Fix: PersistentTileSchedulerParams cluster_shape must be Python ints not MLIR values
This commit is contained in:
@@ -308,7 +308,7 @@ class Nvfp4FusedRouterKernel:
|
||||
|
||||
tile_sched_params = utils.PersistentTileSchedulerParams(
|
||||
(cutlass.Int32(num_M_tiles), cutlass.Int32(num_N_tiles), cutlass.Int32(1)),
|
||||
(cutlass.Int32(self.cluster_shape_mn[0]), cutlass.Int32(self.cluster_shape_mn[1]), cutlass.Int32(1)))
|
||||
(1, 1, 1))
|
||||
|
||||
self._kernel(
|
||||
tiled_mma, tiled_mma_sfb,
|
||||
|
||||
Reference in New Issue
Block a user