[X86] Remove masking from avx512vbmi2 concat and shift by immediate builtins. Use select builtins instead.

llvm-svn: 334577

3 files changed, 182 insertions, 308 deletions

diff --git a/clang/lib/Headers/avx512vbmi2intrin.h b/clang/lib/Headers/avx512vbmi2intrin.h
index e0ada4c16f3..01a75a70bb6 100644
--- a/clang/lib/Headers/avx512vbmi2intrin.h
+++ b/clang/lib/Headers/avx512vbmi2intrin.h
@@ -142,131 +142,89 @@ _mm512_maskz_expandloadu_epi8(__mmask64 __U, void const *__P)
               __U);
 }
 
+#define _mm512_shldi_epi64(A, B, I) \
+  (__m512i)__builtin_ia32_vpshldq512((__v8di)(__m512i)(A), \
+                                     (__v8di)(__m512i)(B), (int)(I))
+
 #define _mm512_mask_shldi_epi64(S, U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshldq512_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v8di)(__m512i)(S), \
-                                          (__mmask8)(U))
+  (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
+                                    (__v8di)_mm512_shldi_epi64((A), (B), (I)), \
+                                    (__v8di)(__m512i)(S))
 
 #define _mm512_maskz_shldi_epi64(U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshldq512_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v8di)_mm512_setzero_si512(), \
-                                          (__mmask8)(U))
+  (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
+                                    (__v8di)_mm512_shldi_epi64((A), (B), (I)), \
+                                    (__v8di)_mm512_setzero_si512())
 
-#define _mm512_shldi_epi64(A, B, I) \
-  (__m512i)__builtin_ia32_vpshldq512_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v8di)_mm512_undefined_epi32(), \
-                                          (__mmask8)-1)
+#define _mm512_shldi_epi32(A, B, I) \
+  (__m512i)__builtin_ia32_vpshldd512((__v16si)(__m512i)(A), \
+                                     (__v16si)(__m512i)(B), (int)(I))
 
 #define _mm512_mask_shldi_epi32(S, U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshldd512_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v16si)(__m512i)(S), \
-                                          (__mmask16)(U))
+  (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
+                                   (__v16si)_mm512_shldi_epi32((A), (B), (I)), \
+                                   (__v16si)(__m512i)(S))
 
 #define _mm512_maskz_shldi_epi32(U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshldd512_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v16si)_mm512_setzero_si512(), \
-                                          (__mmask16)(U))
+  (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
+                                   (__v16si)_mm512_shldi_epi32((A), (B), (I)), \
+                                   (__v16si)_mm512_setzero_si512())
 
-#define _mm512_shldi_epi32(A, B, I) \
-  (__m512i)__builtin_ia32_vpshldd512_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v16si)_mm512_undefined_epi32(), \
-                                          (__mmask16)-1)
+#define _mm512_shldi_epi16(A, B, I) \
+  (__m512i)__builtin_ia32_vpshldw512((__v32hi)(__m512i)(A), \
+                                     (__v32hi)(__m512i)(B), (int)(I))
 
 #define _mm512_mask_shldi_epi16(S, U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshldw512_mask((__v32hi)(__m512i)(A), \
-                                          (__v32hi)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v32hi)(__m512i)(S), \
-                                          (__mmask32)(U))
+  (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
+                                   (__v32hi)_mm512_shldi_epi16((A), (B), (I)), \
+                                   (__v32hi)(__m512i)(S))
 
 #define _mm512_maskz_shldi_epi16(U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshldw512_mask((__v32hi)(__m512i)(A), \
-                                          (__v32hi)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v32hi)_mm512_setzero_si512(), \
-                                          (__mmask32)(U))
+  (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
+                                   (__v32hi)_mm512_shldi_epi16((A), (B), (I)), \
+                                   (__v32hi)_mm512_setzero_si512())
 
-#define _mm512_shldi_epi16(A, B, I) \
-  (__m512i)__builtin_ia32_vpshldw512_mask((__v32hi)(__m512i)(A), \
-                                          (__v32hi)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v32hi)_mm512_undefined_epi32(), \
-                                          (__mmask32)-1)
+#define _mm512_shrdi_epi64(A, B, I) \
+  (__m512i)__builtin_ia32_vpshrdq512((__v8di)(__m512i)(A), \
+                                     (__v8di)(__m512i)(B), (int)(I))
 
