diff --git a/tests/unit/test_fmha_v3_stage_c.py b/tests/unit/test_fmha_v3_stage_c.py index 07e4fa19..37ba7d3d 100644 --- a/tests/unit/test_fmha_v3_stage_c.py +++ b/tests/unit/test_fmha_v3_stage_c.py @@ -181,6 +181,10 @@ class FmhaV3StageCMulti: tVsV,tVgV = cpasync.tma_partition(tma_v,0,b_lay,cute.group_modes(sV,0,3),cute.group_modes(tCgV,0,3)) tAgQ = tAgQ[(None,0,None,0)]; tBgK = tBgK[(None,0,None,0)]; tVgV = tVgV[(None,0,None,0)] + # DEBUG: print shapes and strides to verify kv_tiles dimension has nonzero stride + print(f"tBgK pre-sliced shape: {cute.shape(tBgK)} layout: {tBgK.layout}") + print(f"tVgV pre-sliced shape: {cute.shape(tVgV)} layout: {tVgV.layout}") + tCrQ = qk_mma.make_fragment_A(sQ); tCrK = qk_mma.make_fragment_B(sK) tCrV = pv_mma.make_fragment_B(sV)