Teach more places to use VMOVAPS,VMOVUPS instead of MOVAPS,MOVUPS,

whenever AVX is enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@138849 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 7fa0dba9a01b676b895f51879c857ba9f9fea409..9f4931e221ad860bc2436e53d0e1707c85113d31 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -11215,7 +11215,9 @@ X86TargetLowering::EmitPCMP(MachineInstr *MI, MachineBasicBlock *BB,
     if (!(Op.isReg() && Op.isImplicit()))
       MIB.addOperand(Op);
   }
-  BuildMI(*BB, MI, dl, TII->get(X86::MOVAPSrr), MI->getOperand(0).getReg())
+  BuildMI(*BB, MI, dl,
+    TII->get(Subtarget->hasAVX() ? X86::VMOVAPSrr : X86::MOVAPSrr),
+             MI->getOperand(0).getReg())
     .addReg(X86::XMM0);
 
   MI->eraseFromParent();
@@ -11570,6 +11572,7 @@ X86TargetLowering::EmitVAStartSaveXMMRegsWithCustomInserter(
     MBB->addSuccessor(EndMBB);
   }
 
+  unsigned MOVOpc = Subtarget->hasAVX() ? X86::VMOVAPSmr : X86::MOVAPSmr;
   // In the XMM save block, save all the XMM argument registers.
   for (int i = 3, e = MI->getNumOperands(); i != e; ++i) {
     int64_t Offset = (i - 3) * 16 + VarArgsFPOffset;
@@ -11578,7 +11581,7 @@ X86TargetLowering::EmitVAStartSaveXMMRegsWithCustomInserter(
           MachinePointerInfo::getFixedStack(RegSaveFrameIndex, Offset),
         MachineMemOperand::MOStore,
         /*Size=*/16, /*Align=*/16);
-    BuildMI(XMMSaveMBB, DL, TII->get(X86::MOVAPSmr))
+    BuildMI(XMMSaveMBB, DL, TII->get(MOVOpc))
       .addFrameIndex(RegSaveFrameIndex)
       .addImm(/*Scale=*/1)
       .addReg(/*IndexReg=*/0)

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 674d84f01ffc228faee07cabef76fb51d501a30d..09afb4cb6e49f8b0317d724c9a49119158cf0836 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -1959,7 +1959,8 @@ void X86InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     else
       Opc = X86::MOV8rr;
   } else if (X86::VR128RegClass.contains(DestReg, SrcReg))
-    Opc = X86::MOVAPSrr;
+    Opc = TM.getSubtarget<X86Subtarget>().hasAVX() ?
+          X86::VMOVAPSrr : X86::MOVAPSrr;
   else if (X86::VR256RegClass.contains(DestReg, SrcReg))
     Opc = X86::VMOVAPSYrr;
   else if (X86::VR64RegClass.contains(DestReg, SrcReg))
@@ -2044,13 +2045,19 @@ static unsigned getLoadStoreRegOpcode(unsigned Reg,
   case 10:
     assert(X86::RFP80RegClass.hasSubClassEq(RC) && "Unknown 10-byte regclass");
     return load ? X86::LD_Fp80m : X86::ST_FpP80m;
-  case 16:
+  case 16: {
     assert(X86::VR128RegClass.hasSubClassEq(RC) && "Unknown 16-byte regclass");
+    bool HasAVX = TM.getSubtarget<X86Subtarget>().hasAVX();
     // If stack is realigned we can use aligned stores.
     if (isStackAligned)
-      return load ? X86::MOVAPSrm : X86::MOVAPSmr;
+      return load ?
+        (HasAVX ? X86::VMOVAPSrm : X86::MOVAPSrm) :
+        (HasAVX ? X86::VMOVAPSmr : X86::MOVAPSmr);
     else
-      return load ? X86::MOVUPSrm : X86::MOVUPSmr;
+      return load ?
+        (HasAVX ? X86::VMOVUPSrm : X86::MOVUPSrm) :
+        (HasAVX ? X86::VMOVUPSmr : X86::MOVUPSmr);
+  }
   case 32:
     assert(X86::VR256RegClass.hasSubClassEq(RC) && "Unknown 32-byte regclass");
     // If stack is realigned we can use aligned stores.