 #define _mm512_mask_shrdi_epi64(S, U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdq512_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v8di)(__m512i)(S), \
-                                          (__mmask8)(U))
+  (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
+                                    (__v8di)_mm512_shrdi_epi64((A), (B), (I)), \
+                                    (__v8di)(__m512i)(S))
 
 #define _mm512_maskz_shrdi_epi64(U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdq512_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v8di)_mm512_setzero_si512(), \
-                                          (__mmask8)(U))
+  (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
+                                    (__v8di)_mm512_shrdi_epi64((A), (B), (I)), \
+                                    (__v8di)_mm512_setzero_si512())
 
-#define _mm512_shrdi_epi64(A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdq512_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v8di)_mm512_undefined_epi32(), \
-                                          (__mmask8)-1)
+#define _mm512_shrdi_epi32(A, B, I) \
+  (__m512i)__builtin_ia32_vpshrdd512((__v16si)(__m512i)(A), \
+                                     (__v16si)(__m512i)(B), (int)(I))
 
 #define _mm512_mask_shrdi_epi32(S, U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdd512_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v16si)(__m512i)(S), \
-                                          (__mmask16)(U))
+  (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
+                                   (__v16si)_mm512_shrdi_epi32((A), (B), (I)), \
+                                   (__v16si)(__m512i)(S))
 
 #define _mm512_maskz_shrdi_epi32(U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdd512_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v16si)_mm512_setzero_si512(), \
-                                          (__mmask16)(U))
+  (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
+                                   (__v16si)_mm512_shrdi_epi32((A), (B), (I)), \
+                                   (__v16si)_mm512_setzero_si512())
 
-#define _mm512_shrdi_epi32(A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdd512_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v16si)_mm512_undefined_epi32(), \
-                                          (__mmask16)-1)
+#define _mm512_shrdi_epi16(A, B, I) \
+  (__m512i)__builtin_ia32_vpshrdw512((__v32hi)(__m512i)(A), \
+                                     (__v32hi)(__m512i)(B), (int)(I))
 
 #define _mm512_mask_shrdi_epi16(S, U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdw512_mask((__v32hi)(__m512i)(A), \
-                                          (__v32hi)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v32hi)(__m512i)(S), \
-                                          (__mmask32)(U))
+  (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
+                                   (__v32hi)_mm512_shrdi_epi16((A), (B), (I)), \
+                                   (__v32hi)(__m512i)(S))
 
 #define _mm512_maskz_shrdi_epi16(U, A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdw512_mask((__v32hi)(__m512i)(A), \
-                                          (__v32hi)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v32hi)_mm512_setzero_si512(), \
-                                          (__mmask32)(U))
-
-#define _mm512_shrdi_epi16(A, B, I) \
-  (__m512i)__builtin_ia32_vpshrdw512_mask((__v32hi)(__m512i)(A), \
-                                          (__v32hi)(__m512i)(B), \
-                                          (int)(I), \
-                                          (__v32hi)_mm512_undefined_epi32(), \
-                                          (__mmask32)-1)
+  (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
+                                   (__v32hi)_mm512_shrdi_epi16((A), (B), (I)), \
+                                   (__v32hi)_mm512_setzero_si512())
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_mask_shldv_epi64(__m512i __S, __mmask8 __U, __m512i __A, __m512i __B)
diff --git a/clang/lib/Headers/avx512vlvbmi2intrin.h b/clang/lib/Headers/avx512vlvbmi2intrin.h
index 6dde1e9643e..38ef9b0f849 100644
--- a/clang/lib/Headers/avx512vlvbmi2intrin.h
+++ b/clang/lib/Headers/avx512vlvbmi2intrin.h
@@ -251,257 +251,173 @@ _mm256_maskz_expandloadu_epi8(__mmask32 __U, void const *__P)
               __U);
 }
 
+#define _mm256_shldi_epi64(A, B, I) \
+  (__m256i)__builtin_ia32_vpshldq256((__v4di)(__m256i)(A), \
+                                     (__v4di)(__m256i)(B), (int)(I))
+
 #define _mm256_mask_shldi_epi64(S, U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshldq256_mask((__v4di)(__m256i)(A), \
-                                          (__v4di)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v4di)(__m256i)(S), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
+                                    (__v4di)_mm256_shldi_epi64((A), (B), (I)), \
+                                    (__v4di)(__m256i)(S))
 
