Convert vperm2f128 and vperm2i128 intrinsics back to using llvm intrinsics. Unfortunately, these instructions have behavior that can't be modeled with shuffle vector.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@154906 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 15 insertions, 32 deletions

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 9eddce2427..4aea980a9d 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -412,6 +412,9 @@ BUILTIN(__builtin_ia32_cvtps2pd256, "V4dV4f", "")
 BUILTIN(__builtin_ia32_cvttpd2dq256, "V4iV4d", "")
 BUILTIN(__builtin_ia32_cvtpd2dq256, "V4iV4d", "")
 BUILTIN(__builtin_ia32_cvttps2dq256, "V8iV8f", "")
+BUILTIN(__builtin_ia32_vperm2f128_pd256, "V4dV4dV4dIc", "")
+BUILTIN(__builtin_ia32_vperm2f128_ps256, "V8fV8fV8fIc", "")
+BUILTIN(__builtin_ia32_vperm2f128_si256, "V8iV8iV8iIc", "")
 BUILTIN(__builtin_ia32_vinsertf128_pd256, "V4dV4dV2dIc", "")
 BUILTIN(__builtin_ia32_vinsertf128_ps256, "V8fV8fV4fIc", "")
 BUILTIN(__builtin_ia32_vinsertf128_si256, "V8iV8iV4iIc", "")
@@ -561,6 +564,7 @@ BUILTIN(__builtin_ia32_pbroadcastd128, "V4iV4i", "")
 BUILTIN(__builtin_ia32_pbroadcastq128, "V2LLiV2LLi", "")
 BUILTIN(__builtin_ia32_permvarsi256, "V8iV8iV8i", "")
 BUILTIN(__builtin_ia32_permvarsf256, "V8fV8fV8f", "")
+BUILTIN(__builtin_ia32_permti256, "V4LLiV4LLiV4LLiIc", "")
 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 d6fef59a20..884c46d4d6 100644
--- a/lib/Headers/avx2intrin.h
+++ b/lib/Headers/avx2intrin.h
@@ -841,11 +841,7 @@ _mm256_permutevar8x32_ps(__m256 a, __m256 b)
 #define _mm256_permute2x128_si256(V1, V2, M) __extension__ ({ \
   __m256i __V1 = (V1); \
   __m256i __V2 = (V2); \
-  __builtin_shufflevector(__V1, __V2, \
-                          ((M) & 0x3) * 2, \
-                          ((M) & 0x3) * 2 + 1, \
-                          (((M) & 0x30) >> 4) * 2, \
-                          (((M) & 0x30) >> 4) * 2 + 1); })
+  (__m256i)__builtin_ia32_permti256(__V1, __V2, (M)); })
 
 #define _mm256_extracti128_si256(A, O) __extension__ ({ \
   __m256i __A = (A); \
diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h
index 7a0ec3fbd6..ee7f83572f 100644
--- a/lib/Headers/avxintrin.h
+++ b/lib/Headers/avxintrin.h
@@ -289,37 +289,17 @@ _mm256_permutevar_ps(__m256 a, __m256i c)
 #define _mm256_permute2f128_pd(V1, V2, M) __extension__ ({ \
   __m256d __V1 = (V1); \
   __m256d __V2 = (V2); \
-  (__m256d)__builtin_shufflevector((__v4df)__V1, (__v4df)__V2, \
-                                   ((M) & 0x3) * 2, \
-                                   ((M) & 0x3) * 2 + 1, \
-                                   (((M) & 0x30) >> 4) * 2, \
-                                   (((M) & 0x30) >> 4) * 2 + 1); })
+  (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)__V1, (__v4df)__V2, (M)); })
 
 #define _mm256_permute2f128_ps(V1, V2, M) __extension__ ({ \
   __m256 __V1 = (V1); \
   __m256 __V2 = (V2); \
-  (__m256)__builtin_shufflevector((__v8sf)__V1, (__v8sf)__V2, \
-                                  ((M) & 0x3) * 4, \
-                                  ((M) & 0x3) * 4 + 1, \
-                                  ((M) & 0x3) * 4 + 2, \
-                                  ((M) & 0x3) * 4 + 3, \
-                                  (((M) & 0x30) >> 4) * 4, \
-                                  (((M) & 0x30) >> 4) * 4 + 1, \
-                                  (((M) & 0x30) >> 4) * 4 + 2, \
-                                  (((M) & 0x30) >> 4) * 4 + 3); })
+  (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)__V1, (__v8sf)__V2, (M)); })
 
