[Clang][BuiltIn][AVX512] adding _mm{|256|512}_mask_cvt{s|us|}epi16_storeu_epi8 intrinsics

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

 TARGET_BUILTIN(__builtin_ia32_pbroadcastq128_mask, "V2LLiV2LLiV2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pbroadcastq256_mask, "V4LLiV2LLiV4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pbroadcastb512_mask, "V64cV16cV64cULLi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pbroadcastw512_gpr_mask, "V32shV32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pbroadcastw256_gpr_mask, "V16shV16sUs","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pbroadcastw128_gpr_mask, "V8ssV8sUc","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb512_mask, "V16cV16iV16cUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb512mem_mask, "vV16c*V16iUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmovswb512mem_mask, "vV32c*V32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovsdw512_mask, "V16sV16iV16sUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsdw512mem_mask, "vV16s*V16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsqb512_mask, "V16cV8LLiV16cUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsqb512mem_mask, "vV16c*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsqd512_mask, "V8iV8LLiV8iUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsqd512mem_mask, "vV8i*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsqw512_mask, "V8sV8LLiV8sUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsqw512mem_mask, "vV8s*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb128_mask, "V16cV4iV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb128mem_mask, "vV16c*V4iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovswb128mem_mask, "vV16c*V8sUc","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb256_mask, "V16cV8iV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb256mem_mask, "vV16c*V8iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovswb256mem_mask, "vV16c*V16sUs","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovsdw128_mask, "V8sV4iV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsdw128mem_mask, "vV8s*V4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsdw256_mask, "V8sV8iV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsdw256mem_mask, "vV8s*V8iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqb128_mask, "V16cV2LLiV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqb128mem_mask, "vV16c*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqb256_mask, "V16cV4LLiV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqb256mem_mask, "vV16c*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqd128_mask, "V4iV2LLiV4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqd128mem_mask, "vV4i*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqd256_mask, "V4iV4LLiV4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqd256mem_mask, "vV4i*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqw128_mask, "V8sV2LLiV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqw128mem_mask, "vV8s*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqw256_mask, "V8sV4LLiV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsqw256mem_mask, "vV8s*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusdb512_mask, "V16cV16iV16cUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusdb512mem_mask, "vV16c*V16iUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmovuswb512mem_mask, "vV32c*V32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovusdw512_mask, "V16sV16iV16sUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusdw512mem_mask, "vV16s*V16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusqb512_mask, "V16cV8LLiV16cUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusqb512mem_mask, "vV16c*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusqd512_mask, "V8iV8LLiV8iUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusqd512mem_mask, "vV8i*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw512_mask, "V8sV8LLiV8sUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw512mem_mask, "vV8s*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovusdb128_mask, "V16cV4iV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusdb128mem_mask, "vV16c*V4iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovuswb128mem_mask, "vV16c*V8sUc","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovusdb256_mask, "V16cV8iV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusdb256mem_mask, "vV16c*V8iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovuswb256mem_mask, "vV16c*V16sUs","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovusdw128_mask, "V8sV4iV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusdw128mem_mask, "vV8s*V4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusdw256_mask, "V8sV8iV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusdw256mem_mask, "vV8s*V8iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqb128_mask, "V16cV2LLiV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqb128mem_mask, "vV16c*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqb256_mask, "V16cV4LLiV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqb256mem_mask, "vV16c*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqd128_mask, "V4iV2LLiV4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqd128mem_mask, "vV4i*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqd256_mask, "V4iV4LLiV4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqd256mem_mask, "vV4i*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw128_mask, "V8sV2LLiV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw128mem_mask, "vV8s*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw256_mask, "V8sV4LLiV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw256mem_mask, "vV8s*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdb512_mask, "V16cV16iV16cUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovdb512mem_mask, "vV16c*V16iUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmovwb512mem_mask, "vV32c*V32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovdw512_mask, "V16sV16iV16sUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovdw512mem_mask, "vV16s*V16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqb512_mask, "V16cV8LLiV16cUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqb512mem_mask, "vV16c*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqd512_mask, "V8iV8LLiV8iUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqd512mem_mask, "vV8i*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqw512_mask, "V8sV8LLiV8sUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqw512mem_mask, "vV8s*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovdb128_mask, "V16cV4iV16cUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovwb128mem_mask, "vV16c*V8sUc","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovdb128mem_mask, "vV16c*V4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdb256_mask, "V16cV8iV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdb256mem_mask, "vV16c*V8iUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovwb256mem_mask, "vV16c*V16sUs","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovdw128_mask, "V8sV4iV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdw128mem_mask, "vV8s*V4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdw256_mask, "V8sV8iV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdw256mem_mask, "vV8s*V8iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqb128_mask, "V16cV2LLiV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqb128mem_mask, "vV16c*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqb256_mask, "V16cV4LLiV16cUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqb256mem_mask, "vV16c*V4LLiUc","","avx512vl")

