https://github.com/notnotharsh updated
https://github.com/llvm/llvm-project/pull/156234
>From 202b08f740d2a31df345b6ff63765595b146d5cf Mon Sep 17 00:00:00 2001
From: notnotharsh <harshtiwary2...@gmail.com>
Date: Sun, 31 Aug 2025 03:03:34 -0700
Subject: [PATCH 1/2] [Headers][X86] Allow AVX512 masked blend intrinsics to be
used in constexpr
---
clang/lib/Headers/avx512bwintrin.h | 10 +-
clang/lib/Headers/avx512fintrin.h | 20 ++--
clang/lib/Headers/avx512fp16intrin.h | 2 +-
clang/lib/Headers/avx512vlbwintrin.h | 20 ++--
clang/lib/Headers/avx512vlfp16intrin.h | 7 +-
clang/lib/Headers/avx512vlintrin.h | 32 +++---
clang/test/CodeGen/X86/avx512bw-builtins.c | 46 +++++++++
clang/test/CodeGen/X86/avx512vl-builtins.c | 99 +++++++++++++++++++
clang/test/CodeGen/X86/avx512vlbw-builtins.c | 50 +++++++++-
.../test/CodeGen/X86/avx512vlfp16-builtins.c | 43 ++++++++
10 files changed, 276 insertions(+), 53 deletions(-)
diff --git a/clang/lib/Headers/avx512bwintrin.h
b/clang/lib/Headers/avx512bwintrin.h
index 2d2bf59fc5b76..5ad2856f8b996 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -465,17 +465,15 @@ _mm512_maskz_mullo_epi16(__mmask32 __U, __m512i __A,
__m512i __B) {
(__v32hi)_mm512_setzero_si512());
}
-static __inline__ __m512i __DEFAULT_FN_ATTRS512
-_mm512_mask_blend_epi8 (__mmask64 __U, __m512i __A, __m512i __W)
-{
+static __inline__ __m512i __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_blend_epi8(__mmask64 __U, __m512i __A, __m512i __W) {
return (__m512i) __builtin_ia32_selectb_512 ((__mmask64) __U,
(__v64qi) __W,
(__v64qi) __A);
}
-static __inline__ __m512i __DEFAULT_FN_ATTRS512
-_mm512_mask_blend_epi16 (__mmask32 __U, __m512i __A, __m512i __W)
-{
+static __inline__ __m512i __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_blend_epi16(__mmask32 __U, __m512i __A, __m512i __W) {
return (__m512i) __builtin_ia32_selectw_512 ((__mmask32) __U,
(__v32hi) __W,
(__v32hi) __A);
diff --git a/clang/lib/Headers/avx512fintrin.h
b/clang/lib/Headers/avx512fintrin.h
index e23b1c0381ab1..1ce468c22bdd6 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -3366,33 +3366,29 @@ _mm512_maskz_permutex2var_epi64(__mmask8 __U, __m512i
__A, __m512i __I,
/* Vector Blend */
-static __inline __m512d __DEFAULT_FN_ATTRS512
-_mm512_mask_blend_pd(__mmask8 __U, __m512d __A, __m512d __W)
-{
+static __inline __m512d __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_blend_pd(__mmask8 __U, __m512d __A, __m512d __W) {
return (__m512d) __builtin_ia32_selectpd_512 ((__mmask8) __U,
(__v8df) __W,
(__v8df) __A);
}
-static __inline __m512 __DEFAULT_FN_ATTRS512
-_mm512_mask_blend_ps(__mmask16 __U, __m512 __A, __m512 __W)
-{
+static __inline __m512 __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_blend_ps(__mmask16 __U, __m512 __A, __m512 __W) {
return (__m512) __builtin_ia32_selectps_512 ((__mmask16) __U,
(__v16sf) __W,
(__v16sf) __A);
}
-static __inline __m512i __DEFAULT_FN_ATTRS512
-_mm512_mask_blend_epi64(__mmask8 __U, __m512i __A, __m512i __W)
-{
+static __inline __m512i __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_blend_epi64(__mmask8 __U, __m512i __A, __m512i __W) {
return (__m512i) __builtin_ia32_selectq_512 ((__mmask8) __U,
