llama : reorganize source code + improve CMake (#8006)
* scripts : update sync [no ci] * files : relocate [no ci] * ci : disable kompute build [no ci] * cmake : fixes [no ci] * server : fix mingw build ggml-ci * cmake : minor [no ci] * cmake : link math library [no ci] * cmake : build normal ggml library (not object library) [no ci] * cmake : fix kompute build ggml-ci * make,cmake : fix LLAMA_CUDA + replace GGML_CDEF_PRIVATE ggml-ci * move public backend headers to the public include directory (#8122) * move public backend headers to the public include directory * nix test * spm : fix metal header --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * scripts : fix sync paths [no ci] * scripts : sync ggml-blas.h [no ci] --------- Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
Georgi Gerganov
GitHub
parent
8854044561
commit
f3f65429c4
345 changed files with 2555 additions and 1937 deletions
76
ggml/include/ggml-alloc.h
Normal file
76
ggml/include/ggml-alloc.h
Normal file
|
|
@ -0,0 +1,76 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t;
|
||||
typedef struct ggml_backend_buffer * ggml_backend_buffer_t;
|
||||
typedef struct ggml_backend * ggml_backend_t;
|
||||
|
||||
// Tensor allocator
|
||||
struct ggml_tallocr {
|
||||
ggml_backend_buffer_t buffer;
|
||||
void * base;
|
||||
size_t alignment;
|
||||
size_t offset;
|
||||
};
|
||||
|
||||
GGML_API struct ggml_tallocr ggml_tallocr_new(ggml_backend_buffer_t buffer);
|
||||
GGML_API void ggml_tallocr_alloc(struct ggml_tallocr * talloc, struct ggml_tensor * tensor);
|
||||
|
||||
// Graph allocator
|
||||
/*
|
||||
Example usage:
|
||||
ggml_gallocr_t galloc = ggml_gallocr_new(ggml_bacckend_cpu_buffer_type());
|
||||
|
||||
// optional: create a worst-case graph and reserve the buffers to avoid reallocations
|
||||
ggml_gallocr_reserve(galloc, build_graph(max_batch));
|
||||
|
||||
// allocate the graph
|
||||
struct ggml_cgraph * graph = build_graph(batch);
|
||||
ggml_gallocr_alloc_graph(galloc, graph);
|
||||
|
||||
printf("compute buffer size: %zu bytes\n", ggml_gallocr_get_buffer_size(galloc, 0));
|
||||
|
||||
// evaluate the graph
|
||||
ggml_backend_graph_compute(backend, graph);
|
||||
*/
|
||||
|
||||
// special tensor flags for use with the graph allocator:
|
||||
// ggml_set_input(): all input tensors are allocated at the beginning of the graph in non-overlapping addresses
|
||||
// ggml_set_output(): output tensors are never freed and never overwritten
|
||||
|
||||
typedef struct ggml_gallocr * ggml_gallocr_t;
|
||||
|
||||
GGML_API ggml_gallocr_t ggml_gallocr_new(ggml_backend_buffer_type_t buft);
|
||||
GGML_API ggml_gallocr_t ggml_gallocr_new_n(ggml_backend_buffer_type_t * bufts, int n_bufs);
|
||||
GGML_API void ggml_gallocr_free(ggml_gallocr_t galloc);
|
||||
|
||||
// pre-allocate buffers from a measure graph - does not allocate or modify the graph
|
||||
// call with a worst-case graph to avoid buffer reallocations
|
||||
// not strictly required for single buffer usage: ggml_gallocr_alloc_graph will reallocate the buffers automatically if needed
|
||||
// returns false if the buffer allocation failed
|
||||
GGML_API bool ggml_gallocr_reserve(ggml_gallocr_t galloc, struct ggml_cgraph * graph);
|
||||
GGML_API bool ggml_gallocr_reserve_n(
|
||||
ggml_gallocr_t galloc,
|
||||
struct ggml_cgraph * graph,
|
||||
const int * node_buffer_ids,
|
||||
const int * leaf_buffer_ids);
|
||||
|
||||
// automatic reallocation if the topology changes when using a single buffer
|
||||
// returns false if using multiple buffers and a re-allocation is needed (call ggml_gallocr_reserve_n first to set the node buffers)
