From: Michael Zuckerman Date: Thu, 3 Mar 2016 08:55:20 +0000 (+0000) Subject: [Clang][AVX512][BUILTIN] Adding PSRL{W|WI}{128|256|512} X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=8f01f0c291bee2d7ebe65d5949f014f8f8fa2464;p=clang [Clang][AVX512][BUILTIN] Adding PSRL{W|WI}{128|256|512} Differential Revision: http://reviews.llvm.org/D17754 git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@262593 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def index bfe071a63e..a872555abe 100644 --- a/include/clang/Basic/BuiltinsX86.def +++ b/include/clang/Basic/BuiltinsX86.def @@ -1691,6 +1691,12 @@ TARGET_BUILTIN(__builtin_ia32_psraw128_mask, "V8sV8sV8sV8sUc","","avx512bw,avx51 TARGET_BUILTIN(__builtin_ia32_psraw256_mask, "V16sV16sV8sV16sUs","","avx512bw,avx512vl") TARGET_BUILTIN(__builtin_ia32_psrawi128_mask, "V8sV8sIiV8sUc","","avx512bw,avx512vl") TARGET_BUILTIN(__builtin_ia32_psrawi256_mask, "V16sV16sIiV16sUs","","avx512bw,avx512vl") +TARGET_BUILTIN(__builtin_ia32_psrlw512_mask, "V32sV32sV8sV32sUi","","avx512bw") +TARGET_BUILTIN(__builtin_ia32_psrlwi512_mask, "V32sV32sIiV32sUi","","avx512bw") +TARGET_BUILTIN(__builtin_ia32_psrlw128_mask, "V8sV8sV8sV8sUc","","avx512bw,avx512vl") +TARGET_BUILTIN(__builtin_ia32_psrlw256_mask, "V16sV16sV8sV16sUs","","avx512bw,avx512vl") +TARGET_BUILTIN(__builtin_ia32_psrlwi128_mask, "V8sV8sIiV8sUc","","avx512bw,avx512vl") +TARGET_BUILTIN(__builtin_ia32_psrlwi256_mask, "V16sV16sIiV16sUs","","avx512bw,avx512vl") #undef BUILTIN #undef TARGET_BUILTIN diff --git a/lib/Headers/avx512bwintrin.h b/lib/Headers/avx512bwintrin.h index 7f3f1bf4f0..17ca52ad4d 100644 --- a/lib/Headers/avx512bwintrin.h +++ b/lib/Headers/avx512bwintrin.h @@ -1829,6 +1829,57 @@ __builtin_ia32_psrawi512_mask ((__v32hi)( __A),( __B),\ (__mmask32)( __U));\ }) + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srl_epi16 (__m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A, + (__v8hi) __B, + (__v32hi) + _mm512_setzero_hi (), + (__mmask32) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_srl_epi16 (__m512i __W, __mmask32 __U, __m512i __A, + __m128i __B) +{ + return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A, + (__v8hi) __B, + (__v32hi) __W, + (__mmask32) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_srl_epi16 (__mmask32 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A, + (__v8hi) __B, + (__v32hi) + _mm512_setzero_hi (), + (__mmask32) __U); +} + +#define _mm512_srli_epi16( __A, __imm) __extension__ ({ \ +__builtin_ia32_psrlwi512_mask ((__v32hi)( __A),( __imm),\ + (__v32hi)\ + _mm512_setzero_hi (),\ + (__mmask32) -1);\ +}) + +#define _mm512_mask_srli_epi16( __W, __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psrlwi512_mask ((__v32hi)( __A),( __imm),\ + (__v32hi)( __W),\ + (__mmask32)( __U));\ +}) + +#define _mm512_maskz_srli_epi16( __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psrlwi512_mask ((__v32hi)( __A),( __imm),\ + (__v32hi)\ + _mm512_setzero_hi (),\ + (__mmask32)( __U));\ +}) + #undef __DEFAULT_FN_ATTRS #endif diff --git a/lib/Headers/avx512vlbwintrin.h b/lib/Headers/avx512vlbwintrin.h index a4b8ff9a06..0c213bf6e0 100644 --- a/lib/Headers/avx512vlbwintrin.h +++ b/lib/Headers/avx512vlbwintrin.h @@ -2781,6 +2781,73 @@ __builtin_ia32_psrawi256_mask ((__v16hi)( __A),( __imm),\ (__mmask16)( __U));\ }) +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_srl_epi16 (__m128i __W, __mmask8 __U, __m128i __A, + __m128i __B) +{ + return (__m128i) __builtin_ia32_psrlw128_mask ((__v8hi) __A, + (__v8hi) __B, + (__v8hi) __W, + (__mmask8) __U); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_srl_epi16 (__mmask8 __U, __m128i __A, __m128i __B) +{ + return (__m128i) __builtin_ia32_psrlw128_mask ((__v8hi) __A, + (__v8hi) __B, + (__v8hi) + _mm_setzero_si128 (), + (__mmask8) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_mask_srl_epi16 (__m256i __W, __mmask16 __U, __m256i __A, + __m128i __B) +{ + return (__m256i) __builtin_ia32_psrlw256_mask ((__v16hi) __A, + (__v8hi) __B, + (__v16hi) __W, + (__mmask16) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_maskz_srl_epi16 (__mmask16 __U, __m256i __A, __m128i __B) +{ + return (__m256i) __builtin_ia32_psrlw256_mask ((__v16hi) __A, + (__v8hi) __B, + (__v16hi) + _mm256_setzero_si256 (), + (__mmask16) __U); +} + +#define _mm_mask_srli_epi16( __W, __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psrlwi128_mask ((__v8hi)( __A),( __imm),\ + (__v8hi)( __W),\ + (__mmask8)( __U));\ +}) + +#define _mm_maskz_srli_epi16( __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psrlwi128_mask ((__v8hi)( __A),( __imm),\ + (__v8hi)\ + _mm_setzero_si128 (),\ + (__mmask8)( __U));\ +}) + +#define _mm256_mask_srli_epi16( __W, __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psrlwi256_mask ((__v16hi)( __A),( __imm),\ + (__v16hi)( __W),\ + (__mmask16)( __U));\ +}) + +#define _mm256_maskz_srli_epi16( __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psrlwi256_mask ((__v16hi)( __A),( __imm),\ + (__v16hi)\ + _mm256_setzero_si256 (),\ + (__mmask16)( __U));\ +}) + + #undef __DEFAULT_FN_ATTRS #endif /* __AVX512VLBWINTRIN_H */ diff --git a/test/CodeGen/avx512bw-builtins.c b/test/CodeGen/avx512bw-builtins.c index a1fafe4ce5..77f1007ce9 100644 --- a/test/CodeGen/avx512bw-builtins.c +++ b/test/CodeGen/avx512bw-builtins.c @@ -1239,5 +1239,42 @@ __m512i test_mm512_maskz_srai_epi16(__mmask32 __U, __m512i __A) { return _mm512_maskz_srai_epi16(__U, __A, 5); } +__m512i test_mm512_srl_epi16(__m512i __A, __m128i __B) { + // CHECK-LABEL: @test_mm512_srl_epi16 + // CHECK: @llvm.x86.avx512.mask.psrl.w.512 + return _mm512_srl_epi16(__A, __B); +} + +__m512i test_mm512_mask_srl_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B) { + // CHECK-LABEL: @test_mm512_mask_srl_epi16 + // CHECK: @llvm.x86.avx512.mask.psrl.w.512 + return _mm512_mask_srl_epi16(__W, __U, __A, __B); +} + +__m512i test_mm512_maskz_srl_epi16(__mmask32 __U, __m512i __A, __m128i __B) { + // CHECK-LABEL: @test_mm512_maskz_srl_epi16 + // CHECK: @llvm.x86.avx512.mask.psrl.w.512 + return _mm512_maskz_srl_epi16(__U, __A, __B); +} + +__m512i test_mm512_srli_epi16(__m512i __A) { + // CHECK-LABEL: @test_mm512_srli_epi16 + // CHECK: @llvm.x86.avx512.mask.psrl.wi.512 + return _mm512_srli_epi16(__A, 5); +} + +__m512i test_mm512_mask_srli_epi16(__m512i __W, __mmask32 __U, __m512i __A) { + // CHECK-LABEL: @test_mm512_mask_srli_epi16 + // CHECK: @llvm.x86.avx512.mask.psrl.wi.512 + return _mm512_mask_srli_epi16(__W, __U, __A, 5); +} + +__m512i test_mm512_maskz_srli_epi16(__mmask32 __U, __m512i __A) { + // CHECK-LABEL: @test_mm512_maskz_srli_epi16 + // CHECK: @llvm.x86.avx512.mask.psrl.wi.512 + return _mm512_maskz_srli_epi16(__U, __A, 5); +} + +