[X86][AVX512] lowering shuffle f/i intrinsic - clang part

[jina.nahias]

Summary

The llvm part: https://reviews.llvm.org/D38671

[m_zuckerman]

Please add dependent parent.
LGTM After the parent commit (LLVM side).

[m_zuckerman]

after LLVM-SIDE

[lsaba]

lib/Headers/avx512fintrin.h

7207

This should start at 16 in order to get to B

7213

this should be >>6?

test/CodeGen/avx512f-builtins.c

4491

expected to be
shufflevector <16 x float> %{{.*}}, <16 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 16, i32 17, i32 18, i32 19>
See comment on
#define _mm512_shuffle_f32x4(A, B, imm) extension ({

4496

Same comment as above

lib/Headers/avx512fintrin.h

 _mm512_maskz_srai_epi64(__mmask8 __U, __m512i __A, int __B)
 {
   return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
                                           (__v8di)_mm512_srai_epi64(__A, __B), \
                                           (__v8di)_mm512_setzero_si512());
 }
 
 #define _mm512_shuffle_f32x4(A, B, imm) __extension__ ({ \
-  (__m512)__builtin_ia32_shuf_f32x4_mask((__v16sf)(__m512)(A), \
-                                         (__v16sf)(__m512)(B), (int)(imm), \
-                                         (__v16sf)_mm512_undefined_ps(), \
-                                         (__mmask16)-1); })
+  (__m512)__builtin_shufflevector((__v8df)(__m512)(A), \
+                                  (__v8df)(__m512)(B), \
+                                  0 + ((((imm) >> 0) & 0x3) * 2), \
+                                  1 + ((((imm) >> 0) & 0x3) * 2), \
+                                  0 + ((((imm) >> 2) & 0x3) * 2), \
+                                  1 + ((((imm) >> 2) & 0x3) * 2), \
+                                  8 + ((((imm) >> 4) & 0x3) * 2), \
+                                  9 + ((((imm) >> 4) & 0x3) * 2), \
+                                  8 + ((((imm) >> 6) & 0x3) * 2), \
+                                  9 + ((((imm) >> 6) & 0x3) * 2)); })
 
 #define _mm512_mask_shuffle_f32x4(W, U, A, B, imm) __extension__ ({ \
-  (__m512)__builtin_ia32_shuf_f32x4_mask((__v16sf)(__m512)(A), \
-                                         (__v16sf)(__m512)(B), (int)(imm), \
-                                         (__v16sf)(__m512)(W), \
-                                         (__mmask16)(U)); })
+  (__m512)__builtin_ia32_selectps_512((__mmask16)(U), \
+                                      (__v16sf)_mm512_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v16sf)(__m512)(W)); })
 
 #define _mm512_maskz_shuffle_f32x4(U, A, B, imm) __extension__ ({ \
-  (__m512)__builtin_ia32_shuf_f32x4_mask((__v16sf)(__m512)(A), \
-                                         (__v16sf)(__m512)(B), (int)(imm), \
-                                         (__v16sf)_mm512_setzero_ps(), \
-                                         (__mmask16)(U)); })
+  (__m512)__builtin_ia32_selectps_512((__mmask16)(U), \
+                                      (__v16sf)_mm512_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v16sf)_mm512_setzero_ps()); })
 
 #define _mm512_shuffle_f64x2(A, B, imm) __extension__ ({ \
-  (__m512d)__builtin_ia32_shuf_f64x2_mask((__v8df)(__m512d)(A), \
-                                          (__v8df)(__m512d)(B), (int)(imm), \
-                                          (__v8df)_mm512_undefined_pd(), \
-                                          (__mmask8)-1); })
+  (__m512d)__builtin_shufflevector((__v8df)(__m512d)(A), \
+                                   (__v8df)(__m512d)(B), \
+                                   0 + ((((imm) >> 0) & 0x3) * 2), \
+                                   1 + ((((imm) >> 0) & 0x3) * 2), \
+                                   0 + ((((imm) >> 2) & 0x3) * 2), \
+                                   1 + ((((imm) >> 2) & 0x3) * 2), \
+                                   8 + ((((imm) >> 4) & 0x3) * 2), \
+                                   9 + ((((imm) >> 4) & 0x3) * 2), \
+                                   8 + ((((imm) >> 6) & 0x3) * 2), \
+                                   9 + ((((imm) >> 6) & 0x3) * 2)); })
 
