Implement make_opt_flags function for XPU and enable tests in test_matmul.py
#5051
Conversation
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
anmyachev
force-pushed
the
amyachev/issue4975
branch
from
bd9c534 to
5646fa0
Compare
anmyachev
requested review from
Egor-Krivov,
chengjunlu,
etiotto and
whitneywhtsang
Contributor
Author
|
Looks like I can test these changes using |
chengjunlu
approved these changes
Sep 8, 2025
anmyachev
commented
Sep 8, 2025
| if split_k > 1: | ||
| pytest.skip("splitK hasn't been fully tested on AMD GPU.") | ||
|
|
||
| elif is_xpu(): |
Contributor
Author
There was a problem hiding this comment.
Local result: 980 passed, 2340 skipped.
Contributor
Author
There was a problem hiding this comment.
New results: 8 failed, 1124 passed, 2188 skipped in 1379.15s (0:22:59) (8 cases added to the skiplists)
anmyachev
commented
Sep 8, 2025
| elif is_xpu(): | ||
| if split_k > 1: | ||
| pytest.skip("splitK hasn't been fully tested on INTEL GPU.") | ||
| if "float8_e4m3fn" in act_dtype_str and "float8_e4m3fn" in weight_dtype_str: |
Contributor
Author
There was a problem hiding this comment.
For these cases I see the following:
python/triton_kernels/tests/test_matmul.py::test_op[False-True-True-True-False-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-1-False-None] - AssertionError: ref_y_scale: 0.004773152060806751, tri_y_scale: 0.005022321827709675
anmyachev
commented
Sep 8, 2025
| generate_native_code: bool = False | ||
| advanced_path: bool = False | ||
| enable_tile_load_linear_layout: bool = True | ||
| arch: str = None |
Contributor
Author
There was a problem hiding this comment.
Otherwise, new tests don't work.
anmyachev
commented
Sep 8, 2025
Comment on lines
+58
to
+59
| group_m = 8 | ||
| xcd_swizzle = 1 |
etiotto
reviewed
Sep 9, 2025
| if split_k > 1: | ||
| pytest.skip("splitK hasn't been fully tested on INTEL GPU.") | ||
| if "float8_e4m3fn" in act_dtype_str and "float8_e4m3fn" in weight_dtype_str: | ||
| pytest.skip("FIXME") |
Contributor
There was a problem hiding this comment.
Suggest to create an issue for this and mark te skip with the issue number.
Signed-off-by: Anatoly Myachev <[email protected]>
Signed-off-by: Anatoly Myachev <[email protected]>
Signed-off-by: Anatoly Myachev <[email protected]>
anmyachev
force-pushed
the
amyachev/issue4975
branch
from
40fd5c2 to
2f9b449
Compare
Signed-off-by: Anatoly Myachev <[email protected]>
anmyachev
force-pushed
the
amyachev/issue4975
branch
from
2f9b449 to
d129992
Compare
YangKai0616
reviewed
Sep 10, 2025
|
|
||
| def compute_split_k(block_k: int, k: int | None, grid_size: int) -> int: | ||
| device_props = torch.xpu.get_device_properties(0) | ||
| n_sms = device_props.multi_processor_countgpu_subslice_count |
Contributor
There was a problem hiding this comment.
Should be gpu_subslice_count.
Contributor
Author
There was a problem hiding this comment.
good catch! you're right.
anmyachev
commented
Sep 10, 2025
python/triton_kernels/triton_kernels/matmul_ogs_details/opt_flags_details/opt_flags_intel.py
Outdated
Show resolved
Hide resolved
Signed-off-by: Anatoly Myachev <[email protected]>
Signed-off-by: Anatoly Myachev <[email protected]>
anmyachev
commented
Sep 10, 2025
|
|
||
| TRITON_TEST_SUITE=triton_kernels \ | ||
| run_pytest_command -vvv -n ${PYTEST_MAX_PROCESSES:-8} --device xpu . | ||
| run_pytest_command -vvv -n ${PYTEST_MAX_PROCESSES:-4} --device xpu . |
Contributor
Author
There was a problem hiding this comment.
Otherwise, Python processes start to break.
anmyachev
changed the title
make_opt_flags function for XPUmake_opt_flags function for XPU and enable tests in test_matmul.py
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Note for reviewers: I've left only the most basic heuristics. If you have improvements in mind that will definitely work better without testing, i.e. were already known, then we can make edits directly to this pull request. If you have improvements that need to be tested - that's also good, please also write, but I'd prefer to implement the basic version as quickly as possible and tune it in separate PRs if possible.
Pass rate: 84.11% -> 89.04%