|
||||
GGML_API bool ggml_gallocr_alloc_graph(ggml_gallocr_t galloc, struct ggml_cgraph * graph);
|
||||
|
||||
GGML_API size_t ggml_gallocr_get_buffer_size(ggml_gallocr_t galloc, int buffer_id);
|
||||
|
||||
// Utils
|
||||
// Create a buffer and allocate all the tensors in a ggml_context
|
||||
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_context * ctx, ggml_backend_buffer_type_t buft);
|
||||
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors(struct ggml_context * ctx, ggml_backend_t backend);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
236
ggml/include/ggml-backend.h
Normal file
236
ggml/include/ggml-backend.h
Normal file
|
|
@ -0,0 +1,236 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-alloc.h"
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t;
|
||||
typedef struct ggml_backend_buffer * ggml_backend_buffer_t;
|
||||
typedef struct ggml_backend_event * ggml_backend_event_t;
|
||||
typedef struct ggml_backend * ggml_backend_t;
|
||||
typedef void * ggml_backend_graph_plan_t;
|
||||
|
||||
//
|
||||
// Backend buffer
|
||||
//
|
||||
|
||||
// buffer type
|
||||
GGML_API const char * ggml_backend_buft_name (ggml_backend_buffer_type_t buft);
|
||||
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_buft_alloc_buffer (ggml_backend_buffer_type_t buft, size_t size);
|
||||
GGML_API size_t ggml_backend_buft_get_alignment (ggml_backend_buffer_type_t buft);
|
||||
GGML_API size_t ggml_backend_buft_get_max_size (ggml_backend_buffer_type_t buft);
|
||||
GGML_API GGML_CALL size_t ggml_backend_buft_get_alloc_size (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor);
|
||||
GGML_API bool ggml_backend_buft_is_host (ggml_backend_buffer_type_t buft);
|
||||
|
||||
// buffer
|
||||
enum ggml_backend_buffer_usage {
|
||||
GGML_BACKEND_BUFFER_USAGE_ANY = 0,
|
||||
GGML_BACKEND_BUFFER_USAGE_WEIGHTS = 1,
|
||||
};
|
||||
|
||||
GGML_API const char * ggml_backend_buffer_name (ggml_backend_buffer_t buffer);
|
||||
GGML_API void ggml_backend_buffer_free (ggml_backend_buffer_t buffer);
|
||||
GGML_API void * ggml_backend_buffer_get_base (ggml_backend_buffer_t buffer);
|
||||
GGML_API size_t ggml_backend_buffer_get_size (ggml_backend_buffer_t buffer);
|
||||
GGML_API GGML_CALL void ggml_backend_buffer_init_tensor (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
|
||||
GGML_API size_t ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer);
|
||||
GGML_API size_t ggml_backend_buffer_get_max_size (ggml_backend_buffer_t buffer);
|
||||
GGML_API size_t ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
|
||||
GGML_API void ggml_backend_buffer_clear (ggml_backend_buffer_t buffer, uint8_t value);
|
||||
GGML_API bool ggml_backend_buffer_is_host (ggml_backend_buffer_t buffer);
|
||||
GGML_API void ggml_backend_buffer_set_usage (ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage);
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_buffer_get_type (ggml_backend_buffer_t buffer);
|
||||
GGML_API void ggml_backend_buffer_reset (ggml_backend_buffer_t buffer);
|
||||
|
||||
//
|
||||
// Backend
|
||||
//
|
||||
|
||||
GGML_API ggml_guid_t ggml_backend_guid(ggml_backend_t backend);
|
||||
GGML_API const char * ggml_backend_name(ggml_backend_t backend);
|
||||
GGML_API void ggml_backend_free(ggml_backend_t backend);
|
||||
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_get_default_buffer_type(ggml_backend_t backend);
|
||||
GGML_API ggml_backend_buffer_t ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size);
|
||||
GGML_API size_t ggml_backend_get_alignment(ggml_backend_t backend);