 #define _mm512_mask_shuffle_f64x2(W, U, A, B, imm) __extension__ ({ \
-  (__m512d)__builtin_ia32_shuf_f64x2_mask((__v8df)(__m512d)(A), \
-                                          (__v8df)(__m512d)(B), (int)(imm), \
-                                          (__v8df)(__m512d)(W), \
-                                          (__mmask8)(U)); })
+  (__m512d)__builtin_ia32_selectpd_512((__mmask8)(U), \
+                                       (__v8df)_mm512_shuffle_f64x2((A), (B), (imm)), \
+                                       (__v8df)(__m512d)(W)); })
 
 #define _mm512_maskz_shuffle_f64x2(U, A, B, imm) __extension__ ({ \
-  (__m512d)__builtin_ia32_shuf_f64x2_mask((__v8df)(__m512d)(A), \
-                                          (__v8df)(__m512d)(B), (int)(imm), \
-                                          (__v8df)_mm512_setzero_pd(), \
-                                          (__mmask8)(U)); })
+  (__m512d)__builtin_ia32_selectpd_512((__mmask8)(U), \
+                                       (__v8df)_mm512_shuffle_f64x2((A), (B), (imm)), \
+                                       (__v8df)_mm512_setzero_pd()); })
 
 #define _mm512_shuffle_i32x4(A, B, imm) __extension__ ({ \
-  (__m512i)__builtin_ia32_shuf_i32x4_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), (int)(imm), \
-                                          (__v16si)_mm512_setzero_si512(), \
-                                          (__mmask16)-1); })
+  (__m512i)__builtin_shufflevector((__v8di)(__m512i)(A), \
+                                   (__v8di)(__m512i)(B), \
+                                   0 + ((((imm) >> 0) & 0x3) * 2), \
+                                   1 + ((((imm) >> 0) & 0x3) * 2), \
+                                   0 + ((((imm) >> 2) & 0x3) * 2), \
+                                   1 + ((((imm) >> 2) & 0x3) * 2), \
+                                   8 + ((((imm) >> 4) & 0x3) * 2), \
+                                   9 + ((((imm) >> 4) & 0x3) * 2), \
+                                   8 + ((((imm) >> 6) & 0x3) * 2), \
+                                   9 + ((((imm) >> 6) & 0x3) * 2)); })
 
 #define _mm512_mask_shuffle_i32x4(W, U, A, B, imm) __extension__ ({ \
-  (__m512i)__builtin_ia32_shuf_i32x4_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), (int)(imm), \
-                                          (__v16si)(__m512i)(W), \
-                                          (__mmask16)(U)); })
+  (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
+                                      (__v16si)_mm512_shuffle_i32x4((A), (B), (imm)), \
+                                      (__v16si)(__m512i)(W)); })
 
 #define _mm512_maskz_shuffle_i32x4(U, A, B, imm) __extension__ ({ \
-  (__m512i)__builtin_ia32_shuf_i32x4_mask((__v16si)(__m512i)(A), \
-                                          (__v16si)(__m512i)(B), (int)(imm), \
-                                          (__v16si)_mm512_setzero_si512(), \
-                                          (__mmask16)(U)); })
+  (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
+                                      (__v16si)_mm512_shuffle_i32x4((A), (B), (imm)), \
+                                      (__v16si)_mm512_setzero_si512()); })
 
