Fix vperm2f128_* AVX intrinsics to use "I" (ICE) markings. Fix avxintrin.h to

take them into account.
rdar://10576962

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@146757 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index cfe4e6f5d1f01ac7020c21a2d67b41d203a95810..2eb8b09b47def6a45d5aef8f760f1a13e1be72ec 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -422,9 +422,9 @@ 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, "V4dV4dV4dc", "")
-BUILTIN(__builtin_ia32_vperm2f128_ps256, "V8fV8fV8fc", "")
-BUILTIN(__builtin_ia32_vperm2f128_si256, "V8iV8iV8ic", "")
+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_vpermilps, "V4fV4fc", "")
 BUILTIN(__builtin_ia32_vpermilpd256, "V4dV4dc", "")

diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h
index 85a2c747c2f067fccd39e70f1bb94be0594f91da..3ae5dc76fcd26e3013d94526659a0435548742a0 100644 (file)
--- a/lib/Headers/avxintrin.h
+++ b/lib/Headers/avxintrin.h
@@ -286,23 +286,20 @@ _mm256_permute_ps(__m256 a, const int c)
   return (__m256)__builtin_ia32_vpermilps256((__v8sf)a, c);
 }
 
-static __inline __m256d __attribute__((__always_inline__, __nodebug__))
-_mm256_permute2f128_pd(__m256d a, __m256d b, const int c)
-{
-  return (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)a, (__v4df)b, c);
-}
+#define _mm256_permute2f128_pd(V1, V2, M) __extension__ ({ \
+  __m256d __V1 = (V1); \
+  __m256d __V2 = (V2); \
+  (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)__V1, (__v4df)__V2, M); })
 
-static __inline __m256 __attribute__((__always_inline__, __nodebug__))
-_mm256_permute2f128_ps(__m256 a, __m256 b, const int c)
-{
-  return (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)a, (__v8sf)b, c);
-}
+#define _mm256_permute2f128_ps(V1, V2, M) __extension__ ({ \
+  __m256 __V1 = (V1); \
+  __m256 __V2 = (V2); \
+  (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)__V1, (__v8sf)__V2, M); })
 
-static __inline __m256i __attribute__((__always_inline__, __nodebug__))
-_mm256_permute2f128_si256(__m256i a, __m256i b, const int c)
-{
-  return (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)a, (__v8si)b, c);
-}
+#define _mm256_permute2f128_si256(V1, V2, M) __extension__ ({ \
+  __m256i __V1 = (V1); \
+  __m256i __V2 = (V2); \
+  (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)__V1, (__v8si)__V2, M); })
 
 /* Vector Blend */
 #define _mm256_blend_pd(V1, V2, M) __extension__ ({ \