From 0d67e4b5d6698d3655835207ac42a8e478a4a1d8 Mon Sep 17 00:00:00 2001 From: Michael Zuckerman Date: Thu, 3 Mar 2016 13:43:05 +0000 Subject: [PATCH] [CLANG][AVX512][BUILTIN] movddup{128|256|512} Differential Revision: http://reviews.llvm.org/D17826 llvm-svn: 262617 --- clang/include/clang/Basic/BuiltinsX86.def | 3 +++ clang/lib/Headers/avx512fintrin.h | 28 ++++++++++++++++++++++++++ clang/lib/Headers/avx512vlintrin.h | 33 +++++++++++++++++++++++++++++++ clang/test/CodeGen/avx512f-builtins.c | 19 ++++++++++++++++++ clang/test/CodeGen/avx512vl-builtins.c | 23 +++++++++++++++++++++ 5 files changed, 106 insertions(+) diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def index 7ff41ad..42d3a0f 100644 --- a/clang/include/clang/Basic/BuiltinsX86.def +++ b/clang/include/clang/Basic/BuiltinsX86.def @@ -1716,6 +1716,9 @@ TARGET_BUILTIN(__builtin_ia32_movdquhi128_mask, "V8sV8sV8sUc","","avx512bw,avx51 TARGET_BUILTIN(__builtin_ia32_movdquhi256_mask, "V16sV16sV16sUs","","avx512bw,avx512vl") TARGET_BUILTIN(__builtin_ia32_movdquqi128_mask, "V16cV16cV16cUs","","avx512bw,avx512vl") TARGET_BUILTIN(__builtin_ia32_movdquqi256_mask, "V32cV32cV32cUi","","avx512bw,avx512vl") +TARGET_BUILTIN(__builtin_ia32_movddup512_mask, "V8dV8dV8dUc","","avx512f") +TARGET_BUILTIN(__builtin_ia32_movddup128_mask, "V2dV2dV2dUc","","avx512vl") +TARGET_BUILTIN(__builtin_ia32_movddup256_mask, "V4dV4dV4dUc","","avx512vl") #undef BUILTIN #undef TARGET_BUILTIN diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h index 81565c4..c1fa73c 100644 --- a/clang/lib/Headers/avx512fintrin.h +++ b/clang/lib/Headers/avx512fintrin.h @@ -3668,6 +3668,34 @@ _mm512_mask_store_epi64 (void *__P, __mmask8 __U, __m512i __A) } + +static __inline__ __m512d __DEFAULT_FN_ATTRS +_mm512_movedup_pd (__m512d __A) +{ + return (__m512d) __builtin_ia32_movddup512_mask ((__v8df) __A, + (__v8df) + _mm512_undefined_pd (), + (__mmask8) -1); +} + +static __inline__ __m512d __DEFAULT_FN_ATTRS +_mm512_mask_movedup_pd (__m512d __W, __mmask8 __U, __m512d __A) +{ + return (__m512d) __builtin_ia32_movddup512_mask ((__v8df) __A, + (__v8df) __W, + (__mmask8) __U); +} + +static __inline__ __m512d __DEFAULT_FN_ATTRS +_mm512_maskz_movedup_pd (__mmask8 __U, __m512d __A) +{ + return (__m512d) __builtin_ia32_movddup512_mask ((__v8df) __A, + (__v8df) + _mm512_setzero_pd (), + (__mmask8) __U); +} + + #undef __DEFAULT_FN_ATTRS #endif // __AVX512FINTRIN_H diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h index 873f02d..8d15a22 100644 --- a/clang/lib/Headers/avx512vlintrin.h +++ b/clang/lib/Headers/avx512vlintrin.h @@ -5978,6 +5978,39 @@ _mm256_mask_store_epi64 (void *__P, __mmask8 __U, __m256i __A) (__mmask8) __U); } +static __inline__ __m128d __DEFAULT_FN_ATTRS +_mm_mask_movedup_pd (__m128d __W, __mmask8 __U, __m128d __A) +{ + return (__m128d) __builtin_ia32_movddup128_mask ((__v2df) __A, + (__v2df) __W, + (__mmask8) __U); +} + +static __inline__ __m128d __DEFAULT_FN_ATTRS +_mm_maskz_movedup_pd (__mmask8 __U, __m128d __A) +{ + return (__m128d) __builtin_ia32_movddup128_mask ((__v2df) __A, + (__v2df) + _mm_setzero_pd (), + (__mmask8) __U); +} + +static __inline__ __m256d __DEFAULT_FN_ATTRS +_mm256_mask_movedup_pd (__m256d __W, __mmask8 __U, __m256d __A) +{ + return (__m256d) __builtin_ia32_movddup256_mask ((__v4df) __A, + (__v4df) __W, + (__mmask8) __U); +} + +static __inline__ __m256d __DEFAULT_FN_ATTRS +_mm256_maskz_movedup_pd (__mmask8 __U, __m256d __A) +{ + return (__m256d) __builtin_ia32_movddup256_mask ((__v4df) __A, + (__v4df) + _mm256_setzero_pd (), + (__mmask8) __U); +} #undef __DEFAULT_FN_ATTRS #undef __DEFAULT_FN_ATTRS_BOTH diff --git a/clang/test/CodeGen/avx512f-builtins.c b/clang/test/CodeGen/avx512f-builtins.c index f9e7ea7..0098f3b 100644 --- a/clang/test/CodeGen/avx512f-builtins.c +++ b/clang/test/CodeGen/avx512f-builtins.c @@ -2346,3 +2346,22 @@ void test_mm512_mask_store_epi64(void *__P, __mmask8 __U, __m512i __A) { return _mm512_mask_store_epi64(__P, __U, __A); } +__m512d test_mm512_movedup_pd(__m512d __A) { + // CHECK-LABEL: @test_mm512_movedup_pd + // CHECK: @llvm.x86.avx512.mask.movddup.512 + return _mm512_movedup_pd(__A); +} + +__m512d test_mm512_mask_movedup_pd(__m512d __W, __mmask8 __U, __m512d __A) { + // CHECK-LABEL: @test_mm512_mask_movedup_pd + // CHECK: @llvm.x86.avx512.mask.movddup.512 + return _mm512_mask_movedup_pd(__W, __U, __A); +} + +__m512d test_mm512_maskz_movedup_pd(__mmask8 __U, __m512d __A) { + // CHECK-LABEL: @test_mm512_maskz_movedup_pd + // CHECK: @llvm.x86.avx512.mask.movddup.512 + return _mm512_maskz_movedup_pd(__U, __A); +} + + diff --git a/clang/test/CodeGen/avx512vl-builtins.c b/clang/test/CodeGen/avx512vl-builtins.c index c8c2732..bf29bbb 100644 --- a/clang/test/CodeGen/avx512vl-builtins.c +++ b/clang/test/CodeGen/avx512vl-builtins.c @@ -4008,3 +4008,26 @@ void test_mm256_mask_store_epi64(void *__P, __mmask8 __U, __m256i __A) { return _mm256_mask_store_epi64(__P, __U, __A); } +__m128d test_mm_mask_movedup_pd(__m128d __W, __mmask8 __U, __m128d __A) { + // CHECK-LABEL: @test_mm_mask_movedup_pd + // CHECK: @llvm.x86.avx512.mask.movddup.128 + return _mm_mask_movedup_pd(__W, __U, __A); +} + +__m128d test_mm_maskz_movedup_pd(__mmask8 __U, __m128d __A) { + // CHECK-LABEL: @test_mm_maskz_movedup_pd + // CHECK: @llvm.x86.avx512.mask.movddup.128 + return _mm_maskz_movedup_pd(__U, __A); +} + +__m256d test_mm256_mask_movedup_pd(__m256d __W, __mmask8 __U, __m256d __A) { + // CHECK-LABEL: @test_mm256_mask_movedup_pd + // CHECK: @llvm.x86.avx512.mask.movddup.256 + return _mm256_mask_movedup_pd(__W, __U, __A); +} + +__m256d test_mm256_maskz_movedup_pd(__mmask8 __U, __m256d __A) { + // CHECK-LABEL: @test_mm256_maskz_movedup_pd + // CHECK: @llvm.x86.avx512.mask.movddup.256 + return _mm256_maskz_movedup_pd(__U, __A); +} \ No newline at end of file -- 2.7.4