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

Commit Graph

  • cd6d81fc4b merge: resolve conflict (keep our version) biondizzle 2026-05-24 02:44:09 +00:00
  • 6a909e16d0 D1.5: Replace broken make_cotiled_copy SMEM-P with coordinate-indexed store biondizzle 2026-05-24 02:43:42 +00:00
  • b67668d2bd feat: SMEM-P with make_tiled_copy_tv + manual fill biondizzle 2026-05-24 02:43:12 +00:00
  • fce9a7f4be feat: SMEM-P using make_tiled_copy_tv + logical sP view biondizzle 2026-05-24 02:42:20 +00:00
  • 29231fc75e Merge branch 'master' of ssh://sweetapi.com:2222/biondizzle/nvfp4-megamoe-kernel biondizzle 2026-05-24 02:41:39 +00:00
  • 3803847900 D1.5: Use tCtO_fake layout for epilogue_tma_store (needs STAGE dim) biondizzle 2026-05-24 02:41:32 +00:00
  • 688ca6f737 feat: SMEM-P using make_cotiled_copy (one-row-per-thread) biondizzle 2026-05-24 02:40:38 +00:00
  • 34db0bf802 D1.5: Remove duplicate tTMrO definition (keep unconditional one) biondizzle 2026-05-24 02:40:40 +00:00
  • 093de4e62f merge biondizzle 2026-05-24 02:40:14 +00:00
  • bb4c35facb D1.5: Move tTMrO after O rescale atoms (fix tTMEM_LOADcO reference) biondizzle 2026-05-24 02:39:18 +00:00
  • af8303ba64 fix: reorder tTMrO definition after tTMEM_LOADcO biondizzle 2026-05-24 02:36:08 +00:00
  • e41cf07f50 fix: tTMrO scoping + restore SMEM-P coordinate write biondizzle 2026-05-24 02:34:50 +00:00
  • bfd598b937 D1.5: Move tTMrO def before softmax loop (CuTeDSL scoping) biondizzle 2026-05-24 02:32:39 +00:00
  • 1e8f784bb0 debug: zero-fill sP to check deadlock biondizzle 2026-05-24 02:31:10 +00:00
  • ece137b2c2 D1.5: Use proven Stage C approach - normalize via TMEM round-trip + epilogue_tma_store biondizzle 2026-05-24 02:30:20 +00:00
  • 8010e3dda2 test: add try/except for SMEM-P coord test biondizzle 2026-05-24 02:15:07 +00:00
  • 16bade9e10 test: use FmhaKernel for SMEM-P coord test biondizzle 2026-05-24 01:59:25 +00:00
  • f2d95da4aa test: SMEM-P coordinate verification test biondizzle 2026-05-24 01:58:32 +00:00
  • 394f08601a fix: proper v_major from tensor biondizzle 2026-05-24 01:55:37 +00:00
  • 384de55dfd fix: COL_MAJOR biondizzle 2026-05-24 01:54:30 +00:00
  • 313f2c77cd fix: hardcode v_major for diag test biondizzle 2026-05-24 01:54:20 +00:00
  • 268265ba0b diag: use FmhaKernel setup for cotiled test biondizzle 2026-05-24 01:54:08 +00:00
  • 32d8483443 fix: v_major derivation in cotiled diag biondizzle 2026-05-24 01:53:40 +00:00
  • ce1e2177ec diag: simplified cotiled layout test biondizzle 2026-05-24 01:53:24 +00:00
  • f6e3bddb10 fix: LayoutEnum import in cotiled diag test biondizzle 2026-05-24 01:52:58 +00:00
  • 67a2c3ee72 diag: layout composition test for make_cotiled_copy SMEM-P biondizzle 2026-05-24 01:48:42 +00:00
  • 699c646497 D1.5: Fix bSG_gC slicing - group trailing modes (CUTLASS pattern) biondizzle 2026-05-24 01:41:52 +00:00
  • f2ab5790e8 D1.5: Dynamic slicing for tTR_gC (variable rest dims) biondizzle 2026-05-24 01:40:44 +00:00
  • d28257185f D1.5: Fix flat_divide slice coordinates (4 modes, no STAGE dim) biondizzle 2026-05-24 01:39:21 +00:00
  • 86971dda81 D1.5: Rewrite correction epilogue using CUTLASS pattern (transform_partitioned, flat_divide, paired atoms) biondizzle 2026-05-24 01:37:53 +00:00
  • ec250eccd6 D1.5: Fix TMA store - group_modes on bSG_gC, use flat indexing biondizzle 2026-05-24 01:36:01 +00:00
  • 61c4e107da D1.5: Fix TMA store - use flat_divide on tCgC instead of local_tile on mC biondizzle 2026-05-24 01:35:10 +00:00
  • 6de0d316ef D1.5: Fix TMA store - use 3D tile for local_tile on 3D mC biondizzle 2026-05-24 01:20:33 +00:00
  • f59fd07ba7 D1.5: Use group_modes on sC for 2D TMA view (preserves swizzle) biondizzle 2026-05-24 00:52:57 +00:00
