Draft
Conversation
github-actions
bot
added
in-progress
…yaswann/iris_xops_perf
…yaswann/iris_xops_perf
…yaswann/iris_xops_perf
…yaswann/iris_xops_perf
…yaswann/iris_xops_perf
…e temp files - Restore optional `hint` parameter in `__translate` and all public iris API functions (load, store, get, put, copy, atomic_*) to match main branch pattern. The previous hardcoded `tl.multiple_of(ptr, (32, 32))` assumed 2D pointers and broke all scalar-pointer atomic operations. - Align tritonBLAS commit across pyproject.toml, run_tests.sh, apptainer/iris.def, and docker/Dockerfile to cd119279f. - Remove tracked backup files (iris.py.backup, all_gather_matmul.py.with_chunked) and add gitignore patterns. - Remove unimplemented "chunked" variant from test_all_gather_matmul parametrization. - Fix test_matmul_all_reduce_via_shmem_ops dimensions (N=128->256) to match new default block_size_n=256. - Remove phantom "matmul" from iris/ops/__init__.py __all__. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Resolve conflicts by adopting main's approach: - Remove manual tritonBLAS from containers and run_tests.sh (handled via pyproject.toml dependency) - Use torchrun for test execution (from main) - Keep main's shorter docstring for hint parameter in iris.py Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Examples 28 (matmul_all_reduce) and 30 (matmul_all_gather) used N=128 as default, which is smaller than the new FusedConfig default block_size_n=256. This triggers assertion failures (N >= block_size_n) in CI, crashing all ranks and causing the 8-rank test to hang for 179 minutes waiting for the dead rank. Increase both examples' default N from 128 to 256 to match the new config defaults. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…256" This reverts commit 477b472.
The Triton kernels already handle block_size > dimension via: - tl.cdiv(N, BLOCK_SIZE_N) for grid sizing - mask=(rn < N) on loads/stores - tritonblas GemmContext.reduce_axis handles K masking The assertions were preventing valid configurations (e.g., block_size_n=256 with N=128) that the kernels handle correctly. Removed for_problem() clamping too — it's unnecessary when the kernels already mask. Fixes CI failures on examples 28 and 30 which use N=128 with default FusedConfig block_size_n=256.
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.
Motivation
Add benchmarking capabilities for
iris.ops.