maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[AMDGPU] Fold V_CNDMASK with identical source operands 

Such instructions sometimes appear after lowering and folding.

Differential Revision: https://reviews.llvm.org/D31318

llvm-svn: 298723
This commit is contained in:
Stanislav Mekhanoshin 2017-03-24 18:55:20 +00:00
parent 4986d9fb45
commit 70603dcef2
2 changed files with 63 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
View File

@ -591,6 +591,32 @@ static bool tryConstantFoldOp(MachineRegisterInfo &MRI,
return false;
}
// Try to fold an instruction into a simpler one
static bool tryFoldInst(const SIInstrInfo *TII,
MachineInstr *MI) {
unsigned Opc = MI->getOpcode();
if (Opc == AMDGPU::V_CNDMASK_B32_e32 ||
Opc == AMDGPU::V_CNDMASK_B32_e64 ||
Opc == AMDGPU::V_CNDMASK_B64_PSEUDO) {
const MachineOperand *Src0 = TII->getNamedOperand(*MI, AMDGPU::OpName::src0);
const MachineOperand *Src1 = TII->getNamedOperand(*MI, AMDGPU::OpName::src1);
if (Src1->isIdenticalTo(*Src0)) {
DEBUG(dbgs() << "Folded " << *MI << " into ");
int Src2Idx = AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src2);
if (Src2Idx != -1)
MI->RemoveOperand(Src2Idx);
MI->RemoveOperand(AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src1));
mutateCopyOp(*MI, TII->get(Src0->isReg() ? (unsigned)AMDGPU::COPY
: getMovOpc(false)));
DEBUG(dbgs() << *MI << '\n');
return true;
}
}
return false;
}
void SIFoldOperands::foldInstOperand(MachineInstr &MI,
MachineOperand &OpToFold) const {
// We need mutate the operands of new mov instructions to add implicit
@ -692,6 +718,7 @@ void SIFoldOperands::foldInstOperand(MachineInstr &MI,
}
DEBUG(dbgs() << "Folded source from " << MI << " into OpNo " <<
static_cast<int>(Fold.UseOpNo) << " of " << *Fold.UseMI << '\n');
tryFoldInst(TII, Fold.UseMI);
}
}
}
@ -907,6 +934,8 @@ bool SIFoldOperands::runOnMachineFunction(MachineFunction &MF) {
Next = std::next(I);
MachineInstr &MI = *I;
tryFoldInst(TII, &MI);
if (!isFoldableCopy(MI)) {
if (IsIEEEMode || !tryFoldOMod(MI))
tryFoldClamp(MI);

34
llvm/test/CodeGen/AMDGPU/fold-cndmask.mir Normal file
View File

@ -0,0 +1,34 @@
# RUN: llc -march=amdgcn -run-pass si-fold-operands -verify-machineinstrs -o - %s | FileCheck %s
# CHECK: %1 = V_MOV_B32_e32 0, implicit %exec
# CHECK: %2 = V_MOV_B32_e32 0, implicit %exec
# CHECK: %4 = COPY %3
# CHECK: %5 = V_MOV_B32_e32 0, implicit %exec
# CHECK: %6 = V_MOV_B32_e32 0, implicit %exec
# CHECK: %7 = COPY %3
---
name: fold_cndmask
tracksRegLiveness: true
registers:
- { id: 0, class: sgpr_64 }
- { id: 1, class: vgpr_32 }
- { id: 2, class: vgpr_32 }
- { id: 3, class: vgpr_32 }
- { id: 4, class: vgpr_32 }
- { id: 5, class: vgpr_32 }
- { id: 6, class: vgpr_32 }
- { id: 7, class: vgpr_32 }
body: |
bb.0.entry:
%0 = IMPLICIT_DEF
%1 = V_CNDMASK_B32_e64 0, 0, %0, implicit %exec
%2 = V_CNDMASK_B32_e64 %1, %1, %0, implicit %exec
%3 = IMPLICIT_DEF
%4 = V_CNDMASK_B32_e64 %3, %3, %0, implicit %exec
%5 = COPY %1
%6 = V_CNDMASK_B32_e64 %5, 0, %0, implicit %exec
%vcc = IMPLICIT_DEF
%7 = V_CNDMASK_B32_e32 %3, %3, implicit %exec, implicit %vcc
...