  • 577066bb7f D1.5: Use 2D sC_epi layout from c_smem_s for TMA partition biondizzle 2026-05-24 00:51:18 +00:00
  • 534327f31b D1.5: Fix TMA store - use group_modes on sC and tCgC biondizzle 2026-05-24 00:48:18 +00:00
  • 8048aa4be6 D1.5: Simplify TMA store - use 2D sC_epi and gC_epi views biondizzle 2026-05-24 00:46:52 +00:00
  • 88a5f55120 D1.5: Fix TMA store - use existing gC partition biondizzle 2026-05-24 00:43:35 +00:00
  • b9e91853fb D1.5: Implement correction epilog with paired atoms (get_tmem_load_op + get_smem_store_op) biondizzle 2026-05-24 00:41:27 +00:00
  • e632490682 D1.5: Add TODO for correction epilog - keeping working TMEM round-trip for now biondizzle 2026-05-24 00:37:36 +00:00
  • 9f88db897f D1.5: Revert to pre-epilog backup - correction epilog refactor is complex, will do incrementally biondizzle 2026-05-24 00:35:00 +00:00
  • 501ba7b2a5 D1.5: Fix TMA store - use local_tile with pv_mma_tiler biondizzle 2026-05-24 00:32:35 +00:00
  • d028088dd0 D1.5: Fix TMA store rank mismatch - use 2D sC_epi view biondizzle 2026-05-24 00:31:45 +00:00
  • ea5662ab2b D1.5: Implement correction epilog with get_tmem_load_op + get_smem_store_op paired atoms biondizzle 2026-05-24 00:30:38 +00:00
  • 90131da010 D1.5: Replace TMEM round-trip normalize with correction epilog (one-way: TMEM→reg→SMEM→GMEM) biondizzle 2026-05-24 00:24:24 +00:00
  • e23119ce7e D1.3: Fix LSE tensor layout for weakly congruent store biondizzle 2026-05-24 00:16:22 +00:00
  • d4aeb4e41c D1.3: Add unnormalized debug test to isolate SMEM-P vs O round-trip error biondizzle 2026-05-24 00:15:41 +00:00
  • b2a583a2a8 D1.3: Add SMEM-P write/read diagnostic biondizzle 2026-05-24 00:13:28 +00:00
  • 0fc6530f3f D1.3: Add SMEM-P vs TMEM-P comparison test biondizzle 2026-05-24 00:10:18 +00:00
  • d56e5601bb D1.3: Fix while loop in cotiled diag - precompute num_tmem_alloc_cols biondizzle 2026-05-24 00:07:22 +00:00
  • 7bd857e5d3 D1.3: Fix cotiled diagnostic - use proper MMA construction biondizzle 2026-05-24 00:06:50 +00:00
  • 3fe17495ee D1.3: Add make_cotiled_copy diagnostic test biondizzle 2026-05-24 00:05:48 +00:00
  • ed539653d0 shit left dangling biondizzle 2026-05-23 23:58:57 +00:00
  • 0286d57631 D1.3: Re-enable coordinate-indexed SMEM-P write with identity tensor coords biondizzle 2026-05-23 23:26:46 +00:00
  • 3118bec420 D1.3: Revert to zero-fill for sP - need to verify sP→PV pipeline first biondizzle 2026-05-23 23:26:07 +00:00
  • 0963d949c4 D1.3: Compute (m,k) directly from thread mapping instead of identity tensor biondizzle 2026-05-23 23:24:54 +00:00
  • fca9652719 D1.3: Add debug prints for SMEM-P coordinate mapping biondizzle 2026-05-23 23:24:02 +00:00
  • de869c01c8 D1.3: Add SMEM-P coordinate diagnostic test biondizzle 2026-05-23 23:23:05 +00:00
  • 89393b8352 D1.3: Fix coord extraction - identity tensor stores (m,k) pairs as values biondizzle 2026-05-23 23:21:15 +00:00
  • f74fd75054 D1.3: Fix coordinate indexing - tTMEM_LOADcS first mode is (32,1) nested tuple biondizzle 2026-05-23 23:20:12 +00:00
  • 4b8970d83c D1.3: Direct coordinate-indexed SMEM-P write using tTMEM_LOADcS coords biondizzle 2026-05-23 23:19:21 +00:00
  • 58b4537741 D1.3: Use make_cotiled_copy for SMEM-P — custom TV layout from TMEM-load coords to sP biondizzle 2026-05-23 23:17:30 +00:00
  • 63f662edb3 Add SMEM-P guidance request document for CUTLASS LLM consultation biondizzle 2026-05-23 23:03:35 +00:00
  • 48c1e6d7e9 D1.3: Use const_expr for lse None check biondizzle 2026-05-23 22:30:55 +00:00
  • 82179f1f61 D1.3: Fix LSE with const_expr, always create valid mLSE tensor biondizzle 2026-05-23 22:30:14 +00:00
  • f1341ad76e D1.3: Try make_tiled_copy_C(qk_mma) for SMEM-P copy - zero-fill source for compile test biondizzle 2026-05-23 22:29:10 +00:00
