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

ggml-vulkan: fix Mali-G720 warptile tuning (WM size)

This commit is contained in:
Gong-Mi 2026-01-23 20:49:00 +08:00
parent 73f94ff366
commit 322605cb62
1 changed files with 5 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -3038,10 +3038,11 @@ static void ggml_vk_load_shaders(vk_device& device) {
l_warptile = { 512, 128, 128, 16, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
l_warptile_mmq = { 512, 128, 128, 32, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
} else if (device->vendor_id == VK_VENDOR_ID_ARM && device->subgroup_size >= 16) {
uint32_t wm = 32 / device->subgroup_size;
m_warptile_mmq = m_warptile_mmq_int = { 64, 64, 64, 16, 16, 32, wm, 2, 2, 1, device->subgroup_size };
m_warptile = { 64, 64, 64, 16, 16, 32, wm, 2, 2, 1, device->subgroup_size };
m_warptile_id = m_warptile_mmqid = { 64, 64, 64, 16, 16, 32, wm, 2, 2, 1, device->subgroup_size };
uint32_t wm_iter = 32 / device->subgroup_size;
uint32_t wm_tile = device->subgroup_size * 2;
m_warptile_mmq = m_warptile_mmq_int = { 64, 64, 64, 16, wm_tile, 32, wm_iter, 2, 2, 1, device->subgroup_size };
m_warptile = { 64, 64, 64, 16, wm_tile, 32, wm_iter, 2, 2, 1, device->subgroup_size };
m_warptile_id = m_warptile_mmqid = { 64, 64, 64, 16, wm_tile, 32, wm_iter, 2, 2, 1, device->subgroup_size };
}
l_mmq_wg_denoms = l_wg_denoms = {128, 128, 1 };