Skip to content

Remove z-shape block swizzle #92

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

Merged
zasdfgbnm merged 1 commit into from
Mar 29, 2023
Merged

Remove z-shape block swizzle #92

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
5 changes: 1 addition & 4 deletions csrc/scheduler/matmul.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -181,7 +181,7 @@ void scheduleMatmul(
// [... M,N,K]
scheduler_utils::matmul_utils::makeTile(cc, gemm_tile.cta_tile.toVector());

// Applies swizzle factor on C
// Swizzle block tiles:
if (params.grid_swizzle_factor != 1) {
int factor = std::max(1, params.grid_swizzle_factor); // must be >=1
if (params.rasterization_order ==
Expand Down Expand Up @@ -252,9 +252,6 @@ void scheduleMatmul(
// ------------------------------------------------------------------
// CTA tile:

// Swizzle block tiles:
c->swizzle(Swizzle2DType::ZShape, 0, 1, SwizzleMode::Loop);
Copy link
Contributor

@drzejan2 drzejan2 Mar 29, 2023 •
edited
Loading

Choose a reason for hiding this comment

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

It looks like this is the only place where TensorView::swizzle function is currently used in production.

Do you think it should be kept, or should it be removed, along with tests for it? (test_gpu_tensorcore.cpp and test_gpu_swizzle.cpp). Perhaps there are use cases where it will be required, but nothing comes to my mind right away.

Other than this, PR looks good to me.

Copy link
Collaborator Author

@zasdfgbnm zasdfgbnm Mar 29, 2023

Choose a reason for hiding this comment

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

TensorView::swizzle should also be applied to the shared memory buffer to remove bank conflict. It is now in main branch now, but #41 will add it.


a->computeAt(c, 2);
b->computeAt(c, 2);

Expand Down