[SystemZ] Add CodeGen support for scalar f64 ops in vector registers

[oota-llvm.git] / lib / Target / SystemZ / SystemZShortenInst.cpp

diff --git a/lib/Target/SystemZ/SystemZShortenInst.cpp b/lib/Target/SystemZ/SystemZShortenInst.cpp
index ec7a8c40d18a9f903cd43ce0df245c405e52c29d..d1a17c5500d65c70be0fa8859953b366f4e100f8 100644 (file)
--- a/lib/Target/SystemZ/SystemZShortenInst.cpp
+++ b/lib/Target/SystemZ/SystemZShortenInst.cpp
@@ -15,6 +15,7 @@
 
 #include "SystemZTargetMachine.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 
 using namespace llvm;
 
@@ -36,6 +37,10 @@ public:
 private:
   bool shortenIIF(MachineInstr &MI, unsigned *GPRMap, unsigned LiveOther,
                   unsigned LLIxL, unsigned LLIxH);
+  bool shortenOn0(MachineInstr &MI, unsigned Opcode);
+  bool shortenOn01(MachineInstr &MI, unsigned Opcode);
+  bool shortenOn001(MachineInstr &MI, unsigned Opcode);
+  bool shortenFPConv(MachineInstr &MI, unsigned Opcode);
 
   const SystemZInstrInfo *TII;
 
@@ -97,6 +102,64 @@ bool SystemZShortenInst::shortenIIF(MachineInstr &MI, unsigned *GPRMap,
   return false;
 }
 