(__v8di) __W,
(__v8di) __A);
}
-static __inline __m512i __DEFAULT_FN_ATTRS512
-_mm512_mask_blend_epi32(__mmask16 __U, __m512i __A, __m512i __W)
-{
+static __inline __m512i __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_blend_epi32(__mmask16 __U, __m512i __A, __m512i __W) {
return (__m512i) __builtin_ia32_selectd_512 ((__mmask16) __U,
(__v16si) __W,
(__v16si) __A);
diff --git a/clang/lib/Headers/avx512fp16intrin.h
b/clang/lib/Headers/avx512fp16intrin.h
index 6989b86a7b68c..a6b0820de90cf 100644
--- a/clang/lib/Headers/avx512fp16intrin.h
+++ b/clang/lib/Headers/avx512fp16intrin.h
@@ -3306,7 +3306,7 @@ _mm512_reduce_min_ph(__m512h __V) {
return __builtin_ia32_reduce_fmin_ph512(__V);
}
-static __inline__ __m512h __DEFAULT_FN_ATTRS512
+static __inline__ __m512h __DEFAULT_FN_ATTRS512_CONSTEXPR
_mm512_mask_blend_ph(__mmask32 __U, __m512h __A, __m512h __W) {
return (__m512h)__builtin_ia32_selectph_512((__mmask32)__U, (__v32hf)__W,
(__v32hf)__A);
diff --git a/clang/lib/Headers/avx512vlbwintrin.h
b/clang/lib/Headers/avx512vlbwintrin.h
index f96468c56c392..81a77fa573a90 100644
--- a/clang/lib/Headers/avx512vlbwintrin.h
+++ b/clang/lib/Headers/avx512vlbwintrin.h
@@ -454,33 +454,29 @@ _mm_maskz_mullo_epi16(__mmask8 __U, __m128i __A, __m128i
__B) {
(__v8hi)_mm_setzero_si128());
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_blend_epi8 (__mmask16 __U, __m128i __A, __m128i __W)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_mask_blend_epi8(__mmask16 __U, __m128i __A, __m128i __W) {
return (__m128i) __builtin_ia32_selectb_128 ((__mmask16) __U,
(__v16qi) __W,
(__v16qi) __A);
}
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_blend_epi8 (__mmask32 __U, __m256i __A, __m256i __W)
-{
+static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_mask_blend_epi8(__mmask32 __U, __m256i __A, __m256i __W) {
return (__m256i) __builtin_ia32_selectb_256 ((__mmask32) __U,
(__v32qi) __W,
(__v32qi) __A);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_blend_epi16 (__mmask8 __U, __m128i __A, __m128i __W)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_mask_blend_epi16(__mmask8 __U, __m128i __A, __m128i __W) {
return (__m128i) __builtin_ia32_selectw_128 ((__mmask8) __U,
(__v8hi) __W,
(__v8hi) __A);
}
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_blend_epi16 (__mmask16 __U, __m256i __A, __m256i __W)
-{
+static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_mask_blend_epi16(__mmask16 __U, __m256i __A, __m256i __W) {
return (__m256i) __builtin_ia32_selectw_256 ((__mmask16) __U,
(__v16hi) __W,
(__v16hi) __A);
diff --git a/clang/lib/Headers/avx512vlfp16intrin.h
b/clang/lib/Headers/avx512vlfp16intrin.h
index 98ad9b54eef39..6be559d9a1a33 100644
--- a/clang/lib/Headers/avx512vlfp16intrin.h
+++ b/clang/lib/Headers/avx512vlfp16intrin.h
@@ -1984,14 +1984,13 @@ _mm256_maskz_fmadd_pch(__mmask8 __U, __m256h __A,
__m256h __B, __m256h __C) {
(__v8sf)__C,
(__mmask8)__U);
}
-static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_mask_blend_ph(__mmask8 __U,
- __m128h __A,
- __m128h __W)
{
+static __inline__ __m128h __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_mask_blend_ph(__mmask8 __U, __m128h __A, __m128h __W) {
