vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix to support multiple GPUs, fix set single device, unify id/device_id/device_index

Browse source
This commit is contained in:
Jianyu Zhang 2024-06-19 22:54:15 +08:00 • committed by Neo Zhang
parent 9cbf82e65a
commit 0f98e44cd0
2 changed files with 11 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

13
ggml/src/ggml-sycl/common.hpp
View file

@ -48,6 +48,11 @@ static int g_ggml_sycl_debug = 0;
} \
}()
// #define DEBUG_SYCL_MALLOC
static int g_work_group_size = -1;
// typedef sycl::half ggml_fp16_t;
#define __SYCL_ARCH__ DPCT_COMPATIBILITY_TEMP
#define VER_4VEC 610 // todo for hardward optimize.
#define VER_GEN9 700 // todo for hardward optimize.
@ -163,6 +168,8 @@ int get_current_device_id();
(void)bad_arch; // suppress unused function warning
}
int get_current_device_id();
inline dpct::err0 ggml_sycl_set_device(const int device_id) try {
int current_device_id;
@ -217,11 +224,11 @@ struct ggml_sycl_device_info {
bool sycl_visible_devices_existed = false;
struct sycl_device_info {
int cc; // compute capability
int cc; // compute capability
// int nsm; // number of streaming multiprocessors
// size_t smpb; // max. shared memory per block
bool vmm; // virtual memory support
size_t total_vram;
bool vmm; // virtual memory support
size_t total_vram;
};
sycl_device_info devices[GGML_SYCL_MAX_DEVICES] = {};

1
src/llama.cpp
View file

@ -17620,6 +17620,7 @@ struct llama_context * llama_new_context_with_model(
#elif defined(GGML_USE_SYCL)
// with split_mode LLAMA_SPLIT_MODE_NONE or LLAMA_SPLIT_MODE_ROW, only the main GPU backend is used
if (model->split_mode == LLAMA_SPLIT_MODE_NONE || model->split_mode == LLAMA_SPLIT_MODE_ROW) {
ggml_sycl_set_single_device(model->main_gpu);
ggml_backend_t backend = ggml_backend_sycl_init(model->main_gpu);
if (backend == nullptr) {
LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d backend\n", __func__, model->main_gpu);