CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

opencl: add hs=40 to FA (#15758)

Browse Source
This commit is contained in:
rmatif
2025-09-04 08:30:28 +02:00
committed by GitHub
parent 239b60e898
commit 820bc98531
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-opencl/ggml-opencl.cpp
View File

@@ -1339,7 +1339,7 @@ static void load_cl_kernels(ggml_backend_opencl_context *backend_ctx, ggml_cl_ve
if (!kernel_src_f16.empty() && !kernel_src_f32.empty() && !kernel_src_f32_f16.empty()) {
const struct { int dk; int dv; int bm; int bn; } fa_dims[] = {
{ 64, 64, 64, 64}, { 80, 80, 64, 32}, { 96, 96, 64, 32},
{ 40, 40, 32, 32}, { 64, 64, 64, 64}, { 80, 80, 64, 32}, { 96, 96, 64, 32},
{112, 112, 32, 32}, {128, 128, 32, 32}, {192, 128, 16, 16},
{192, 192, 16, 16}, {256, 256, 16, 16},
};
@@ -2784,7 +2784,7 @@ static bool ggml_opencl_supports_op(ggml_backend_dev_t dev, const struct ggml_te
const int dv = v->ne[0];
const struct { int dk; int dv; } supported_dims[] = {
{ 64, 64}, { 80, 80}, { 96, 96},
{ 40, 40}, { 64, 64}, { 80, 80}, { 96, 96},
{112, 112}, {128, 128}, {192, 128},
{192, 192}, {256, 256},
};