 #define _mm512_shuffle_i64x2(A, B, imm) __extension__ ({ \
-  (__m512i)__builtin_ia32_shuf_i64x2_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), (int)(imm), \
-                                          (__v8di)_mm512_setzero_si512(), \
-                                          (__mmask8)-1); })
+  (__m512i)__builtin_shufflevector((__v8di)(__m512i)(A), \
+                                   (__v8di)(__m512i)(B), \
+                                   0 + ((((imm) >> 0) & 0x3) * 2), \
+                                   1 + ((((imm) >> 0) & 0x3) * 2), \
+                                   0 + ((((imm) >> 2) & 0x3) * 2), \
+                                   1 + ((((imm) >> 2) & 0x3) * 2), \
+                                   8 + ((((imm) >> 4) & 0x3) * 2), \
+                                   9 + ((((imm) >> 4) & 0x3) * 2), \
+                                   8 + ((((imm) >> 6) & 0x3) * 2), \
+                                   9 + ((((imm) >> 6) & 0x3) * 2)); })
 
 #define _mm512_mask_shuffle_i64x2(W, U, A, B, imm) __extension__ ({ \
-  (__m512i)__builtin_ia32_shuf_i64x2_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), (int)(imm), \
-                                          (__v8di)(__m512i)(W), \
-                                          (__mmask8)(U)); })
+  (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
+                                      (__v8di)_mm512_shuffle_i64x2((A), (B), (imm)), \
+                                      (__v8di)(__m512i)(W)); })
 
 #define _mm512_maskz_shuffle_i64x2(U, A, B, imm) __extension__ ({ \
-  (__m512i)__builtin_ia32_shuf_i64x2_mask((__v8di)(__m512i)(A), \
-                                          (__v8di)(__m512i)(B), (int)(imm), \
-                                          (__v8di)_mm512_setzero_si512(), \
-                                          (__mmask8)(U)); })
+  (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
+                                      (__v8di)_mm512_shuffle_i64x2((A), (B), (imm)), \
+                                      (__v8di)_mm512_setzero_si512()); })
 
 #define _mm512_shuffle_pd(A, B, M) __extension__ ({ \
   (__m512d)__builtin_shufflevector((__v8df)(__m512d)(A), \
                                    (__v8df)(__m512d)(B), \
                                    0  + (((M) >> 0) & 0x1), \
                                    8  + (((M) >> 1) & 0x1), \
                                    2  + (((M) >> 2) & 0x1), \
                                    10 + (((M) >> 3) & 0x1), \

lib/Headers/avx512vlintrin.h

#define _mm256_maskz_ternarylogic_epi64(U, A, B, C, imm) __extension__ ({ \
   (__m256i)__builtin_ia32_pternlogq256_maskz((__v4di)(__m256i)(A), \
                                              (__v4di)(__m256i)(B), \
                                              (__v4di)(__m256i)(C), (int)(imm), \
                                              (__mmask8)(U)); })
 
 
 
 #define _mm256_shuffle_f32x4(A, B, imm) __extension__ ({ \
-  (__m256)__builtin_ia32_shuf_f32x4_256_mask((__v8sf)(__m256)(A), \
-                                             (__v8sf)(__m256)(B), (int)(imm), \
-                                             (__v8sf)_mm256_setzero_ps(), \
-                                             (__mmask8)-1); })
+  (__m256)__builtin_shufflevector((__v4df)(__m256)(A), \
+                                  (__v4df)(__m256)(B), \
+                                  0 + ((((imm) >> 0) & 0x1) * 2), \
+                                  1 + ((((imm) >> 0) & 0x1) * 2), \
+                                  4 + ((((imm) >> 1) & 0x1) * 2), \
+                                  5 + ((((imm) >> 1) & 0x1) * 2)); })
 
 #define _mm256_mask_shuffle_f32x4(W, U, A, B, imm) __extension__ ({ \
-  (__m256)__builtin_ia32_shuf_f32x4_256_mask((__v8sf)(__m256)(A), \
-                                             (__v8sf)(__m256)(B), (int)(imm), \
-                                             (__v8sf)(__m256)(W), \
-                                             (__mmask8)(U)); })
+  (__m256)__builtin_ia32_selectps_256((__mmask8)(U), \
+                                      (__v8sf)_mm256_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v8sf)(__m256)(W)); })
 