|
||||
GGML_API size_t ggml_backend_get_max_size(ggml_backend_t backend);
|
||||
|
||||
GGML_API void ggml_backend_tensor_set_async(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
|
||||
GGML_API void ggml_backend_tensor_get_async(ggml_backend_t backend, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
|
||||
|
||||
GGML_API GGML_CALL void ggml_backend_tensor_set( struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
|
||||
GGML_API GGML_CALL void ggml_backend_tensor_get(const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
|
||||
|
||||
GGML_API void ggml_backend_synchronize(ggml_backend_t backend);
|
||||
|
||||
GGML_API ggml_backend_graph_plan_t ggml_backend_graph_plan_create(ggml_backend_t backend, struct ggml_cgraph * cgraph);
|
||||
GGML_API void ggml_backend_graph_plan_free (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
|
||||
|
||||
GGML_API enum ggml_status ggml_backend_graph_plan_compute (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
|
||||
GGML_API enum ggml_status ggml_backend_graph_compute (ggml_backend_t backend, struct ggml_cgraph * cgraph);
|
||||
GGML_API enum ggml_status ggml_backend_graph_compute_async(ggml_backend_t backend, struct ggml_cgraph * cgraph);
|
||||
GGML_API bool ggml_backend_supports_op(ggml_backend_t backend, const struct ggml_tensor * op);
|
||||
GGML_API bool ggml_backend_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type_t buft);
|
||||
GGML_API bool ggml_backend_offload_op(ggml_backend_t backend, const struct ggml_tensor * op);
|
||||
|
||||
// tensor copy between different backends
|
||||
GGML_API void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst);
|
||||
|
||||
// asynchronous copy
|
||||
// the copy is performed after all the currently queued operations in backend_src
|
||||
// backend_dst will wait for the copy to complete before performing other operations
|
||||
// automatic fallback to sync copy if async is not supported
|
||||
GGML_API void ggml_backend_tensor_copy_async(ggml_backend_t backend_src, ggml_backend_t backend_dst, struct ggml_tensor * src, struct ggml_tensor * dst);
|
||||
|
||||
// events
|
||||
GGML_API ggml_backend_event_t ggml_backend_event_new (ggml_backend_t backend);
|
||||
GGML_API void ggml_backend_event_free (ggml_backend_event_t event);
|
||||
GGML_API void ggml_backend_event_record (ggml_backend_event_t event);
|
||||
GGML_API void ggml_backend_event_synchronize(ggml_backend_event_t event);
|
||||
GGML_API void ggml_backend_event_wait (ggml_backend_t backend, ggml_backend_event_t event);
|
||||
|
||||
//
|
||||
// CPU backend
|
||||
//
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_cpu_init(void);
|
||||
|
||||
GGML_API GGML_CALL bool ggml_backend_is_cpu (ggml_backend_t backend);
|
||||
GGML_API void ggml_backend_cpu_set_n_threads (ggml_backend_t backend_cpu, int n_threads);
|
||||
GGML_API void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
|
||||
|
||||
// Create a backend buffer from an existing pointer
|
||||
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size);
|
||||
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void);
|
||||
|
||||
#ifdef GGML_USE_CPU_HBM
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
|
||||
#endif
|
||||
|
||||
//
|
||||
// Backend registry
|
||||
//
|
||||
|
||||
// The backend registry is a registry of all the available backends, and allows initializing backends in a generic way
|
||||
|
||||
GGML_API size_t ggml_backend_reg_get_count(void);
|
||||
GGML_API size_t ggml_backend_reg_find_by_name(const char * name);
|
||||
