biondizzle

0 Followers · 0 Following

• Joined on 2025-12-10

Repositories

25

Projects Packages Public Activity Starred Repositories

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:37:48 +00:00

60cabb186d fix: always provide valid gP tensor

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:37:21 +00:00

49f54aef2d fix: const_expr for SMEM-P tma_p creation

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:36:40 +00:00

6f0475f0db fix: guard tma_p creation with gP check

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:36:13 +00:00

6068077e09 feat: SMEM-P via gP→TMA→sP (gP as caller param)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:35:33 +00:00

3063dbd5da D1: SMEM-P test at hd=128

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:34:24 +00:00

d9eb02fd8b merge

f550c7dd33 D1: SMEM-P coordinate-indexed store with dynamic sP shape extraction

Compare 2 commits »

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:33:43 +00:00

9ea7f511b0 fix: dummy tma_p reuse tma_q for non-SMEM-P

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:32:56 +00:00

00c9991396 merge

ae4c501903 merge: keep our fmha.py

5e154770de D1: SMEM-P using make_tiled_copy_C + retile (correct CUTLASS pattern)

Compare 3 commits »

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:32:17 +00:00

edacdae017 fix: use make_tiled_tma_atom_A for P TMA

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:31:28 +00:00

0f943ab48c feat: SMEM-P via gP→TMA→sP path (register→GMEM→SMEM)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:30:30 +00:00

d0aec403e4 D1: Layout diagnostic for SMEM-P

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:29:15 +00:00

dd7356afc6 D1: Simplified debug test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:28:28 +00:00

1f1b16ad07 D1: Add hd=128 debug test

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:27:15 +00:00

0f52c3453c D1: Fix SMEM-P (coordinate store), LSE (FP32), add TMEM-P-only test

52570d94cb D1: Fix SMEM-P - coordinate-indexed store (replaces make_tiled_copy_C)

Compare 2 commits »

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:23:56 +00:00

70c9d93d28 feat: SMEM-P make_tiled_copy_C + zero-fill dest tensor

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:23:24 +00:00

86b7ff8b36 Merge branch 'master' of ssh://sweetapi.com:2222/biondizzle/nvfp4-megamoe-kernel

e1a5077fda D1: LSE diagnostic at various hd

Compare 2 commits »

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:23:00 +00:00

7cf9704584 feat: SMEM-P using make_tiled_copy_C(qk_mma) approach

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:22:26 +00:00

efe6f4dff8 D1: Add diagnostic test (TMEM-P vs SMEM-P at various hd)

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:21:07 +00:00

847372976f Merge branch 'master' of ssh://sweetapi.com:2222/biondizzle/nvfp4-megamoe-kernel

7142e459cc D1 test: compare un-norm O + norm using ref row_sum + LSE verification

Compare 2 commits »

biondizzle pushed to master at biondizzle/nvfp4-megamoe-kernel 2026-05-24 03:20:28 +00:00

fe47a5f882 fix: LSE type mismatch Float32→BFloat16