diff options
| author | Juergen Ributzka <juergen@apple.com> | 2015-03-03 17:22:53 +0000 |
|---|---|---|
| committer | Juergen Ributzka <juergen@apple.com> | 2015-03-03 17:22:53 +0000 |
| commit | 9baa03fc0793afd4d4cc6da553c13aa0b75ee91b (patch) | |
| tree | ad7c8a3cd00bc04ec7626abd96fc6b2e628853e1 | |
| parent | 773fa2cb0361d5dc27a7e6b13d65d8416be92d37 (diff) | |
| download | bcm5719-llvm-9baa03fc0793afd4d4cc6da553c13aa0b75ee91b.tar.gz bcm5719-llvm-9baa03fc0793afd4d4cc6da553c13aa0b75ee91b.zip | |
Lower _mm256_broadcastsi128_si256 directly to a vector shuffle.
Originally we were using the same GCC builtins to lower this AVX2 vector
intrinsic. Instead we will now lower it directly to a vector shuffle.
This will not only allow LLVM to generate better code, but it will also allow us
to remove the GCC intrinsics.
Reviewed by Andrea
This is related to rdar://problem/18742778.
llvm-svn: 231081
| -rw-r--r-- | clang/include/clang/Basic/BuiltinsX86.def | 1 | ||||
| -rw-r--r-- | clang/lib/CodeGen/CGBuiltin.cpp | 7 | ||||
| -rw-r--r-- | clang/lib/Headers/avx2intrin.h | 2 | ||||
| -rw-r--r-- | clang/test/CodeGen/avx2-builtins.c | 2 |
4 files changed, 2 insertions, 10 deletions
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def index ce8dced1bc0..acf2eb6acbb 100644 --- a/clang/include/clang/Basic/BuiltinsX86.def +++ b/clang/include/clang/Basic/BuiltinsX86.def @@ -582,7 +582,6 @@ BUILTIN(__builtin_ia32_movntdqa256, "V4LLiV4LLi*", "") BUILTIN(__builtin_ia32_vbroadcastss_ps, "V4fV4f", "") BUILTIN(__builtin_ia32_vbroadcastss_ps256, "V8fV4f", "") BUILTIN(__builtin_ia32_vbroadcastsd_pd256, "V4dV2d", "") -BUILTIN(__builtin_ia32_vbroadcastsi256, "V4LLiV2LLi", "") BUILTIN(__builtin_ia32_pbroadcastb256, "V32cV16c", "") BUILTIN(__builtin_ia32_pbroadcastw256, "V16sV8s", "") BUILTIN(__builtin_ia32_pbroadcastd256, "V8iV4i", "") diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp index bf7d86fb12f..d3d1e22b4bb 100644 --- a/clang/lib/CodeGen/CGBuiltin.cpp +++ b/clang/lib/CodeGen/CGBuiltin.cpp @@ -6090,13 +6090,6 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID, Builder.CreateStore(Builder.CreateExtractValue(Call, 0), Ops[0]); return Builder.CreateExtractValue(Call, 1); } - // AVX2 broadcast - case X86::BI__builtin_ia32_vbroadcastsi256: { - Value *VecTmp = CreateMemTemp(E->getArg(0)->getType()); - Builder.CreateStore(Ops[0], VecTmp); - Value *F = CGM.getIntrinsic(Intrinsic::x86_avx2_vbroadcasti128); - return Builder.CreateCall(F, Builder.CreateBitCast(VecTmp, Int8PtrTy)); - } // SSE comparison intrisics case X86::BI__builtin_ia32_cmpeqps: case X86::BI__builtin_ia32_cmpltps: diff --git a/clang/lib/Headers/avx2intrin.h b/clang/lib/Headers/avx2intrin.h index 394fdfee965..af43bec0efd 100644 --- a/clang/lib/Headers/avx2intrin.h +++ b/clang/lib/Headers/avx2intrin.h @@ -771,7 +771,7 @@ _mm256_broadcastsd_pd(__m128d __X) static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) _mm256_broadcastsi128_si256(__m128i __X) { - return (__m256i)__builtin_ia32_vbroadcastsi256(__X); + return (__m256i)__builtin_shufflevector(__X, __X, 0, 1, 0, 1); } #define _mm_blend_epi32(V1, V2, M) __extension__ ({ \ diff --git a/clang/test/CodeGen/avx2-builtins.c b/clang/test/CodeGen/avx2-builtins.c index 27ee91e4ada..371f9c6ee24 100644 --- a/clang/test/CodeGen/avx2-builtins.c +++ b/clang/test/CodeGen/avx2-builtins.c @@ -612,7 +612,7 @@ __m256d test_mm256_broadcastsd_pd(__m128d a) { } __m256i test_mm256_broadcastsi128_si256(__m128i a) { - // CHECK: @llvm.x86.avx2.vbroadcasti128 + // CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 0, i32 1> return _mm256_broadcastsi128_si256(a); } |
