This patch optimizes shuffle instruction - generates 2 instructions instead of 4.

[oota-llvm.git] / lib / Target / X86 / X86InstrInfo.cpp

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 846e8b934abb82c7999e27846b2343ad2f94f034..4f3d824b4a7638c6cbeeb5fd279e09eae6ff91a0 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -1110,6 +1110,36 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VPUNPCKLWDYrr,     X86::VPUNPCKLWDYrm,      TB_ALIGN_32 },
     { X86::VPXORYrr,          X86::VPXORYrm,           TB_ALIGN_32 },
     // FIXME: add AVX 256-bit foldable instructions
+
+    // FMA4 foldable patterns
+    { X86::VFMADDSS4rr,       X86::VFMADDSS4mr,        TB_ALIGN_16 },
+    { X86::VFMADDSD4rr,       X86::VFMADDSD4mr,        TB_ALIGN_16 },
+    { X86::VFMADDPS4rr,       X86::VFMADDPS4mr,        TB_ALIGN_16 },
+    { X86::VFMADDPD4rr,       X86::VFMADDPD4mr,        TB_ALIGN_16 },
+    { X86::VFMADDPS4rrY,      X86::VFMADDPS4mrY,       TB_ALIGN_32 },
+    { X86::VFMADDPD4rrY,      X86::VFMADDPD4mrY,       TB_ALIGN_32 },
+    { X86::VFNMADDPS4rr,      X86::VFNMADDPS4mr,       TB_ALIGN_16 },
+    { X86::VFNMADDPD4rr,      X86::VFNMADDPD4mr,       TB_ALIGN_16 },
+    { X86::VFNMADDPS4rrY,     X86::VFNMADDPS4mrY,      TB_ALIGN_32 },
+    { X86::VFNMADDPD4rrY,     X86::VFNMADDPD4mrY,      TB_ALIGN_32 },
+    { X86::VFMSUBSS4rr,       X86::VFMSUBSS4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBSD4rr,       X86::VFMSUBSD4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBPS4rr,       X86::VFMSUBPS4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBPD4rr,       X86::VFMSUBPD4mr,        TB_ALIGN_16 },
+    { X86::VFMSUBPS4rrY,      X86::VFMSUBPS4mrY,       TB_ALIGN_32 },
+    { X86::VFMSUBPD4rrY,      X86::VFMSUBPD4mrY,       TB_ALIGN_32 },
+    { X86::VFNMSUBPS4rr,      X86::VFNMSUBPS4mr,       TB_ALIGN_16 },
+    { X86::VFNMSUBPD4rr,      X86::VFNMSUBPD4mr,       TB_ALIGN_16 },
+    { X86::VFNMSUBPS4rrY,     X86::VFNMSUBPS4mrY,      TB_ALIGN_32 },
+    { X86::VFNMSUBPD4rrY,     X86::VFNMSUBPD4mrY,      TB_ALIGN_32 },
+    { X86::VFMADDSUBPS4rr,    X86::VFMADDSUBPS4mr,     TB_ALIGN_16 },
+    { X86::VFMADDSUBPD4rr,    X86::VFMADDSUBPD4mr,     TB_ALIGN_16 },
+    { X86::VFMADDSUBPS4rrY,   X86::VFMADDSUBPS4mrY,    TB_ALIGN_32 },
+    { X86::VFMADDSUBPD4rrY,   X86::VFMADDSUBPD4mrY,    TB_ALIGN_32 },
+    { X86::VFMSUBADDPS4rr,    X86::VFMSUBADDPS4mr,     TB_ALIGN_16 },
+    { X86::VFMSUBADDPD4rr,    X86::VFMSUBADDPD4mr,     TB_ALIGN_16 },
+    { X86::VFMSUBADDPS4rrY,   X86::VFMSUBADDPS4mrY,    TB_ALIGN_32 },
+    { X86::VFMSUBADDPD4rrY,   X86::VFMSUBADDPD4mrY,    TB_ALIGN_32 },
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl2); i != e; ++i) {
@@ -1237,6 +1267,36 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VFMSUBADDPDr132rY,     X86::VFMSUBADDPDr132mY,     TB_ALIGN_32 },
     { X86::VFMSUBADDPSr213rY,     X86::VFMSUBADDPSr213mY,     TB_ALIGN_32 },
     { X86::VFMSUBADDPDr213rY,     X86::VFMSUBADDPDr213mY,     TB_ALIGN_32 },
+
+    // FMA4 foldable patterns
+    { X86::VFMADDSS4rr,           X86::VFMADDSS4rm,           TB_ALIGN_16 },
+    { X86::VFMADDSD4rr,           X86::VFMADDSD4rm,           TB_ALIGN_16 },
+    { X86::VFMADDPS4rr,           X86::VFMADDPS4rm,           TB_ALIGN_16 },
+    { X86::VFMADDPD4rr,           X86::VFMADDPD4rm,           TB_ALIGN_16 },
+    { X86::VFMADDPS4rrY,          X86::VFMADDPS4rmY,          TB_ALIGN_32 },
+    { X86::VFMADDPD4rrY,          X86::VFMADDPD4rmY,          TB_ALIGN_32 },
+    { X86::VFNMADDPS4rr,          X86::VFNMADDPS4rm,          TB_ALIGN_16 },