GGML_API ggml_backend_t ggml_backend_reg_init_backend_from_str(const char * backend_str); // str is backend_name:params (params is optional)
|
||||
GGML_API const char * ggml_backend_reg_get_name(size_t i);
|
||||
GGML_API ggml_backend_t ggml_backend_reg_init_backend(size_t i, const char * params); // params is backend-specific
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_reg_get_default_buffer_type(size_t i);
|
||||
GGML_API ggml_backend_buffer_t ggml_backend_reg_alloc_buffer(size_t i, size_t size);
|
||||
|
||||
//
|
||||
// Backend scheduler
|
||||
//
|
||||
|
||||
// The backend scheduler allows for multiple backends to be used together
|
||||
// Handles compute buffer allocation, assignment of tensors to backends, and copying of tensors between backends
|
||||
// The backends are selected based on:
|
||||
// - the backend that supports the operation
|
||||
// - the location of the pre-allocated tensors (e.g. the weights)
|
||||
/*
|
||||
Example usage:
|
||||
|
||||
// operations that use tensors allocated in a buffer with USAGE_WEIGHTS will be assigned
|
||||
// preferrably to run on the same backend as the buffer
|
||||
ggml_backend_buffer_set_usage(buf_weights, GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
|
||||
|
||||
sched = ggml_backend_sched_new({backend_gpu, backend_gpu2, backend_cpu}, NULL, num_backends, GGML_DEFAULT_GRAPH_SIZE, false);
|
||||
|
||||
// initialize buffers from a max size graph (optional)
|
||||
reserve_graph = build_graph(sched, max_batch_size);
|
||||
|
||||
// manually assign nodes to a backend (optional, should not be needed in most cases)
|
||||
struct ggml_tensor * node = ggml_mul_mat(ctx, ...);
|
||||
ggml_backend_sched_set_tensor_backend(sched, node, backend_gpu);
|
||||
|
||||
ggml_backend_sched_reserve(sched, reserve_graph);
|
||||
|
||||
// compute
|
||||
graph = build_graph(sched);
|
||||
ggml_backend_sched_graph_compute(sched, graph);
|
||||
|
||||
// if there are graph inputs:
|
||||
ggml_backend_sched_reset(sched);
|
||||
ggml_backend_sched_alloc_graph(sched, graph);
|
||||
ggml_backend_tensor_set(input_tensor, ...);
|
||||
ggml_backend_sched_graph_compute(sched, graph);
|
||||
}
|
||||
*/
|
||||
|
||||
struct ggml_backend_sched;
|
||||
typedef struct ggml_backend_sched * ggml_backend_sched_t;
|
||||
|
||||
// when ask == true, the scheduler wants to know if the user wants to observe this node
|
||||
// this allows the scheduler to batch nodes together in order to evaluate them in a single call
|
||||
//
|
||||
// when ask == false, the scheduler is passing the node tensor to the user for observation
|
||||
// if the user returns false, the scheduler will cancel the graph compute
|
||||
//
|
||||
typedef bool (*ggml_backend_sched_eval_callback)(struct ggml_tensor * t, bool ask, void * user_data);
|
||||
|
||||
// Initialize a backend scheduler
|
||||
GGML_API ggml_backend_sched_t ggml_backend_sched_new(ggml_backend_t * backends, ggml_backend_buffer_type_t * bufts, int n_backends, size_t graph_size, bool parallel);
|
||||
GGML_API void ggml_backend_sched_free(ggml_backend_sched_t sched);
|
||||
|
||||
// Initialize backend buffers from a measure graph
|
||||
GGML_API bool ggml_backend_sched_reserve(ggml_backend_sched_t sched, struct ggml_cgraph * measure_graph);
|
||||
|
||||
GGML_API int ggml_backend_sched_get_n_backends(ggml_backend_sched_t sched);
|
||||
GGML_API ggml_backend_t ggml_backend_sched_get_backend(ggml_backend_sched_t sched, int i);
|
||||
|
||||
// Get the number of splits of the last graph
|
||||
GGML_API int ggml_backend_sched_get_n_splits(ggml_backend_sched_t sched);
|
||||
