Fix assertion in ScanLowering for num_ctas>1 #5680
Conversation
|
|
|
Added the test cases, they pass if the scan is within a CTA and fail if it is across multiple CTAs. I didn't find any logic which performs accumulation across CTAs in the code. I could try to look into the logic and test coverage for cross-CTA scans in a different PR, but I think fixing the assertion for within-CTA scans can be independent from that. How should I proceed? |
For now it should happen only within a CTA. Please do not work on anything across CTAs |
This reverts commit 33ca052.
This reverts commit 44ddc3e.
Assume the clusterCTAId along the scan axis (cctaIdAxis) is ==0, raise runtime assertion otherwise. Combine the clusterCTAId across the scan axis (cctaIdParallel) into the flatIdParallel and compute numParallelLane per CGA instead of per CTA.
Test BlockedLayout for
- thread_size=4, num_warps=4, num_ctas=1
- thread_size=4, num_warps=1, num_ctas=4: CTASplitNum=[1,1]
- thread_size=4, num_warps=1, num_ctas=4: CTASplitNum=CTAsPerCGA
- thread_size=1, num_warps=4, num_ctas=4: CTASplitNum=[1,1]
- thread_size=1, num_warps=4, num_ctas=4: CTASplitNum=CTAsPerCGA
…fix-scanlowering-cga
safelix
changed the title
|
The initial fix was only superficial, so I implemented within-CTA scan from scratch. I changed testing from end-to-end tests to layout tests to make sure all edge cases are caught. Pytest output for I'm not sure if the tests are too extensive now, maybe we could just test the combinations with Is there anything else I can do to get this merged? |
Enable within-CTA scans if
num_ctas>1andcluster_dims[axis]==1on Hopper or later.Assume the
clusterCTAIdalong the scan axis iscctaIdAxis==0, raise runtime assertion otherwise. Combine theclusterCTAIdacross the scan axis (cctaIdParallel) into theflatIdParalleland computenumParallelLaneper CGA instead of per CTA.This fixes assertions when
num_ctas > 1:triton/lib/Conversion/TritonGPUToLLVM/ScanOpToLLVM.cpp
Line 150 in cea35da
triton/lib/Conversion/TritonGPUToLLVM/ScanOpToLLVM.cpp
Line 254 in cea35da
New contributor declaration
I am not making a trivial change, such as fixing a typo in a comment.
I have written a PR description following these
rules.
I have run
pre-commit run --from-ref origin/main --to-ref HEAD.Select one of the following.
/testforlittests/unittestfor C++ tests/python/testfor end-to-end testsSelect one of the following.
littests.littests I have added follow these best practices,including the "tests should be minimal" section. (Usually running Python code
and using the instructions it generates is not minimal.)