CUDA: experimental native mxfp4 support for blackwell#17906

Merged

am17an merged 15 commits intoggml-org:masterfrom

Dec 24, 2025

Commits on Dec 20, 2025

CUDA: experimental native mxfp4 support for blackwell
Aman Gupta
committed
optimize load_tiles
Aman Gupta
committed
optimize quantize_mxfp4
Aman Gupta
committed
cleanup
Aman Gupta
committed
first pass review: formatting
Aman Gupta
committed
use interleaved layout for mma
Aman Gupta
committed
mmq: add assert for size
Aman Gupta
committed
use __nv_fp4x4_e2m1
Aman Gupta
committed
use iter_k as 512, cleanup
Aman Gupta
committed
Use 1200 as blackwell instead of 1000
Aman Gupta
committed
address review comments
Aman Gupta
committed
mmq: fix stride
Aman Gupta
committed

Commits on Dec 22, 2025

quantize.cu: use reference impl of e8m0 scale
Aman Gupta
committed
address review comments
Aman Gupta
committed

Commits on Dec 24, 2025

add 120f-virtual + minor fixes
Aman Gupta
committed