[Clang][AVX512][BuiltIn] Adding avx512 ( store ) builtin to clang

cfe/trunk/include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_loaddqudi128_mask, "V2LLiV2LLi*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loaddqudi256_mask, "V4LLiV4LLi*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loaddqusi128_mask, "V4iV4i*V4iUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_loaddqusi256_mask, "V8iV8i*V8iUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_loadupd128_mask, "V2dV2d*V2dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadupd256_mask, "V4dV4d*V4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadups128_mask, "V4fV4f*V4fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadups256_mask, "V8fV8f*V8fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storedquhi512_mask, "vV32s*V32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_storedquqi512_mask, "vV64c*V64cULLi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_storedquhi128_mask, "vV8s*V8sUc","","avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_storedquhi256_mask, "vV16s*V16sUs","","avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_storedquqi128_mask, "vV16c*V16cUs","","avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_storedquqi256_mask, "vV32c*V32cUi","","avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_storeapd128_mask, "vV2d*V2dUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storeapd256_mask, "vV4d*V4dUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storeaps128_mask, "vV4f*V4fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storeaps256_mask, "vV8f*V8fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storedqudi128_mask, "vV2LLi*V2LLiUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storedqudi256_mask, "vV4LLi*V4LLiUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storedqusi128_mask, "vV4i*V4iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storedqusi256_mask, "vV8i*V8iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storeupd128_mask, "vV2d*V2dUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storeupd256_mask, "vV4d*V4dUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storeups128_mask, "vV4f*V4fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storeups256_mask, "vV8f*V8fUc","","avx512vl")
 
 #undef BUILTIN
 #undef TARGET_BUILTIN

cfe/trunk/lib/Headers/avx512bwintrin.h

 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_maskz_loadu_epi8 (__mmask64 __U, void const *__P)
 {
   return (__m512i) __builtin_ia32_loaddquqi512_mask ((__v64qi *) __P,
                  (__v64qi)
                  _mm512_setzero_hi (),
                  (__mmask64) __U);
 }
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm512_mask_storeu_epi16 (void *__P, __mmask32 __U, __m512i __A)
+{
+  __builtin_ia32_storedquhi512_mask ((__v32hi *) __P,
+             (__v32hi) __A,
+             (__mmask32) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm512_mask_storeu_epi8 (void *__P, __mmask64 __U, __m512i __A)
+{
+  __builtin_ia32_storedquqi512_mask ((__v64qi *) __P,
+             (__v64qi) __A,
+             (__mmask64) __U);
+}
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif

cfe/trunk/lib/Headers/avx512vlbwintrin.h

 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_maskz_loadu_epi8 (__mmask32 __U, void const *__P)
 {
   return (__m256i) __builtin_ia32_loaddquqi256_mask ((__v32qi *) __P,
                  (__v32qi)
                  _mm256_setzero_si256 (),
                  (__mmask32) __U);
 }
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_storeu_epi16 (void *__P, __mmask8 __U, __m128i __A)
+{
+  __builtin_ia32_storedquhi128_mask ((__v8hi *) __P,
+             (__v8hi) __A,
+             (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_storeu_epi16 (void *__P, __mmask16 __U, __m256i __A)
+{
+  __builtin_ia32_storedquhi256_mask ((__v16hi *) __P,
+             (__v16hi) __A,
+             (__mmask16) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_storeu_epi8 (void *__P, __mmask16 __U, __m128i __A)
+{
+  __builtin_ia32_storedquqi128_mask ((__v16qi *) __P,
+             (__v16qi) __A,
+             (__mmask16) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_storeu_epi8 (void *__P, __mmask32 __U, __m256i __A)
+{
+  __builtin_ia32_storedquqi256_mask ((__v32qi *) __P,
+             (__v32qi) __A,
+             (__mmask32) __U);
+}
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif /* __AVX512VLBWINTRIN_H */