 #define _mm256_maskz_shldi_epi64(U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshldq256_mask((__v4di)(__m256i)(A), \
-                                          (__v4di)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v4di)_mm256_setzero_si256(), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
+                                    (__v4di)_mm256_shldi_epi64((A), (B), (I)), \
+                                    (__v4di)_mm256_setzero_si256())
 
-#define _mm256_shldi_epi64(A, B, I) \
-  (__m256i)__builtin_ia32_vpshldq256_mask((__v4di)(__m256i)(A), \
-                                          (__v4di)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v4di)_mm256_undefined_si256(), \
-                                          (__mmask8)-1)
+#define _mm_shldi_epi64(A, B, I) \
+  (__m128i)__builtin_ia32_vpshldq128((__v2di)(__m128i)(A), \
+                                     (__v2di)(__m128i)(B), (int)(I))
 
 #define _mm_mask_shldi_epi64(S, U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshldq128_mask((__v2di)(__m128i)(A), \
-                                          (__v2di)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v2di)(__m128i)(S), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectq_128((__mmask8)(U), \
+                                      (__v2di)_mm_shldi_epi64((A), (B), (I)), \
+                                      (__v2di)(__m128i)(S))
 
 #define _mm_maskz_shldi_epi64(U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshldq128_mask((__v2di)(__m128i)(A), \
-                                          (__v2di)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v2di)_mm_setzero_si128(), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectq_128((__mmask8)(U), \
+                                      (__v2di)_mm_shldi_epi64((A), (B), (I)), \
+                                      (__v2di)_mm_setzero_si128())
 
-#define _mm_shldi_epi64(A, B, I) \
-  (__m128i)__builtin_ia32_vpshldq128_mask((__v2di)(__m128i)(A), \
-                                          (__v2di)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v2di)_mm_undefined_si128(), \
-                                          (__mmask8)-1)
+#define _mm256_shldi_epi32(A, B, I) \
+  (__m256i)__builtin_ia32_vpshldd256((__v8si)(__m256i)(A), \
+                                     (__v8si)(__m256i)(B), (int)(I))
 
 #define _mm256_mask_shldi_epi32(S, U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshldd256_mask((__v8si)(__m256i)(A), \
-                                          (__v8si)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v8si)(__m256i)(S), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
+                                    (__v8si)_mm256_shldi_epi32((A), (B), (I)), \
+                                    (__v8si)(__m256i)(S))
 
 #define _mm256_maskz_shldi_epi32(U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshldd256_mask((__v8si)(__m256i)(A), \
-                                          (__v8si)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v8si)_mm256_setzero_si256(), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
+                                    (__v8si)_mm256_shldi_epi32((A), (B), (I)), \
+                                    (__v8si)_mm256_setzero_si256())
 
-#define _mm256_shldi_epi32(A, B, I) \
-  (__m256i)__builtin_ia32_vpshldd256_mask((__v8si)(__m256i)(A), \
-                                          (__v8si)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v8si)_mm256_undefined_si256(), \
-                                          (__mmask8)-1)
+#define _mm_shldi_epi32(A, B, I) \
+  (__m128i)__builtin_ia32_vpshldd128((__v4si)(__m128i)(A), \
+                                     (__v4si)(__m128i)(B), (int)(I))
 
 #define _mm_mask_shldi_epi32(S, U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshldd128_mask((__v4si)(__m128i)(A), \
-                                          (__v4si)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v4si)(__m128i)(S), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
+                                      (__v4si)_mm_shldi_epi32((A), (B), (I)), \
+                                      (__v4si)(__m128i)(S))
 
 #define _mm_maskz_shldi_epi32(U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshldd128_mask((__v4si)(__m128i)(A), \
-                                          (__v4si)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v4si)_mm_setzero_si128(), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
+                                      (__v4si)_mm_shldi_epi32((A), (B), (I)), \
+                                      (__v4si)_mm_setzero_si128())
 
-#define _mm_shldi_epi32(A, B, I) \
-  (__m128i)__builtin_ia32_vpshldd128_mask((__v4si)(__m128i)(A), \
-                                          (__v4si)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v4si)_mm_undefined_si128(), \
-                                          (__mmask8)-1)
+#define _mm256_shldi_epi16(A, B, I) \
+  (__m256i)__builtin_ia32_vpshldw256((__v16hi)(__m256i)(A), \
+                                     (__v16hi)(__m256i)(B), (int)(I))
 
