nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 23 Packages Projects Releases Wiki Activity

Added comments explaining thread block size selection logic based on row count and column size, derived from historical commit context (#18212)

Browse Source
This commit is contained in:
Aadeshveer Singh
2025-12-20 19:28:57 +08:00
committed by GitHub
parent 408616adbd
commit 10b4f82d44
1 changed files with 3 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/mean.cu
+3
View File
@@ -63,6 +63,9 @@ void ggml_cuda_op_mean(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
const int id = ggml_cuda_get_device(); const int id = ggml_cuda_get_device();
const int nsm = ggml_cuda_info().devices[id].nsm; const int nsm = ggml_cuda_info().devices[id].nsm;
// Heuristic for block size selection to optimize occupancy.
// See discussion in: https://github.com/ggml-org/llama.cpp/pull/15132
if ((nrows / nsm) < 2) { if ((nrows / nsm) < 2) {
const dim3 block_dims(512, 1, 1); const dim3 block_dims(512, 1, 1);
reduce_rows_f32</*norm=*/true><<<block_nums, block_dims, 0, stream>>>(src0_d, dst_d, ncols); reduce_rows_f32</*norm=*/true><<<block_nums, block_dims, 0, stream>>>(src0_d, dst_d, ncols);