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: move ARM/Qualcomm mul_mat_l restriction to init block 

Moves the logic that disables large matrix multiplication for ARM and Qualcomm devices from ggml_vk_load_shaders to the device initialization switch block.

This fixes stability issues (silent calculation errors) on Mali G720/Immortalis MC12 while adhering to the code structure requested in PR #18493 discussion.
This commit is contained in:
Gong-Mi 2026-02-02 01:11:08 +08:00
parent e6c76f7771
commit cef48b4cbf
1 changed files with 10 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -3084,10 +3084,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
device->mul_mat_l[i] = false;
}
if (device->vendor_id == VK_VENDOR_ID_ARM || device->vendor_id == VK_VENDOR_ID_QUALCOMM) {
device->mul_mat_l[i] = false;
device->mul_mat_id_l[i] = false;
}
// Disable mul_mat_id if not enough shared memory is available
if (!ggml_vk_matmul_shmem_support(device, s_warptile_mmqid, true, t)) {
@ -5195,6 +5192,15 @@ static vk_device ggml_vk_get_device(size_t idx) {
device->mul_mat_id_m[i] = true;
device->mul_mat_id_s[i] = true;
break;
case VK_VENDOR_ID_ARM:
case VK_VENDOR_ID_QUALCOMM:
device->mul_mat_l[i] = false;
device->mul_mat_id_l[i] = false;
device->mul_mat_m[i] = true;
device->mul_mat_s[i] = true;
device->mul_mat_id_m[i] = true;
device->mul_mat_id_s[i] = true;
break;
case VK_VENDOR_ID_APPLE:
device->mul_mat_l[i] = false;
device->mul_mat_m[i] = true;