Skip to content

feat: swiglu simple math changes for perf upgrades #77

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
hannahli-nv merged 4 commits into from
Mar 17, 2026
Merged

feat: swiglu simple math changes for perf upgrades #77

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
cc37563
feat: swiglu simple math changes for perf upgrades
aghilann Mar 13, 2026
d18aaac
feat: swiglu simple math changes for perf upgrades
aghilann Mar 13, 2026
e6f4145
Update src/tilegym/ops/cutile/swiglu.py
aghilann Mar 17, 2026
baa63e7
Merge branch 'main' into swg-bwd
aghilann Mar 17, 2026
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: 1 addition & 6 deletions src/tilegym/ops/cutile/swiglu.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,12 +90,7 @@ def swiglu_backward_kernel(dc, a, b, da, db, TILE_SIZE: ct.Constant[int]):
a_tile_f32 = a_tile.astype(ct.float32)
b_tile_f32 = b_tile.astype(ct.float32)

# NOTE: sigmoid is intentionally inlined here to preserve current backward
# kernel behavior and benchmark baselines. Forward already uses
# the shared `sigmoid()` helper; backward will switch to it in a follow-up
# optimization PR that re-benchmarks backward performance.
# Compute sigmoid(a) and silu(a)
sigmoid_a = 1.0 / (1.0 + ct.exp(-a_tile_f32))
sigmoid_a = sigmoid(a_tile_f32)
Copy link
Copy Markdown
Contributor Author

@aghilann aghilann Mar 13, 2026

Choose a reason for hiding this comment

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

For such a small change, I choose not to move everything into experimental

silu_a = a_tile_f32 * sigmoid_a

# db = dc * silu(a)
Expand Down
Loading