happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

vulkan: Use BK=32 for coopmat2 mul_mat_id (#18332)

This commit is contained in:
Jeff Bolz 2025-12-26 11:15:02 -06:00 committed by GitHub
parent cb999704fb
commit 9bf20d8ac3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -1565,7 +1565,7 @@ class vk_perf_logger {
total_op_times += time;
}
std::cerr << t.first << ": " << t.second.size() << " x " << (total_op_times / t.second.size() / 1000.0)
<< " us";
<< " us = " << (total_op_times / 1000.0) << " us";
// If we have as many flops entries as timing entries for the op, then compute and log the flops/S.
auto it = flops.find(t.first);
@ -2830,9 +2830,9 @@ static void ggml_vk_load_shaders(vk_device& device) {
s_mmq_wg_denoms_k = { 32, 64, 1 };
// spec constants and tile sizes for quant matmul_id
l_warptile_mmqid = { 256, 128, 128, 16, 1, device->subgroup_size };
m_warptile_mmqid = { 256, 128, 64, 16, 0, device->subgroup_size };
s_warptile_mmqid = { 256, 128, 64, 16, 0, device->subgroup_size };
l_warptile_mmqid = { 256, 128, 128, 32, 1, device->subgroup_size };
m_warptile_mmqid = { 256, 128, 64, 32, 0, device->subgroup_size };
s_warptile_mmqid = { 256, 128, 64, 32, 0, device->subgroup_size };
l_mmqid_wg_denoms = { 128, 128, 1 };
m_mmqid_wg_denoms = { 128, 64, 1 };
s_mmqid_wg_denoms = { 128, 64, 1 };