[AVX-512] Remove masked 128/256-bit builtins for vpmaddwd and vpmaddubsw. Replace with unmasked builtins and select.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@285516 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 33 insertions, 42 deletions

diff --git a/lib/Headers/avx512vlbwintrin.h b/lib/Headers/avx512vlbwintrin.h
index 427dbeae6b..ed92efa9ba 100644
--- a/lib/Headers/avx512vlbwintrin.h
+++ b/lib/Headers/avx512vlbwintrin.h
@@ -1664,69 +1664,60 @@ _mm256_maskz_permutex2var_epi16 (__mmask16 __U, __m256i __A,
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_maddubs_epi16 (__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
-  return (__m128i) __builtin_ia32_pmaddubsw128_mask ((__v16qi) __X,
-               (__v16qi) __Y,
-               (__v8hi) __W,
-               (__mmask8) __U);
+_mm_mask_maddubs_epi16(__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+                                            (__v8hi)_mm_maddubs_epi16(__X, __Y),
+                                            (__v8hi)__W);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_maddubs_epi16 (__mmask8 __U, __m128i __X, __m128i __Y) {
-  return (__m128i) __builtin_ia32_pmaddubsw128_mask ((__v16qi) __X,
-               (__v16qi) __Y,
-              (__v8hi) _mm_setzero_si128(),
-               (__mmask8) __U);
+_mm_maskz_maddubs_epi16(__mmask8 __U, __m128i __X, __m128i __Y) {
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+                                            (__v8hi)_mm_maddubs_epi16(__X, __Y),
+                                            (__v8hi)_mm_setzero_si128());
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_maddubs_epi16 (__m256i __W, __mmask16 __U, __m256i __X,
-         __m256i __Y) {
-  return (__m256i) __builtin_ia32_pmaddubsw256_mask ((__v32qi) __X,
-               (__v32qi) __Y,
-               (__v16hi) __W,
-               (__mmask16) __U);
+_mm256_mask_maddubs_epi16(__m256i __W, __mmask16 __U, __m256i __X,
+                          __m256i __Y) {
+  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+                                        (__v16hi)_mm256_maddubs_epi16(__X, __Y),
+                                        (__v16hi)__W);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_maddubs_epi16 (__mmask16 __U, __m256i __X, __m256i __Y) {
-  return (__m256i) __builtin_ia32_pmaddubsw256_mask ((__v32qi) __X,
-               (__v32qi) __Y,
-               (__v16hi) _mm256_setzero_si256(),
-               (__mmask16) __U);
+_mm256_maskz_maddubs_epi16(__mmask16 __U, __m256i __X, __m256i __Y) {
+  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+                                        (__v16hi)_mm256_maddubs_epi16(__X, __Y),
+                                        (__v16hi)_mm256_setzero_si256());
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_madd_epi16 (__m128i __W, __mmask8 __U, __m128i __A,
-         __m128i __B) {
-  return (__m128i) __builtin_ia32_pmaddwd128_mask ((__v8hi) __A,
-               (__v8hi) __B,
-               (__v4si) __W,
-               (__mmask8) __U);
+_mm_mask_madd_epi16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U,
+                                             (__v4si)_mm_madd_epi16(__A, __B),
+                                             (__v4si)__W);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_madd_epi16 (__mmask8 __U, __m128i __A, __m128i __B) {
-  return (__m128i) __builtin_ia32_pmaddwd128_mask ((__v8hi) __A,
-               (__v8hi) __B,
-               (__v4si) _mm_setzero_si128(),
-               (__mmask8) __U);
+_mm_maskz_madd_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U,
+                                             (__v4si)_mm_madd_epi16(__A, __B),
+                                             (__v4si)_mm_setzero_si128());
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_madd_epi16 (__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
-  return (__m256i) __builtin_ia32_pmaddwd256_mask ((__v16hi) __A,
-               (__v16hi) __B,
-               (__v8si) __W,
-               (__mmask8) __U);
+_mm256_mask_madd_epi16(__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U,
+                                            (__v8si)_mm256_madd_epi16(__A, __B),
+                                            (__v8si)__W);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_madd_epi16 (__mmask8 __U, __m256i __A, __m256i __B) {
-  return (__m256i) __builtin_ia32_pmaddwd256_mask ((__v16hi) __A,
-               (__v16hi) __B,
-               (__v8si) _mm256_setzero_si256(),
-               (__mmask8) __U);
+_mm256_maskz_madd_epi16(__mmask8 __U, __m256i __A, __m256i __B) {
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U,
+                                            (__v8si)_mm256_madd_epi16(__A, __B),
+                                            (__v8si)_mm256_setzero_si256());
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS