diff options
Diffstat (limited to 'llvm/lib/Target/X86/X86FastISel.cpp')
| -rw-r--r-- | llvm/lib/Target/X86/X86FastISel.cpp | 50 |
1 files changed, 34 insertions, 16 deletions
diff --git a/llvm/lib/Target/X86/X86FastISel.cpp b/llvm/lib/Target/X86/X86FastISel.cpp index 17704da91c2..0771cbeb4a2 100644 --- a/llvm/lib/Target/X86/X86FastISel.cpp +++ b/llvm/lib/Target/X86/X86FastISel.cpp @@ -433,6 +433,10 @@ bool X86FastISel::X86FastEmitLoad(EVT VT, X86AddressMode &AM, bool X86FastISel::X86FastEmitStore(EVT VT, unsigned ValReg, bool ValIsKill, X86AddressMode &AM, MachineMemOperand *MMO, bool Aligned) { + bool HasSSE2 = Subtarget->hasSSE2(); + bool HasAVX = Subtarget->hasAVX(); + bool IsNonTemporal = MMO && MMO->isNonTemporal(); + // Get opcode and regclass of the output for the given store instruction. unsigned Opc = 0; switch (VT.getSimpleVT().SimpleTy) { @@ -449,35 +453,49 @@ bool X86FastISel::X86FastEmitStore(EVT VT, unsigned ValReg, bool ValIsKill, // FALLTHROUGH, handling i1 as i8. case MVT::i8: Opc = X86::MOV8mr; break; case MVT::i16: Opc = X86::MOV16mr; break; - case MVT::i32: Opc = X86::MOV32mr; break; - case MVT::i64: Opc = X86::MOV64mr; break; // Must be in x86-64 mode. + case MVT::i32: + Opc = (IsNonTemporal && HasSSE2) ? X86::MOVNTImr : X86::MOV32mr; + break; + case MVT::i64: + // Must be in x86-64 mode. + Opc = (IsNonTemporal && HasSSE2) ? X86::MOVNTI_64mr : X86::MOV64mr; + break; case MVT::f32: - Opc = X86ScalarSSEf32 ? - (Subtarget->hasAVX() ? X86::VMOVSSmr : X86::MOVSSmr) : X86::ST_Fp32m; + Opc = X86ScalarSSEf32 ? + (HasAVX ? X86::VMOVSSmr : X86::MOVSSmr) : X86::ST_Fp32m; break; case MVT::f64: Opc = X86ScalarSSEf64 ? - (Subtarget->hasAVX() ? X86::VMOVSDmr : X86::MOVSDmr) : X86::ST_Fp64m; + (HasAVX ? X86::VMOVSDmr : X86::MOVSDmr) : X86::ST_Fp64m; break; case MVT::v4f32: - if (Aligned) - Opc = Subtarget->hasAVX() ? X86::VMOVAPSmr : X86::MOVAPSmr; - else - Opc = Subtarget->hasAVX() ? X86::VMOVUPSmr : X86::MOVUPSmr; + if (Aligned) { + if (IsNonTemporal) + Opc = HasAVX ? X86::VMOVNTPSmr : X86::MOVNTPSmr; + else + Opc = HasAVX ? X86::VMOVAPSmr : X86::MOVAPSmr; + } else + Opc = HasAVX ? X86::VMOVUPSmr : X86::MOVUPSmr; break; case MVT::v2f64: - if (Aligned) - Opc = Subtarget->hasAVX() ? X86::VMOVAPDmr : X86::MOVAPDmr; - else - Opc = Subtarget->hasAVX() ? X86::VMOVUPDmr : X86::MOVUPDmr; + if (Aligned) { + if (IsNonTemporal) + Opc = HasAVX ? X86::VMOVNTPDmr : X86::MOVNTPDmr; + else + Opc = HasAVX ? X86::VMOVAPDmr : X86::MOVAPDmr; + } else + Opc = HasAVX ? X86::VMOVUPDmr : X86::MOVUPDmr; break; case MVT::v4i32: case MVT::v2i64: case MVT::v8i16: case MVT::v16i8: - if (Aligned) - Opc = Subtarget->hasAVX() ? X86::VMOVDQAmr : X86::MOVDQAmr; - else + if (Aligned) { + if (IsNonTemporal) + Opc = HasAVX ? X86::VMOVNTDQmr : X86::MOVNTDQmr; + else + Opc = HasAVX ? X86::VMOVDQAmr : X86::MOVDQAmr; + } else Opc = Subtarget->hasAVX() ? X86::VMOVDQUmr : X86::MOVDQUmr; break; } |