return (__m128h)__builtin_ia32_selectph_128((__mmask8)__U, (__v8hf)__W,
(__v8hf)__A);
}
-static __inline__ __m256h __DEFAULT_FN_ATTRS256
+static __inline__ __m256h __DEFAULT_FN_ATTRS256_CONSTEXPR
_mm256_mask_blend_ph(__mmask16 __U, __m256h __A, __m256h __W) {
return (__m256h)__builtin_ia32_selectph_256((__mmask16)__U, (__v16hf)__W,
(__v16hf)__A);
diff --git a/clang/lib/Headers/avx512vlintrin.h
b/clang/lib/Headers/avx512vlintrin.h
index 5a5b09e274563..f92a01293d1e9 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -1631,57 +1631,57 @@ _mm256_maskz_add_ps(__mmask8 __U, __m256 __A, __m256
__B) {
(__v8sf)_mm256_setzero_ps());
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_blend_epi32 (__mmask8 __U, __m128i __A, __m128i __W) {
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_mask_blend_epi32(__mmask8 __U, __m128i __A, __m128i __W) {
return (__m128i) __builtin_ia32_selectd_128 ((__mmask8) __U,
(__v4si) __W,
(__v4si) __A);
}
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_blend_epi32 (__mmask8 __U, __m256i __A, __m256i __W) {
+static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_mask_blend_epi32(__mmask8 __U, __m256i __A, __m256i __W) {
return (__m256i) __builtin_ia32_selectd_256 ((__mmask8) __U,
(__v8si) __W,
(__v8si) __A);
}
-static __inline__ __m128d __DEFAULT_FN_ATTRS128
-_mm_mask_blend_pd (__mmask8 __U, __m128d __A, __m128d __W) {
+static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_mask_blend_pd(__mmask8 __U, __m128d __A, __m128d __W) {
return (__m128d) __builtin_ia32_selectpd_128 ((__mmask8) __U,
(__v2df) __W,
(__v2df) __A);
}
-static __inline__ __m256d __DEFAULT_FN_ATTRS256
-_mm256_mask_blend_pd (__mmask8 __U, __m256d __A, __m256d __W) {
+static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_mask_blend_pd(__mmask8 __U, __m256d __A, __m256d __W) {
return (__m256d) __builtin_ia32_selectpd_256 ((__mmask8) __U,
(__v4df) __W,
(__v4df) __A);
}
-static __inline__ __m128 __DEFAULT_FN_ATTRS128
-_mm_mask_blend_ps (__mmask8 __U, __m128 __A, __m128 __W) {
+static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_mask_blend_ps(__mmask8 __U, __m128 __A, __m128 __W) {
return (__m128) __builtin_ia32_selectps_128 ((__mmask8) __U,
(__v4sf) __W,
(__v4sf) __A);
}
-static __inline__ __m256 __DEFAULT_FN_ATTRS256
-_mm256_mask_blend_ps (__mmask8 __U, __m256 __A, __m256 __W) {
+static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_mask_blend_ps(__mmask8 __U, __m256 __A, __m256 __W) {
return (__m256) __builtin_ia32_selectps_256 ((__mmask8) __U,
(__v8sf) __W,
(__v8sf) __A);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_blend_epi64 (__mmask8 __U, __m128i __A, __m128i __W) {
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_mask_blend_epi64(__mmask8 __U, __m128i __A, __m128i __W) {
return (__m128i) __builtin_ia32_selectq_128 ((__mmask8) __U,
(__v2di) __W,
(__v2di) __A);
}
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_blend_epi64 (__mmask8 __U, __m256i __A, __m256i __W) {
+static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_mask_blend_epi64(__mmask8 __U, __m256i __A, __m256i __W) {
return (__m256i) __builtin_ia32_selectq_256 ((__mmask8) __U,
(__v4di) __W,
(__v4di) __A);
