|
|
fd6b1e82d8
|
TMA shape diag: pure Python, no JIT
|
2026-05-22 21:55:03 +00:00 |
|
|
|
2f670e33d1
|
TMA shape diagnostic: exact setup from example9 + shape prints
|
2026-05-22 21:54:22 +00:00 |
|
|
|
b64227e5b6
|
Fix group_modes range in TMA shape diag
|
2026-05-22 21:53:22 +00:00 |
|
|
|
be27720cb2
|
Add TMA shape diagnostic
|
2026-05-22 21:52:51 +00:00 |
|
|
|
845ad98b22
|
Fix TMA indexing: 4-mode tensors, kt at mode 2 (GMEM tile dim)
|
2026-05-22 21:51:33 +00:00 |
|
|
|
61b0501a8b
|
Fix test_fmha_v3_stage_c.py: 8-mode TMA indexing + O rescale (from example9)
|
2026-05-22 21:50:35 +00:00 |
|
|
|
0996ffc1ba
|
Add fmha_v3_stage_c_example10: 8-mode TMA + O rescale + paired-atom epilogue
|
2026-05-22 21:43:11 +00:00 |
|
|
|
328f9b0080
|
Test n=384
|
2026-05-22 21:36:22 +00:00 |
|
|
|
2da0a452d1
|
Quick test n=128,256
|
2026-05-22 21:36:00 +00:00 |
|
|
|
0d3caced47
|
Add: O rescale (correction_rescale) in softmax loop + remove pk from TMA/MMA
|
2026-05-22 21:35:39 +00:00 |
|
|
|
c47d229e6a
|
Sweep test
|
2026-05-22 21:32:18 +00:00 |
|
|
|
a751b3baf7
|
Sweep test: n=128,256,384,512,1024
|
2026-05-22 21:31:15 +00:00 |
|
|
|
beaf60db5c
|
DOCUMENT: TMA 8-mode indexing — the bug that cost us a full day. README + inline comments.
|
2026-05-22 21:28:58 +00:00 |
|
|
|
27116110ab
|
Fix identity diag: same 8D TMA indexing fix
|
2026-05-22 21:21:52 +00:00 |
|
|
|
bb92af5b0c
|
FIX: Use full 8D indexing for tBgK/tVgV — mode 4 is the GMEM tile dim
|
2026-05-22 21:21:23 +00:00 |
|
|
|
2a9f764f8b
|
Diagnostic: check tBgK/tVgV layout strides for degenerate dims
|
2026-05-22 21:20:46 +00:00 |
|
|
|
ae173d3963
|
Test identity diag multi-tile
|
2026-05-22 21:14:42 +00:00 |
|
|
|
a35fb1b077
|
Minimal reference FMHA test: n=256 only
|
2026-05-22 21:13:37 +00:00 |
|
|
|
4fc264e034
|
Test reference FMHA with proper API
|
2026-05-22 21:12:49 +00:00 |
|
|
|
a800a83d5c
|
Test: CUTLASS reference FMHA on B200 multi-tile
|
2026-05-22 21:11:58 +00:00 |
|
|
|
2a14c2dd18
|
REVERT to working baseline (n=128 cos 0.999998). Multi-tile TMA is a CuTeDSL JIT limitation.
|
2026-05-22 20:37:21 +00:00 |
|
|
|
1ab326f2d2
|
Test: use kvh.index (pipeline state) as TMA GMEM coordinate
|
2026-05-22 20:36:21 +00:00 |
|
|
|
7b8b022e23
|
SMEM counter: separate allocate_tensor instead of struct field
|
2026-05-22 20:35:42 +00:00 |
|
|
|
462778efcf
|
Fix SMEM counter type: cutlass.Int32 for MemRange
|
2026-05-22 20:35:17 +00:00 |
|
|
|
f5c827d0b9
|
SMEM-backed kv_coord counter — JIT can't constant-fold SMEM reads
|
2026-05-22 20:34:52 +00:00 |
|
|
|
215282971c
|
DEBUG: hardcoded Int32(1) to test if TMA can read tile 1
|
2026-05-22 20:34:21 +00:00 |
|
|
|
79ebe20a39
|
DEBUG: use Int32(kt) directly to test if coordinate matters
|
2026-05-22 20:34:03 +00:00 |
|
|
|
b3778896b9
|
Test: kv_coord = warp_idx() * 0 — force SSA from runtime value
|
2026-05-22 20:33:40 +00:00 |
|
|
|
1de848c5ca
|
DEBUG: add cute.printf for kv_coord runtime value
|
2026-05-22 20:33:03 +00:00 |
|
|
|
587c16679c
|
Test: Python range() instead of cutlass.range() for TMA loop
|
2026-05-22 20:32:44 +00:00 |
|
|
|
91230fe5e6
|
Test example9: drop try_acquire/pk, single loop-carried kv_coord
|
2026-05-22 20:32:25 +00:00 |
|
|
|
3c0451a3e5
|
REVERT to working example7 (n=128 cos 0.999998). Example8 TMA fix didn't work.
|
2026-05-22 20:28:15 +00:00 |
|
|
|
880bd9ef81
|
Update stage_c test to example8: SSA kv_coord + per-tile O rescale
|
2026-05-22 20:27:58 +00:00 |
|
|
|
c395b279d2
|
Clean up tests: archive superseded files, keep only essential unit tests
Kept in tests/unit/:
- test_fmha_v3.py (stages A+B)
- test_fmha_v3_diag.py (identity softmax, n=128+256)
- test_fmha_v3_stage_c.py (real softmax, n=128 cos 0.999998)
- layertest.py + cudagraph_test.py (required for every change)
- infrastructure: cache, custom_op, cutedsl, router, fp4, fused, interleave
Archived: 19 superseded unit tests + 10 root-level scratch files
Root level: only fmha_v3_stage_c_example7.py remains (now in unit/)
|
2026-05-22 20:25:27 +00:00 |
|
|
|
a54a241052
|
Revert TMA to kt pattern (n=128 works), multi-tile TMA is separate bug
|
2026-05-22 20:10:35 +00:00 |
|
|
|
956ca1ecfd
|
TMA: use self.n_kv_tiles + kv_coord pattern from working diag test
|
2026-05-22 20:08:07 +00:00 |
|
|
|
242ffebcd9
|
REVERT to 0bdcdc0 — the version that passed n=128 cos 0.999998
|
2026-05-22 20:07:18 +00:00 |
|
|
|
3814838107
|
DEBUG: disable O rescale + normalize, test if corr setup alone causes regression
|
2026-05-22 19:57:53 +00:00 |
|
|
|
2c51f82382
|
Shared corr tensors for O rescale + final normalize, fix softmax loop
|
2026-05-22 19:55:06 +00:00 |
|
|
|
f165257c50
|
Add O rescale with correction_rescale pattern + fix TMA to working diag pattern
|
2026-05-22 19:51:53 +00:00 |
|
|
|
0bdcdc0efd
|
O normalize: exact CUTLASS correction_rescale pattern with 2D reg tensor
|
2026-05-22 19:50:37 +00:00 |
|
|
|
009cf9f80d
|
O normalize: TMEM round-trip with paired Ld/St atoms + standard epilogue_tma_store
|
2026-05-22 19:49:51 +00:00 |
|
|
|
f821dd00fe
|
Fix: use NamedBarrier instead of mbarrier_arrive/wait
|
2026-05-22 19:49:07 +00:00 |
|
|
|
3bd406e925
|
Fix: barrier_wait → mbarrier_wait, barrier_arrive → mbarrier_arrive
|
2026-05-22 19:48:31 +00:00 |
|
|
|
544f0ca52b
|
Fix epilogue: corr_tile_size=16, proper epi_subtile tuple, match CUTLASS reference
|
2026-05-22 19:47:57 +00:00 |
|
|
|
d526aa04fb
|
Fix example7: K slice (None,None,0,0) and softmax scale_log2 double-bug
|
2026-05-22 19:45:59 +00:00 |
|
|
|
558aac0581
|
Fix: fence_async_shared -> fence_view_async_shared
|
2026-05-22 19:25:05 +00:00 |
|
|
|
d62e6fc9ca
|
Clean v2: real softmax P, no O TMEM modify, standard epilogue. Baseline for custom epilogue work.
|
2026-05-22 19:24:25 +00:00 |
|
|
|
183292d919
|
O normalize using tmem_ptr base (same as epilogue) + CUTLASS sub-tile pattern
|
2026-05-22 19:18:16 +00:00 |
|
|
|
365e8f53af
|
O normalize with full layout (no sub-tiling), Repetition(64)
|
2026-05-22 19:17:03 +00:00 |
|
