Revert r146797, which was a partial revert of r146791; It was correct in the

first place.  The permutevar_* (note the *var*) intrinsics use ymm/mem.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@146807 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index eefefb5fe15192236a1c988f6e7fa3a26ed9c269..832b4163206bbc55570330f5f21057fc8771b126 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -425,9 +425,9 @@ 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_vpermilpd, "V2dV2dc", "")
+BUILTIN(__builtin_ia32_vpermilpd, "V2dV2dIc", "")
 BUILTIN(__builtin_ia32_vpermilps, "V4fV4fc", "")
-BUILTIN(__builtin_ia32_vpermilpd256, "V4dV4dc", "")
+BUILTIN(__builtin_ia32_vpermilpd256, "V4dV4dIc", "")
 BUILTIN(__builtin_ia32_vpermilps256, "V8fV8fc", "")
 BUILTIN(__builtin_ia32_vinsertf128_pd256, "V4dV4dV2dIc", "")
 BUILTIN(__builtin_ia32_vinsertf128_ps256, "V8fV8fV4fIc", "")

diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h
index fc606b37770f3b16ac6c7b66fc0ae0a6bc1bcbaa..64ab67b5859c8245a639dcc1d8a70c3c76e13435 100644 (file)
--- a/lib/Headers/avxintrin.h
+++ b/lib/Headers/avxintrin.h
@@ -258,17 +258,13 @@ _mm256_permutevar_ps(__m256 a, __m256i c)
                                                  (__v8si)c);
 }
 
-static __inline __m128d __attribute__((__always_inline__, __nodebug__))
-_mm_permute_pd(__m128d a, const int c)
-{
-  return (__m128d)__builtin_ia32_vpermilpd((__v2df)a, c);
-}
+#define _mm_permute_pd(A, C) __extension__ ({ \
+  __m128d __A = (A); \
+  (__m128d)__builtin_ia32_vpermilpd((__v2df)__A, C); })
 
-static __inline __m256d __attribute__((__always_inline__, __nodebug__))
-_mm256_permute_pd(__m256d a, const int c)
-{
-  return (__m256d)__builtin_ia32_vpermilpd256((__v4df)a, c);
-}
+#define _mm256_permute_pd(A, C) __extension__ ({ \
+  __m256d __A = (A); \
+  (__m256d)__builtin_ia32_vpermilpd256((__v4df)__A, C); })
 
 static __inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_permute_ps(__m128 a, const int c)