biondizzle 97656a5cd1 Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong ...

Key fixes:
- PipelineUmmaAsync consumer group: 32*4=128 threads (not 4 warps)
- TMEM offsets computed from find_tmem_tensor_col_offset (not hardcoded)
- P fragment from p_tmem_s.outer + make_fragment_A (matching fmha.py)
- V SMEM aliasing via recast_ptr

Status:
- Stage A: cosine 0.999999 ✅
- Stage B: runs without crash, identity softmax cosine -0.02 ❌
- Diagnostics: TMEM layout inspection, bisection results

2026-05-20 20:26:25 +00:00

12 KiB

Raw Blame History

View Raw