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

Always try to lower standalone lmhlo.transpose to custom call for better performance #1282

Merged
merged 3 commits into from
Feb 26, 2024
Merged

Always try to lower standalone lmhlo.transpose to custom call for better performance #1282

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9f78a57
add one-hop buffer reuse propogation
eedalong Jan 22, 2024
28960f0
add ut
eedalong Jan 22, 2024
99cc4f4
always try to lower standalone transposeOp to custom call
eedalong Feb 22, 2024
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
5 changes: 2 additions & 3 deletions tao_compiler/mlir/disc/transforms/disc_lower_to_library_call.cc
100644 → 100755
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -494,7 +494,7 @@ struct TransposeConverter : public OpRewritePattern<lmhlo::TransposeOp> {
if (rank != 2 && rank != 3) return failure();
// only rewriter custom library when switch 1 and 2 dimensions of
// a 3d tensor, that means permute = [0, 2, 1]
if (rank == 3 && permutation[1] != 2 && permutation[2] != 1)
if (rank == 3 && (permutation[1] != 2 || permutation[2] != 1))
return failure();
bool on_gpu = placement_utils::isGpuMemRef(op->getOperand(0));
// TODO: support other device
Expand Down Expand Up @@ -914,8 +914,7 @@ struct DiscLowerToLibraryCallPass
SendOutputOpConvertor
>(context);
// clang-format on
if (enableTransposeLibraryCall())
patterns.insert<TransposeConverter>(context);
patterns.insert<TransposeConverter>(context);

// GPU copy related ops
patterns.insert<GpuCopyOpConvertor<H2DOp>>(context, "h2d");
Expand Down
Loading