cfe/trunk/lib/Headers/avx512bwintrin.h

 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi16_epi8 (__mmask32 __M, __m512i __A) {
   return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
               (__v32qi) _mm256_setzero_si256(),
               __M);
 }
 
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm512_mask_cvtepi16_storeu_epi8 (void * __P, __mmask32 __M, __m512i __A)
+{
+  __builtin_ia32_pmovwb512mem_mask ((__v32qi *) __P, (__v32hi) __A, __M);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm512_mask_cvtsepi16_storeu_epi8 (void * __P, __mmask32 __M, __m512i __A)
+{
+  __builtin_ia32_pmovswb512mem_mask ((__v32qi *) __P, (__v32hi) __A, __M);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm512_mask_cvtusepi16_storeu_epi8 (void * __P, __mmask32 __M, __m512i __A)
+{
+  __builtin_ia32_pmovuswb512mem_mask ((__v32qi *) __P, (__v32hi) __A, __M);
+}
+
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_unpackhi_epi8(__m512i __A, __m512i __B) {
   return (__m512i)__builtin_shufflevector((__v64qi)__A, (__v64qi)__B,
                                           8,  64+8,   9, 64+9,
                                           10, 64+10, 11, 64+11,
                                           12, 64+12, 13, 64+13,
                                           14, 64+14, 15, 64+15,
                                           24, 64+24, 25, 64+25,

cfe/trunk/lib/Headers/avx512vlbwintrin.h

 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_maskz_cvtepi16_epi8 (__mmask8 __M, __m128i __A) {
   return (__m128i) __builtin_ia32_pmovwb128_mask ((__v8hi) __A,
                (__v16qi) _mm_setzero_si128(),
                __M);
 }
 
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_cvtepi16_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
+{
+  __builtin_ia32_pmovwb128mem_mask ((__v16qi *) __P, (__v8hi) __A, __M);
+}
+
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_cvtsepi16_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
+{
+  __builtin_ia32_pmovswb128mem_mask ((__v16qi *) __P, (__v8hi) __A, __M);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_cvtusepi16_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
+{
+  __builtin_ia32_pmovuswb128mem_mask ((__v16qi *) __P, (__v8hi) __A, __M);
+}
+
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_cvtepi16_epi8 (__m256i __A) {
   return (__m128i) __builtin_ia32_pmovwb256_mask ((__v16hi) __A,
                (__v16qi) _mm_setzero_si128(),
                (__mmask16) -1);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepi16_epi8 (__m128i __O, __mmask16 __M, __m256i __A) {
   return (__m128i) __builtin_ia32_pmovwb256_mask ((__v16hi) __A,
                (__v16qi) __O,
                __M);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepi16_epi8 (__mmask16 __M, __m256i __A) {
   return (__m128i) __builtin_ia32_pmovwb256_mask ((__v16hi) __A,
                (__v16qi) _mm_setzero_si128(),
                __M);
 }
 