GGML_API int ggml_backend_sched_get_n_copies(ggml_backend_sched_t sched);
|
||||
|
||||
GGML_API size_t ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend);
|
||||
|
||||
GGML_API void ggml_backend_sched_set_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend);
|
||||
GGML_API ggml_backend_t ggml_backend_sched_get_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node);
|
||||
|
||||
// Allocate and compute graph on the backend scheduler
|
||||
GGML_API bool ggml_backend_sched_alloc_graph(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
|
||||
GGML_API enum ggml_status ggml_backend_sched_graph_compute(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
|
||||
GGML_API enum ggml_status ggml_backend_sched_graph_compute_async(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
|
||||
GGML_API void ggml_backend_sched_synchronize(ggml_backend_sched_t sched);
|
||||
|
||||
// Reset all assignments and allocators - must be called before changing the node backends
|
||||
GGML_API void ggml_backend_sched_reset(ggml_backend_sched_t sched);
|
||||
|
||||
// Set a callback to be called for each resulting node during graph compute
|
||||
GGML_API void ggml_backend_sched_set_eval_callback(ggml_backend_sched_t sched, ggml_backend_sched_eval_callback callback, void * user_data);
|
||||
|
||||
//
|
||||
// Utils
|
||||
//
|
||||
|
||||
struct ggml_backend_graph_copy {
|
||||
ggml_backend_buffer_t buffer;
|
||||
struct ggml_context * ctx_allocated;
|
||||
struct ggml_context * ctx_unallocated;
|
||||
struct ggml_cgraph * graph;
|
||||
};
|
||||
|
||||
// Copy a graph to a different backend
|
||||
GGML_API struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, struct ggml_cgraph * graph);
|
||||
GGML_API void ggml_backend_graph_copy_free(struct ggml_backend_graph_copy copy);
|
||||
|
||||
typedef bool (*GGML_CALL ggml_backend_eval_callback)(int node_index, struct ggml_tensor * t1, struct ggml_tensor * t2, void * user_data);
|
||||
|
||||
// Compare the output of two backends
|
||||
GGML_API bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data);
|
||||
|
||||
// Tensor initialization
|
||||
GGML_API void ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr);
|
||||
GGML_API void ggml_backend_view_init(struct ggml_tensor * tensor);
|
||||
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
23
ggml/include/ggml-blas.h
Normal file
23
ggml/include/ggml-blas.h
Normal file
|
|
@ -0,0 +1,23 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
// backend API
|
||||
GGML_API GGML_CALL ggml_backend_t ggml_backend_blas_init(void);
|
||||
|
||||
GGML_API GGML_CALL bool ggml_backend_is_blas(ggml_backend_t backend);
|
||||
|
||||
// number of threads used for conversion to float
|
||||
// for openblas and blis, this will also set the number of threads used for blas operations
|
||||
GGML_API GGML_CALL void ggml_backend_blas_set_n_threads(ggml_backend_t backend_blas, int n_threads);
|
||||
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
44
ggml/include/ggml-cuda.h
Normal file
44
ggml/include/ggml-cuda.h
Normal file
|
|
@ -0,0 +1,44 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
#ifdef GGML_USE_HIPBLAS
|
||||
#define GGML_CUDA_NAME "ROCm"
|
||||
#define GGML_CUBLAS_NAME "hipBLAS"
|
||||
#else
|
||||
#define GGML_CUDA_NAME "CUDA"
|
||||
#define GGML_CUBLAS_NAME "cuBLAS"
|
||||
#endif
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#define GGML_CUDA_MAX_DEVICES 16
|
||||
|
||||
// backend API
|
||||
GGML_API GGML_CALL ggml_backend_t ggml_backend_cuda_init(int device);
|
||||
|
||||
GGML_API GGML_CALL bool ggml_backend_is_cuda(ggml_backend_t backend);
