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

SMEM-P: debug with test pattern (k+j)*0.01

Browse Source
This commit is contained in:
biondizzle
2026-05-23 19:52:02 +00:00
parent ca1fb7a1ca
commit 111b50b62a
1 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
dsv4/kernels/attention/fmha.py
View File

@@ -359,8 +359,12 @@ class FmhaKernel:
n2 = n // 64
pv_coord = ((m, n0), 0, (n1, n2), 0)
# Convert Float32 → BF16 and write to SMEM
p_val_bf16 = tTMEM_LOADrS_frg[k, j].to(self.q_dtype)
# DEBUG: Write simple test pattern (k+j)*0.01
# This helps verify coordinate mapping
# k and j are loop indices (0-31, 0-3)
pattern_val = Float32(k + j) * Float32(0.01)
p_val_bf16 = pattern_val.to(self.q_dtype)
# Original: p_val_bf16 = tTMEM_LOADrS_frg[k, j].to(self.q_dtype)
sP[pv_coord] = p_val_bf16
row_sum = row_sum + tTMEM_LOADrS_frg[k, j]