vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-09 18:34:39 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[SYCL] remove global variables (#7710)

Browse source 
* separate DPCT helpers outside

* replace global variables with context

* remove useless extra

* update mul_mat condition

* remove duplicate buft initialization

* remove duplicate extra and global work group size

* remove useless backend check

* remove duplicated extras

* use macro for group_size and remove cuda-related
This commit is contained in:
Meng, Hengyu 2024-06-15 14:05:10 +08:00 • committed by GitHub
parent f8ec8877b7
commit 7b2f4a7d19
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 4142 additions and 4835 deletions
Download patch file Download diff file Expand all files Collapse all files

13
llama.cpp
View file

@ -6625,16 +6625,6 @@ static int llama_model_load(const std::string & fname, llama_model & model, llam
}
#endif
#ifdef GGML_USE_SYCL
if (params.split_mode == LLAMA_SPLIT_MODE_NONE) {
ggml_backend_sycl_set_single_device_mode(params.main_gpu);
//SYCL use device index (0, 1, 2) directly, uer input device id, then convert to device index.
params.main_gpu = ggml_backend_sycl_get_device_index(params.main_gpu);
} else {
ggml_backend_sycl_set_mul_device_mode();
}
#endif
if (!llm_load_tensors(
ml, model, params.n_gpu_layers, params.split_mode, params.main_gpu, params.tensor_split, params.use_mlock,
params.progress_callback, params.progress_callback_user_data
@ -16241,8 +16231,7 @@ struct llama_context * llama_new_context_with_model(
if (model->split_mode == LLAMA_SPLIT_MODE_NONE || model->split_mode == LLAMA_SPLIT_MODE_ROW) {
ggml_backend_t backend = ggml_backend_sycl_init(model->main_gpu);
if (backend == nullptr) {
int main_gpu_id = ggml_backend_sycl_get_device_id(model->main_gpu);
LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d (index %d) backend\n", __func__, main_gpu_id, model->main_gpu);
LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d backend\n", __func__, model->main_gpu);
llama_free(ctx);
return nullptr;
}