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

Try kt (cutlass.range induction) with correct (None,0,None,0) pre-slice

Browse Source
This commit is contained in:
biondizzle
2026-05-22 22:27:58 +00:00
parent d656598185
commit be103deb6d
1 changed files with 2 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
tests/unit/test_fmha_v3_stage_c.py
View File

@@ -214,17 +214,10 @@ class FmhaV3StageCMulti:
cute.copy(tma_q, tAgQ[(None, Int32(0))], tAsQ[(None, qh.index)], tma_bar_ptr=qh.barrier)
qp.tail()
kvp.reset(); pk = kvp.try_acquire()
# CUTLASS reference initializes kv_coord from a runtime parameter
# (seqlen_kv_loop_start), not a Python literal. A Python 0 gets
# constant-folded by the JIT, so kv_coord += 1 never propagates.
# n_kv_tiles - n_kv_tiles is a runtime expression that evaluates to 0
# but the JIT can't fold it, forcing kv_coord to be a tracked SSA reg.
kv_coord = n_kv_tiles - n_kv_tiles
for kt in cutlass.range(0, n_kv_tiles, 1, unroll=1):
kvh = kvp.acquire_and_advance(pk)
cute.copy(tma_k, tBgK[(None, kv_coord)], tBsK[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
cute.copy(tma_v, tVgV[(None, kv_coord)], tVsV[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
kv_coord = kv_coord + 1
cute.copy(tma_k, tBgK[(None, kt)], tBsK[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
cute.copy(tma_v, tVgV[(None, kt)], tVsV[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
pk = cutlass.Boolean(1)
kvp.tail()