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

modify warptile tuning for xe3

This commit is contained in:
Viraj Wadhwa 2025-12-14 19:50:15 -08:00
parent 4aced7a631
commit 0f82d0a404
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -2805,11 +2805,12 @@ static void ggml_vk_load_shaders(vk_device& device) {
const uint32_t tk_s = device->coopmat_support ? device->coopmat_k : 1;
l_warptile = { 128, 128, 128, 16, subgroup_size_8 * 2, 64, 2, tm_l, tn_l, tk_l, subgroup_size_8 };
m_warptile = { 128, 64, 64, 16, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
m_warptile = { 512, 128, 128, 16, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
s_warptile = { subgroup_size_16, 32, 32, 16, 32, 32, 2, tm_s, tn_s, tk_s, subgroup_size_8 };
l_warptile_mmq = { 128, 128, 128, 32, subgroup_size_8 * 2, 64, 2, tm_l, tn_l, tk_l, subgroup_size_8 };
m_warptile_mmq = { 128, 64, 64, 32, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
m_warptile_mmq = { 512, 128, 128, 32, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
// (BLOCK SIZE, BM, BN, BK, WM, WN, WMITER, TM, TN, TK, WARP)
s_warptile_mmq = { subgroup_size_32, 32, 32, 32, 32, 32, 2, tm_s, tn_s, tk_s, subgroup_size_8 };
// Integer MMQ has a smaller shared memory profile, but heavier register use
@ -2845,7 +2846,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
}
l_mmq_wg_denoms = l_wg_denoms = {128, 128, 1 };
m_mmq_wg_denoms = m_wg_denoms = { 64, 64, 1 };
m_mmq_wg_denoms = m_wg_denoms = { 128, 128, 1 };
s_mmq_wg_denoms = s_wg_denoms = { 32, 32, 1 };
l_align = 128;
m_align = 64;