[Clang][AVX512][BUILTIN] Adding PSRL{W|WI}{128|256|512}

[m_zuckerman]

ping

[m_zuckerman]

ping

[delena]

LGTM

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_psllw128_mask, "V8sV8sV8sV8sUc","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_psllw256_mask, "V16sV16sV8sV16sUs","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_psllwi128_mask, "V8sV8siV8sUc","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_psllwi256_mask, "V16sV16siV16sUs","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_psllv2di_mask, "V2LLiV2LLiV2LLiV2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_psllv4di_mask, "V4LLiV4LLiV4LLiV4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_psllv4si_mask, "V4iV4iV4iV4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_psllv8si_mask, "V8iV8iV8iV8iUc","","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

lib/Headers/avx512bwintrin.h

 
 #define _mm512_maskz_slli_epi16( __U, __A, __B) __extension__ ({ \
 __builtin_ia32_psllwi512_mask ((__v32hi)( __A),( __B),\
               (__v32hi)\
               _mm512_setzero_hi (),\
               (__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

lib/Headers/avx512vlbwintrin.h

 
 #define _mm256_maskz_slli_epi16( __U, __A, __B) __extension__ ({ \
 __builtin_ia32_psllwi256_mask ((__v16hi)( __A),( __B),\
               (__v16hi)\
               _mm256_setzero_si256 (),\
               (__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 */

test/CodeGen/avx512bw-builtins.c

 }
 
 __m512i test_mm512_maskz_slli_epi16(__mmask32 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_slli_epi16
   // CHECK: @llvm.x86.avx512.mask.psll.wi.512
   return _mm512_maskz_slli_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); 
+}
+
+
 

test/CodeGen/avx512vlbw-builtins.c

 }
 
 __m256i test_mm256_maskz_slli_epi16(__mmask16 __U, __m256i __A) {
   // CHECK-LABEL: @test_mm256_maskz_slli_epi16
   // CHECK: @llvm.x86.avx512.mask.psll.wi.256
   return _mm256_maskz_slli_epi16(__U, __A, 5); 
 }
 
+__m128i test_mm_mask_srl_epi16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+  // CHECK-LABEL: @test_mm_mask_srl_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.w.128
+  return _mm_mask_srl_epi16(__W, __U, __A, __B); 
+}
+
+__m128i test_mm_maskz_srl_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
+  // CHECK-LABEL: @test_mm_maskz_srl_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.w.128
+  return _mm_maskz_srl_epi16(__U, __A, __B); 
+}
+
+__m256i test_mm256_mask_srl_epi16(__m256i __W, __mmask16 __U, __m256i __A, __m128i __B) {
+  // CHECK-LABEL: @test_mm256_mask_srl_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.w.256
+  return _mm256_mask_srl_epi16(__W, __U, __A, __B); 
+}
+
+__m256i test_mm256_maskz_srl_epi16(__mmask16 __U, __m256i __A, __m128i __B) {
+  // CHECK-LABEL: @test_mm256_maskz_srl_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.w.256
+  return _mm256_maskz_srl_epi16(__U, __A, __B); 
+}
+
+__m128i test_mm_mask_srli_epi16(__m128i __W, __mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_srli_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.wi.128
+  return _mm_mask_srli_epi16(__W, __U, __A, 5); 
+}
+
+__m128i test_mm_maskz_srli_epi16(__mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_srli_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.wi.128
+  return _mm_maskz_srli_epi16(__U, __A, 5); 
+}
+
+__m256i test_mm256_mask_srli_epi16(__m256i __W, __mmask16 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_mask_srli_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.wi.256
+  return _mm256_mask_srli_epi16(__W, __U, __A, 5); 
+}
+
+__m256i test_mm256_maskz_srli_epi16(__mmask16 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_srli_epi16
+  // CHECK: @llvm.x86.avx512.mask.psrl.wi.256
+  return _mm256_maskz_srli_epi16(__U, __A, 5); 
+}
+