public_git/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/whisper.cpp.git synced 2026-10-12 09:15:38 +02:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: disable large matmul tile on Samsung GPUs with 32KB shared memory (llama/28531)

Browse Source 
Assisted-by: Claude Opus
This commit is contained in:
Jiwoong Song
2026-10-06 10:38:17 +03:00
committed by Georgi Gerganov
parent c4051a33e6
commit 95a05ed4da
2 changed files with 10 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/ggml-vulkan-types.h
+2 -1
View File
@@ -209,6 +209,8 @@ static bool is_pow2(uint32_t x) { return x > 1 && (x & (x-1)) == 0; }
#define VK_VENDOR_ID_QUALCOMM 0x5143
#define VK_VENDOR_ID_SAMSUNG 0x144d
#define VK_DEVICE_DESCRIPTOR_POOL_SIZE 256
#define VK_CHECK(err, msg, dev) \
@@ -1451,4 +1453,3 @@ struct ggml_backend_vk_device_context {
std::string pci_bus_id;
int op_offload_min_batch_size;
};
ggml/src/ggml-vulkan/ggml-vulkan.cpp
+8
View File
@@ -4823,6 +4823,14 @@ vk_device ggml_vk_get_device(size_t idx) {
device->mul_mat_id_m[i] = true;
device->mul_mat_id_s[i] = !device->coopmat_support;
break;
case VK_VENDOR_ID_SAMSUNG:
device->mul_mat_l[i] = false;
device->mul_mat_m[i] = true;
device->mul_mat_s[i] = true;
device->mul_mat_id_l[i] = false;
device->mul_mat_id_m[i] = true;
device->mul_mat_id_s[i] = true;
break;
#endif
default:
device->mul_mat_l[i] = true;