[Conv] Enable bwd weight splitk autodeduction with cap

[johannes-graner]

Proposed changes

Enables split-k autodeduction for old CK bwd weight convolutions. Since the autodeduction can produce very large split-k values that lead to inaccuracy, the deduced value is capped at 128. Under the assumption that performance is a convex function of split-k and the auto-deduced value is correct, capping at 128 results in the best reasonable value if the reasonable values are {1, ..., 128}.

Enabling autodeduction led to tests using larger split-k values, which exposed a bug in the profiler error threshold calculations. This was also addressed since it made the tests fail.

Checklist

Please put an x into the boxes that apply. You can also fill these out after creating the PR. If you're not sure, please don't hesitate to ask.

• I have added tests relevant to the introduced functionality, and the unit tests are passing locally
• I have added the test to REGRESSION_TESTS list defined at the top of CMakeLists.txt in tests/CMakeLists.txt, IF the test takes more than 30 seconds to run.
• I have added inline documentation which enables the maintainers with understanding the motivation
• I have removed the stale documentation which is no longer relevant after this pull request
• (If this change is user-facing) I have added release notes which provide the end users with a brief summary of the improvement from this pull request
• I have run clang-format on all changed files
• Any dependent changes have been merged

Discussion

If this is a relatively large or complex change, feel free to start a discussion by explaining why you chose the solution you did and what alternatives you considered

[vpietila-amd]

Looks good to me. One minor comment regarding the WMMA multi-D kernel, but we can fix this separately.

[vpietila-amd]

Can you modify this such that we return false if BlkGemmPipelineVer == BlockGemmPipelineVersion::v4 since the autodeducted split-K value is probably too low in this case.