Skip to content

[Dlight] Enhance vectorization for gpu matmul #16894

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
merged 2 commits into from
Apr 17, 2024
Merged

[Dlight] Enhance vectorization for gpu matmul #16894

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
449b3d7
[Dlight] Enhance vectorization for gpu matmul
vinx13 Apr 16, 2024
f52ffa7
fix
vinx13 Apr 17, 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
7 changes: 5 additions & 2 deletions python/tvm/dlight/gpu/matmul.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -873,7 +873,10 @@ def apply( # pylint: disable=too-many-locals,missing-docstring
x, [None, config.vthread_x, config.block_size_x, config.micro_size_x]
)
ko, ki = sch.split(k, factors=[None, config.micro_size_k])
sch.reorder(by, bx, vy, vx, ty, tx, ko, ki, yi, xi)
reordered_loops = [by, bx, vy, vx, ty, tx, ko, ki] + (
[yi, xi] if config.inner_x else [xi, yi]
)
sch.reorder(*reordered_loops)
by = sch.fuse(batch, by)
sch.bind(bx, "blockIdx.x")
sch.bind(by, "blockIdx.y")
Expand All @@ -883,7 +886,7 @@ def apply( # pylint: disable=too-many-locals,missing-docstring
sch.bind(tx, "threadIdx.x")
inner_loop = config.micro_size_x if config.inner_x else config.micro_size_y
if inner_loop % config.vector_size == 0:
_, v = sch.split(xi, [None, config.vector_size])
_, v = sch.split(reordered_loops[-1], [None, config.vector_size])
sch.vectorize(v)

if config.unroll > 0:
Expand Down
Loading