+// Change MI's opcode to Opcode if register operand 0 has a 4-bit encoding.
+bool SystemZShortenInst::shortenOn0(MachineInstr &MI, unsigned Opcode) {
+  if (SystemZMC::getFirstReg(MI.getOperand(0).getReg()) < 16) {
+    MI.setDesc(TII->get(Opcode));
+    return true;
+  }
+  return false;
+}
+
+// Change MI's opcode to Opcode if register operands 0 and 1 have a
+// 4-bit encoding.
+bool SystemZShortenInst::shortenOn01(MachineInstr &MI, unsigned Opcode) {
+  if (SystemZMC::getFirstReg(MI.getOperand(0).getReg()) < 16 &&
+      SystemZMC::getFirstReg(MI.getOperand(1).getReg()) < 16) {
+    MI.setDesc(TII->get(Opcode));
+    return true;
+  }
+  return false;
+}
+
+// Change MI's opcode to Opcode if register operands 0, 1 and 2 have a
+// 4-bit encoding and if operands 0 and 1 are tied.
+bool SystemZShortenInst::shortenOn001(MachineInstr &MI, unsigned Opcode) {
+  if (SystemZMC::getFirstReg(MI.getOperand(0).getReg()) < 16 &&
+      MI.getOperand(1).getReg() == MI.getOperand(0).getReg() &&
+      SystemZMC::getFirstReg(MI.getOperand(2).getReg()) < 16) {
+    MI.setDesc(TII->get(Opcode));
+    return true;
+  }
+  return false;
+}
+
+// MI is a vector-style conversion instruction with the operand order:
+// destination, source, exact-suppress, rounding-mode.  If both registers
+// have a 4-bit encoding then change it to Opcode, which has operand order:
+// destination, rouding-mode, source, exact-suppress.
+bool SystemZShortenInst::shortenFPConv(MachineInstr &MI, unsigned Opcode) {
+  if (SystemZMC::getFirstReg(MI.getOperand(0).getReg()) < 16 &&
+      SystemZMC::getFirstReg(MI.getOperand(1).getReg()) < 16) {
+    MachineOperand Dest(MI.getOperand(0));
+    MachineOperand Src(MI.getOperand(1));
+    MachineOperand Suppress(MI.getOperand(2));
+    MachineOperand Mode(MI.getOperand(3));
+    MI.RemoveOperand(3);
+    MI.RemoveOperand(2);
+    MI.RemoveOperand(1);
+    MI.RemoveOperand(0);
+    MI.setDesc(TII->get(Opcode));
+    MachineInstrBuilder(*MI.getParent()->getParent(), &MI)
+      .addOperand(Dest)
+      .addOperand(Mode)
+      .addOperand(Src)
+      .addOperand(Suppress);
+    return true;
+  }
+  return false;
+}
+
 // Process all instructions in MBB.  Return true if something changed.
 bool SystemZShortenInst::processBlock(MachineBasicBlock &MBB) {
   bool Changed = false;
@@ -117,13 +180,83 @@ bool SystemZShortenInst::processBlock(MachineBasicBlock &MBB) {
   // Iterate backwards through the block looking for instructions to change.
   for (auto MBBI = MBB.rbegin(), MBBE = MBB.rend(); MBBI != MBBE; ++MBBI) {
     MachineInstr &MI = *MBBI;
-    unsigned Opcode = MI.getOpcode();
-    if (Opcode == SystemZ::IILF)
+    switch (MI.getOpcode()) {
+    case SystemZ::IILF:
       Changed |= shortenIIF(MI, LowGPRs, LiveHigh, SystemZ::LLILL,
                             SystemZ::LLILH);
-    else if (Opcode == SystemZ::IIHF)
+      break;
+
+    case SystemZ::IIHF:
       Changed |= shortenIIF(MI, HighGPRs, LiveLow, SystemZ::LLIHL,
                             SystemZ::LLIHH);
+      break;
+
+    case SystemZ::WFADB:
+      Changed |= shortenOn001(MI, SystemZ::ADBR);
+      break;
+
+    case SystemZ::WFDDB:
+      Changed |= shortenOn001(MI, SystemZ::DDBR);
+      break;
+
+    case SystemZ::WFIDB:
+      Changed |= shortenFPConv(MI, SystemZ::FIDBRA);
+      break;
+
+    case SystemZ::WLDEB:
+      Changed |= shortenOn01(MI, SystemZ::LDEBR);
+      break;
+
+    case SystemZ::WLEDB:
+      Changed |= shortenFPConv(MI, SystemZ::LEDBRA);
+      break;
+
+    case SystemZ::WFMDB:
+      Changed |= shortenOn001(MI, SystemZ::MDBR);
+      break;
+
+    case SystemZ::WFLCDB:
+      Changed |= shortenOn01(MI, SystemZ::LCDBR);
+      break;
+
+    case SystemZ::WFLNDB:
+      Changed |= shortenOn01(MI, SystemZ::LNDBR);
+      break;
+
+    case SystemZ::WFLPDB:
+      Changed |= shortenOn01(MI, SystemZ::LPDBR);
+      break;
+
+    case SystemZ::WFSQDB:
+      Changed |= shortenOn01(MI, SystemZ::SQDBR);
+      break;
+
+    case SystemZ::WFSDB:
+      Changed |= shortenOn001(MI, SystemZ::SDBR);
+      break;
+
+    case SystemZ::WFCDB:
+      Changed |= shortenOn01(MI, SystemZ::CDBR);
+      break;
+
+    case SystemZ::VL32:
+      // For z13 we prefer LDE over LE to avoid partial register dependencies.
+      Changed |= shortenOn0(MI, SystemZ::LDE32);
+      break;
+
+    case SystemZ::VST32:
+      Changed |= shortenOn0(MI, SystemZ::STE);
+      break;
+
+    case SystemZ::VL64:
+      Changed |= shortenOn0(MI, SystemZ::LD);
+      break;
+
+    case SystemZ::VST64:
+      Changed |= shortenOn0(MI, SystemZ::STD);
+      break;
+    }
+
     unsigned UsedLow = 0;
     unsigned UsedHigh = 0;
     for (auto MOI = MI.operands_begin(), MOE = MI.operands_end();