[Clang][BuiltIn][avx512] Adding intrinsics for vpshufd instruction set

[delena]

include/clang/Basic/BuiltinsX86.def

2188

Where is the mask version of sad?

[m_zuckerman]

I delete the SAD instruction from this review and enter him into another review.

[m_zuckerman]

ping

[AsafBadouh]

lgtm

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_kandhi, "UsUsUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_kandnhi, "UsUsUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_korhi, "UsUsUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_kortestchi, "iUsUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_kortestzhi, "iUsUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_kunpckhi, "UsUsUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_kxnorhi, "UsUsUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_kxorhi, "UsUsUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psadbw512, "V8LLiV64cV64c","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pshufd512_mask, "V16iV16iCsV16iUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pshufd256_mask, "V8iV8iCsV8iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pshufd128_mask, "V4iV4iCsV4iUc","","avx512vl")
 
 #undef BUILTIN
 #undef TARGET_BUILTIN

lib/Headers/avx512bwintrin.h

_mm512_mask_permutexvar_epi16 (__m512i __W, __mmask32 __M, __m512i __A,
              __m512i __B)
 {
   return (__m512i) __builtin_ia32_permvarhi512_mask ((__v32hi) __B,
                  (__v32hi) __A,
                  (__v32hi) __W,
                  (__mmask32) __M);
 }
 
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_sad_epu8 (__m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_psadbw512 ((__v64qi) __A,
+               (__v64qi) __B);
+}
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif

lib/Headers/avx512fintrin.h

 }
 
 static __inline__ __mmask16 __DEFAULT_FN_ATTRS
 _mm512_kxor (__mmask16 __A, __mmask16 __B)
 {
   return (__mmask16) __builtin_ia32_kxorhi ((__mmask16) __A, (__mmask16) __B);
 }
 
+#define _mm512_shuffle_epi32( __A, __I) __extension__ ({ \
+__builtin_ia32_pshufd512_mask ((__v16si)( __A),\
+              ( __I),\
+              (__v16si) _mm512_undefined_epi32 (),\
+              (__mmask16) -1);\
+})
+
+#define _mm512_mask_shuffle_epi32( __W, __U, __A, __I) __extension__ ({ \
+__builtin_ia32_pshufd512_mask ((__v16si)( __A),\
+              ( __I),\
+              (__v16si)( __W),\
+              (__mmask16)( __U));\
+})
+
+#define _mm512_maskz_shuffle_epi32( __U, __A, __I) __extension__ ({ \
+__builtin_ia32_pshufd512_mask ((__v16si)( __A),\
+              ( __I),\
+              (__v16si) _mm512_setzero_si512 (),\
+              (__mmask16)( __U));\
+})
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif // __AVX512FINTRIN_H

lib/Headers/avx512vlintrin.h

 _mm256_permutexvar_epi32 (__m256i __X, __m256i __Y)
 {
   return (__m256i) __builtin_ia32_permvarsi256_mask ((__v8si) __Y,
                  (__v8si) __X,
                  (__v8si) _mm256_undefined_si256(),
                  (__mmask8) -1);
 }
 
+#define _mm256_mask_shuffle_epi32( __W, __U, __A, __I) __extension__({\
+__builtin_ia32_pshufd256_mask((__v8si) (__A), (__I),\
+              (__v8si) (__W), (__mmask8) __U);\
+})
+
+#define _mm256_maskz_shuffle_epi32( __U,  __A, __I) __extension__({\
+__builtin_ia32_pshufd256_mask((__v8si) (__A), (__I),\
+              (__v8si) _mm256_setzero_si256 (),\
+              (__mmask8) (__U));\
+})
+
+#define _mm_mask_shuffle_epi32( __W, __U, __A, __I) __extension__({\
+  __builtin_ia32_pshufd128_mask ((__v4si) (__A), (__I),\
+              (__v4si) (__W), (__mmask8) __U);\
+})
+
+#define _mm_maskz_shuffle_epi32( __U,  __A, __I) __extension__({\
+  __builtin_ia32_pshufd128_mask ((__v4si) (__A), (__I),\
+              (__v4si)\
+              _mm_setzero_si128 (),\
+              (__mmask8) (__U));\
+})
+
 #undef __DEFAULT_FN_ATTRS
 #undef __DEFAULT_FN_ATTRS_BOTH
 
 #endif /* __AVX512VLINTRIN_H */

test/CodeGen/avx512bw-builtins.c

// CHECK-LABEL: @test_mm512_maskz_permutexvar_epi16
   return _mm512_maskz_permutexvar_epi16(__M, __A, __B); 
 }
 
 __m512i test_mm512_mask_permutexvar_epi16(__m512i __W, __mmask32 __M, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_permutexvar_epi16
   // CHECK: @llvm.x86.avx512.mask.permvar.hi.512
   return _mm512_mask_permutexvar_epi16(__W, __M, __A, __B); 
 }
+
+__m512i test_mm512_sad_epu8(__m512i __A, __m512i __B) {
+  // CHECK-LABEL: @test_mm512_sad_epu8
+  // CHECK: @llvm.x86.avx512.psad.bw.512
+  return _mm512_sad_epu8(__A, __B); 
+}
+

test/CodeGen/avx512f-builtins.c

__mmask16 test_mm512_kxnor(__mmask16 __A, __mmask16 __B) {
   return _mm512_kxnor(__A, __B); 
 }
 
 __mmask16 test_mm512_kxor(__mmask16 __A, __mmask16 __B) {
   // CHECK-LABEL: @test_mm512_kxor
   // CHECK: @llvm.x86.avx512.kxor.w
   return _mm512_kxor(__A, __B); 
 }
+
+__m512i test_mm512_shuffle_epi32(__m512i __A) {
+  // CHECK-LABEL: @test_mm512_shuffle_epi32
+  // CHECK: @llvm.x86.avx512.mask.pshuf.d.512
+  return _mm512_shuffle_epi32(__A, 1); 
+}
+
+__m512i test_mm512_mask_shuffle_epi32(__m512i __W, __mmask16 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_mask_shuffle_epi32
+  // CHECK: @llvm.x86.avx512.mask.pshuf.d.512
+  return _mm512_mask_shuffle_epi32(__W, __U, __A, 1); 
+}
+
+__m512i test_mm512_maskz_shuffle_epi32(__mmask16 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_shuffle_epi32
+  // CHECK: @llvm.x86.avx512.mask.pshuf.d.512
+  return _mm512_maskz_shuffle_epi32(__U, __A, 1); 
+}
+