RPCS3/llvm
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm.git synced 2025-01-07 04:21:39 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

R600/SI: Update tablegen defs to avoid restoring spilled sgprs to m0

Browse Source 
We had code to do this in SIRegisterInfo::eliminateFrameIndex(), but
it is easier to just change the definition of SI_SPILL_S32_RESTORE to
only allow numbered sgprs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237143 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Tom Stellard 2015-05-12 15:00:53 +00:00
parent f36ad4aa04
commit a52fdfb7c7
2 changed files with 4 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
lib/Target/R600/SIInstructions.td
View File

@ -2047,7 +2047,10 @@ multiclass SI_SPILL_SGPR <RegisterClass sgpr_class> {
} // End UseNamedOperandTable = 1
}
defm SI_SPILL_S32 : SI_SPILL_SGPR <SReg_32>;
// It's unclear whether you can use M0 as the output of v_readlane_b32
// instructions, so use SGPR_32 register class for spills to prevent
// this from happening.
defm SI_SPILL_S32 : SI_SPILL_SGPR <SGPR_32>;
defm SI_SPILL_S64 : SI_SPILL_SGPR <SReg_64>;
defm SI_SPILL_S128 : SI_SPILL_SGPR <SReg_128>;
defm SI_SPILL_S256 : SI_SPILL_SGPR <SReg_256>;

8
lib/Target/R600/SIRegisterInfo.cpp
View File

@ -245,7 +245,6 @@ void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
&AMDGPU::SGPR_32RegClass, i);
bool isM0 = SubReg == AMDGPU::M0;
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
@ -254,19 +253,12 @@ void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
Ctx.emitError("Ran out of VGPRs for spilling SGPR");
}
if (isM0)
SubReg = RS->scavengeRegister(&AMDGPU::SGPR_32RegClass, MI, 0);
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_READLANE_B32),
SubReg)
.addReg(Spill.VGPR)
.addImm(Spill.Lane)
.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
if (isM0) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
.addReg(SubReg);
}
}
// TODO: only do this when it is needed