diff options
Diffstat (limited to 'clang/lib')
| -rw-r--r-- | clang/lib/CodeGen/CGBuiltin.cpp | 22 | ||||
| -rw-r--r-- | clang/lib/Headers/xmmintrin.h | 13 |
2 files changed, 5 insertions, 30 deletions
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp index 8a565985f53..3c7c5e53987 100644 --- a/clang/lib/CodeGen/CGBuiltin.cpp +++ b/clang/lib/CodeGen/CGBuiltin.cpp @@ -764,28 +764,6 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID, llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_sse2_cmp_sd); return Builder.CreateCall(F, &Ops[0], &Ops[0] + Ops.size(), "cmpsd"); } - case X86::BI__builtin_ia32_loadlps: - case X86::BI__builtin_ia32_loadhps: { - // FIXME: This should probably be represented as - // shuffle (dst, (v4f32 (insert undef, (load i64), 0)), shuf mask hi/lo) - const llvm::Type *EltTy = llvm::Type::DoubleTy; - const llvm::Type *VecTy = llvm::VectorType::get(EltTy, 2); - const llvm::Type *OrigTy = Ops[0]->getType(); - unsigned Index = BuiltinID == X86::BI__builtin_ia32_loadlps ? 0 : 1; - llvm::Value *Idx = llvm::ConstantInt::get(llvm::Type::Int32Ty, Index); - Ops[1] = Builder.CreateBitCast(Ops[1], llvm::PointerType::getUnqual(EltTy)); - Ops[1] = Builder.CreateLoad(Ops[1], "tmp"); - Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast"); - Ops[0] = Builder.CreateInsertElement(Ops[0], Ops[1], Idx, "loadps"); - return Builder.CreateBitCast(Ops[0], OrigTy, "loadps"); - } - case X86::BI__builtin_ia32_loadlpd: - case X86::BI__builtin_ia32_loadhpd: { - Ops[1] = Builder.CreateLoad(Ops[1], "tmp"); - unsigned Index = BuiltinID == X86::BI__builtin_ia32_loadlpd ? 0 : 1; - llvm::Value *Idx = llvm::ConstantInt::get(llvm::Type::Int32Ty, Index); - return Builder.CreateInsertElement(Ops[0], Ops[1], Idx, "loadpd"); - } case X86::BI__builtin_ia32_storehps: case X86::BI__builtin_ia32_storelps: { const llvm::Type *EltTy = llvm::Type::Int64Ty; diff --git a/clang/lib/Headers/xmmintrin.h b/clang/lib/Headers/xmmintrin.h index 7291f88979e..2903049170c 100644 --- a/clang/lib/Headers/xmmintrin.h +++ b/clang/lib/Headers/xmmintrin.h @@ -464,20 +464,19 @@ _mm_cvtss_f32(__m128 a) static inline __m128 __attribute__((__always_inline__, __nodebug__)) _mm_loadh_pi(__m128 a, __m64 const *p) { - return __builtin_ia32_loadhps(a, (__v2si *)p); + __m128 b; + b[0] = *(float*)p; + b[1] = *((float*)p+1); + return __builtin_shufflevector(a, b, 0, 1, 4, 5); } static inline __m128 __attribute__((__always_inline__, __nodebug__)) _mm_loadl_pi(__m128 a, __m64 const *p) { -#if 0 - // FIXME: This should work, but gives really crappy code at the moment __m128 b; b[0] = *(float*)p; b[1] = *((float*)p+1); - return __builtin_shufflevector(a, b, 0, 1, 4, 5); -#endif - return __builtin_ia32_loadlps(a, (__v2si *)p); + return __builtin_shufflevector(a, b, 4, 5, 2, 3); } static inline __m128 __attribute__((__always_inline__, __nodebug__)) @@ -899,8 +898,6 @@ do { \ (row3) = _mm_movelh_ps(tmp3, tmp1); \ } while (0) -#include <emmintrin.h> - #endif /* __SSE__ */ #endif /* __XMMINTRIN_H */ |
