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

SMEM-P: test permutation 1 (swap m↔n0)

Browse Source
This commit is contained in:
biondizzle
2026-05-23 20:14:21 +00:00
parent d5081fe6f0
commit 5a11f7c09a
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
dsv4/kernels/attention/fmha.py
View File

@@ -24,7 +24,7 @@ class FmhaKernel:
self.use_smem_p = use_smem_p if use_smem_p is not None else (head_dim > 64)
self.debug_p_one = False # DEBUG: write constant P=1.0 to verify mapping
self.debug_swap_mn = False # DEBUG: try swapping m and n0 in coordinate mapping
self.debug_permute = 0 # DEBUG: try different coordinate permutations
self.debug_permute = 1 # DEBUG: try different coordinate permutations (0=original,1=swap m↔n0)
self.acc_dtype = Float32; self.qk_acc_dtype = Float32
self.q_dtype = BFloat16; self.o_dtype = BFloat16; self.c_dtype = BFloat16
self.use_2cta_instrs = False; self.epilog_sync_bar_id = 1