtin_reduce_min((__v16si)__V); } -static __inline__ unsigned int __DEFAULT_FN_ATTRS512 +static __inline__ unsigned int __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_min_epu32(__m512i __V) { return __builtin_reduce_min((__v16su)__V); } diff --git a/clang/test/CodeGen/X86/avx512-reduceIntrin.c b/clang/test/CodeGen/X86/avx512-reduceIntrin.c index 2ceac3a8b2670..0f781fc9aacca 100644 --- a/clang/test/CodeGen/X86/avx512-reduceIntrin.c +++ b/clang/test/CodeGen/X86/avx512-reduceIntrin.c @@ -1,30 +1,35 @@ // RUN: %clang_cc1 -ffreestanding %s -O0 -triple=x86_64-apple-darwin -target-cpu skylake-avx512 -emit-llvm -o - -Wall -Werror | FileCheck %s #include <immintrin.h> +#include "builtin_test_helpers.h" long long test_mm512_reduce_add_epi64(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_add_epi64( // CHECK: call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_add_epi64(__W); } +TEST_CONSTEXPR(_mm512_reduce_add_epi64((__m512i)(__v8di){-4, -3, -2, -1, 0, 1, 2, 3}) == -4); long long test_mm512_reduce_mul_epi64(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_mul_epi64( // CHECK: call i64 @llvm.vector.reduce.mul.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_mul_epi64(__W); } +TEST_CONSTEXPR(_mm512_reduce_mul_epi64((__m512i)(__v8di){1, 2, 3, 4, 5, 6, 7, 8}) == 40320); long long test_mm512_reduce_or_epi64(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_or_epi64( // CHECK: call i64 @llvm.vector.reduce.or.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_or_epi64(__W); } +TEST_CONSTEXPR(_mm512_reduce_or_epi64((__m512i)(__v8di){0x100, 0x200, 0x400, 0x800, 0, 0, 0, 0}) == 0xF00); long long test_mm512_reduce_and_epi64(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_and_epi64( // CHECK: call i64 @llvm.vector.reduce.and.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_and_epi64(__W); } +TEST_CONSTEXPR(_mm512_reduce_and_epi64((__m512i)(__v8di){0xFFFF, 0xFF00, 0x00FF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFF00, 0x00FF}) == 0x0000); long long test_mm512_mask_reduce_add_epi64(__mmask8 __M, __m512i __W){ // CHECK-LABEL: @test_mm512_mask_reduce_add_epi64( @@ -59,23 +64,27 @@ int test_mm512_reduce_add_epi32(__m512i __W){ // CHECK: call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_add_epi32(__W); } +TEST_CONSTEXPR(_mm512_reduce_add_epi32((__m512i)(__v16si){-8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7}) == -8); int test_mm512_reduce_mul_epi32(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_mul_epi32( // CHECK: call i32 @llvm.vector.reduce.mul.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_mul_epi32(__W); } +TEST_CONSTEXPR(_mm512_reduce_mul_epi32((__m512i)(__v16si){1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 1, 1, -3, 1, 1}) == -36); int test_mm512_reduce_or_epi32(__m512i __W){ // CHECK: call i32 @llvm.vector.reduce.or.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_or_epi32(__W); } +TEST_CONSTEXPR(_mm512_reduce_or_epi32((__m512i)(__v16si){0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0, 0, 0, 0, 0, 0, 0, 0}) == 0xFF); int test_mm512_reduce_and_epi32(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_and_epi32( // CHECK: call i32 @llvm.vector.reduce.and.