Binyang Li 9b42ad3062 Support FP8 NVLS multimem reduction (#826) ...

## Summary
- add FP16 accumulation support for FP8 NVLS multimem load-reduce
- probe and gate FP8 NVLS algorithms on device support
- keep MNNVL algorithm sizing and buffer-pool changes out of this sub-PR

## Validation
- ./tools/lint.sh
- cmake --build build --target mscclpp_collectives mscclpp_nccl
-j$(nproc)

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

2026-07-07 08:56:35 -07:00

..

mscclpp

Support FP8 NVLS multimem reduction (#826)

2026-07-07 08:56:35 -07:00

CMakeLists.txt

Torch integration (#692)

2026-01-21 20:32:24 -08:00