AMDGPU: Set mem operands for spill instructions

[oota-llvm.git] / lib / Target / AMDGPU / SIInstrInfo.cpp

diff --git a/lib/Target/AMDGPU/SIInstrInfo.cpp b/lib/Target/AMDGPU/SIInstrInfo.cpp
index d6f70a6da46c01bcc9ed595b7002d3831baea25f..52d0fc3ac2431ee26f3725885747bd57f7ff1d2d 100644 (file)
--- a/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -508,14 +508,23 @@ void SIInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
   }
 
   if (Opcode != -1) {
+    MachinePointerInfo PtrInfo
+      = MachinePointerInfo::getFixedStack(*MF, FrameIndex);
+    unsigned Size = FrameInfo->getObjectSize(FrameIndex);
+    unsigned Align = FrameInfo->getObjectAlignment(FrameIndex);
+    MachineMemOperand *MMO
+      = MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
+                                 Size, Align);
+
     FrameInfo->setObjectAlignment(FrameIndex, 4);
     BuildMI(MBB, MI, DL, get(Opcode))
-            .addReg(SrcReg)
-            .addFrameIndex(FrameIndex)
-            // Place-holder registers, these will be filled in by
-            // SIPrepareScratchRegs.
-            .addReg(AMDGPU::SGPR0_SGPR1_SGPR2_SGPR3, RegState::Undef)
-            .addReg(AMDGPU::SGPR0, RegState::Undef);
+      .addReg(SrcReg)
+      .addFrameIndex(FrameIndex)
+      // Place-holder registers, these will be filled in by
+      // SIPrepareScratchRegs.
+      .addReg(AMDGPU::SGPR0_SGPR1_SGPR2_SGPR3, RegState::Undef)
+      .addReg(AMDGPU::SGPR0, RegState::Undef)
+      .addMemOperand(MMO);
   } else {
     LLVMContext &Ctx = MF->getFunction()->getContext();
     Ctx.emitError("SIInstrInfo::storeRegToStackSlot - Do not know how to"
@@ -556,14 +565,22 @@ void SIInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
   }
 
   if (Opcode != -1) {
-    FrameInfo->setObjectAlignment(FrameIndex, 4);
-    BuildMI(MBB, MI, DL, get(Opcode), DestReg)
-            .addFrameIndex(FrameIndex)
-            // Place-holder registers, these will be filled in by
-            // SIPrepareScratchRegs.
-            .addReg(AMDGPU::SGPR0_SGPR1_SGPR2_SGPR3, RegState::Undef)
-            .addReg(AMDGPU::SGPR0, RegState::Undef);
+    unsigned Align = 4;
+    FrameInfo->setObjectAlignment(FrameIndex, Align);
+    unsigned Size = FrameInfo->getObjectSize(FrameIndex);
 
+    MachinePointerInfo PtrInfo
+      = MachinePointerInfo::getFixedStack(*MF, FrameIndex);
+    MachineMemOperand *MMO = MF->getMachineMemOperand(
+      PtrInfo, MachineMemOperand::MOLoad, Size, Align);
+
+    BuildMI(MBB, MI, DL, get(Opcode), DestReg)
+      .addFrameIndex(FrameIndex)
+      // Place-holder registers, these will be filled in by
+      // SIPrepareScratchRegs.
+      .addReg(AMDGPU::SGPR0_SGPR1_SGPR2_SGPR3, RegState::Undef)
+      .addReg(AMDGPU::SGPR0, RegState::Undef)
+      .addMemOperand(MMO);
   } else {
     LLVMContext &Ctx = MF->getFunction()->getContext();
     Ctx.emitError("SIInstrInfo::loadRegFromStackSlot - Do not know how to"
@@ -1588,36 +1605,6 @@ MachineOperand SIInstrInfo::buildExtractSubRegOrImm(
   return MachineOperand::CreateReg(SubReg, false);
 }
 
-unsigned SIInstrInfo::split64BitImm(SmallVectorImpl<MachineInstr *> &Worklist,
-                                    MachineBasicBlock::iterator MI,
-                                    MachineRegisterInfo &MRI,
-                                    const TargetRegisterClass *RC,
-                                    const MachineOperand &Op) const {
-  MachineBasicBlock *MBB = MI->getParent();
-  DebugLoc DL = MI->getDebugLoc();
-  unsigned LoDst = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
-  unsigned HiDst = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
-  unsigned Dst = MRI.createVirtualRegister(RC);
-
-  MachineInstr *Lo = BuildMI(*MBB, MI, DL, get(AMDGPU::S_MOV_B32),
-                             LoDst)
-    .addImm(Op.getImm() & 0xFFFFFFFF);
-  MachineInstr *Hi = BuildMI(*MBB, MI, DL, get(AMDGPU::S_MOV_B32),
-                             HiDst)
-    .addImm(Op.getImm() >> 32);
-
-  BuildMI(*MBB, MI, DL, get(TargetOpcode::REG_SEQUENCE), Dst)
-    .addReg(LoDst)
-    .addImm(AMDGPU::sub0)
-    .addReg(HiDst)
-    .addImm(AMDGPU::sub1);
-
-  Worklist.push_back(Lo);
-  Worklist.push_back(Hi);
-
-  return Dst;
-}
-
 // Change the order of operands from (0, 1, 2) to (0, 2, 1)
 void SIInstrInfo::swapOperands(MachineBasicBlock::iterator Inst) const {
   assert(Inst->getNumExplicitOperands() == 3);
@@ -1918,17 +1905,18 @@ void SIInstrInfo::legalizeOperands(MachineInstr *MI) const {
       // Create the new instruction.
       unsigned Addr64Opcode = AMDGPU::getAddr64Inst(MI->getOpcode());
       MachineInstr *Addr64 =
-          BuildMI(MBB, MI, MI->getDebugLoc(), get(Addr64Opcode))
-                  .addOperand(*VData)
-                  .addReg(AMDGPU::NoRegister) // Dummy value for vaddr.
-                                              // This will be replaced later
-                                              // with the new value of vaddr.
-                  .addOperand(*SRsrc)
-                  .addOperand(*SOffset)
-                  .addOperand(*Offset)
-                  .addImm(0) // glc
-                  .addImm(0) // slc
-                  .addImm(0); // tfe
+        BuildMI(MBB, MI, MI->getDebugLoc(), get(Addr64Opcode))
+        .addOperand(*VData)
+        .addReg(AMDGPU::NoRegister) // Dummy value for vaddr.
+                                    // This will be replaced later
+                                    // with the new value of vaddr.
+        .addOperand(*SRsrc)
+        .addOperand(*SOffset)
+        .addOperand(*Offset)
+        .addImm(0) // glc
+        .addImm(0) // slc
+        .addImm(0) // tfe
+        .setMemRefs(MI->memoperands_begin(), MI->memoperands_end());
 
       MI->removeFromParent();
       MI = Addr64;
@@ -2170,30 +2158,6 @@ void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
         moveSMRDToVALU(Inst, MRI);
       }
       break;
-    case AMDGPU::S_MOV_B64: {
-      DebugLoc DL = Inst->getDebugLoc();
-
-      // If the source operand is a register we can replace this with a
-      // copy.
-      if (Inst->getOperand(1).isReg()) {
-        MachineInstr *Copy = BuildMI(*MBB, Inst, DL, get(TargetOpcode::COPY))
-          .addOperand(Inst->getOperand(0))
-          .addOperand(Inst->getOperand(1));
-        Worklist.push_back(Copy);
-      } else {
-        // Otherwise, we need to split this into two movs, because there is
-        // no 64-bit VALU move instruction.
-        unsigned Reg = Inst->getOperand(0).getReg();
-        unsigned Dst = split64BitImm(Worklist,
-                                     Inst,
-                                     MRI,
-                                     MRI.getRegClass(Reg),
-                                     Inst->getOperand(1));
-        MRI.replaceRegWith(Reg, Dst);
-      }
-      Inst->eraseFromParent();
-      continue;
-    }
     case AMDGPU::S_AND_B64:
       splitScalar64BitBinaryOp(Worklist, Inst, AMDGPU::V_AND_B32_e64);
       Inst->eraseFromParent();