 #define _mm256_permute2f128_si256(V1, V2, M) __extension__ ({ \
   __m256i __V1 = (V1); \
   __m256i __V2 = (V2); \
-  (__m256i)__builtin_shufflevector((__v8si)__V1, (__v8si)__V2, \
-                                   ((M) & 0x3) * 4, \
-                                   ((M) & 0x3) * 4 + 1, \
-                                   ((M) & 0x3) * 4 + 2, \
-                                   ((M) & 0x3) * 4 + 3, \
-                                   (((M) & 0x30) >> 4) * 4, \
-                                   (((M) & 0x30) >> 4) * 4 + 1, \
-                                   (((M) & 0x30) >> 4) * 4 + 2, \
-                                   (((M) & 0x30) >> 4) * 4 + 3); })
+  (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)__V1, (__v8si)__V2, (M)); })
 
 /* Vector Blend */
 #define _mm256_blend_pd(V1, V2, M) __extension__ ({ \
diff --git a/test/CodeGen/avx-shuffle-builtins.c b/test/CodeGen/avx-shuffle-builtins.c
index 538ae50c93..d071f825aa 100644
--- a/test/CodeGen/avx-shuffle-builtins.c
+++ b/test/CodeGen/avx-shuffle-builtins.c
@@ -48,18 +48,18 @@ __m256 test_mm256_permute_ps(__m256 a) {
 
 __m256d test_mm256_permute2f128_pd(__m256d a, __m256d b) {
   // Check if the mask is correct
-  // CHECK: shufflevector{{.*}}<i32 2, i32 3, i32 6, i32 7>
+  // CHECK: @llvm.x86.avx.vperm2f128.pd.256
   return _mm256_permute2f128_pd(a, b, 0x31);
 }
 
 __m256 test_mm256_permute2f128_ps(__m256 a, __m256 b) {
   // Check if the mask is correct
-  // CHECK: shufflevector{{.*}}<i32 12, i32 13, i32 14, i32 15, i32 4, i32 5, i32 6, i32 7>
+  // CHECK: @llvm.x86.avx.vperm2f128.ps.256
   return _mm256_permute2f128_ps(a, b, 0x13);
 }
 
 __m256i test_mm256_permute2f128_si256(__m256i a, __m256i b) {
   // Check if the mask is correct
-  // CHECK: shufflevector{{.*}}<i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
+  // CHECK: @llvm.x86.avx.vperm2f128.si.256
   return _mm256_permute2f128_si256(a, b, 0x20);
 }
diff --git a/test/CodeGen/avx2-builtins.c b/test/CodeGen/avx2-builtins.c
index e56e55d205..7d166b5626 100644
--- a/test/CodeGen/avx2-builtins.c
+++ b/test/CodeGen/avx2-builtins.c
@@ -677,7 +677,7 @@ __m256i test_mm256_permute4x64_epi64(__m256i a) {
 }
 
 __m256i test_mm256_permute2x128_si256(__m256i a, __m256i b) {
-  // CHECK: shufflevector{{.*}}<i32 2, i32 3, i32 6, i32 7>
+  // CHECK: @llvm.x86.avx2.vperm2i128
   return _mm256_permute2x128_si256(a, b, 0x31);
 }
 
diff --git a/test/CodeGen/builtins-x86.c b/test/CodeGen/builtins-x86.c
index 30138d6374..acb5554db4 100644
--- a/test/CodeGen/builtins-x86.c
+++ b/test/CodeGen/builtins-x86.c
@@ -414,6 +414,9 @@ void f0() {
   tmp_V4i = __builtin_ia32_cvttpd2dq256(tmp_V4d);
   tmp_V4i = __builtin_ia32_cvtpd2dq256(tmp_V4d);
   tmp_V8i = __builtin_ia32_cvttps2dq256(tmp_V8f);
+  tmp_V4d = __builtin_ia32_vperm2f128_pd256(tmp_V4d, tmp_V4d, 0x7);
+  tmp_V8f = __builtin_ia32_vperm2f128_ps256(tmp_V8f, tmp_V8f, 0x7);
+  tmp_V8i = __builtin_ia32_vperm2f128_si256(tmp_V8i, tmp_V8i, 0x7);
   tmp_V4d = __builtin_ia32_vinsertf128_pd256(tmp_V4d, tmp_V2d, 0x7);
   tmp_V8f = __builtin_ia32_vinsertf128_ps256(tmp_V8f, tmp_V4f, 0x7);
   tmp_V8i = __builtin_ia32_vinsertf128_si256(tmp_V8i, tmp_V4i, 0x7);