|
||||
|
||||
// device buffer
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_buffer_type(int device);
|
||||
|
||||
// split tensor buffer that splits matrices by rows across multiple devices
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_split_buffer_type(const float * tensor_split);
|
||||
|
||||
// pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_host_buffer_type(void);
|
||||
|
||||
GGML_API GGML_CALL int ggml_backend_cuda_get_device_count(void);
|
||||
GGML_API GGML_CALL void ggml_backend_cuda_get_device_description(int device, char * description, size_t description_size);
|
||||
GGML_API GGML_CALL void ggml_backend_cuda_get_device_memory(int device, size_t * free, size_t * total);
|
||||
|
||||
GGML_API GGML_CALL bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size);
|
||||
GGML_API GGML_CALL void ggml_backend_cuda_unregister_host_buffer(void * buffer);
|
||||
|
||||
GGML_API void ggml_backend_cuda_log_set_callback(ggml_log_callback log_callback, void * user_data);
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
46
ggml/include/ggml-kompute.h
Normal file
46
ggml/include/ggml-kompute.h
Normal file
|
|
@ -0,0 +1,46 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
#include <stdbool.h>
|
||||
#include <stddef.h>
|
||||
#include <stdint.h>
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
struct ggml_vk_device {
|
||||
int index;
|
||||
int type; // same as VkPhysicalDeviceType
|
||||
size_t heapSize;
|
||||
const char * name;
|
||||
const char * vendor;
|
||||
int subgroupSize;
|
||||
uint64_t bufferAlignment;
|
||||
uint64_t maxAlloc;
|
||||
};
|
||||
|
||||
struct ggml_vk_device * ggml_vk_available_devices(size_t memoryRequired, size_t * count);
|
||||
bool ggml_vk_get_device(struct ggml_vk_device * device, size_t memoryRequired, const char * name);
|
||||
bool ggml_vk_has_vulkan(void);
|
||||
bool ggml_vk_has_device(void);
|
||||
struct ggml_vk_device ggml_vk_current_device(void);
|
||||
|
||||
//
|
||||
// backend API
|
||||
//
|
||||
|
||||
// forward declaration
|
||||
typedef struct ggml_backend * ggml_backend_t;
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_kompute_init(int device);
|
||||
|
||||
GGML_API bool ggml_backend_is_kompute(ggml_backend_t backend);
|
||||
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_kompute_buffer_type(int device);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
66
ggml/include/ggml-metal.h
Normal file
66
ggml/include/ggml-metal.h
Normal file
|
|
@ -0,0 +1,66 @@
|
|||
// An interface allowing to compute ggml_cgraph with Metal
|
||||
//
|
||||
// This is a fully functional interface that extends ggml with GPU support for Apple devices.
|
||||
// A similar interface can be created for other GPU backends (e.g. Vulkan, CUDA, etc.)
|
||||
//
|
||||
// How it works?
|
||||
//
|
||||
// As long as your program can create and evaluate a ggml_cgraph on the CPU, you can use this
|
||||
// interface to evaluate the same graph on the GPU. Instead of using ggml_graph_compute(), you
|
||||
// use ggml_metal_graph_compute() (or ggml_vulkan_graph_compute(), etc.)
|
||||
//
|
||||
// You only need to make sure that all memory buffers that you used during the graph creation
|
||||
// are mapped to the device memory with the ggml_metal_add_buffer() function. This mapping is
|
||||
// used during the graph evaluation to determine the arguments of the compute kernels.
|
||||
//
|
||||
// Synchronization between device and host memory (for example for input and output tensors)
|
||||
// is done with the ggml_metal_set_tensor() and ggml_metal_get_tensor() functions.