 #define _mm256_maskz_shuffle_f32x4(U, A, B, imm) __extension__ ({ \
-  (__m256)__builtin_ia32_shuf_f32x4_256_mask((__v8sf)(__m256)(A), \
-                                             (__v8sf)(__m256)(B), (int)(imm), \
-                                             (__v8sf)_mm256_setzero_ps(), \
-                                             (__mmask8)(U)); })
+  (__m256)__builtin_ia32_selectps_256((__mmask8)(U), \
+                                      (__v8sf)_mm256_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v8sf)_mm256_setzero_ps()); })
 
 #define _mm256_shuffle_f64x2(A, B, imm) __extension__ ({ \
-  (__m256d)__builtin_ia32_shuf_f64x2_256_mask((__v4df)(__m256d)(A), \
-                                              (__v4df)(__m256d)(B), \
-                                              (int)(imm), \
-                                              (__v4df)_mm256_setzero_pd(), \
-                                              (__mmask8)-1); })
+  (__m256d)__builtin_shufflevector((__v4df)(__m256d)(A), \
+                                   0 + ((((imm) >> 0) & 0x1) * 2), \
+                                   1 + ((((imm) >> 0) & 0x1) * 2), \
+                                   4 + ((((imm) >> 1) & 0x1) * 2), \
+                                   5 + ((((imm) >> 1) & 0x1) * 2)); })
 
 #define _mm256_mask_shuffle_f64x2(W, U, A, B, imm) __extension__ ({ \
-  (__m256d)__builtin_ia32_shuf_f64x2_256_mask((__v4df)(__m256d)(A), \
-                                              (__v4df)(__m256d)(B), \
-                                              (int)(imm), \
-                                              (__v4df)(__m256d)(W), \
-                                              (__mmask8)(U)); })
+  (__m256d)__builtin_ia32_selectpd_256((__mmask8)(U), \
+                                      (__v4df)_mm256_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v4df)(__m256)(W)); })
 
 #define _mm256_maskz_shuffle_f64x2(U, A, B, imm) __extension__ ({ \
-  (__m256d)__builtin_ia32_shuf_f64x2_256_mask((__v4df)(__m256d)(A), \
-                                              (__v4df)(__m256d)(B), \
-                                              (int)(imm), \
-                                              (__v4df)_mm256_setzero_pd(), \
-                                              (__mmask8)(U)); })
+  (__m256d)__builtin_ia32_selectpd_256((__mmask8)(U), \
+                                      (__v4df)_mm256_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v4df)_mm256_setzero_pd()); })
 
 #define _mm256_shuffle_i32x4(A, B, imm) __extension__ ({ \
-  (__m256i)__builtin_ia32_shuf_i32x4_256_mask((__v8si)(__m256i)(A), \
-                                              (__v8si)(__m256i)(B), \
-                                              (int)(imm), \
-                                              (__v8si)_mm256_setzero_si256(), \
-                                              (__mmask8)-1); })
+  (__m256i)__builtin_shufflevector((__v4di)(__m256i)(A), \
+                                   (__v4di)(__m256i)(B), \
+                                   0 + ((((imm) >> 0) & 0x1) * 2), \
+                                   1 + ((((imm) >> 0) & 0x1) * 2), \
+                                   4 + ((((imm) >> 1) & 0x1) * 2), \
+                                   5 + ((((imm) >> 1) & 0x1) * 2)); })
 
 #define _mm256_mask_shuffle_i32x4(W, U, A, B, imm) __extension__ ({ \
-  (__m256i)__builtin_ia32_shuf_i32x4_256_mask((__v8si)(__m256i)(A), \
-                                              (__v8si)(__m256i)(B), \
-                                              (int)(imm), \
-                                              (__v8si)(__m256i)(W), \
-                                              (__mmask8)(U)); })
+  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
+                                      (__v8si)_mm256_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v8si)(__m256)(W)); })
 