 #define _mm256_mask_shldi_epi16(S, U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshldw256_mask((__v16hi)(__m256i)(A), \
-                                          (__v16hi)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v16hi)(__m256i)(S), \
-                                          (__mmask16)(U))
+  (__m256i)__builtin_ia32_selectw_256((__mmask16)(U), \
+                                   (__v16hi)_mm256_shldi_epi16((A), (B), (I)), \
+                                   (__v16hi)(__m256i)(S))
 
 #define _mm256_maskz_shldi_epi16(U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshldw256_mask((__v16hi)(__m256i)(A), \
-                                          (__v16hi)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v16hi)_mm256_setzero_si256(), \
-                                          (__mmask16)(U))
+  (__m256i)__builtin_ia32_selectw_256((__mmask16)(U), \
+                                   (__v16hi)_mm256_shldi_epi16((A), (B), (I)), \
+                                   (__v16hi)_mm256_setzero_si256())
 
-#define _mm256_shldi_epi16(A, B, I) \
-  (__m256i)__builtin_ia32_vpshldw256_mask((__v16hi)(__m256i)(A), \
-                                          (__v16hi)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v16hi)_mm256_undefined_si256(), \
-                                          (__mmask16)-1)
+#define _mm_shldi_epi16(A, B, I) \
+  (__m128i)__builtin_ia32_vpshldw128((__v8hi)(__m128i)(A), \
+                                     (__v8hi)(__m128i)(B), (int)(I))
 
 #define _mm_mask_shldi_epi16(S, U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshldw128_mask((__v8hi)(__m128i)(A), \
-                                          (__v8hi)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v8hi)(__m128i)(S), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectw_128((__mmask8)(U), \
+                                      (__v8hi)_mm_shldi_epi16((A), (B), (I)), \
+                                      (__v8hi)(__m128i)(S))
 
 #define _mm_maskz_shldi_epi16(U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshldw128_mask((__v8hi)(__m128i)(A), \
-                                          (__v8hi)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v8hi)_mm_setzero_si128(), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectw_128((__mmask8)(U), \
+                                      (__v8hi)_mm_shldi_epi16((A), (B), (I)), \
+                                      (__v8hi)_mm_setzero_si128())
 
-#define _mm_shldi_epi16(A, B, I) \
-  (__m128i)__builtin_ia32_vpshldw128_mask((__v8hi)(__m128i)(A), \
-                                          (__v8hi)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v8hi)_mm_undefined_si128(), \
-                                          (__mmask8)-1)
+#define _mm256_shrdi_epi64(A, B, I) \
+  (__m256i)__builtin_ia32_vpshrdq256((__v4di)(__m256i)(A), \
+                                     (__v4di)(__m256i)(B), (int)(I))
 
 #define _mm256_mask_shrdi_epi64(S, U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdq256_mask((__v4di)(__m256i)(A), \
-                                          (__v4di)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v4di)(__m256i)(S), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
+                                    (__v4di)_mm256_shrdi_epi64((A), (B), (I)), \
+                                    (__v4di)(__m256i)(S))
 
 #define _mm256_maskz_shrdi_epi64(U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdq256_mask((__v4di)(__m256i)(A), \
-                                          (__v4di)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v4di)_mm256_setzero_si256(), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
+                                    (__v4di)_mm256_shrdi_epi64((A), (B), (I)), \
+                                    (__v4di)_mm256_setzero_si256())
 
-#define _mm256_shrdi_epi64(A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdq256_mask((__v4di)(__m256i)(A), \
-                                          (__v4di)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v4di)_mm256_undefined_si256(), \
-                                          (__mmask8)-1)
+#define _mm_shrdi_epi64(A, B, I) \
+  (__m128i)__builtin_ia32_vpshrdq128((__v2di)(__m128i)(A), \
+                                     (__v2di)(__m128i)(B), (int)(I))
 
 #define _mm_mask_shrdi_epi64(S, U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdq128_mask((__v2di)(__m128i)(A), \
-                                          (__v2di)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v2di)(__m128i)(S), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectq_128((__mmask8)(U), \
+                                      (__v2di)_mm_shrdi_epi64((A), (B), (I)), \
+                                      (__v2di)(__m128i)(S))
 
