Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[AMDGPU] Merge the conditions used for deciding CS spills for amdgpu_cs_chain[_preserve] #109911

Merged
merged 1 commit into from
Sep 26, 2024
Merged

[AMDGPU] Merge the conditions used for deciding CS spills for amdgpu_cs_chain[_preserve] #109911

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 4 additions & 14 deletions llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1342,20 +1342,10 @@ void SIFrameLowering::processFunctionBeforeFrameFinalized(
SIMachineFunctionInfo *FuncInfo = MF.getInfo<SIMachineFunctionInfo>();

// Allocate spill slots for WWM reserved VGPRs.
// For chain functions, we only need to do this if we have calls to
// llvm.amdgcn.cs.chain (otherwise there's no one to save them for, since
// chain functions do not return) and the function did not contain a call to
// llvm.amdgcn.init.whole.wave (since in that case there are no inactive lanes
// when entering the function).
bool IsChainWithoutRestores =
FuncInfo->isChainFunction() &&
(!MF.getFrameInfo().hasTailCall() || FuncInfo->hasInitWholeWave());
if (!FuncInfo->isEntryFunction() && !IsChainWithoutRestores) {
for (Register Reg : FuncInfo->getWWMReservedRegs()) {
const TargetRegisterClass *RC = TRI->getPhysRegBaseClass(Reg);
FuncInfo->allocateWWMSpill(MF, Reg, TRI->getSpillSize(*RC),
TRI->getSpillAlign(*RC));
}
for (Register Reg : FuncInfo->getWWMReservedRegs()) {
const TargetRegisterClass *RC = TRI->getPhysRegBaseClass(Reg);
FuncInfo->allocateWWMSpill(MF, Reg, TRI->getSpillSize(*RC),
TRI->getSpillAlign(*RC));
}

const bool SpillVGPRToAGPR = ST.hasMAIInsts() && FuncInfo->hasSpilledVGPRs()
Expand Down
10 changes: 8 additions & 2 deletions llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -287,8 +287,14 @@ void SIMachineFunctionInfo::allocateWWMSpill(MachineFunction &MF, Register VGPR,
// amdgpu_cs_chain_preserve calling convention and this is a scratch register.
// We never need to allocate a spill for these because we don't even need to
// restore the inactive lanes for them (they're scratchier than the usual
// scratch registers).
if (isChainFunction() && SIRegisterInfo::isChainScratchRegister(VGPR))
// scratch registers). We only need to do this if we have calls to
// llvm.amdgcn.cs.chain (otherwise there's no one to save them for, since
// chain functions do not return) and the function did not contain a call to
// llvm.amdgcn.init.whole.wave (since in that case there are no inactive lanes
// when entering the function).
if (isChainFunction() &&
(SIRegisterInfo::isChainScratchRegister(VGPR) ||
!MF.getFrameInfo().hasTailCall() || hasInitWholeWave()))
return;

WWMSpills.insert(std::make_pair(
Expand Down
Loading