diff options
| author | Matt Arsenault <Matthew.Arsenault@amd.com> | 2019-05-03 15:21:53 +0000 |
|---|---|---|
| committer | Matt Arsenault <Matthew.Arsenault@amd.com> | 2019-05-03 15:21:53 +0000 |
| commit | cfd0ca38b09b50fab0b1dfd6533b16bc2993c676 (patch) | |
| tree | 9fab2a14155d4824c31e2cd0559b371dfecf3950 /llvm/lib/Target/AMDGPU | |
| parent | e28ab93546539a2b75bd2d2956b4de6131c39189 (diff) | |
| download | bcm5719-llvm-cfd0ca38b09b50fab0b1dfd6533b16bc2993c676.tar.gz bcm5719-llvm-cfd0ca38b09b50fab0b1dfd6533b16bc2993c676.zip | |
AMDGPU: Support shrinking add with FI in SIFoldOperands
Avoids test regression in a future patch
llvm-svn: 359898
Diffstat (limited to 'llvm/lib/Target/AMDGPU')
| -rw-r--r-- | llvm/lib/Target/AMDGPU/SIFoldOperands.cpp | 72 |
1 files changed, 37 insertions, 35 deletions
diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp index dcfc39e6755..7fdbf97fe40 100644 --- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp +++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp @@ -216,53 +216,55 @@ static bool updateOperand(FoldCandidate &Fold, break; } } + } - if (Fold.needsShrink()) { - MachineBasicBlock *MBB = MI->getParent(); - auto Liveness = MBB->computeRegisterLiveness(&TRI, AMDGPU::VCC, MI); - if (Liveness != MachineBasicBlock::LQR_Dead) - return false; + if ((Fold.isImm() || Fold.isFI()) && Fold.needsShrink()) { + MachineBasicBlock *MBB = MI->getParent(); + auto Liveness = MBB->computeRegisterLiveness(&TRI, AMDGPU::VCC, MI); + if (Liveness != MachineBasicBlock::LQR_Dead) + return false; - MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo(); - int Op32 = Fold.getShrinkOpcode(); - MachineOperand &Dst0 = MI->getOperand(0); - MachineOperand &Dst1 = MI->getOperand(1); - assert(Dst0.isDef() && Dst1.isDef()); + MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo(); + int Op32 = Fold.getShrinkOpcode(); + MachineOperand &Dst0 = MI->getOperand(0); + MachineOperand &Dst1 = MI->getOperand(1); + assert(Dst0.isDef() && Dst1.isDef()); - bool HaveNonDbgCarryUse = !MRI.use_nodbg_empty(Dst1.getReg()); + bool HaveNonDbgCarryUse = !MRI.use_nodbg_empty(Dst1.getReg()); - const TargetRegisterClass *Dst0RC = MRI.getRegClass(Dst0.getReg()); - unsigned NewReg0 = MRI.createVirtualRegister(Dst0RC); + const TargetRegisterClass *Dst0RC = MRI.getRegClass(Dst0.getReg()); + unsigned NewReg0 = MRI.createVirtualRegister(Dst0RC); - MachineInstr *Inst32 = TII.buildShrunkInst(*MI, Op32); + MachineInstr *Inst32 = TII.buildShrunkInst(*MI, Op32); - if (HaveNonDbgCarryUse) { - BuildMI(*MBB, MI, MI->getDebugLoc(), TII.get(AMDGPU::COPY), Dst1.getReg()) - .addReg(AMDGPU::VCC, RegState::Kill); - } - - // Keep the old instruction around to avoid breaking iterators, but - // replace it with a dummy instruction to remove uses. - // - // FIXME: We should not invert how this pass looks at operands to avoid - // this. Should track set of foldable movs instead of looking for uses - // when looking at a use. - Dst0.setReg(NewReg0); - for (unsigned I = MI->getNumOperands() - 1; I > 0; --I) - MI->RemoveOperand(I); - MI->setDesc(TII.get(AMDGPU::IMPLICIT_DEF)); - - if (Fold.isCommuted()) - TII.commuteInstruction(*Inst32, false); - return true; + if (HaveNonDbgCarryUse) { + BuildMI(*MBB, MI, MI->getDebugLoc(), TII.get(AMDGPU::COPY), Dst1.getReg()) + .addReg(AMDGPU::VCC, RegState::Kill); } - Old.ChangeToImmediate(Fold.ImmToFold); + // Keep the old instruction around to avoid breaking iterators, but + // replace it with a dummy instruction to remove uses. + // + // FIXME: We should not invert how this pass looks at operands to avoid + // this. Should track set of foldable movs instead of looking for uses + // when looking at a use. + Dst0.setReg(NewReg0); + for (unsigned I = MI->getNumOperands() - 1; I > 0; --I) + MI->RemoveOperand(I); + MI->setDesc(TII.get(AMDGPU::IMPLICIT_DEF)); + + if (Fold.isCommuted()) + TII.commuteInstruction(*Inst32, false); return true; } assert(!Fold.needsShrink() && "not handled"); + if (Fold.isImm()) { + Old.ChangeToImmediate(Fold.ImmToFold); + return true; + } + if (Fold.isFI()) { Old.ChangeToFrameIndex(Fold.FrameIndexToFold); return true; @@ -363,7 +365,7 @@ static bool tryAddToFoldList(SmallVectorImpl<FoldCandidate> &FoldList, if ((Opc == AMDGPU::V_ADD_I32_e64 || Opc == AMDGPU::V_SUB_I32_e64 || Opc == AMDGPU::V_SUBREV_I32_e64) && // FIXME - OpToFold->isImm()) { + (OpToFold->isImm() || OpToFold->isFI())) { MachineRegisterInfo &MRI = MI->getParent()->getParent()->getRegInfo(); // Verify the other operand is a VGPR, otherwise we would violate the |