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c
b/clang/test/CodeGen/X86/avx512bw-builtins.c
index 6e10af2c282e0..0d01a26077bf9 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -847,11 +847,57 @@ __m512i test_mm512_mask_blend_epi8(__mmask64 __U, __m512i
__A, __m512i __W) {
// CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
return _mm512_mask_blend_epi8(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v64qi(
+ _mm512_mask_blend_epi8(
+ (__mmask64) 0x00000001,
+ (__m512i)(__v64qi) {2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2},
+ (__m512i)(__v64qi){
10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,
10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25}
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
+));
__m512i test_mm512_mask_blend_epi16(__mmask32 __U, __m512i __A, __m512i __W) {
// CHECK-LABEL: test_mm512_mask_blend_epi16
// CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_mask_blend_epi16(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v32hi(
+ _mm512_mask_blend_epi16(
+ (__mmask32) 0x00000001,
+ (__m512i)(__v32hi) {2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2},
+ (__m512i)(__v32hi){
10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25}
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2
+));
+
+__m512i test_mm512_mask_blend_epi32(__mmask16 __U, __m512i __A, __m512i __W) {
+ // CHECK-LABEL: test_mm512_mask_blend_epi32
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
+ return _mm512_mask_blend_epi32(__U, __A, __W);
+}
+TEST_CONSTEXPR(match_v16si(
+ _mm512_mask_blend_epi32(
+ (__mmask16) 0x0001,
+ (__m512i)(__v16si) {2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2},
+ (__m512i)(__v16si){ 10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25}
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
+));
+
+__m512i test_mm512_mask_blend_epi64(__mmask8 __U, __m512i __A, __m512i __W) {
+ // CHECK-LABEL: test_mm512_mask_blend_epi64
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
+ return _mm512_mask_blend_epi64(__U, __A, __W);
+}
+
+TEST_CONSTEXPR(match_v8di(
+ _mm512_mask_blend_epi64(
+ (__mmask8)0x01,
+ (__m512i)(__v8di){2, 2, 2, 2, 2, 2, 2, 2},
+ (__m512i)(__v8di){10, 11, 12, 13, 14, 15, 16, 17}
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2
+));
+
__m512i test_mm512_abs_epi8(__m512i __A) {
// CHECK-LABEL: test_mm512_abs_epi8
// CHECK: [[ABS:%.*]] = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %{{.*}}, i1
false)
diff --git a/clang/test/CodeGen/X86/avx512vl-builtins.c
b/clang/test/CodeGen/X86/avx512vl-builtins.c
index 6858f114c29c4..55a1d7fc184fe 100644
--- a/clang/test/CodeGen/X86/avx512vl-builtins.c
+++ b/clang/test/CodeGen/X86/avx512vl-builtins.c
@@ -3521,41 +3521,140 @@ __m128i test_mm_mask_blend_epi32(__mmask8 __U, __m128i
__A, __m128i __W) {
// CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
return _mm_mask_blend_epi32(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v4si(
+ _mm_mask_blend_epi32(
+ (__mmask8)0x01,
+ (__m128i)(__v4si){2, 2, 2, 2},
+ (__m128i)(__v4si){ 10,11,12,13 }
+ ),
+ 10, 2, 2, 2
+));
__m256i test_mm256_mask_blend_epi32(__mmask8 __U, __m256i __A, __m256i __W) {
// CHECK-LABEL: test_mm256_mask_blend_epi32
// CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
return _mm256_mask_blend_epi32(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v8si(
+ _mm256_mask_blend_epi32(
+ (__mmask8)0x01,
+ (__m256i)(__v8si){2, 2, 2, 2, 2, 2, 2, 2},
+ (__m256i)(__v8si){ 10,11,12,13,14,15,16,17 }
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2
+));
__m128d test_mm_mask_blend_pd(__mmask8 __U, __m128d __A, __m128d __W) {
// CHECK-LABEL: test_mm_mask_blend_pd
// CHECK: select <2 x i1> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}
return _mm_mask_blend_pd(__U,__A,__W);
}
+TEST_CONSTEXPR(match_m128d(
+ _mm_mask_blend_pd(
+ (__mmask8)0x01,
+ (__m128d)(__v2df){2.0, 2.0},
+ (__m128d)(__v2df){10.0, 20.0}
+ ),
+ 10.0, 2.0
+));
__m256d test_mm256_mask_blend_pd(__mmask8 __U, __m256d __A, __m256d __W) {
// CHECK-LABEL: test_mm256_mask_blend_pd
// CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
return _mm256_mask_blend_pd(__U,__A,__W);
}
+TEST_CONSTEXPR(match_m256d(
+ _mm256_mask_blend_pd(
+ (__mmask8)0x01,
+ (__m256d)(__v4df){2.0, 2.0, 2.0, 2.0},
+ (__m256d)(__v4df){10.0, 11.0, 12.0, 13.0}
+ ),
+ 10.0, 2.0, 2.0, 2.0
+));
+
+__m512d test_mm512_mask_blend_pd(__mmask8 __U, __m512d __A, __m512d __W) {
+ // CHECK-LABEL: test_mm512_mask_blend_pd
+ // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
+ return _mm512_mask_blend_pd(__U, __A, __W);
+}
+
+TEST_CONSTEXPR(match_m512d(
+ _mm512_mask_blend_pd(
+ (__mmask8)0x01,
+ (__m512d)(__v8df){2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0},
+ (__m512d)(__v8df){10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0, 17.0}
+ ),
+ 10.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0
+));
+
__m128 test_mm_mask_blend_ps(__mmask8 __U, __m128 __A, __m128 __W) {
// CHECK-LABEL: test_mm_mask_blend_ps
// CHECK: select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
return _mm_mask_blend_ps(__U,__A,__W);
}
+TEST_CONSTEXPR(match_m128(
+ _mm_mask_blend_ps(
+ (__mmask8)0x01,
+ (__m128)(__v4sf){2.0f, 2.0f, 2.0f, 2.0f},
+ (__m128)(__v4sf){10.0f, 11.0f, 12.0f, 13.0f}
+ ),
+ 10.0f, 2.0f, 2.0f, 2.0f
+));
+
__m256 test_mm256_mask_blend_ps(__mmask8 __U, __m256 __A, __m256 __W) {
// CHECK-LABEL: test_mm256_mask_blend_ps
// CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
return _mm256_mask_blend_ps(__U,__A,__W);
}
+TEST_CONSTEXPR(match_m256(
+ _mm256_mask_blend_ps(
+ (__mmask8)0x01,
+ (__m256)(__v8sf){2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f},
+ (__m256)(__v8sf){10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f, 17.0f}
+ ),
+ 10.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f
+));
+
+__m512 test_mm512_mask_blend_ps(__mmask8 __U, __m512 __A, __m512 __W) {
+ // CHECK-LABEL: test_mm512_mask_blend_ps
+ // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float>
%{{.*}}
+ return _mm512_mask_blend_ps(__U, __A, __W);
+}
+TEST_CONSTEXPR(match_m512(
+ _mm512_mask_blend_ps(
+ (__mmask16)0x01,
+ (__m512)(__v16sf){2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f},
+ (__m512)(__v16sf){10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f, 17.0f,
+ 18.0f, 19.0f, 20.0f, 21.0f, 22.0f, 23.0f, 24.0f, 25.0f}
+ ),
+ 10.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f