 #define _mm_maskz_shrdi_epi64(U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdq128_mask((__v2di)(__m128i)(A), \
-                                          (__v2di)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v2di)_mm_setzero_si128(), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectq_128((__mmask8)(U), \
+                                      (__v2di)_mm_shrdi_epi64((A), (B), (I)), \
+                                      (__v2di)_mm_setzero_si128())
 
-#define _mm_shrdi_epi64(A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdq128_mask((__v2di)(__m128i)(A), \
-                                          (__v2di)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v2di)_mm_undefined_si128(), \
-                                          (__mmask8)-1)
+#define _mm256_shrdi_epi32(A, B, I) \
+  (__m256i)__builtin_ia32_vpshrdd256((__v8si)(__m256i)(A), \
+                                     (__v8si)(__m256i)(B), (int)(I))
 
 #define _mm256_mask_shrdi_epi32(S, U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdd256_mask((__v8si)(__m256i)(A), \
-                                          (__v8si)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v8si)(__m256i)(S), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
+                                    (__v8si)_mm256_shrdi_epi32((A), (B), (I)), \
+                                    (__v8si)(__m256i)(S))
 
 #define _mm256_maskz_shrdi_epi32(U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdd256_mask((__v8si)(__m256i)(A), \
-                                          (__v8si)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v8si)_mm256_setzero_si256(), \
-                                          (__mmask8)(U))
+  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
+                                    (__v8si)_mm256_shrdi_epi32((A), (B), (I)), \
+                                    (__v8si)_mm256_setzero_si256())
 
-#define _mm256_shrdi_epi32(A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdd256_mask((__v8si)(__m256i)(A), \
-                                          (__v8si)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v8si)_mm256_undefined_si256(), \
-                                          (__mmask8)-1)
+#define _mm_shrdi_epi32(A, B, I) \
+  (__m128i)__builtin_ia32_vpshrdd128((__v4si)(__m128i)(A), \
+                                     (__v4si)(__m128i)(B), (int)(I))
 
 #define _mm_mask_shrdi_epi32(S, U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdd128_mask((__v4si)(__m128i)(A), \
-                                          (__v4si)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v4si)(__m128i)(S), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
+                                      (__v4si)_mm_shrdi_epi32((A), (B), (I)), \
+                                      (__v4si)(__m128i)(S))
 
 #define _mm_maskz_shrdi_epi32(U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdd128_mask((__v4si)(__m128i)(A), \
-                                          (__v4si)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v4si)_mm_setzero_si128(), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
+                                      (__v4si)_mm_shrdi_epi32((A), (B), (I)), \
+                                      (__v4si)_mm_setzero_si128())
 
-#define _mm_shrdi_epi32(A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdd128_mask((__v4si)(__m128i)(A), \
-                                          (__v4si)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v4si)_mm_undefined_si128(), \
-                                          (__mmask8)-1)
+#define _mm256_shrdi_epi16(A, B, I) \
+  (__m256i)__builtin_ia32_vpshrdw256((__v16hi)(__m256i)(A), \
+                                     (__v16hi)(__m256i)(B), (int)(I))
 
 #define _mm256_mask_shrdi_epi16(S, U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdw256_mask((__v16hi)(__m256i)(A), \
-                                          (__v16hi)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v16hi)(__m256i)(S), \
-                                          (__mmask16)(U))
+  (__m256i)__builtin_ia32_selectw_256((__mmask16)(U), \
+                                   (__v16hi)_mm256_shrdi_epi16((A), (B), (I)), \
+                                   (__v16hi)(__m256i)(S))
 
 #define _mm256_maskz_shrdi_epi16(U, A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdw256_mask((__v16hi)(__m256i)(A), \
-                                          (__v16hi)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v16hi)_mm256_setzero_si256(), \
-                                          (__mmask16)(U))
+  (__m256i)__builtin_ia32_selectw_256((__mmask16)(U), \
+                                   (__v16hi)_mm256_shrdi_epi16((A), (B), (I)), \
+                                   (__v16hi)_mm256_setzero_si256())
 
-#define _mm256_shrdi_epi16(A, B, I) \
-  (__m256i)__builtin_ia32_vpshrdw256_mask((__v16hi)(__m256i)(A), \
-                                          (__v16hi)(__m256i)(B), \
-                                          (int)(I), \
-                                          (__v16hi)_mm256_undefined_si256(), \
-                                          (__mmask16)-1)
+#define _mm_shrdi_epi16(A, B, I) \
+  (__m128i)__builtin_ia32_vpshrdw128((__v8hi)(__m128i)(A), \
+                                     (__v8hi)(__m128i)(B), (int)(I))
 
