Remove 256-bit unpack built-ins and make the AVX intrinsic call llvm __builtin_shuffl...

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@110768 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 71f25cabc311daf57c951250f3fe1f517d9d6b3d..3e54135ab0959da27f023dfc8f107716788c394d 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -377,10 +377,6 @@ BUILTIN(__builtin_ia32_rsqrtps_nr256, "V8fV8f", "")
 BUILTIN(__builtin_ia32_rcpps256, "V8fV8f", "")
 BUILTIN(__builtin_ia32_roundpd256, "V4dV4di", "")
 BUILTIN(__builtin_ia32_roundps256, "V8fV8fi", "")
-BUILTIN(__builtin_ia32_unpckhpd256, "V4dV4dV4d", "")
-BUILTIN(__builtin_ia32_unpcklpd256, "V4dV4dV4d", "")
-BUILTIN(__builtin_ia32_unpckhps256, "V8fV8fV8f", "")
-BUILTIN(__builtin_ia32_unpcklps256, "V8fV8fV8f", "")
 BUILTIN(__builtin_ia32_si256_si, "V8iV4i", "")
 BUILTIN(__builtin_ia32_ps256_ps, "V8fV4f", "")
 BUILTIN(__builtin_ia32_pd256_pd, "V4dV2d", "")

diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h
index 7beb5ba1c1849c5868adedb3de13e2a77bed3969..6c1e32d2405cc3cf0ebedc6723ff546502d39175 100644 (file)
--- a/lib/Headers/avxintrin.h
+++ b/lib/Headers/avxintrin.h
@@ -598,25 +598,25 @@ _mm256_movedup_pd(__m256d a)
 static __inline __m256d __attribute__((__always_inline__, __nodebug__))
 _mm256_unpackhi_pd(__m256d a, __m256d b)
 {
-  return (__m256d)__builtin_ia32_unpckhpd256((__v4df)a, (__v4df)b);
+  return __builtin_shufflevector(a, b, 1, 5, 1+2, 5+2);
 }
 
 static __inline __m256d __attribute__((__always_inline__, __nodebug__))
 _mm256_unpacklo_pd(__m256d a, __m256d b)
 {
-  return (__m256d)__builtin_ia32_unpcklpd256((__v4df)a, (__v4df)b);
+  return __builtin_shufflevector(a, b, 0, 4, 0+2, 4+2);
 }
 
 static __inline __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_unpackhi_ps(__m256 a, __m256 b)
 {
-  return (__m256)__builtin_ia32_unpckhps256((__v8sf)a, (__v8sf)b);
+  return __builtin_shufflevector(a, b, 2, 10, 2+1, 10+1, 6, 14, 6+1, 14+1);
 }
 
 static __inline __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_unpacklo_ps(__m256 a, __m256 b)
 {
-  return (__m256)__builtin_ia32_unpcklps256((__v8sf)a, (__v8sf)b);
+  return __builtin_shufflevector(a, b, 0, 8, 0+1, 8+1, 4, 12, 4+1, 12+1);
 }
 
 /* Bit Test */

diff --git a/test/CodeGen/builtins-x86.c b/test/CodeGen/builtins-x86.c
index 0e0a253755b70b4eac3b15415aa0e39c6cd87259..b08fc5f9aceeb6c49e910310f82a55ba9f89ab29 100644 (file)
--- a/test/CodeGen/builtins-x86.c
+++ b/test/CodeGen/builtins-x86.c
@@ -440,10 +440,6 @@ void f0() {
   tmp_V8f = __builtin_ia32_rcpps256(tmp_V8f);
   tmp_V4d = __builtin_ia32_roundpd256(tmp_V4d, tmp_i);
   tmp_V8f = __builtin_ia32_roundps256(tmp_V8f, tmp_i);
-  tmp_V4d = __builtin_ia32_unpckhpd256(tmp_V4d, tmp_V4d);
-  tmp_V4d = __builtin_ia32_unpcklpd256(tmp_V4d, tmp_V4d);
-  tmp_V8f = __builtin_ia32_unpckhps256(tmp_V8f, tmp_V8f);
-  tmp_V8f = __builtin_ia32_unpcklps256(tmp_V8f, tmp_V8f);
   tmp_V8i = __builtin_ia32_si256_si(tmp_V4i);
   tmp_V8f = __builtin_ia32_ps256_ps(tmp_V4f);
   tmp_V4d = __builtin_ia32_pd256_pd(tmp_V2d);