+));
+
__m128i test_mm_mask_blend_epi64(__mmask8 __U, __m128i __A, __m128i __W) {
// CHECK-LABEL: test_mm_mask_blend_epi64
// CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_mask_blend_epi64(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v2di(
+ _mm_mask_blend_epi64(
+ (__mmask8)0x01,
+ (__m128i)(__v2di){2, 2},
+ (__m128i)(__v2di){ 10,11 }
+ ),
+ 10, 2
+));
__m256i test_mm256_mask_blend_epi64(__mmask8 __U, __m256i __A, __m256i __W) {
// CHECK-LABEL: test_mm256_mask_blend_epi64
// CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_mask_blend_epi64(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v4di(
+ _mm256_mask_blend_epi64(
+ (__mmask8)0x01,
+ (__m256i)(__v4di){2, 2, 2, 2},
+ (__m256i)(__v4di){ 10,11,12,13 }
+ ),
+ 10, 2, 2, 2
+));
__m128d test_mm_mask_compress_pd(__m128d __W, __mmask8 __U, __m128d __A) {
// CHECK-LABEL: test_mm_mask_compress_pd
// CHECK: @llvm.x86.avx512.mask.compress
diff --git a/clang/test/CodeGen/X86/avx512vlbw-builtins.c
b/clang/test/CodeGen/X86/avx512vlbw-builtins.c
index 5d697440bfc1b..59e37c47e2eb3 100644
--- a/clang/test/CodeGen/X86/avx512vlbw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512vlbw-builtins.c
@@ -882,23 +882,56 @@ __m128i test_mm_mask_blend_epi8(__mmask16 __U, __m128i
__A, __m128i __W) {
// CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
return _mm_mask_blend_epi8(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v16qi(
+ _mm_mask_blend_epi8(
+ (__mmask16)0x0001,
+ (__m128i)(__v16qi){2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2},
+ (__m128i)(__v16qi){ 10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25 }
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
+));
+
__m256i test_mm256_mask_blend_epi8(__mmask32 __U, __m256i __A, __m256i __W) {
// CHECK-LABEL: test_mm256_mask_blend_epi8
// CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
return _mm256_mask_blend_epi8(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v32qi(
+ _mm256_mask_blend_epi8(
+ (__mmask32) 0x00000001,
+ (__m256i)(__v32qi) {2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2},
+ (__m256i)(__v32qi){
10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25}
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2
+));
__m128i test_mm_mask_blend_epi16(__mmask8 __U, __m128i __A, __m128i __W) {
// CHECK-LABEL: test_mm_mask_blend_epi16
// CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
return _mm_mask_blend_epi16(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v8hi(
+ _mm_mask_blend_epi16(
+ (__mmask8)0x01,
+ (__m128i)(__v8hi){2, 2, 2, 2, 2, 2, 2, 2},
+ (__m128i)(__v8hi){ 10,11,12,13,14,15,16,17 }
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2
+));
__m256i test_mm256_mask_blend_epi16(__mmask16 __U, __m256i __A, __m256i __W) {
// CHECK-LABEL: test_mm256_mask_blend_epi16
// CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
return _mm256_mask_blend_epi16(__U,__A,__W);
}
+TEST_CONSTEXPR(match_v16hi(
+ _mm256_mask_blend_epi16(
+ (__mmask16)0x0001,
+ (__m256i)(__v16hi){2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2},
+ (__m256i)(__v16hi){ 10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25 }
+ ),
