Skip to content

Insert iglp(10) in computer cluster as the marker for better instruction interleaving #820

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

Open
wants to merge 1 commit into
base: shared/triton-gfx950-launch
Choose a base branch
from
Open

Insert iglp(10) in computer cluster as the marker for better instruction interleaving #820

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
2 changes: 2 additions & 0 deletions third_party/amd/lib/TritonAMDGPUTransforms/BlockPingpong.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1025,6 +1025,7 @@ LogicalResult Pingponger::transformFAv3(OpBuilder &builder, Location loc) {

// dot cluster 0 operations here.

appendOp(builder.create<ROCDL::IglpOpt>(loc, 10));
updateOpInsertion(asyncWaitOps[0]);
prependOp(builder.create<ROCDL::SetPrioOp>(loc, highPriority), false);
appendOp(builder.create<ROCDL::SchedBarrier>(loc, 0));
Expand All @@ -1033,6 +1034,7 @@ LogicalResult Pingponger::transformFAv3(OpBuilder &builder, Location loc) {

updateOpInsertion(dotOps[1]);
// below ops are inserted backward
prependOp(builder.create<ROCDL::IglpOpt>(loc, 10), true);
prependOp(builder.create<ROCDL::SetPrioOp>(loc, lowPriority), true);
prependOp(builder.create<ROCDL::SBarrierOp>(loc), true);
prependOp(builder.create<ROCDL::SchedBarrier>(loc, 0), true);
Expand Down
Loading