happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

FIX: backend buffer type issue

This commit is contained in:
Yu, Zijun 2025-05-13 17:45:47 +08:00 committed by Mustafa Cavus
parent 4c905b2b25
commit a0b30529bf
2 changed files with 5 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/src/ggml-backend-reg.cpp
View File

@ -630,6 +630,7 @@ void ggml_backend_load_all_from_path(const char * dir_path) {
ggml_backend_load_best("opencl", silent, dir_path); ggml_backend_load_best("opencl", silent, dir_path);
ggml_backend_load_best("hexagon", silent, dir_path); ggml_backend_load_best("hexagon", silent, dir_path);
ggml_backend_load_best("musa", silent, dir_path); ggml_backend_load_best("musa", silent, dir_path);
ggml_backend_load_best("openvino", silent, dir_path);
ggml_backend_load_best("cpu", silent, dir_path); ggml_backend_load_best("cpu", silent, dir_path);
// check the environment variable GGML_BACKEND_PATH to load an out-of-tree backend // check the environment variable GGML_BACKEND_PATH to load an out-of-tree backend
const char * backend_path = std::getenv("GGML_BACKEND_PATH"); const char * backend_path = std::getenv("GGML_BACKEND_PATH");

15
ggml/src/ggml-openvino/ggml-openvino.cpp
View File

@ -46,17 +46,11 @@ static const char * ggml_backend_openvino_get_name(ggml_backend_t backend) {
GGML_UNUSED(backend); GGML_UNUSED(backend);
} }
static ggml_backend_buffer_type_t ggml_backend_openvino_get_default_buffer_type(ggml_backend_t backend) {
return ggml_backend_cpu_buffer_type();
GGML_UNUSED(backend);
}
static enum ggml_status static enum ggml_status
ggml_backend_openvino_graph_compute(ggml_backend_t backend, struct ggml_cgraph *cgraph) { ggml_backend_openvino_graph_compute(ggml_backend_t backend, struct ggml_cgraph *cgraph) {
openvino_frontend_compute(backend, cgraph); openvino_frontend_compute(backend, cgraph);
return GGML_STATUS_SUCCESS; return GGML_STATUS_SUCCESS;
GGML_UNUSED(backend);
} }
static const ggml_backend_i ggml_backend_openvino_interface = { static const ggml_backend_i ggml_backend_openvino_interface = {
@ -108,14 +102,14 @@ GGML_BACKEND_API ggml_backend_t ggml_backend_openvino_init(int device) {
} }
GGML_BACKEND_API bool ggml_backend_is_openvino(ggml_backend_t backend) { GGML_BACKEND_API bool ggml_backend_is_openvino(ggml_backend_t backend) {
GGML_ASSERT(backend->context != nullptr); return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_openvino_guid());
return true;
} }
// device buffer // device buffer
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_openvino_buffer_type(int device) { GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_openvino_buffer_type(int device) {
GGML_ASSERT(device >= 0); GGML_ASSERT(device >= 0);
return nullptr; return ggml_backend_cpu_buffer_type();
GGML_UNUSED(device);
} }
// split tensor buffer that splits matrices by rows across multiple devices // split tensor buffer that splits matrices by rows across multiple devices
@ -184,8 +178,7 @@ static void ggml_backend_openvino_device_get_memory(ggml_backend_dev_t dev, size
static enum ggml_backend_dev_type ggml_backend_openvino_device_get_type(ggml_backend_dev_t dev) { static enum ggml_backend_dev_type ggml_backend_openvino_device_get_type(ggml_backend_dev_t dev) {
GGML_UNUSED(dev); GGML_UNUSED(dev);
return GGML_BACKEND_DEVICE_TYPE_CPU; return GGML_BACKEND_DEVICE_TYPE_ACCEL;
// return GGML_BACKEND_DEVICE_TYPE_GPU_FULL;
} }
static void ggml_backend_openvino_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) { static void ggml_backend_openvino_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) {