+ 10, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
+));
__m128i test_mm_mask_abs_epi8(__m128i __W, __mmask16 __U, __m128i __A) {
// CHECK-LABEL: test_mm_mask_abs_epi8
@@ -908,7 +941,14 @@ __m128i test_mm_mask_abs_epi8(__m128i __W, __mmask16 __U,
__m128i __A) {
// CHECK: select <16 x i1> %{{.*}}, <16 x i8> [[ABS]], <16 x i8> %{{.*}}
return _mm_mask_abs_epi8(__W,__U,__A);
}
-TEST_CONSTEXPR(match_v16qi(_mm_mask_abs_epi8((__m128i)(__v16qi){99, 99, 99,
99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99}, (__mmask16)0x0001,
(__m128i)(__v16qi){(char)-1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2}), 1,
99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99));
+TEST_CONSTEXPR(match_v16qi(
+ _mm_mask_abs_epi8(
+ (__m128i)(__v16qi){99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99,
99, 99},
+ (__mmask16)0x0001,
+ (__m128i)(__v16qi){(char)-1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2}
+ ),
+ 1, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99
+));
__m128i test_mm_maskz_abs_epi8(__mmask16 __U, __m128i __A) {
// CHECK-LABEL: test_mm_maskz_abs_epi8
@@ -938,7 +978,13 @@ __m256i test_mm256_maskz_abs_epi8(__mmask32 __U, __m256i
__A) {
// CHECK: select <32 x i1> %{{.*}}, <32 x i8> [[ABS]], <32 x i8> %{{.*}}
return _mm256_maskz_abs_epi8(__U,__A);
}
-TEST_CONSTEXPR(match_v32qi(_mm256_maskz_abs_epi8((__mmask32)0x55555555,
(__m256i)(__v32qi){(char)-1, 2, (char)-3, 4, (char)-5, 6, (char)-7, 8,
(char)-9, 10, (char)-11, 12, (char)-13, 14, (char)-15, 16, (char)-17, 18,
(char)-19, 20, (char)-21, 22, (char)-23, 24, (char)-25, 26, (char)-27, 28,
(char)-29, 30, (char)-31, 32}), 1, 0, 3, 0, 5, 0, 7, 0, 9, 0, 11, 0, 13, 0, 15,
0, 17, 0, 19, 0, 21, 0, 23, 0, 25, 0, 27, 0, 29, 0, 31, 0));
+TEST_CONSTEXPR(match_v32qi(
+ _mm256_maskz_abs_epi8(
+ (__mmask32)0x55555555,
+ (__m256i)(__v32qi){(char)-1, 2, (char)-3, 4, (char)-5, 6, (char)-7, 8,
(char)-9, 10, (char)-11, 12, (char)-13, 14, (char)-15, 16, (char)-17, 18,
(char)-19, 20, (char)-21, 22, (char)-23, 24, (char)-25, 26, (char)-27, 28,
(char)-29, 30, (char)-31, 32}
+ ),
+ 1, 0, 3, 0, 5, 0, 7, 0, 9, 0, 11, 0, 13, 0, 15, 0, 17, 0, 19, 0, 21, 0, 23,
0, 25, 0, 27, 0, 29, 0, 31, 0
+));
__m128i test_mm_mask_abs_epi16(__m128i __W, __mmask8 __U, __m128i __A) {
// CHECK-LABEL: test_mm_mask_abs_epi16
diff --git a/clang/test/CodeGen/X86/avx512vlfp16-builtins.c
b/clang/test/CodeGen/X86/avx512vlfp16-builtins.c
index 8dd4904ae94c9..b234cd195aa6e 100644
--- a/clang/test/CodeGen/X86/avx512vlfp16-builtins.c
+++ b/clang/test/CodeGen/X86/avx512vlfp16-builtins.c
@@ -2988,6 +2988,14 @@ __m128h test_mm_mask_blend_ph(__mmask8 __U, __m128h __A,
__m128h __W) {
// CHECK: %{{.*}} = select <8 x i1> %{{.*}}, <8 x half> %{{.*}}, <8 x half>
%{{.*}}
return _mm_mask_blend_ph(__U, __A, __W);
}
+TEST_CONSTEXPR(match_m128h(
+ _mm_mask_blend_ph(
+ (__mmask8)0x01,
+ (__m128h)(__v8hf){2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f},
+ (__m128h)(__v8hf){10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f, 17.0f}
+ ),
+ 10.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f
+));
__m256h test_mm256_mask_blend_ph(__mmask16 __U, __m256h __A, __m256h __W) {
// CHECK-LABEL: test_mm256_mask_blend_ph
