ROCm / Tensile

Stretching GPU performance for GEMMs and tensor contractions.
MIT License
208 stars 142 forks source link

Refactor generateLogicDataAndSolutions #1957

Closed ellosel closed 1 month ago

ellosel commented 1 month ago

The goal of this PR is to refactor the logic associated with the generateLogicDataAndSolutions into smaller testable functions with documentation. This will enable finer grained profile results that will help identify precisely where performance bottlenecks occur. Further, by testing the new functions, code coverage will improve and provide stability in the presence of future changes.

In addition to refactoring the function, we eliminated the need to carry multiple master solution libraries (i.e. masterLibraries and fullMasterLibrary) forward in TensileCreateLibrary which simplifies logic in downstream code.