+    { X86::VFNMADDPD4rr,          X86::VFNMADDPD4rm,          TB_ALIGN_16 },
+    { X86::VFNMADDPS4rrY,         X86::VFNMADDPS4rmY,         TB_ALIGN_32 },
+    { X86::VFNMADDPD4rrY,         X86::VFNMADDPD4rmY,         TB_ALIGN_32 },
+    { X86::VFMSUBSS4rr,           X86::VFMSUBSS4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBSD4rr,           X86::VFMSUBSD4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBPS4rr,           X86::VFMSUBPS4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBPD4rr,           X86::VFMSUBPD4rm,           TB_ALIGN_16 },
+    { X86::VFMSUBPS4rrY,          X86::VFMSUBPS4rmY,          TB_ALIGN_32 },
+    { X86::VFMSUBPD4rrY,          X86::VFMSUBPD4rmY,          TB_ALIGN_32 },
+    { X86::VFNMSUBPS4rr,          X86::VFNMSUBPS4rm,          TB_ALIGN_16 },
+    { X86::VFNMSUBPD4rr,          X86::VFNMSUBPD4rm,          TB_ALIGN_16 },
+    { X86::VFNMSUBPS4rrY,         X86::VFNMSUBPS4rmY,         TB_ALIGN_32 },
+    { X86::VFNMSUBPD4rrY,         X86::VFNMSUBPD4rmY,         TB_ALIGN_32 },
+    { X86::VFMADDSUBPS4rr,        X86::VFMADDSUBPS4rm,        TB_ALIGN_16 },
+    { X86::VFMADDSUBPD4rr,        X86::VFMADDSUBPD4rm,        TB_ALIGN_16 },
+    { X86::VFMADDSUBPS4rrY,       X86::VFMADDSUBPS4rmY,       TB_ALIGN_32 },
+    { X86::VFMADDSUBPD4rrY,       X86::VFMADDSUBPD4rmY,       TB_ALIGN_32 },
+    { X86::VFMSUBADDPS4rr,        X86::VFMSUBADDPS4rm,        TB_ALIGN_16 },
+    { X86::VFMSUBADDPD4rr,        X86::VFMSUBADDPD4rm,        TB_ALIGN_16 },
+    { X86::VFMSUBADDPS4rrY,       X86::VFMSUBADDPS4rmY,       TB_ALIGN_32 },
+    { X86::VFMSUBADDPD4rrY,       X86::VFMSUBADDPD4rmY,       TB_ALIGN_32 },
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl3); i != e; ++i) {
@@ -3432,6 +3492,10 @@ bool X86InstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
   case X86::AVX_SET0:
     assert(HasAVX && "AVX not supported");
     return Expand2AddrUndef(MI, get(X86::VXORPSYrr));
+  case X86::V_SETALLONES:
+    return Expand2AddrUndef(MI, get(HasAVX ? X86::VPCMPEQDrr : X86::PCMPEQDrr));
+  case X86::AVX2_SETALLONES:
+    return Expand2AddrUndef(MI, get(X86::VPCMPEQDYrr));
   case X86::TEST8ri_NOREX:
     MI->setDesc(get(X86::TEST8ri));
     return true;
@@ -3789,7 +3853,6 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
       break;
     case X86::V_SET0:
     case X86::V_SETALLONES:
-    case X86::AVX_SETALLONES:
       Alignment = 16;
       break;
     case X86::FsFLD0SD:
@@ -3825,7 +3888,6 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
   switch (LoadMI->getOpcode()) {
   case X86::V_SET0:
   case X86::V_SETALLONES:
-  case X86::AVX_SETALLONES:
   case X86::AVX2_SETALLONES:
   case X86::AVX_SET0:
   case X86::FsFLD0SD:
@@ -3864,8 +3926,7 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
     else
       Ty = VectorType::get(Type::getInt32Ty(MF.getFunction()->getContext()), 4);
 
-    bool IsAllOnes = (Opc == X86::V_SETALLONES || Opc == X86::AVX_SETALLONES ||
-                      Opc == X86::AVX2_SETALLONES);
+    bool IsAllOnes = (Opc == X86::V_SETALLONES || Opc == X86::AVX2_SETALLONES);
     const Constant *C = IsAllOnes ? Constant::getAllOnesValue(Ty) :
                                     Constant::getNullValue(Ty);
     unsigned CPI = MCP.getConstantPoolIndex(C, Alignment);
@@ -3940,6 +4001,8 @@ bool X86InstrInfo::canFoldMemoryOperand(const MachineInstr *MI,
     OpcodeTablePtr = &RegOp2MemOpTable1;
   } else if (OpNum == 2) {
     OpcodeTablePtr = &RegOp2MemOpTable2;
+  } else if (OpNum == 3) {
+    OpcodeTablePtr = &RegOp2MemOpTable3;
   }
 
   if (OpcodeTablePtr && OpcodeTablePtr->count(Opc))