cfe/trunk/lib/Headers/avx512vlintrin.h

 static __inline__ __m256 __DEFAULT_FN_ATTRS
 _mm256_maskz_loadu_ps (__mmask8 __U, void const *__P)
 {
   return (__m256) __builtin_ia32_loadups256_mask ((__v8sf *) __P,
               (__v8sf)
               _mm256_setzero_ps (),
               (__mmask8) __U);
 }
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_store_pd (void *__P, __mmask8 __U, __m128d __A)
+{
+  __builtin_ia32_storeapd128_mask ((__v2df *) __P,
+           (__v2df) __A,
+           (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_store_pd (void *__P, __mmask8 __U, __m256d __A)
+{
+  __builtin_ia32_storeapd256_mask ((__v4df *) __P,
+           (__v4df) __A,
+           (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_store_ps (void *__P, __mmask8 __U, __m128 __A)
+{
+  __builtin_ia32_storeaps128_mask ((__v4sf *) __P,
+           (__v4sf) __A,
+           (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_store_ps (void *__P, __mmask8 __U, __m256 __A)
+{
+  __builtin_ia32_storeaps256_mask ((__v8sf *) __P,
+           (__v8sf) __A,
+           (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_storeu_epi64 (void *__P, __mmask8 __U, __m128i __A)
+{
+  __builtin_ia32_storedqudi128_mask ((__v2di *) __P,
+             (__v2di) __A,
+             (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_storeu_epi64 (void *__P, __mmask8 __U, __m256i __A)
+{
+  __builtin_ia32_storedqudi256_mask ((__v4di *) __P,
+             (__v4di) __A,
+             (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_storeu_epi32 (void *__P, __mmask8 __U, __m128i __A)
+{
+  __builtin_ia32_storedqusi128_mask ((__v4si *) __P,
+             (__v4si) __A,
+             (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_storeu_epi32 (void *__P, __mmask8 __U, __m256i __A)
+{
+  __builtin_ia32_storedqusi256_mask ((__v8si *) __P,
+             (__v8si) __A,
+             (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_storeu_pd (void *__P, __mmask8 __U, __m128d __A)
+{
+  __builtin_ia32_storeupd128_mask ((__v2df *) __P,
+           (__v2df) __A,
+           (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_storeu_pd (void *__P, __mmask8 __U, __m256d __A)
+{
+  __builtin_ia32_storeupd256_mask ((__v4df *) __P,
+           (__v4df) __A,
+           (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_storeu_ps (void *__P, __mmask8 __U, __m128 __A)
+{
+  __builtin_ia32_storeups128_mask ((__v4sf *) __P,
+           (__v4sf) __A,
+           (__mmask8) __U);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_storeu_ps (void *__P, __mmask8 __U, __m256 __A)
+{
+  __builtin_ia32_storeups256_mask ((__v8sf *) __P,
+           (__v8sf) __A,
+           (__mmask8) __U);
+}
+
 #undef __DEFAULT_FN_ATTRS
 #undef __DEFAULT_FN_ATTRS_BOTH
 
 #endif /* __AVX512VLINTRIN_H */

cfe/trunk/test/CodeGen/avx512bw-builtins.c

__m512i test_mm512_mask_loadu_epi8(__m512i __W, __mmask64 __U, void const *__P) {
   return _mm512_mask_loadu_epi8(__W, __U, __P); 
 }
 
 __m512i test_mm512_maskz_loadu_epi8(__mmask64 __U, void const *__P) {
   // CHECK-LABEL: @test_mm512_maskz_loadu_epi8
   // CHECK: @llvm.x86.avx512.mask.loadu.b.512
   return _mm512_maskz_loadu_epi8(__U, __P); 
 }
+void test_mm512_mask_storeu_epi16(void *__P, __mmask32 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_mask_storeu_epi16
+  // CHECK: @llvm.x86.avx512.mask.storeu.w.
+  return _mm512_mask_storeu_epi16(__P, __U, __A); 
+}
 
-
+void test_mm512_mask_storeu_epi8(void *__P, __mmask64 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_mask_storeu_epi8
+  // CHECK: @llvm.x86.avx512.mask.storeu.b.
+  return _mm512_mask_storeu_epi8(__P, __U, __A); 
+}

cfe/trunk/test/CodeGen/avx512vl-builtins.c

 }
 
 __m256 test_mm256_maskz_loadu_ps(__mmask8 __U, void const *__P) {
   // CHECK-LABEL: @test_mm256_maskz_loadu_ps
   // CHECK: @llvm.x86.avx512.mask.loadu.ps.256
   return _mm256_maskz_loadu_ps(__U, __P); 
 }
 
