Skip to content

CUDA: fix half2 -> half conversion for HIP #15529

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
IMbackK merged 1 commit into from
Aug 23, 2025
Merged

CUDA: fix half2 -> half conversion for HIP #15529

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion ggml/src/ggml-cuda/fattn-tile-f16.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -258,7 +258,7 @@ static __global__ void flash_attn_tile_ext_f16(
const half val = hexp(sink - kqmax[j0/nwarps]);
kqsum[j0/nwarps] = kqsum[j0/nwarps] * KQ_max_scale;
if (threadIdx.x == 0) {
kqsum[j0/nwarps].x = __hadd(kqsum[j0/nwarps].x, val);
kqsum[j0/nwarps].x = __hadd(__low2half(kqsum[j0/nwarps]), val);
}

#pragma unroll
Expand Down
Loading