  • bafcfa658f D1.3: Define SMEM-P copy atoms unconditionally (CuTeDSL scoping) biondizzle 2026-05-23 22:28:12 +00:00
  • 8d226a6243 D1.3: Use full sP (4D) for make_tiled_copy_D partition biondizzle 2026-05-23 22:27:11 +00:00
  • fa2e513168 D1.3: SMEM-P via get_smem_store_op + make_tiled_copy_D biondizzle 2026-05-23 22:26:09 +00:00
  • d10cab7a8e D1.3: Enhanced diagnostic - test QK C-fragment as source for make_tiled_copy_C biondizzle 2026-05-23 22:24:15 +00:00
  • a90fe41b6b D1.3: Skip fragment creation in diagnostic, just print layouts biondizzle 2026-05-23 22:21:31 +00:00
  • b871e6874b D1.3: Fix diagnostic - use dummy ptr 0 for shape analysis biondizzle 2026-05-23 22:20:16 +00:00
  • 0b3d3bcd2b D1.3: Fix sP allocation - p_smem_s.outer is already a layout biondizzle 2026-05-23 22:19:11 +00:00
  • fc4106f37e D1.3: Fix layout diagnostic - compute c_major outside kernel biondizzle 2026-05-23 22:17:54 +00:00
  • 6afa610b0d D1.3: Layout diagnostic v2 - run inside JIT-compiled kernel biondizzle 2026-05-23 22:16:57 +00:00
  • b185ac2080 D1.3: Fix layout diagnostic - remove JIT-dependent code biondizzle 2026-05-23 22:15:47 +00:00
  • d265264f74 D1.3: Layout diagnostic - print all QK C-fragment and PV A-operand shapes biondizzle 2026-05-23 22:14:35 +00:00
  • 98e5b48470 Update all .md files with D5a/D5b progress, tOrP0 fix, LSE formula biondizzle 2026-05-23 22:07:53 +00:00
  • 53efb0c95e Update STAGE_D.md with D5b results: merge cos 0.961, LSE err=0.0 biondizzle 2026-05-23 21:45:22 +00:00
  • b1152acd88 D5b: Fix reference computation - use logsumexp for stable LSE, fix o_unnorm definition biondizzle 2026-05-23 21:43:04 +00:00
  • 4ed2b46020 D5b MILESTONE: SWA+sink merge works! cos 0.969 biondizzle 2026-05-23 21:36:26 +00:00
  • b77ad244a2 D5b: Use normalized O + LSE for merge (correct formula), always output LSE biondizzle 2026-05-23 21:35:40 +00:00
  • 84200ca557 D5b: Clean up merge test - stable formula for both ref and kernel biondizzle 2026-05-23 21:33:45 +00:00
  • 909f880cc2 D5b: Use reference per-row LSE for proper O normalization biondizzle 2026-05-23 21:31:52 +00:00
  • fef7e90c0a D5b: Fix kernel_obj reference biondizzle 2026-05-23 21:30:59 +00:00
  • d245342a7a D5b: Fix syntax error biondizzle 2026-05-23 21:30:00 +00:00
  • 4a8fd64c4e D5b: Debug reference formula mismatch, add numerically stable merge biondizzle 2026-05-23 21:19:25 +00:00
  • e0201e581d D5b: Python SWA+sink merge test biondizzle 2026-05-23 21:18:06 +00:00
  • a629babb6a Update STAGE_D.md: D5a done, CG-2/CG-3 status updated, tOrP0 offset rule added biondizzle 2026-05-23 21:16:52 +00:00
  • aff208fb4c D5a: Fix LSE formula - lse = ln(row_sum) + row_max * ln(2) biondizzle 2026-05-23 21:15:14 +00:00
  • a5061a24b9 D5a: Use tensor indexing for LSE write biondizzle 2026-05-23 21:13:52 +00:00
  • 7a87c634fb D5a: Use cute.store for LSE write biondizzle 2026-05-23 21:13:07 +00:00
  • 7c38bd5522 D5a: Fix LSE - compute row_max_safe from final row_max, remove mLSE None check biondizzle 2026-05-23 21:12:29 +00:00
  • aecf6671d9 D5a: Fix - add normalize param to __init__ biondizzle 2026-05-23 21:11:37 +00:00
  • b273d4a7e8 D5a: Add normalize flag + LSE output biondizzle 2026-05-23 21:10:40 +00:00
  • aa72779a7a D1.3: Use const_expr if for tOrP0 compile-time selection biondizzle 2026-05-23 21:07:58 +00:00
  • 61cb8f002d D1.3: Pre-compute tOrP0_offset in _setup, use const_expr for compile-time selection biondizzle 2026-05-23 21:07:10 +00:00
  • 77035f8a54 D1.3: Use const_expr for tOrP0 offset (compile-time conditional) biondizzle 2026-05-23 21:06:16 +00:00
  • 2e0db67c80 D1.3: Use MLIR-compatible expression for tOrP0 offset (same as Stage C) biondizzle 2026-05-23 21:05:12 +00:00
  • 399a241e84 D1.3: Initialize tOrP0 before conditional for CuTeDSL scoping biondizzle 2026-05-23 21:03:53 +00:00