Mhc large m - #3651

Merged

valarLip merged 16 commits into

LiuYinfeng01:mhc_large_m

Jun 12, 2026

Commits on Jun 11, 2026

Add gfx950 large-M hybrid path for mhc_fused_post_pre (M>=1024).
LiuYinfeng01
committed
850b3a6
Fix atom docker benchmark to pip install editable aiter.
LiuYinfeng01
committed
da9e52b
Fix mhc_pre_big_fuse_rmsnorm use_nt signature for hybrid compile.
LiuYinfeng01
committed
bfc0387
Add use_nt pybind for mhc_pre_big_fuse_rmsnorm.
LiuYinfeng01
committed
b105b39
Fix large-M hybrid gemm_out padding to match mhc_pre layout.
LiuYinfeng01
committed
d75b173
Add gfx950 large-M benchmark results for mhc_large_m branch.
LiuYinfeng01
committed
f229e1b
Update large-M benchmark results and PR comparison tables.
LiuYinfeng01
committed
0caaebb
Move large-M mhc_post_pre testing to op_tests and drop EvoKernel paths.
LiuYinfeng01
committed
bee37d8
Refactor gfx950 large-M path into additive kernels only.
LiuYinfeng01
committed
203b3c4
Fix Black formatting in mhc op tests.
LiuYinfeng01
committed
9b31ee8
Remove __main__ guards from mhc op tests.
LiuYinfeng01
committed
c9371f2
Drop PR_mhc_large_m.md and restore op_tests/test_mhc.py to upstream.
LiuYinfeng01
committed
c433700
Migrate gfx950 large-M cache policy into upstream kernels.
LiuYinfeng01
committed
5dacd83
Slim large-M path: drop C++ orchestrator, use mhc_post + mhc_pre in Python.
LiuYinfeng01
committed
2c09880
Merge gfx950 large-M test into test_mhc.py with --largeM flag.
LiuYinfeng01
committed
b2581c1
Fix Black formatting in test_mhc.py --largeM logging.
LiuYinfeng01
committed
3abdf5e