From 1ad03e7f01124041865000cccc99b54f0a2e853a Mon Sep 17 00:00:00 2001 From: Michael Zuckerman Date: Thu, 3 Mar 2016 11:34:52 +0000 Subject: [PATCH] [CLANG][AVX512][BUILTIN] movdqu{qi|hi} {128|256|512} Differential Revision: http://reviews.llvm.org/D17814 llvm-svn: 262609 --- clang/include/clang/Basic/BuiltinsX86.def | 6 +++ clang/lib/Headers/avx512bwintrin.h | 37 ++++++++++++++++ clang/lib/Headers/avx512vlbwintrin.h | 71 +++++++++++++++++++++++++++++++ clang/test/CodeGen/avx512bw-builtins.c | 22 ++++++++++ clang/test/CodeGen/avx512vlbw-builtins.c | 48 +++++++++++++++++++++ 5 files changed, 184 insertions(+) diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def index 4155424..7ff41ad 100644 --- a/clang/include/clang/Basic/BuiltinsX86.def +++ b/clang/include/clang/Basic/BuiltinsX86.def @@ -1710,6 +1710,12 @@ TARGET_BUILTIN(__builtin_ia32_movdqa64load128_mask, "V2LLivC*V2LLiUc","","avx512 TARGET_BUILTIN(__builtin_ia32_movdqa64load256_mask, "V4LLivC*V4LLiUc","","avx512vl") TARGET_BUILTIN(__builtin_ia32_movdqa64store128_mask, "vV2LLi*V2LLiUc","","avx512f") TARGET_BUILTIN(__builtin_ia32_movdqa64store256_mask, "vV4LLi*V4LLiUc","","avx512f") +TARGET_BUILTIN(__builtin_ia32_movdquhi512_mask, "V32sV32sV32sUi","","avx512bw") +TARGET_BUILTIN(__builtin_ia32_movdquqi512_mask, "V64cV64cV64cULLi","","avx512bw") +TARGET_BUILTIN(__builtin_ia32_movdquhi128_mask, "V8sV8sV8sUc","","avx512bw,avx512vl") +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") #undef BUILTIN #undef TARGET_BUILTIN diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h index 17ca52a..6ae786b 100644 --- a/clang/lib/Headers/avx512bwintrin.h +++ b/clang/lib/Headers/avx512bwintrin.h @@ -1880,6 +1880,43 @@ __builtin_ia32_psrlwi512_mask ((__v32hi)( __A),( __imm),\ (__mmask32)( __U));\ }) + + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_mov_epi16 (__m512i __W, __mmask32 __U, __m512i __A) +{ + return (__m512i) __builtin_ia32_movdquhi512_mask ((__v32hi) __A, + (__v32hi) __W, + (__mmask32) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_mov_epi16 (__mmask32 __U, __m512i __A) +{ + return (__m512i) __builtin_ia32_movdquhi512_mask ((__v32hi) __A, + (__v32hi) + _mm512_setzero_hi (), + (__mmask32) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_mov_epi8 (__m512i __W, __mmask64 __U, __m512i __A) +{ + return (__m512i) __builtin_ia32_movdquqi512_mask ((__v64qi) __A, + (__v64qi) __W, + (__mmask64) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_mov_epi8 (__mmask64 __U, __m512i __A) +{ + return (__m512i) __builtin_ia32_movdquqi512_mask ((__v64qi) __A, + (__v64qi) + _mm512_setzero_hi (), + (__mmask64) __U); +} + + #undef __DEFAULT_FN_ATTRS #endif diff --git a/clang/lib/Headers/avx512vlbwintrin.h b/clang/lib/Headers/avx512vlbwintrin.h index 0c213bf..569acb0 100644 --- a/clang/lib/Headers/avx512vlbwintrin.h +++ b/clang/lib/Headers/avx512vlbwintrin.h @@ -2848,6 +2848,77 @@ __builtin_ia32_psrlwi256_mask ((__v16hi)( __A),( __imm),\ }) + + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_mov_epi16 (__m128i __W, __mmask8 __U, __m128i __A) +{ + return (__m128i) __builtin_ia32_movdquhi128_mask ((__v8hi) __A, + (__v8hi) __W, + (__mmask8) __U); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_mov_epi16 (__mmask8 __U, __m128i __A) +{ + return (__m128i) __builtin_ia32_movdquhi128_mask ((__v8hi) __A, + (__v8hi) + _mm_setzero_hi (), + (__mmask8) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_mask_mov_epi16 (__m256i __W, __mmask16 __U, __m256i __A) +{ + return (__m256i) __builtin_ia32_movdquhi256_mask ((__v16hi) __A, + (__v16hi) __W, + (__mmask16) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_maskz_mov_epi16 (__mmask16 __U, __m256i __A) +{ + return (__m256i) __builtin_ia32_movdquhi256_mask ((__v16hi) __A, + (__v16hi) + _mm256_setzero_si256 (), + (__mmask16) __U); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_mov_epi8 (__m128i __W, __mmask16 __U, __m128i __A) +{ + return (__m128i) __builtin_ia32_movdquqi128_mask ((__v16qi) __A, + (__v16qi) __W, + (__mmask16) __U); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_mov_epi8 (__mmask16 __U, __m128i __A) +{ + return (__m128i) __builtin_ia32_movdquqi128_mask ((__v16qi) __A, + (__v16qi) + _mm_setzero_hi (), + (__mmask16) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_mask_mov_epi8 (__m256i __W, __mmask32 __U, __m256i __A) +{ + return (__m256i) __builtin_ia32_movdquqi256_mask ((__v32qi) __A, + (__v32qi) __W, + (__mmask32) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_maskz_mov_epi8 (__mmask32 __U, __m256i __A) +{ + return (__m256i) __builtin_ia32_movdquqi256_mask ((__v32qi) __A, + (__v32qi) + _mm256_setzero_si256 (), + (__mmask32) __U); +} + + #undef __DEFAULT_FN_ATTRS #endif /* __AVX512VLBWINTRIN_H */ diff --git a/clang/test/CodeGen/avx512bw-builtins.c b/clang/test/CodeGen/avx512bw-builtins.c index 77f1007..a78712e 100644 --- a/clang/test/CodeGen/avx512bw-builtins.c +++ b/clang/test/CodeGen/avx512bw-builtins.c @@ -1275,6 +1275,28 @@ __m512i test_mm512_maskz_srli_epi16(__mmask32 __U, __m512i __A) { return _mm512_maskz_srli_epi16(__U, __A, 5); } +__m512i test_mm512_mask_mov_epi16(__m512i __W, __mmask32 __U, __m512i __A) { + // CHECK-LABEL: @test_mm512_mask_mov_epi16 + // CHECK: @llvm.x86.avx512.mask.movu.w.512 + return _mm512_mask_mov_epi16(__W, __U, __A); +} + +__m512i test_mm512_maskz_mov_epi16(__mmask32 __U, __m512i __A) { + // CHECK-LABEL: @test_mm512_maskz_mov_epi16 + // CHECK: @llvm.x86.avx512.mask.movu.w.512 + return _mm512_maskz_mov_epi16(__U, __A); +} +__m512i test_mm512_mask_mov_epi8(__m512i __W, __mmask64 __U, __m512i __A) { + // CHECK-LABEL: @test_mm512_mask_mov_epi8 + // CHECK: @llvm.x86.avx512.mask.movu.b.512 + return _mm512_mask_mov_epi8(__W, __U, __A); +} + +__m512i test_mm512_maskz_mov_epi8(__mmask64 __U, __m512i __A) { + // CHECK-LABEL: @test_mm512_maskz_mov_epi8 + // CHECK: @llvm.x86.avx512.mask.movu.b.512 + return _mm512_maskz_mov_epi8(__U, __A); +} diff --git a/clang/test/CodeGen/avx512vlbw-builtins.c b/clang/test/CodeGen/avx512vlbw-builtins.c index 4a00244..5bc7a7c 100644 --- a/clang/test/CodeGen/avx512vlbw-builtins.c +++ b/clang/test/CodeGen/avx512vlbw-builtins.c @@ -1957,3 +1957,51 @@ __m256i test_mm256_maskz_slli_epi16(__mmask16 __U, __m256i __A) { return _mm256_maskz_slli_epi16(__U, __A, 5); } +__m128i test_mm_mask_mov_epi16(__m128i __W, __mmask8 __U, __m128i __A) { + // CHECK-LABEL: @test_mm_mask_mov_epi16 + // CHECK: @llvm.x86.avx512.mask.mov.128 + return _mm_mask_mov_epi16(__W, __U, __A); +} + +__m128i test_mm_maskz_mov_epi16(__mmask8 __U, __m128i __A) { + // CHECK-LABEL: @test_mm_maskz_mov_epi16 + // CHECK: @llvm.x86.avx512.mask.mov.128 + return _mm_maskz_mov_epi16(__U, __A); +} + +__m256i test_mm256_mask_mov_epi16(__m256i __W, __mmask16 __U, __m256i __A) { + // CHECK-LABEL: @test_mm256_mask_mov_epi16 + // CHECK: @llvm.x86.avx512.mask.mov.256 + return _mm256_mask_mov_epi16(__W, __U, __A); +} + +__m256i test_mm256_maskz_mov_epi16(__mmask16 __U, __m256i __A) { + // CHECK-LABEL: @test_mm256_maskz_mov_epi16 + // CHECK: @llvm.x86.avx512.mask.mov.256 + return _mm256_maskz_mov_epi16(__U, __A); +} + +__m128i test_mm_mask_mov_epi8(__m128i __W, __mmask16 __U, __m128i __A) { + // CHECK-LABEL: @test_mm_mask_mov_epi8 + // CHECK: @llvm.x86.avx512.mask.mov.128 + return _mm_mask_mov_epi8(__W, __U, __A); +} + +__m128i test_mm_maskz_mov_epi8(__mmask16 __U, __m128i __A) { + // CHECK-LABEL: @test_mm_maskz_mov_epi8 + // CHECK: @llvm.x86.avx512.mask.mov.128 + return _mm_maskz_mov_epi8(__U, __A); +} + +__m256i test_mm256_mask_mov_epi8(__m256i __W, __mmask32 __U, __m256i __A) { + // CHECK-LABEL: @test_mm256_mask_mov_epi8 + // CHECK: @llvm.x86.avx512.mask.mov.256 + return _mm256_mask_mov_epi8(__W, __U, __A); +} + +__m256i test_mm256_maskz_mov_epi8(__mmask32 __U, __m256i __A) { + // CHECK-LABEL: @test_mm256_maskz_mov_epi8 + // CHECK: @llvm.x86.avx512.mask.mov.256 + return _mm256_maskz_mov_epi8(__U, __A); +} + -- 2.7.4