added fp16 kernel support#1

Open

bssrdf wants to merge 4 commits intoetasnadi:conv2d-cuda-optfrom

bssrdf:add-fp16-conv2d-mm

Commits on Sep 18, 2025

added fp16 kernel support
bssrdf
committed

Commits on Sep 19, 2025

use fastdiv equivalents already in cuda backend
bssrdf
committed

Commits on Sep 25, 2025

change symbol copy to async copy
bssrdf
committed

Commits on Oct 17, 2025

add sd.cpp test cases and revert back to cudaMemcpyToSymbol
bssrdf
committed