Skip to content

[AMDGPU] A SCHED_BARRIER in a bundle should not prevent other SCHED_BARRIERs to be considered #152627

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

Closed
wants to merge 2 commits into from
Closed

[AMDGPU] A SCHED_BARRIER in a bundle should not prevent other SCHED_BARRIERs to be considered #152627

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f06e57f
A SCHED_BARRIER in a bundle should not prevent other SCHED_BARRIER to…
yoonseoch Aug 8, 2025
c1b4e37
Simplifying scanning inside a bundle
yoonseoch Aug 8, 2025
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
15 changes: 10 additions & 5 deletions llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2502,13 +2502,18 @@ bool SchedGroup::canAddSU(SUnit &SU) const {
return canAddMI(MI);

// Special case for bundled MIs.
// Return true if all of the bundled MIs can be added to this group.
// A meta instruction in a bundle is an exception.
const MachineBasicBlock *MBB = MI.getParent();
MachineBasicBlock::instr_iterator B = MI.getIterator(), E = ++B;
while (E != MBB->end() && E->isBundledWithPred())
++E;
// Initially, iterator is on a bundler header.
MachineBasicBlock::instr_iterator B = std::next(MI.getIterator());
while (B != MBB->end() && B->isBundledWithPred()) {
if (!B->isMetaInstruction() && !canAddMI(*B))
Copy link
Contributor

@arsenm arsenm Aug 12, 2025

Choose a reason for hiding this comment

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

Should also remove the isMetaInstruction handling inside canAddMI

Copy link
Contributor Author

@yoonseoch yoonseoch Aug 12, 2025

Choose a reason for hiding this comment

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

@arsenm, Did you mean cannAddMI return true on isMetaInstruction()? Or, a NFC change of removing lines 3 and 4 below?

1 bool SchedGroup::canAddMI(const MachineInstr &MI) const {
2  bool Result = false;
3  if (MI.isMetaInstruction())
4    Result = false;
5  ...
6  return Result;

As I addressed in my last comment, #152627 (comment), making canAddMI return true on meta instructions affected the results of tests related to sched_group_barrier/iglp_opt. This is due to lines 2501 and 2502.

In effect, making canAddMI return true on meta instructions will add artificial dependence edges among sched_barriers/sched_group_barriers/iglp_opts. Let me know if those potential changes are alright. Then, I will make canAddMI return true on meta instructions and further simplify line 2511 above.

Copy link
Contributor

@arsenm arsenm Aug 13, 2025

Choose a reason for hiding this comment

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

You also need to adjust the above usage in the unbundled case, but I meant make it not consider metainstructions at all (i.e. fully pull the meta instruction handling into this function here)

return false;
++B;
}

// Return true if all of the bundled MIs can be added to this group.
return std::all_of(B, E, [this](MachineInstr &MI) { return canAddMI(MI); });
return true;
}

void SchedGroup::initSchedGroup() {
Expand Down
Loading
Loading