Closed vgokhale closed 1 month ago
Changed all block pointers to tensor pointers. Also puggybacked some minor code cleanup.
This change was necessary because block pointers are not well supported currently for non-TMA things.
All tests pass.
Performance before / after delta < 1%.
Changed all block pointers to tensor pointers. Also puggybacked some minor code cleanup.
This change was necessary because block pointers are not well supported currently for non-TMA things.
All tests pass.
Performance before / after delta < 1%.