New dynamic mode to run large problems at slightly reduced CU count if it improves work division and power. New mode can be enabled by setting environment variable TENSILE_STREAMK_DYNAMIC_GRID=2.
Option is still being benchmarked and evaluated for best use, but initial tests indicate this option should improve stream-k kernel performance on gfx942.
New dynamic mode to run large problems at slightly reduced CU count if it improves work division and power. New mode can be enabled by setting environment variable TENSILE_STREAMK_DYNAMIC_GRID=2. Option is still being benchmarked and evaluated for best use, but initial tests indicate this option should improve stream-k kernel performance on gfx942.