mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-11-11 21:39:52 +00:00
4760e7cc0b
* sync : ggml (backend v2) (wip) * sync : migrate examples and llama.cpp to dynamic graphs (wip) * sync : update tests + fix max op params to 64 ggml-ci * sync : ggml-cuda ggml-ci * llama : fix save/load state context size ggml-ci * sync : try to fix build on tvOS * sync : pass custom graph sizes in training examples * sync : update graph copies to new ggml API * sync : update sync-ggml.sh with new files * scripts : fix header in sync script * train : fix context size calculations * llama : increase inference graph size up to 4096 nodes * train : allocate grads for backward graphs * train : allocate grads for gb_tmp
88 lines
3.2 KiB
C
88 lines
3.2 KiB
C
#pragma once
|
|
|
|
// ggml-backend internal header
|
|
|
|
#include "ggml-backend.h"
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
//
|
|
// Backend buffer
|
|
//
|
|
|
|
typedef void * ggml_backend_buffer_context_t;
|
|
|
|
struct ggml_backend_buffer_i {
|
|
void (*free_buffer) (ggml_backend_buffer_t buffer);
|
|
void * (*get_base) (ggml_backend_buffer_t buffer); // get base pointer
|
|
size_t (*get_alloc_size)(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); // pre-allocation callback
|
|
void (*init_tensor) (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); // post-allocation callback
|
|
void (*free_tensor) (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor); // pre-free callback
|
|
};
|
|
|
|
struct ggml_backend_buffer {
|
|
struct ggml_backend_buffer_i iface;
|
|
|
|
ggml_backend_t backend;
|
|
ggml_backend_buffer_context_t context;
|
|
|
|
size_t size;
|
|
};
|
|
|
|
GGML_API ggml_backend_buffer_t ggml_backend_buffer_init(
|
|
struct ggml_backend * backend,
|
|
struct ggml_backend_buffer_i iface,
|
|
ggml_backend_buffer_context_t context,
|
|
size_t size);
|
|
|
|
//
|
|
// Backend
|
|
//
|
|
|
|
typedef void * ggml_backend_context_t;
|
|
|
|
struct ggml_backend_i {
|
|
const char * (*get_name)(ggml_backend_t backend);
|
|
|
|
void (*free)(ggml_backend_t backend);
|
|
|
|
// buffer allocation
|
|
ggml_backend_buffer_t (*alloc_buffer)(ggml_backend_t backend, size_t size);
|
|
|
|
// get buffer alignment
|
|
size_t (*get_alignment)(ggml_backend_t backend);
|
|
|
|
// tensor data access
|
|
// these functions can be asynchronous, helper functions are provided for synchronous access that automatically call synchronize
|
|
void (*set_tensor_async)(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
|
|
void (*get_tensor_async)(ggml_backend_t backend, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
|
|
void (*synchronize) (ggml_backend_t backend);
|
|
|
|
// (optional) copy tensor between different backends, allow for single-copy tranfers
|
|
void (*cpy_tensor_from)(ggml_backend_t backend, struct ggml_tensor * src, struct ggml_tensor * dst);
|
|
void (*cpy_tensor_to) (ggml_backend_t backend, struct ggml_tensor * src, struct ggml_tensor * dst);
|
|
|
|
// compute graph with a plan
|
|
ggml_backend_graph_plan_t (*graph_plan_create) (ggml_backend_t backend, 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);
|
|
|
|
// compute graph without a plan
|
|
void (*graph_compute)(ggml_backend_t backend, struct ggml_cgraph * cgraph);
|
|
|
|
// check if the backend supports an operation
|
|
bool (*supports_op)(ggml_backend_t backend, const struct ggml_tensor * op);
|
|
};
|
|
|
|
struct ggml_backend {
|
|
struct ggml_backend_i iface;
|
|
|
|
ggml_backend_context_t context;
|
|
};
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|