add gtest for conv gpu SAME_UPPER autopad - #35
Merged
Conversation
Contributor
Author
|
While working on this PR, I noticed SAME_UPPER and SAME_LOWER only differ when the total padding is odd and for even padding they resolve to the same pads. ROOT's SameUpper model (input 5, k3, s1) is an even case, so it's effectively identical to SameLower. The odd case would something be like having a 3x3 stride-2 conv on an even-sized input(this gives odd padding) and the TF/Keras "SAME" maps to SAME_UPPER. So it kinda seemed worth covering to me. Does this odd-split path actually come up in the models SOFIE targets or is the even case all we really need here? Any thoughts or guidance would be helpful. @sanjibansg |
harz05
force-pushed
the
feat/conv-autopad-upper-test
branch
from
August 13, 2026 00:54
e432fd9 to
d39a038
Compare
Contributor
Author
|
rebased the pr; pls review, thanks |
Member
|
/runci |
|
/runci: GPU Unit Tests ❌ failed — view run |
Member
|
/runci |
|
/runci: GPU Unit Tests ❌ failed — view run |
harz05
force-pushed
the
feat/conv-autopad-upper-test
branch
from
August 13, 2026 10:29
d39a038 to
005e940
Compare
Member
|
/runci |
|
/runci: GPU Unit Tests ❌ failed — view run |
Member
|
/runtest |
1 similar comment
Member
|
/runtest |
Member
|
/runtest h100-47gb |
|
/runtest (h100-47gb): GPU Unit Tests ❌ failed — view run |
harz05
force-pushed
the
feat/conv-autopad-upper-test
branch
from
August 13, 2026 15:08
005e940 to
ee85673
Compare
Member
|
/runtest h100-47gb |
|
/runtest: triggered — view run |
|
/runtest (h100-47gb): GPU Unit Tests ✅ passed — view run |
Member
|
/runbenchmark h100-47gb |
|
/runbenchmark: triggered — view run |
Benchmark results (h100-47gb)Comparing against |
|
/runbenchmark (h100-47gb): Benchmark ✅ completed — results updated above. view run |
sanjibansg
approved these changes
Aug 13, 2026
sanjibansg
left a comment
sanjibansg
left a comment
Member
There was a problem hiding this comment.
LGTM! Thanks for this implementation.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #33.
Adds a GPU test for the SAME_UPPER autopad branch, which the alpaka tests didn't cover (only SAME_LOWER was tested).
SAME_UPPER and SAME_LOWER only differ when the total padding is odd, so the test uses x[1,1,4,4], kernel 3, stride 2 (total pad 1): SAME_UPPER puts the extra pad at the end (begin 0), SAME_LOWER at the start (begin 1). The output differs from what SAME_LOWER gives, and the generated im2col confirms the begin pad (- 0 vs - 1), so the test actually exercises the upper split rather than just the code path.
Tested on Colab T4: ConvWithAutopadSameUpper passes and all existing Conv tests still pass.