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_cvtepi16_storeu_epi8 (void * __P, __mmask16 __M, __m256i __A)
+{
+  __builtin_ia32_pmovwb256mem_mask ((__v16qi *) __P, (__v16hi) __A, __M);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_cvtsepi16_storeu_epi8 (void * __P, __mmask16 __M, __m256i __A)
+{
+  __builtin_ia32_pmovswb256mem_mask ((__v16qi *) __P, (__v16hi) __A, __M);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm256_mask_cvtusepi16_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
+{
+  __builtin_ia32_pmovuswb256mem_mask ((__v16qi*) __P, (__v16hi) __A, __M);
+}
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_mask_mulhrs_epi16 (__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
   return (__m128i) __builtin_ia32_pmulhrsw128_mask ((__v8hi) __X,
                (__v8hi) __Y,
                (__v8hi) __W,
                (__mmask8) __U);
 }
 

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

 }
 
 __mmask32 test_mm512_movepi16_mask(__m512i __A) {
   // CHECK-LABEL: @test_mm512_movepi16_mask
   // CHECK: @llvm.x86.avx512.cvtw2mask.512
   return _mm512_movepi16_mask(__A); 
 }
 
+void test_mm512_mask_cvtepi16_storeu_epi8 (void * __P, __mmask32 __M, __m512i __A)
+{
+ // CHECK-LABEL: @test_mm512_mask_cvtepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmov.wb.mem.512
+ __builtin_ia32_pmovwb512mem_mask ( __P,  __A, __M);
+}
+
+void test_mm512_mask_cvtsepi16_storeu_epi8 (void * __P, __mmask32 __M, __m512i __A)
+{
+ // CHECK-LABEL: @test_mm512_mask_cvtsepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmovs.wb.mem.512
+ __builtin_ia32_pmovswb512mem_mask ( __P,  __A, __M);
+}
+
+void test_mm512_mask_cvtusepi16_storeu_epi8 (void * __P, __mmask32 __M, __m512i __A)
+{
+ // CHECK-LABEL: @test_mm512_mask_cvtusepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmovus.wb.mem.512
+ __builtin_ia32_pmovuswb512mem_mask ( __P, __A, __M);
+}

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

 }
 
 __m256i test_mm256_maskz_shufflelo_epi16(__mmask32 __U, __m256i __A) {
   // CHECK-LABEL: @test_mm256_maskz_shufflelo_epi16
   // CHECK: shufflevector <16 x i16> %{{.*}}, <16 x i16> %{{.*}}, <16 x i32> <i32 1, i32 1, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 9, i32 9, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15>
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_maskz_shufflelo_epi16(__U, __A, 5); 
 }
+
+void test_mm_mask_cvtepi16_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
+{
+ // CHECK-LABEL:@test_mm_mask_cvtepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmov.wb.mem.128
+ _mm_mask_cvtepi16_storeu_epi8 (__P, __M, __A);
+}
+
+void test_mm_mask_cvtsepi16_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
+{
+ // CHECK-LABEL:@test_mm_mask_cvtsepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmovs.wb.mem.128
+  _mm_mask_cvtsepi16_storeu_epi8 ( __P,  __M, __A);
+}
+
+void test_mm_mask_cvtusepi16_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
+{
+ // CHECK-LABEL:@test_mm_mask_cvtusepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmovus.wb.mem.128
+  _mm_mask_cvtusepi16_storeu_epi8 (__P, __M, __A);
+}
+
+void test_mm256_mask_cvtusepi16_storeu_epi8 (void * __P, __mmask16 __M, __m256i __A)
+{
+ // CHECK-LABEL:@test_mm256_mask_cvtusepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmovus.wb.mem.256
+  _mm256_mask_cvtusepi16_storeu_epi8 ( __P, __M, __A);
+}
+
+void test_mm256_mask_cvtepi16_storeu_epi8 (void * __P, __mmask16 __M, __m256i __A)
+{
+ // CHECK-LABEL:@test_mm256_mask_cvtepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmov.wb.mem.256
+  _mm256_mask_cvtepi16_storeu_epi8 ( __P,  __M, __A);
+}
+
+void test_mm256_mask_cvtsepi16_storeu_epi8 (void * __P, __mmask16 __M, __m256i __A)
+{
+ // CHECK-LABEL:@test_mm256_mask_cvtsepi16_storeu_epi8
+ // CHECK: @llvm.x86.avx512.mask.pmovs.wb.mem.256
+ _mm256_mask_cvtsepi16_storeu_epi8 ( __P, __M, __A);
+}