Skip to content

ggml-cuda: optimize single-token MMVQ dispatch for RDNA3 architecture #28003

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

Draft
maci0 wants to merge 1 commit into
base: master
Choose a base branch
from
+11 −0
Draft

ggml-cuda: optimize single-token MMVQ dispatch for RDNA3 architecture #28003

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
11 changes: 11 additions & 0 deletions ggml/src/ggml-cuda/mmvq.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1024,6 +1024,17 @@ static void mul_mat_vec_q_switch_ncols_dst(
stream);
};

// Fast-path for RDNA3 single-token vector multiplication
if (table_id == MMVQ_PARAMETERS_RDNA3_0 && !has_ids && nchannels_dst == 1) {
const std::pair<dim3, dim3> dims = calc_launch_params<type>(1, nrows_x, 1, 1, warp_size, table_id, false, false);
mul_mat_vec_q_switch_fusion<type, 1, false, false>(
vx, vy, ids, fusion, dst, ncols_x, nchannels_y_fd, stride_row_x, stride_col_y, stride_col_dst,
channel_ratio_fd, stride_channel_x, stride_channel_y, stride_channel_dst, sample_ratio_fd,
stride_sample_x, stride_sample_y, stride_sample_dst, dims.first, dims.second, 0, ids_stride,
stream);
break;
}

if (should_use_small_k(c_ncols_dst)) {
launch(std::true_type{}, std::false_type{});
} else if (should_halve_iters()) {
Expand Down