frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 12:54:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update the UT of test_decompose_mm_cpu (#154100)

Browse Source 
**Summary**
Fixes #153616
Based on the latest decomposed heuristic in daca611465/torch/_inductor/fx_passes/decompose_mem_bound_mm.py (L79-L82), for the shape in this test case `[m=1, k=64, n=32]`, the result should be decomposed. The previous CI didn't capture this failure due to the UT skip described in https://github.com/pytorch/pytorch/pull/153245. So this PR should be verified in CI after https://github.com/pytorch/pytorch/pull/153245 landed.

**Test Plan**
```
python -u -m pytest -s -v test/inductor/test_decompose_mem_bound_mm.py -k test_decompose_mm_cpu
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/154100
Approved by: https://github.com/jansel
This commit is contained in:
leslie-fang-intel
2025-06-25 00:09:08 +00:00
committed by PyTorch MergeBot
parent f5f4beaf56
commit 194c221e0a
1 changed files with 1 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/inductor/test_decompose_mem_bound_mm.py
View File

@ -278,11 +278,9 @@ class TestDecomposeMemMM(TestCase):
)
counters.clear()
# (1, 64, 32, False) vesrion fails
@unittest.skip
@parametrize(
"m,k,n, should_decompose",
[(1, 64, 16, True), (2, 64, 16, False), (1, 64, 32, False)],
[(1, 64, 16, True), (2, 64, 16, False), (1, 64, 32, True)],
)
def test_decompose_mm_cpu(self, m, n, k, should_decompose):
torch._logging.set_logs(inductor=logging.DEBUG)