|
||||
//
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
#include <stddef.h>
|
||||
#include <stdbool.h>
|
||||
|
||||
// max memory buffers that can be mapped to the device
|
||||
#define GGML_METAL_MAX_BUFFERS 64
|
||||
|
||||
struct ggml_tensor;
|
||||
struct ggml_cgraph;
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
//
|
||||
// backend API
|
||||
// user-code should use only these functions
|
||||
//
|
||||
|
||||
GGML_API void ggml_backend_metal_log_set_callback(ggml_log_callback log_callback, void * user_data);
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_metal_init(void);
|
||||
|
||||
GGML_API bool ggml_backend_is_metal(ggml_backend_t backend);
|
||||
|
||||
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size);
|
||||
|
||||
GGML_API void ggml_backend_metal_set_n_cb(ggml_backend_t backend, int n_cb);
|
||||
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void);
|
||||
|
||||
// helper to check if the device supports a specific family
|
||||
// ideally, the user code should be doing these checks
|
||||
// ref: https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf
|
||||
GGML_API bool ggml_backend_metal_supports_family(ggml_backend_t backend, int family);
|
||||
|
||||
// capture all command buffers committed the next time `ggml_backend_graph_compute` is called
|
||||
GGML_API void ggml_backend_metal_capture_next_compute(ggml_backend_t backend);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
||||
24
ggml/include/ggml-rpc.h
Normal file
24
ggml/include/ggml-rpc.h
Normal file
|
|
@ -0,0 +1,24 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#define GGML_RPC_MAX_SERVERS 16
|
||||
|
||||
// backend API
|
||||
GGML_API GGML_CALL ggml_backend_t ggml_backend_rpc_init(const char * endpoint);
|
||||
GGML_API GGML_CALL bool ggml_backend_is_rpc(ggml_backend_t backend);
|
||||
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint);
|
||||
|
||||
GGML_API GGML_CALL void ggml_backend_rpc_get_device_memory(const char * endpoint, size_t * free, size_t * total);
|
||||
|
||||
GGML_API GGML_CALL void start_rpc_server(ggml_backend_t backend, const char * endpoint, size_t free_mem, size_t total_mem);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
42
ggml/include/ggml-sycl.h
Normal file
42
ggml/include/ggml-sycl.h
Normal file
|
|
@ -0,0 +1,42 @@
|
|||
//
|
||||
// MIT license
|
||||
// Copyright (C) 2024 Intel Corporation
|
||||
// SPDX-License-Identifier: MIT
|
||||
//
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
#define GGML_SYCL_NAME "SYCL"
|
||||
#define GGML_SYCL_MAX_DEVICES 48
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_sycl_init(int device);
|
||||
|
||||
// devide buffer
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device);
|
||||
|
||||
// split tensor buffer that splits matrices by rows across multiple devices
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split);
|
||||
|
||||
// pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type(void);
|
||||
|
||||
GGML_API void ggml_backend_sycl_print_sycl_devices(void);
|
||||
GGML_API GGML_CALL void ggml_sycl_get_gpu_list(int *id_list, int max_len);
|
||||
GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
|
||||
GGML_API GGML_CALL int ggml_backend_sycl_get_device_count();
|
||||
GGML_API GGML_CALL void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
|
||||
|
||||
// SYCL doesn't support registering host memory, keep here for reference
|
||||
// GGML_API GGML_CALL bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size);
|
||||
// GGML_API GGML_CALL void ggml_backend_sycl_unregister_host_buffer(void * buffer);
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
29
ggml/include/ggml-vulkan.h
Normal file
29
ggml/include/ggml-vulkan.h
Normal file
|
|
@ -0,0 +1,29 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#define GGML_VK_NAME "Vulkan"
|
||||
#define GGML_VK_MAX_DEVICES 16
|
||||
|
||||
GGML_API void ggml_vk_instance_init(void);
|
||||
|
||||
// backend API
|
||||
GGML_API GGML_CALL ggml_backend_t ggml_backend_vk_init(size_t dev_num);
|
||||
|
||||
GGML_API GGML_CALL bool ggml_backend_is_vk(ggml_backend_t backend);
|
||||
GGML_API GGML_CALL int ggml_backend_vk_get_device_count(void);
|
||||
GGML_API GGML_CALL void ggml_backend_vk_get_device_description(int device, char * description, size_t description_size);
|
||||
GGML_API GGML_CALL void ggml_backend_vk_get_device_memory(int device, size_t * free, size_t * total);
|
||||
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_vk_buffer_type(size_t dev_num);
|
||||
// pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
|
||||
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_vk_host_buffer_type(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
2427
ggml/include/ggml.h
Normal file
2427
ggml/include/ggml.h
Normal file
File diff suppressed because it is too large
Load diff
Loading…
Add table
Add a link
Reference in a new issue