 #define _mm_mask_shrdi_epi16(S, U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdw128_mask((__v8hi)(__m128i)(A), \
-                                          (__v8hi)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v8hi)(__m128i)(S), \
-                                          (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectw_128((__mmask8)(U), \
+                                      (__v8hi)_mm_shrdi_epi16((A), (B), (I)), \
+                                      (__v8hi)(__m128i)(S))
 
 #define _mm_maskz_shrdi_epi16(U, A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdw128_mask((__v8hi)(__m128i)(A), \
-                                          (__v8hi)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v8hi)_mm_setzero_si128(), \
-                                          (__mmask8)(U))
-
-#define _mm_shrdi_epi16(A, B, I) \
-  (__m128i)__builtin_ia32_vpshrdw128_mask((__v8hi)(__m128i)(A), \
-                                          (__v8hi)(__m128i)(B), \
-                                          (int)(I), \
-                                          (__v8hi)_mm_undefined_si128(), \
-                                          (__mmask8)-1)
+  (__m128i)__builtin_ia32_selectw_128((__mmask8)(U), \
+                                      (__v8hi)_mm_shrdi_epi16((A), (B), (I)), \
+                                      (__v8hi)_mm_setzero_si128())
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_mask_shldv_epi64(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B)
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index d5945ef6f62..6a1cbdf6210 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -2841,24 +2841,24 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_dbpsadbw128:
   case X86::BI__builtin_ia32_dbpsadbw256:
   case X86::BI__builtin_ia32_dbpsadbw512:
-  case X86::BI__builtin_ia32_vpshldd128_mask:
-  case X86::BI__builtin_ia32_vpshldd256_mask:
-  case X86::BI__builtin_ia32_vpshldd512_mask:
-  case X86::BI__builtin_ia32_vpshldq128_mask:
-  case X86::BI__builtin_ia32_vpshldq256_mask:
-  case X86::BI__builtin_ia32_vpshldq512_mask:
-  case X86::BI__builtin_ia32_vpshldw128_mask:
-  case X86::BI__builtin_ia32_vpshldw256_mask:
-  case X86::BI__builtin_ia32_vpshldw512_mask:
-  case X86::BI__builtin_ia32_vpshrdd128_mask:
-  case X86::BI__builtin_ia32_vpshrdd256_mask:
-  case X86::BI__builtin_ia32_vpshrdd512_mask:
-  case X86::BI__builtin_ia32_vpshrdq128_mask:
-  case X86::BI__builtin_ia32_vpshrdq256_mask:
-  case X86::BI__builtin_ia32_vpshrdq512_mask:
-  case X86::BI__builtin_ia32_vpshrdw128_mask:
-  case X86::BI__builtin_ia32_vpshrdw256_mask:
-  case X86::BI__builtin_ia32_vpshrdw512_mask:
+  case X86::BI__builtin_ia32_vpshldd128:
+  case X86::BI__builtin_ia32_vpshldd256:
+  case X86::BI__builtin_ia32_vpshldd512:
+  case X86::BI__builtin_ia32_vpshldq128:
+  case X86::BI__builtin_ia32_vpshldq256:
+  case X86::BI__builtin_ia32_vpshldq512:
+  case X86::BI__builtin_ia32_vpshldw128:
+  case X86::BI__builtin_ia32_vpshldw256:
+  case X86::BI__builtin_ia32_vpshldw512:
+  case X86::BI__builtin_ia32_vpshrdd128:
+  case X86::BI__builtin_ia32_vpshrdd256:
+  case X86::BI__builtin_ia32_vpshrdd512:
+  case X86::BI__builtin_ia32_vpshrdq128:
+  case X86::BI__builtin_ia32_vpshrdq256:
+  case X86::BI__builtin_ia32_vpshrdq512:
+  case X86::BI__builtin_ia32_vpshrdw128:
+  case X86::BI__builtin_ia32_vpshrdw256:
+  case X86::BI__builtin_ia32_vpshrdw512:
     i = 2; l = 0; u = 255;
     break;
   case X86::BI__builtin_ia32_fixupimmpd512_mask: