Change _mm256_permute4x64_epi64 and _mm256_permute4x64_pd to use builtin_shufflevecto...

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@154777 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index f44aed6f9548e9cd057ce988a4533dc8a585fd06..9eddce2427d6eb278abe425dcb41d1f395c0d3bb 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -560,9 +560,7 @@ BUILTIN(__builtin_ia32_pbroadcastw128, "V8sV8s", "")
 BUILTIN(__builtin_ia32_pbroadcastd128, "V4iV4i", "")
 BUILTIN(__builtin_ia32_pbroadcastq128, "V2LLiV2LLi", "")
 BUILTIN(__builtin_ia32_permvarsi256, "V8iV8iV8i", "")
-BUILTIN(__builtin_ia32_permdf256, "V4dV4dIc", "")
 BUILTIN(__builtin_ia32_permvarsf256, "V8fV8fV8f", "")
-BUILTIN(__builtin_ia32_permdi256, "V4LLiV4LLiIc", "")
 BUILTIN(__builtin_ia32_extract128i256, "V2LLiV4LLiIc", "")
 BUILTIN(__builtin_ia32_insert128i256, "V4LLiV4LLiV2LLiIc", "")
 BUILTIN(__builtin_ia32_maskloadd256, "V8iV8iC*V8i", "")

diff --git a/lib/Headers/avx2intrin.h b/lib/Headers/avx2intrin.h
index d165f1fbe2dd10d7946857aae725dace272a448f..d6fef59a20a5a336759b2265a1e9b81c71b7acd6 100644 (file)
--- a/lib/Headers/avx2intrin.h
+++ b/lib/Headers/avx2intrin.h
@@ -822,7 +822,9 @@ _mm256_permutevar8x32_epi32(__m256i a, __m256i b)
 
 #define _mm256_permute4x64_pd(V, M) __extension__ ({ \
   __m256d __V = (V); \
-  (__m256d)__builtin_ia32_permdf256((__v4df)__V, (M)); })
+  (__m256d)__builtin_shufflevector((__v4df)__V, (__v4df) _mm256_setzero_pd(), \
+                                   (M) & 0x3, ((M) & 0xc) >> 2, \
+                                   ((M) & 0x30) >> 4, ((M) & 0xc0) >> 6); })
 
 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_permutevar8x32_ps(__m256 a, __m256 b)
@@ -832,7 +834,9 @@ _mm256_permutevar8x32_ps(__m256 a, __m256 b)
 
 #define _mm256_permute4x64_epi64(V, M) __extension__ ({ \
   __m256i __V = (V); \
-  (__m256i)__builtin_ia32_permdi256(__V, (M)); })
+  (__m256i)__builtin_shufflevector((__v4di)__V, (__v4di) _mm256_setzero_si256(), \
+                                   (M) & 0x3, ((M) & 0xc) >> 2, \
+                                   ((M) & 0x30) >> 4, ((M) & 0xc0) >> 6); })
 
 #define _mm256_permute2x128_si256(V1, V2, M) __extension__ ({ \
   __m256i __V1 = (V1); \

diff --git a/test/CodeGen/avx2-builtins.c b/test/CodeGen/avx2-builtins.c
index a5ddc8eb2ed82e1740677cc64f5092c9c7bcb1c7..e56e55d205ae15649586989b1263306de332ed5f 100644 (file)
--- a/test/CodeGen/avx2-builtins.c
+++ b/test/CodeGen/avx2-builtins.c
@@ -662,7 +662,7 @@ __m256i test_mm256_permutevar8x32_epi32(__m256i a, __m256i b) {
 }
 
 __m256d test_mm256_permute4x64_pd(__m256d a) {
-  // CHECK: @llvm.x86.avx2.permpd
+  // CHECK: shufflevector{{.*}}<i32 1, i32 2, i32 1, i32 0>
   return _mm256_permute4x64_pd(a, 25);
 }
 
@@ -672,7 +672,7 @@ __m256 test_mm256_permutevar8x32_ps(__m256 a, __m256 b) {
 }
 
 __m256i test_mm256_permute4x64_epi64(__m256i a) {
-  // CHECK:  @llvm.x86.avx2.permq
+  // CHECK: shufflevector{{.*}}<i32 3, i32 0, i32 2, i32 0>
   return _mm256_permute4x64_epi64(a, 35);
 }