Skip to content
This repository was archived by the owner on Mar 28, 2023. It is now read-only.

[SYCL][CUDA] Enable sub-group load_store tests #90

Merged
merged 3 commits into from
Jan 13, 2021

Conversation

Pennycook
Copy link

Tests associated with new functionality from intel/llvm#2763.

The test assumed that it was safe to load/store off the end of a
buffer as long as the results of those accesses were ignored.
Some backends (e.g. CUDA) will fail because of an out-of-bounds
access, so it's safer to pad the buffers instead.

Signed-off-by: John Pennycook <[email protected]>
@Pennycook Pennycook requested a review from vladimirlaz January 11, 2021 15:31
Using // caused comments to run together.
Using #if 0 allows them to remain formatted like real lines.

Signed-off-by: John Pennycook <[email protected]>
@vladimirlaz
Copy link

@Pennycook could you please remove tests in intel/llvm#2763 to resolve conflicts.
Otherwise tests for compatible change in llvm-test-suite and compiler will fail on compiler merge stage.

@Pennycook
Copy link
Author

@Pennycook could you please remove tests in intel/llvm#2763 to resolve conflicts.
Otherwise tests for compatible change in llvm-test-suite and compiler will fail on compiler merge stage.

Oh, woops. I'd done that locally but forgot to push. Should be fixed now.

@vladimirlaz vladimirlaz merged commit 942abad into intel:intel Jan 13, 2021
aelovikov-intel pushed a commit to aelovikov-intel/llvm that referenced this pull request Mar 27, 2023
)

* Prevent out-of-bounds access in load_store. The test assumed that it was safe to load/store off the end of a
buffer as long as the results of those accesses were ignored. Some backends (e.g. CUDA) will fail because of an out-of-bounds access, so it's safer to pad the buffers instead.
* Enable sub-group load_store tests

Signed-off-by: John Pennycook <[email protected]>
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants