biondizzle/nvfp4-megamoe-kernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

D1.4: Fix merge test - use use_smem_p=False for hd=256 kernel (SMEM budget)

Browse Source
This commit is contained in:
biondizzle
2026-05-24 16:36:48 +00:00
parent d70f083e17
commit bd08bfee8e
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/unit/test_d1_hd512_merge.py
View File

@@ -37,7 +37,7 @@ def test():
# Use the hd=256 kernel (no k_sub path) with k_tile=256
# Call once per k_sub tile, merge results via online softmax
kernel = FmhaKernel(head_dim=k_tile, s_k=n, normalize=False)
kernel = FmhaKernel(head_dim=k_tile, s_k=n, use_smem_p=False, normalize=False)
pv_n_tile = kernel.pv_n_tile
n_pv_tiles = kernel.n_pv_tiles
stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream)