biondizzle/nvfp4-megamoe-kernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
1,976 Commits 1 Branch 17 Tags
cb2ca8591fb87dce58b5cd4e1ff3a4e36963efdf
Commit Graph

491 Commits

Author SHA1 Message Date
biondizzle
14524295c7 SMEM-P: add debug_swap_mn flag to test swapped coordinate mapping 2026-05-23 20:10:39 +00:00
biondizzle
aafbe16a81 SMEM-P: add debug_p_one flag to write constant P=1.0 2026-05-23 20:09:47 +00:00
biondizzle
75b3d0f4b4 auto: pre-test commit 2026-05-23 20:08:31 +00:00
biondizzle
b551287123 SMEM-P: implement two-phase softmax with normalization before SMEM write 2026-05-23 20:08:29 +00:00
biondizzle
c88ec626e0 Revert TMEM layout change (caused error) 2026-05-23 20:07:02 +00:00
biondizzle
bb7f53702f SMEM-P: try using PV A-operand layout directly for TMEM-P 2026-05-23 20:06:23 +00:00
biondizzle
cbb85cf091 SMEM-P: debug compute offset for known coordinate 2026-05-23 20:05:02 +00:00
biondizzle
64aa69c421 SMEM-P: add crd2idx debug attempt 2026-05-23 20:04:28 +00:00
biondizzle
1854fa19d5 SMEM-P: test pattern based on fragment indices (k,j) 2026-05-23 20:03:22 +00:00
biondizzle
f21167f029 SMEM-P: add shape debug prints 2026-05-23 20:02:32 +00:00
biondizzle
d2d0eec33a SMEM-P: add debug prints for coordinates 2026-05-23 20:00:33 +00:00
biondizzle
8edf2d434c SMEM-P: implement CUTLASS LLM fixes - dynamic frg_tile, local coordinate conversion 2026-05-23 19:59:52 +00:00
biondizzle
2b51837465 SMEM-P: try transposed mapping (swap m/n) 2026-05-23 19:53:42 +00:00
biondizzle
01fe51b175 SMEM-P: current state - working but mapping wrong (cos 0.02) 2026-05-23 19:53:25 +00:00
biondizzle
3d044b4747 SMEM-P: debug with linear index pattern m*128+n 2026-05-23 19:52:46 +00:00
biondizzle
81630037bd SMEM-P: debug with test pattern (k+j)*0.01 2026-05-23 19:52:02 +00:00
biondizzle
1eb0c1b47a SMEM-P: fix coordinate access - qk_coord is (m,n) not ((m,n),0,0) 2026-05-23 19:38:11 +00:00
biondizzle
dd42245478 SMEM-P: fix scoping - define tTMEM_LOADcS_frg unconditionally 2026-05-23 19:37:34 +00:00
biondizzle
e09c8057be SMEM-P: implement full 128-value write in softmax loop using coordinate mapping 2026-05-23 19:36:56 +00:00
biondizzle
e118ad967d SMEM-P: fix BF16 value creation (use constant) 2026-05-23 19:33:29 +00:00
biondizzle
58639aa634 SMEM-P: implement CUTLASS LLM coordinate mapping pattern (minimal test) 2026-05-23 19:32:11 +00:00
biondizzle
dfe1d3803b SMEM-P: fix thread_idx tuple access 2026-05-23 19:30:09 +00:00
biondizzle
ec84d35cf8 SMEM-P: add debug to understand thread partitioning 2026-05-23 19:29:27 +00:00
biondizzle
e0407793af SMEM-P: implement simple test pattern instead of coord lookup 2026-05-23 19:21:31 +00:00
biondizzle
9b72411ca7 Start implementing manual SMEM-P addressing (helpers are a trap) 2026-05-23 19:20:40 +00:00
biondizzle
e0a2d272f4 Implement manual SMEM-P copy instead of cute.copy (helpers are a trap) 2026-05-23 19:14:44 +00:00
biondizzle
341527977c Try flattening sP and rP_bf16_qk with group_modes to fix rank mismatch 2026-05-23 19:13:59 +00:00
biondizzle
c9448aca03 Add debug prints for SMEM-P partition layouts 2026-05-23 19:13:07 +00:00
biondizzle
b4b11db0fa Fix SMEM-P: use BF16 copy atom and BF16 source with QK C-fragment layout 2026-05-23 19:12:13 +00:00
biondizzle
7b65adf7a3 Fix SMEM-P copy: use tcgen05.copy.St32x32bOp with Float32 and copy from rP_words (Float32) not rP_bf16 2026-05-23 19:11:08 +00:00
biondizzle
4bf3c435b5 Fix rP scope issue: use rP_bf16.iterator instead of rP.iterator 2026-05-23 09:36:22 +00:00
biondizzle
431b8e0abe Fix duplicate else: line in SMEM-P block 2026-05-23 09:35:47 +00:00
biondizzle
d2bb02a331 SMEM-P: Use QK C-fragment layout instead of TMEM layout to fix rank mismatch 2026-05-23 09:35:24 +00:00
biondizzle
63c9a5ce82 Fix sP_2d definition for tSMEM_CPYsP 2026-05-23 09:34:50 +00:00
biondizzle
c7006b0969 Remove debug print lines referencing deleted sP_2d 2026-05-23 09:34:09 +00:00
biondizzle
b09c432942 Remove duplicate sP_2d line causing indentation error 2026-05-23 09:33:40 +00:00
biondizzle
518dce37f0 SMEM-P: Implement rank mismatch fix by reshaping source tensor 2026-05-23 09:33:24 +00:00
biondizzle
c9dda47971 Add more debug prints for sP shapes 2026-05-23 09:26:30 +00:00
biondizzle
2283de1cfc Add debug prints to SMEM-P path to understand rank mismatch 2026-05-23 09:25:48 +00:00
biondizzle
7c350e6a18 Fix SMEM-P copy rank mismatch (use rP_bf16 directly instead of group_modes) 2026-05-23 09:21:13 +00:00
biondizzle
cb2849bff5 D1.3: Implement SMEM-P path (write P to SMEM via tiled_smem_copy instead of zeroing sP) 2026-05-23 09:20:37 +00:00
biondizzle
2c36cd0d32 Stage D1: Multi-PV-tile support for hd>256 (tcgen05 MMA max N=256) 2026-05-23 09:04:01 +00:00
biondizzle
f556060ddf Fix v_fmha layout to use pv_n_tile instead of head_dim for multi-PV-tile support 2026-05-23 09:02:01 +00:00
biondizzle
f1ad264da6 D1.4: Add pv_n_tile and n_pv_tiles for multi-PV-tile support (tcgen05 MMA max N=256) 2026-05-23 09:00:18 +00:00
biondizzle
73fa8a2b70 shit carmine left dangling 2026-05-23 06:55:22 +00:00
biondizzle
5012703bad fix: add SwiGLU clamping to fused kernel (paper §4.2.3, CG-1) 
The fused SwiGLU kernel stored swiglu_limit but never applied it.
Paper §4.2.3: gate capped at swiglu_limit, linear clamped to [-limit, +limit].
Non-fused reference path already applies clamping correctly.
Fix: add fmin/fmax clamping in FP32 before BF16 conversion.
 2026-05-23 06:32:54 +00:00
biondizzle
580d2f6999 STAGE_D.md: restructure with correctness gaps, TMEM budget, execution order 2026-05-23 06:31:37 +00:00
biondizzle
df43c3232d D1.1: Fix make_fragment_A — use sP for SMEM source pv_mma 2026-05-23 06:04:44 +00:00
biondizzle
80434d0284 D1.1: Fix PV A-operand construction — compile-time branch for TMEM vs SMEM 2026-05-23 06:03:27 +00:00
biondizzle
d36b727898 D1.1: Add SMEM-P path behind use_smem_p flag (stub: zero sP) 2026-05-23 06:01:02 +00:00