Add support for AVX to materialize +0.0 when doing scalar FP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@121415 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 06c35b4..73654d3 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -2420,9 +2420,11 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
       Alignment = 16;
       break;
     case X86::FsFLD0SD:
+    case X86::VFsFLD0SD:
       Alignment = 8;
       break;
     case X86::FsFLD0SS:
+    case X86::VFsFLD0SS:
       Alignment = 4;
       break;
     default:
@@ -2486,9 +2488,9 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
     MachineConstantPool &MCP = *MF.getConstantPool();
     const Type *Ty;
     unsigned Opc = LoadMI->getOpcode();
-    if (Opc == X86::FsFLD0SS)
+    if (Opc == X86::FsFLD0SS || Opc == X86::VFsFLD0SS)
       Ty = Type::getFloatTy(MF.getFunction()->getContext());
-    else if (Opc == X86::FsFLD0SD)
+    else if (Opc == X86::FsFLD0SD || Opc == X86::VFsFLD0SD)
       Ty = Type::getDoubleTy(MF.getFunction()->getContext());
     else if (Opc == X86::AVX_SET0PSY || Opc == X86::AVX_SET0PDY)
       Ty = VectorType::get(Type::getFloatTy(MF.getFunction()->getContext()), 8);

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 73a6202..b954299 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -1440,6 +1440,12 @@ def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
                  [(set FR64:$dst, fpimm0)]>,
                Requires<[HasSSE2]>, TB, OpSize;
+def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
+                  [(set FR32:$dst, fp32imm0)]>,
+                  Requires<[HasAVX]>, TB, OpSize, VEX_4V;
+def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
+                  [(set FR64:$dst, fpimm0)]>,
+                  Requires<[HasAVX]>, TB, OpSize, VEX_4V;
 }
 
 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper

diff --git a/lib/Target/X86/X86MCInstLower.cpp b/lib/Target/X86/X86MCInstLower.cpp
index 3ca87ed..cbe6db2 100644 (file)
--- a/lib/Target/X86/X86MCInstLower.cpp
+++ b/lib/Target/X86/X86MCInstLower.cpp
@@ -374,6 +374,8 @@ ReSimplify:
   case X86::MOV32r0:      LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); break;
   case X86::FsFLD0SS:      LowerUnaryToTwoAddr(OutMI, X86::PXORrr); break;
   case X86::FsFLD0SD:      LowerUnaryToTwoAddr(OutMI, X86::PXORrr); break;
+  case X86::VFsFLD0SS:     LowerUnaryToTwoAddr(OutMI, X86::VPXORrr); break;
+  case X86::VFsFLD0SD:     LowerUnaryToTwoAddr(OutMI, X86::VPXORrr); break;
   case X86::V_SET0PS:      LowerUnaryToTwoAddr(OutMI, X86::XORPSrr); break;
   case X86::V_SET0PD:      LowerUnaryToTwoAddr(OutMI, X86::XORPDrr); break;
   case X86::V_SET0PI:      LowerUnaryToTwoAddr(OutMI, X86::PXORrr); break;