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_and_epi32(__W); } +TEST_CONSTEXPR(_mm512_reduce_and_epi32((__m512i)(__v16si){0xFF, 0xF0, 0x0F, 0xFF, 0xFF, 0xFF, 0xF0, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0xF0, 0x0F, 0x0F}) == 0x00); int test_mm512_mask_reduce_add_epi32(__mmask16 __M, __m512i __W){ // CHECK-LABEL: @test_mm512_mask_reduce_add_epi32( diff --git a/clang/test/CodeGen/X86/avx512-reduceMinMaxIntrin.c b/clang/test/CodeGen/X86/avx512-reduceMinMaxIntrin.c index 3e33ec566d281..4f7b7b0cf9ae3 100644 --- a/clang/test/CodeGen/X86/avx512-reduceMinMaxIntrin.c +++ b/clang/test/CodeGen/X86/avx512-reduceMinMaxIntrin.c @@ -1,18 +1,21 @@ // RUN: %clang_cc1 -ffreestanding %s -O0 -triple=x86_64-apple-darwin -target-cpu skylake-avx512 -emit-llvm -o - -Wall -Werror | FileCheck %s #include <immintrin.h> +#include "builtin_test_helpers.h" long long test_mm512_reduce_max_epi64(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_max_epi64( // CHECK: call i64 @llvm.vector.reduce.smax.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_max_epi64(__W); } +TEST_CONSTEXPR(_mm512_reduce_max_epi64((__m512i)(__v8di){-4, -3, -2, -1, 0, 1, 2, 3}) == 3); unsigned long long test_mm512_reduce_max_epu64(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_max_epu64( // CHECK: call i64 @llvm.vector.reduce.umax.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_max_epu64(__W); } +TEST_CONSTEXPR(_mm512_reduce_max_epu64((__m512i)(__v8du){0, 1, 2, 3, 4, 5, 6, 7}) == 7); double test_mm512_reduce_max_pd(__m512d __W, double ExtraAddOp){ // CHECK-LABEL: @test_mm512_reduce_max_pd( @@ -27,12 +30,14 @@ long long test_mm512_reduce_min_epi64(__m512i __W){ // CHECK: call i64 @llvm.vector.reduce.smin.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_min_epi64(__W); } +TEST_CONSTEXPR(_mm512_reduce_min_epi64((__m512i)(__v8di){-4, -3, -2, -1, 0, 1, 2, 3}) == -4); unsigned long long test_mm512_reduce_min_epu64(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_min_epu64( // CHECK: call i64 @llvm.vector.reduce.umin.v8i64(<8 x i64> %{{.*}}) return _mm512_reduce_min_epu64(__W); } +TEST_CONSTEXPR(_mm512_reduce_min_epu64((__m512i)(__v8du){0, 1, 2, 3, 4, 5, 6, 7}) == 0); double test_mm512_reduce_min_pd(__m512d __W, double ExtraMulOp){ // CHECK-LABEL: @test_mm512_reduce_min_pd( @@ -89,12 +94,14 @@ int test_mm512_reduce_max_epi32(__m512i __W){ // CHECK: call i32 @llvm.vector.reduce.smax.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_max_epi32(__W); } +TEST_CONSTEXPR(_mm512_reduce_max_epi32((__m512i)(__v16si){-8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7}) == 7); unsigned int test_mm512_reduce_max_epu32(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_max_epu32( // CHECK: call i32 @llvm.vector.reduce.umax.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_max_epu32(__W); } +TEST_CONSTEXPR(_mm512_reduce_max_epu32((__m512i)(__v16su){0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}) == 15); float test_mm512_reduce_max_ps(__m512 __W){ // CHECK-LABEL: @test_mm512_reduce_max_ps( @@ -107,12 +114,14 @@ int test_mm512_reduce_min_epi32(__m512i __W){ // CHECK: call i32 @llvm.vector.reduce.smin.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_min_epi32(__W); } +TEST_CONSTEXPR(_mm512_reduce_min_epi32((__m512i)(__v16si){-8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7}) == -8); unsigned int test_mm512_reduce_min_epu32(__m512i __W){ // CHECK-LABEL: @test_mm512_reduce_min_epu32( // CHECK: call i32 @llvm.vector.reduce.umin.v16i32(<16 x i32> %{{.*}}) return _mm512_reduce_min_epu32(__W); } +TEST_CONSTEXPR(_mm512_reduce_min_epu32((__m512i)(__v16su){0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}) == 0); float test_mm512_reduce_min_ps(__m512 __W){ // CHECK-LABEL: @test_mm512_reduce_min_ps( _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