 #define _mm256_maskz_shuffle_i32x4(U, A, B, imm) __extension__ ({ \
-  (__m256i)__builtin_ia32_shuf_i32x4_256_mask((__v8si)(__m256i)(A), \
-                                              (__v8si)(__m256i)(B), \
-                                              (int)(imm), \
-                                              (__v8si)_mm256_setzero_si256(), \
-                                              (__mmask8)(U)); })
+  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
+                                      (__v8si)_mm256_shuffle_f32x4((A), (B), (imm)), \
+                                      (__v8si)_mm256_setzero_si256()); })
 
 #define _mm256_shuffle_i64x2(A, B, imm) __extension__ ({ \
-  (__m256i)__builtin_ia32_shuf_i64x2_256_mask((__v4di)(__m256i)(A), \
-                                              (__v4di)(__m256i)(B), \
-                                              (int)(imm), \
-                                              (__v4di)_mm256_setzero_si256(), \
-                                              (__mmask8)-1); })
+  (__m256i)__builtin_shufflevector((__v4di)(__m256i)(A), \
+                                   0 + ((((imm) >> 0) & 0x1) * 2), \
+                                   1 + ((((imm) >> 0) & 0x1) * 2), \
+                                   4 + ((((imm) >> 1) & 0x1) * 2), \
+                                   5 + ((((imm) >> 1) & 0x1) * 2)); })
 
 #define _mm256_mask_shuffle_i64x2(W, U, A, B, imm) __extension__ ({ \
-  (__m256i)__builtin_ia32_shuf_i64x2_256_mask((__v4di)(__m256i)(A), \
-                                              (__v4di)(__m256i)(B), \
-                                              (int)(imm), \
-                                              (__v4di)(__m256i)(W), \
-                                              (__mmask8)(U)); })
+  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
+                                      (__v4di)_mm256_shuffle_f64x2((A), (B), (imm)), \
+                                      (__v4di)(__m256)(W)); })
+
 
 #define _mm256_maskz_shuffle_i64x2(U, A, B, imm) __extension__ ({ \
-  (__m256i)__builtin_ia32_shuf_i64x2_256_mask((__v4di)(__m256i)(A), \
-                                              (__v4di)(__m256i)(B), \
-                                              (int)(imm), \
-                                              (__v4di)_mm256_setzero_si256(), \
-                                              (__mmask8)(U)); })
+  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
+                                      (__v4di)_mm256_shuffle_f64x2((A), (B), (imm)), \
+                                      (__v4di)_mm256_setzero_si256()); })
 
 #define _mm_mask_shuffle_pd(W, U, A, B, M) __extension__ ({ \
   (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \
                                        (__v2df)_mm_shuffle_pd((A), (B), (M)), \
                                        (__v2df)(__m128d)(W)); })
 
 #define _mm_maskz_shuffle_pd(U, A, B, M) __extension__ ({ \
   (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \

test/CodeGen/avx512f-builtins.c

 __m512i test_mm512_maskz_ternarylogic_epi64(__mmask8 __U, __m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: @test_mm512_maskz_ternarylogic_epi64
   // CHECK: @llvm.x86.avx512.maskz.pternlog.q.512
   return _mm512_maskz_ternarylogic_epi64(__U, __A, __B, __C, 4); 
 }
 
 __m512 test_mm512_shuffle_f32x4(__m512 __A, __m512 __B) {
   // CHECK-LABEL: @test_mm512_shuffle_f32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.f32x4
+  // CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
   return _mm512_shuffle_f32x4(__A, __B, 4); 
 }
 
 __m512 test_mm512_mask_shuffle_f32x4(__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
   // CHECK-LABEL: @test_mm512_mask_shuffle_f32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.f32x4
+  // CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
   return _mm512_mask_shuffle_f32x4(__W, __U, __A, __B, 4); 
 }
 