@@ -2995,6 +3003,41 @@ __m256h test_mm256_mask_blend_ph(__mmask16 __U, __m256h
__A, __m256h __W) {
// CHECK: %{{.*}} = select <16 x i1> %{{.*}}, <16 x half> %{{.*}}, <16 x
half> %{{.*}}
return _mm256_mask_blend_ph(__U, __A, __W);
}
+TEST_CONSTEXPR(match_m256h(
+ _mm256_mask_blend_ph(
+ (__mmask16)0x0001,
+ (__m256h)(__v16hf){2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f},
+ (__m256h)(__v16hf){10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f, 17.0f,
+ 18.0f, 19.0f, 20.0f, 21.0f, 22.0f, 23.0f, 24.0f, 25.0f}
+ ),
+ 10.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f
+));
+
+__m512h test_mm512_mask_blend_ph(__mmask32 __U, __m512h __A, __m512h __W) {
+ // CHECK-LABEL: test_mm512_mask_blend_ph
+ // CHECK: %{{.*}} = bitcast i32 %{{.*}} to <32 x i1>
+ // CHECK: %{{.*}} = select <32 x i1> %{{.*}}, <32 x half> %{{.*}}, <32 x
half> %{{.*}}
+ return _mm512_mask_blend_ph(__U, __A, __W);
+}
+TEST_CONSTEXPR(match_m512h(
+ _mm512_mask_blend_ph(
+ (__mmask32)0x00000001,
+ (__m512h)(__v32hf){2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f},
+ (__m512h)(__v32hf){10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f, 17.0f,
+ 18.0f, 19.0f, 20.0f, 21.0f, 22.0f, 23.0f, 24.0f, 25.0f,
+ 26.0f, 27.0f, 28.0f, 29.0f, 30.0f, 31.0f, 32.0f, 33.0f,
+ 34.0f, 35.0f, 36.0f, 37.0f, 38.0f, 39.0f, 40.0f, 41.0f}
+ ),
+ 10.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f,
+ 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f
+));
__m128h test_mm_permutex2var_ph(__m128h __A, __m128i __I, __m128h __B) {
// CHECK-LABEL: test_mm_permutex2var_ph
>From b94585a7b5df2e5c24c67dfa52207b54f6112c12 Mon Sep 17 00:00:00 2001
From: notnotharsh <harshtiwary2...@gmail.com>
Date: Wed, 10 Sep 2025 23:56:42 -0700
Subject: [PATCH 2/2] [Headers][X86] Allow AVX512 masked blend intrinsics to be
used in constexpr
---
clang/test/CodeGen/X86/avx512vlbw-builtins.c | 9 +--------
1 file changed, 1 insertion(+), 8 deletions(-)
diff --git a/clang/test/CodeGen/X86/avx512vlbw-builtins.c
b/clang/test/CodeGen/X86/avx512vlbw-builtins.c
index 3f1df68d1b9b9..f2424a1fb745e 100644
--- a/clang/test/CodeGen/X86/avx512vlbw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512vlbw-builtins.c
@@ -952,14 +952,7 @@ __m128i test_mm_mask_abs_epi8(__m128i __W, __mmask16 __U,
__m128i __A) {
// CHECK: select <16 x i1> %{{.*}}, <16 x i8> [[ABS]], <16 x i8> %{{.*}}
return _mm_mask_abs_epi8(__W,__U,__A);
}
-TEST_CONSTEXPR(match_v16qi(
- _mm_mask_abs_epi8(
- (__m128i)(__v16qi){99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99,
99, 99},
- (__mmask16)0x0001,
- (__m128i)(__v16qi){(char)-1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2}
- ),
- 1, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99
-));
+TEST_CONSTEXPR(match_v16qi(_mm_mask_abs_epi8((__m128i)(__v16qi){99, 99, 99,
99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99}, (__mmask16)0x0001,
(__m128i)(__v16qi){(char)-1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2}), 1,
99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99));
__m128i test_mm_maskz_abs_epi8(__mmask16 __U, __m128i __A) {
// CHECK-LABEL: test_mm_maskz_abs_epi8
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
