test: size test_int64_upsample3d largeTensorTest guard to actual peak (64GB)#3457
Open
fjankovi wants to merge 1 commit into
Open
test: size test_int64_upsample3d largeTensorTest guard to actual peak (64GB)#3457fjankovi wants to merge 1 commit into
fjankovi wants to merge 1 commit into
Conversation
… (64GB) The test allocates a 7.03 GiB bf16 input plus a 56.25 GiB nearest scale_factor=2 output simultaneously (~63.3 GiB peak), but guarded itself with largeTensorTest(56GB). When free VRAM was in the 56-63 GiB window the gate admitted the test and it OOMed. Raise the guard to 64GB (ceil of the true peak) so the check matches the real footprint. Refs: ROCm/TheRock#6567
|
Jenkins build for e666b775034d22beaf2fc58adcebe6a59fdbb537 commit finished as ABORTED |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
test_int64_upsample3d(test/test_torch.py) guards itself with@largeTensorTest('56GB'), but its true peak footprint is ~63.3 GiB: a 7.03 GiB bf16 input(1, 256, 16, 720, 1280)plus the 56.25 GiBscale_factor=2nearest-mode output held at the same time.largeTensorTestonly skips when free VRAM at test start is below the stated size. So when free memory lands in the 56–63 GiB window, the gate admits the test and it then fails withHIP out of memory(56.25 GiB allocation). This is the intermittent gfx1151 failure tracked in ROCm/TheRock#6567:The "skip on py3.13" was never Python-version logic — just runtime memory luck.
Fix
Raise the guard to
64GB(ceil of the ~63.3 GiB real peak) so the memory check matches the actual footprint. The test is now deterministically skipped when the device cannot hold it, and only runs when it will actually fit — removing the OOM flake across all Python versions without special-casing any of them.Notes
expectedFailure(upstream's MI200 approach for this test) is unsuitable here: the test genuinely passes when the card has enough free memory, so an xfail would flip to an erroring unexpected-pass on idle runners.Refs: ROCm/TheRock#6567