+void test_mm_mask_store_pd(void *__P, __mmask8 __U, __m128d __A) {
+  // CHECK-LABEL: @test_mm_mask_store_pd
+  // CHECK: @llvm.x86.avx512.mask.store.pd.128
+  return _mm_mask_store_pd(__P, __U, __A); 
+}
+
+void test_mm256_mask_store_pd(void *__P, __mmask8 __U, __m256d __A) {
+  // CHECK-LABEL: @test_mm256_mask_store_pd
+  // CHECK: @llvm.x86.avx512.mask.store.pd.256
+  return _mm256_mask_store_pd(__P, __U, __A); 
+}
+
+void test_mm_mask_store_ps(void *__P, __mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_mask_store_ps
+  // CHECK: @llvm.x86.avx512.mask.store.ps.128
+  return _mm_mask_store_ps(__P, __U, __A); 
+}
+
+void test_mm256_mask_store_ps(void *__P, __mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_mask_store_ps
+  // CHECK: @llvm.x86.avx512.mask.store.ps.256
+  return _mm256_mask_store_ps(__P, __U, __A); 
+}
+
+void test_mm_mask_storeu_epi64(void *__P, __mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_storeu_epi64
+  // CHECK: @llvm.x86.avx512.mask.storeu
+  return _mm_mask_storeu_epi64(__P, __U, __A); 
+}
+
+void test_mm256_mask_storeu_epi64(void *__P, __mmask8 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_mask_storeu_epi64
+  // CHECK: @llvm.x86.avx512.mask.storeu
+  return _mm256_mask_storeu_epi64(__P, __U, __A); 
+}
+
+void test_mm_mask_storeu_epi32(void *__P, __mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_storeu_epi32
+  // CHECK: @llvm.x86.avx512.mask.storeu
+  return _mm_mask_storeu_epi32(__P, __U, __A); 
+}
+
+void test_mm256_mask_storeu_epi32(void *__P, __mmask8 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_mask_storeu_epi32
+  // CHECK: @llvm.x86.avx512.mask.storeu
+  return _mm256_mask_storeu_epi32(__P, __U, __A); 
+}
+
+void test_mm_mask_storeu_pd(void *__P, __mmask8 __U, __m128d __A) {
+  // CHECK-LABEL: @test_mm_mask_storeu_pd
+  // CHECK: @llvm.x86.avx512.mask.storeu.pd.128
+  return _mm_mask_storeu_pd(__P, __U, __A); 
+}
+
+void test_mm256_mask_storeu_pd(void *__P, __mmask8 __U, __m256d __A) {
+  // CHECK-LABEL: @test_mm256_mask_storeu_pd
+  // CHECK: @llvm.x86.avx512.mask.storeu.pd.256
+  return _mm256_mask_storeu_pd(__P, __U, __A); 
+}
+
+void test_mm_mask_storeu_ps(void *__P, __mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_mask_storeu_ps
+  // CHECK: @llvm.x86.avx512.mask.storeu.ps.128
+  return _mm_mask_storeu_ps(__P, __U, __A); 
+}
+
+void test_mm256_mask_storeu_ps(void *__P, __mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_mask_storeu_ps
+  // CHECK: @llvm.x86.avx512.mask.storeu.ps.256
+  return _mm256_mask_storeu_ps(__P, __U, __A); 
+}
+

cfe/trunk/test/CodeGen/avx512vlbw-builtins.c

 }
 
 __m256i test_mm256_maskz_loadu_epi8(__mmask32 __U, void const *__P) {
   // CHECK-LABEL: @test_mm256_maskz_loadu_epi8
   // CHECK: @llvm.x86.avx512.mask.loadu.b.256
   return _mm256_maskz_loadu_epi8(__U, __P); 
 }
 
+void test_mm_mask_storeu_epi16(void *__P, __mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_storeu_epi16
+  // CHECK: @llvm.x86.avx512.mask.storeu.w.128
+  return _mm_mask_storeu_epi16(__P, __U, __A); 
+}
+
+void test_mm256_mask_storeu_epi16(void *__P, __mmask16 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_mask_storeu_epi16
+  // CHECK: @llvm.x86.avx512.mask.storeu.w.256
+  return _mm256_mask_storeu_epi16(__P, __U, __A); 
+}
+
+void test_mm_mask_storeu_epi8(void *__P, __mmask16 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_storeu_epi8
+  // CHECK: @llvm.x86.avx512.mask.storeu.b.128
+  return _mm_mask_storeu_epi8(__P, __U, __A); 
+}
+
+void test_mm256_mask_storeu_epi8(void *__P, __mmask32 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_mask_storeu_epi8
+  // CHECK: @llvm.x86.avx512.mask.storeu.b.256
+  return _mm256_mask_storeu_epi8(__P, __U, __A); 
+}