Complete multi-PV-tile fixes: pv_n_tile, v_fmha layout, MMA construction, n_corr_tiles
This commit is contained in:
@@ -19,6 +19,8 @@ class FmhaKernel:
|
||||
self.head_dim = head_dim
|
||||
self.s_k = s_k
|
||||
self.n_kv_tiles = s_k // 128
|
||||
self.pv_n_tile = min(head_dim, 256) # tcgen05 MMA max N=256
|
||||
self.n_pv_tiles = head_dim // self.pv_n_tile
|
||||
|
||||
self.pv_n_tile = min(head_dim, 256) # tcgen05 MMA max N=256
|
||||
self.n_pv_tiles = head_dim // self.pv_n_tile
|
||||
|
||||
Reference in New Issue
Block a user