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

PR #4766

Browse source
This commit is contained in:
Jared Van Bortel 2024-01-10 11:29:04 -05:00
parent 3773e1afe7
commit 1eb8804c18
18 changed files with 2183 additions and 2081 deletions
Download patch file Download diff file Expand all files Collapse all files

29
ggml-backend-impl.h
View file

@ -16,9 +16,10 @@ extern "C" {
typedef void * ggml_backend_buffer_type_context_t;
struct ggml_backend_buffer_type_i {
const char * (*get_name) (ggml_backend_buffer_type_t buft);
ggml_backend_buffer_t (*alloc_buffer) (ggml_backend_buffer_type_t buft, size_t size);
size_t (*get_alignment) (ggml_backend_buffer_type_t buft); // tensor alignment
size_t (*get_alloc_size) (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor); // data size needed to allocate the tensor, including padding
size_t (*get_alloc_size) (ggml_backend_buffer_type_t buft, const struct ggml_tensor * tensor); // data size needed to allocate the tensor, including padding
bool (*supports_backend)(ggml_backend_buffer_type_t buft, ggml_backend_t backend); // check if the buffer type is usable by the backend
// check if tensor data is in host memory
// should be equivalent to supports_backend(buft, ggml_backend_cpu_init())
@ -34,16 +35,17 @@ extern "C" {
typedef void * ggml_backend_buffer_context_t;
struct ggml_backend_buffer_i {
void (*free_buffer) (ggml_backend_buffer_t buffer);
//void (*reset) (ggml_backend_buffer_t buffer); // reset any internal state due to tensor initialization, such as tensor extras
void * (*get_base) (ggml_backend_buffer_t buffer);
void (*init_tensor) (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
void (*set_tensor) (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
void (*get_tensor) (ggml_backend_buffer_t buffer, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
const char * (*get_name) (ggml_backend_buffer_t buffer);
void (*free_buffer) (ggml_backend_buffer_t buffer);
void * (*get_base) (ggml_backend_buffer_t buffer);
void (*init_tensor) (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
void (*set_tensor) (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
void (*get_tensor) (ggml_backend_buffer_t buffer, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
// (optional) copy tensor between different buffer-type, allow for single-copy tranfers
void (*cpy_tensor_from)(ggml_backend_buffer_t buffer, struct ggml_tensor * src, struct ggml_tensor * dst);
void (*cpy_tensor_to) (ggml_backend_buffer_t buffer, struct ggml_tensor * src, struct ggml_tensor * dst);
void (*clear) (ggml_backend_buffer_t buffer, uint8_t value);
void (*cpy_tensor_from)(ggml_backend_buffer_t buffer, const struct ggml_tensor * src, struct ggml_tensor * dst);
void (*cpy_tensor_to) (ggml_backend_buffer_t buffer, const struct ggml_tensor * src, struct ggml_tensor * dst);
void (*clear) (ggml_backend_buffer_t buffer, uint8_t value);
void (*reset) (ggml_backend_buffer_t buffer); // reset any internal state due to tensor initialization, such as tensor extras
};
struct ggml_backend_buffer {
@ -51,6 +53,7 @@ extern "C" {
ggml_backend_buffer_type_t buft;
ggml_backend_buffer_context_t context;
size_t size;
enum ggml_backend_buffer_usage usage;
};
ggml_backend_buffer_t ggml_backend_buffer_init(
@ -79,13 +82,13 @@ extern "C" {
void (*get_tensor_async)(ggml_backend_t backend, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
// (optional) asynchroneous tensor copy
void (*cpy_tensor_from_async)(ggml_backend_t backend, struct ggml_tensor * src, struct ggml_tensor * dst);
void (*cpy_tensor_to_async) (ggml_backend_t backend, struct ggml_tensor * src, struct ggml_tensor * dst);
void (*cpy_tensor_from_async)(ggml_backend_t backend, const struct ggml_tensor * src, struct ggml_tensor * dst);
void (*cpy_tensor_to_async) (ggml_backend_t backend, const struct ggml_tensor * src, struct ggml_tensor * dst);
void (*synchronize)(ggml_backend_t backend);
// compute graph with a plan
ggml_backend_graph_plan_t (*graph_plan_create) (ggml_backend_t backend, struct ggml_cgraph * cgraph);
ggml_backend_graph_plan_t (*graph_plan_create) (ggml_backend_t backend, const struct ggml_cgraph * cgraph);
void (*graph_plan_free) (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
void (*graph_plan_compute)(ggml_backend_t backend, ggml_backend_graph_plan_t plan);