 __m512 test_mm512_maskz_shuffle_f32x4(__mmask16 __U, __m512 __A, __m512 __B) {
   // CHECK-LABEL: @test_mm512_maskz_shuffle_f32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.f32x4
+  // CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
   return _mm512_maskz_shuffle_f32x4(__U, __A, __B, 4); 
 }
 
 __m512d test_mm512_shuffle_f64x2(__m512d __A, __m512d __B) {
   // CHECK-LABEL: @test_mm512_shuffle_f64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.f64x2
+  // CHECK: shufflevector <8 x double> %0, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
   return _mm512_shuffle_f64x2(__A, __B, 4); 
 }
 
 __m512d test_mm512_mask_shuffle_f64x2(__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
   // CHECK-LABEL: @test_mm512_mask_shuffle_f64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.f64x2
+  // CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
   return _mm512_mask_shuffle_f64x2(__W, __U, __A, __B, 4); 
 }
 
 __m512d test_mm512_maskz_shuffle_f64x2(__mmask8 __U, __m512d __A, __m512d __B) {
   // CHECK-LABEL: @test_mm512_maskz_shuffle_f64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.f64x2
+  // CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
   return _mm512_maskz_shuffle_f64x2(__U, __A, __B, 4); 
 }
 
 __m512i test_mm512_shuffle_i32x4(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_shuffle_i32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.i32x4
+  // CHECK: shufflevector <8 x i64> %0, <8 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
   return _mm512_shuffle_i32x4(__A, __B, 4); 
 }
 
 __m512i test_mm512_mask_shuffle_i32x4(__m512i __W, __mmask16 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_shuffle_i32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.i32x4
+  // CHECK: shufflevector <8 x i64> %{{.*}}, <8 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_shuffle_i32x4(__W, __U, __A, __B, 4); 
 }
 
 __m512i test_mm512_maskz_shuffle_i32x4(__mmask16 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_shuffle_i32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.i32x4
+  // CHECK: shufflevector <8 x i64> %{{.*}}, <8 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_shuffle_i32x4(__U, __A, __B, 4); 
 }
 
 __m512i test_mm512_shuffle_i64x2(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_shuffle_i64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.i64x2
+  // CHECK: shufflevector <8 x i64> %0, <8 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
   return _mm512_shuffle_i64x2(__A, __B, 4); 
 }
 
 __m512i test_mm512_mask_shuffle_i64x2(__m512i __W, __mmask8 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_shuffle_i64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.i64x2
+  // CHECK: shufflevector <8 x i64> %{{.*}}, <8 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
   return _mm512_mask_shuffle_i64x2(__W, __U, __A, __B, 4); 
 }
 
 __m512i test_mm512_maskz_shuffle_i64x2(__mmask8 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_shuffle_i64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.i64x2
+  // CHECK: shufflevector <8 x i64> %{{.*}}, <8 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 8, i32 9>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
   return _mm512_maskz_shuffle_i64x2(__U, __A, __B, 4); 
 }
 
 __m512d test_mm512_shuffle_pd(__m512d __M, __m512d __V) {
   // CHECK-LABEL: @test_mm512_shuffle_pd
   // CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 8, i32 3, i32 10, i32 4, i32 12, i32 6, i32 14>
   return _mm512_shuffle_pd(__M, __V, 4); 
 }

test/CodeGen/avx512vl-builtins.c

 
 __m256i test_mm256_maskz_ternarylogic_epi64(__mmask8 __U, __m256i __A, __m256i __B, __m256i __C) {
   // CHECK-LABEL: @test_mm256_maskz_ternarylogic_epi64
   // CHECK: @llvm.x86.avx512.maskz.pternlog.q.256
   return _mm256_maskz_ternarylogic_epi64(__U, __A, __B, __C, 4); 
 }
 __m256 test_mm256_shuffle_f32x4(__m256 __A, __m256 __B) {
   // CHECK-LABEL: @test_mm256_shuffle_f32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.f32x4
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
   return _mm256_shuffle_f32x4(__A, __B, 3); 
 }
 
