Skip to content

[RISCV] Mark all registers marked isConstant as reserved #96002

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.

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

Merged
merged 1 commit into from
Jun 23, 2024
Merged
Changes from all commits
Commits
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
8 changes: 5 additions & 3 deletions llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -104,14 +104,17 @@ BitVector RISCVRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
BitVector Reserved(getNumRegs());
auto &Subtarget = MF.getSubtarget<RISCVSubtarget>();

// Mark any registers requested to be reserved as such
for (size_t Reg = 0; Reg < getNumRegs(); Reg++) {
Copy link
Collaborator

@topperc topperc Jun 19, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I find it odd that this loop was looping over more than just the GPRs originally since those are the only ones that can be reserved. We could maybe save some compile time by not checking every register, but this patch now requires us to check every register.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good point!
So I think we should iterate through all GPRs explicitly here.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I went with iterating from RISCV::NoRegister to RISCV::NUM_TARGET_REGS, with a GPR check before calling Subtarget.isRegisterReservedByUser.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To be honest, I like your first patch better. As long as RISCVSubtarget uses std::bitset<RISCV::NUM_TARGET_REGS> to track reserved registers, we should continue iterating all of the registers.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah I kind of agree. I reverted it back to the initial patch. I don't think calling isRegisterReservedByUser with non-GPRs is an issue.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe isRegisterReservedByUser should be replaced by isGPRRegisterReservedByUser. My main point was that as the number of registers increases with new extensions, this loop gets more and more iterations, making the function slower even if the new registers don't need to be reserved.

// Mark any GPRs requested to be reserved as such
if (Subtarget.isRegisterReservedByUser(Reg))
markSuperRegs(Reserved, Reg);

// Mark all the registers defined as constant in TableGen as reserved.
if (isConstantPhysReg(Reg))
markSuperRegs(Reserved, Reg);
}

// Use markSuperRegs to ensure any register aliases are also reserved
markSuperRegs(Reserved, RISCV::X0); // zero
markSuperRegs(Reserved, RISCV::X2); // sp
markSuperRegs(Reserved, RISCV::X3); // gp
markSuperRegs(Reserved, RISCV::X4); // tp
Expand All @@ -136,7 +139,6 @@ BitVector RISCVRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
markSuperRegs(Reserved, RISCV::VTYPE);
markSuperRegs(Reserved, RISCV::VXSAT);
markSuperRegs(Reserved, RISCV::VXRM);
markSuperRegs(Reserved, RISCV::VLENB); // vlenb (constant)

// Floating point environment registers.
markSuperRegs(Reserved, RISCV::FRM);
Expand Down
Loading