 __m256 test_mm256_mask_shuffle_f32x4(__m256 __W, __mmask8 __U, __m256 __A, __m256 __B) {
   // CHECK-LABEL: @test_mm256_mask_shuffle_f32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.f32x4
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm256_mask_shuffle_f32x4(__W, __U, __A, __B, 3); 
 }
 
 __m256 test_mm256_maskz_shuffle_f32x4(__mmask8 __U, __m256 __A, __m256 __B) {
   // CHECK-LABEL: @test_mm256_maskz_shuffle_f32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.f32x4
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm256_maskz_shuffle_f32x4(__U, __A, __B, 3); 
 }
 
 __m256d test_mm256_shuffle_f64x2(__m256d __A, __m256d __B) {
   // CHECK-LABEL: @test_mm256_shuffle_f64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.f64x2
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
   return _mm256_shuffle_f64x2(__A, __B, 3); 
 }
 
 __m256d test_mm256_mask_shuffle_f64x2(__m256d __W, __mmask8 __U, __m256d __A, __m256d __B) {
   // CHECK-LABEL: @test_mm256_mask_shuffle_f64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.f64x2
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: shufflevector <8 x i1> %{{.*}}, <8 x i1> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
   return _mm256_mask_shuffle_f64x2(__W, __U, __A, __B, 3); 
 }
 
 __m256d test_mm256_maskz_shuffle_f64x2(__mmask8 __U, __m256d __A, __m256d __B) {
   // CHECK-LABEL: @test_mm256_maskz_shuffle_f64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.f64x2
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: shufflevector <8 x i1> %{{.*}}1, <8 x i1> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
   return _mm256_maskz_shuffle_f64x2(__U, __A, __B, 3); 
 }
 
 __m256i test_mm256_shuffle_i32x4(__m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_shuffle_i32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.i32x4
+  // CHECK: shufflevector <4 x i64> %{{.*}}, <4 x i64> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
   return _mm256_shuffle_i32x4(__A, __B, 3); 
 }
 
 __m256i test_mm256_mask_shuffle_i32x4(__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_shuffle_i32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.i32x4
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_shuffle_i32x4(__W, __U, __A, __B, 3); 
 }
 
 __m256i test_mm256_maskz_shuffle_i32x4(__mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_shuffle_i32x4
-  // CHECK: @llvm.x86.avx512.mask.shuf.i32x4
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_shuffle_i32x4(__U, __A, __B, 3); 
 }
 
 __m256i test_mm256_shuffle_i64x2(__m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_shuffle_i64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.i64x2
+  // CHECK: shufflevector <4 x i64> %{{.*}}, <4 x i64> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
   return _mm256_shuffle_i64x2(__A, __B, 3); 
 }
 
 __m256i test_mm256_mask_shuffle_i64x2(__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_shuffle_i64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.i64x2
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: shufflevector <8 x i1> %{{.*}}, <8 x i1> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
   return _mm256_mask_shuffle_i64x2(__W, __U, __A, __B, 3); 
 }
 
 __m256i test_mm256_maskz_shuffle_i64x2(__mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_shuffle_i64x2
-  // CHECK: @llvm.x86.avx512.mask.shuf.i64x2
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
+  // CHECK: shufflevector <8 x i1> %{{.*}}, <8 x i1> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
   return _mm256_maskz_shuffle_i64x2(__U, __A, __B, 3); 
 }
 
 __m128d test_mm_mask_shuffle_pd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) {
   // CHECK-LABEL: @test_mm_mask_shuffle_pd
   // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x i32> <i32 1, i32 3>
   // CHECK: select <2 x i1> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}
   return _mm_mask_shuffle_pd(__W, __U, __A, __B, 3);