mirror of
https://github.com/ggerganov/whisper.cpp.git
synced 2025-06-16 05:48:09 +00:00
ggml : build backends as libraries (llama/10256)
* ggml : build backends as libraries --------- Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Co-authored-by: R0CKSTAR <xiaodong.ye@mthreads.com>
This commit is contained in:
Diego Devesa
committed by
Georgi Gerganov
Georgi Gerganov
parent
5f7e094ccb
commit
746bf2596f
@ -9,16 +9,16 @@ extern "C" {
|
||||
#endif
|
||||
|
||||
// buffer_type API
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_amx_buffer_type(void);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_amx_buffer_type(void);
|
||||
|
||||
GGML_API bool ggml_backend_is_amx(ggml_backend_t backend);
|
||||
GGML_BACKEND_API bool ggml_backend_is_amx(ggml_backend_t backend);
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_amx_init(void);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_amx_init(void);
|
||||
|
||||
GGML_API void ggml_backend_amx_set_n_threads(ggml_backend_t backend_amx, int n_threads);
|
||||
GGML_BACKEND_API void ggml_backend_amx_set_n_threads(ggml_backend_t backend_amx, int n_threads);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_amx_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_amx_reg(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
|
||||
@ -3,6 +3,20 @@
|
||||
#include "ggml.h"
|
||||
#include "ggml-alloc.h"
|
||||
|
||||
#ifdef GGML_BACKEND_SHARED
|
||||
# if defined(_WIN32) && !defined(__MINGW32__)
|
||||
# ifdef GGML_BACKEND_BUILD
|
||||
# define GGML_BACKEND_API __declspec(dllexport) extern
|
||||
# else
|
||||
# define GGML_BACKEND_API __declspec(dllimport) extern
|
||||
# endif
|
||||
# else
|
||||
# define GGML_BACKEND_API __attribute__ ((visibility ("default"))) extern
|
||||
# endif
|
||||
#else
|
||||
# define GGML_BACKEND_API extern
|
||||
#endif
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
@ -9,15 +9,15 @@ extern "C" {
|
||||
#endif
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_blas_init(void);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_blas_init(void);
|
||||
|
||||
GGML_API bool ggml_backend_is_blas(ggml_backend_t backend);
|
||||
GGML_BACKEND_API 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 void ggml_backend_blas_set_n_threads(ggml_backend_t backend_blas, int n_threads);
|
||||
GGML_BACKEND_API void ggml_backend_blas_set_n_threads(ggml_backend_t backend_blas, int n_threads);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_blas_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_blas_reg(void);
|
||||
|
||||
|
||||
#ifdef __cplusplus
|
||||
|
||||
@ -34,7 +34,7 @@ extern "C" {
|
||||
*/
|
||||
#define GGML_CANN_MAX_DEVICES 16
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_cann_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_cann_reg(void);
|
||||
|
||||
/**
|
||||
* @brief Initializes the CANN backend for a specified device.
|
||||
@ -46,7 +46,7 @@ GGML_API ggml_backend_reg_t ggml_backend_cann_reg(void);
|
||||
* @param device The index of the device to initialize.
|
||||
* @return A pointer to the initialized backend instance, or nullptr on failure.
|
||||
*/
|
||||
GGML_API ggml_backend_t ggml_backend_cann_init(int32_t device);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_cann_init(int32_t device);
|
||||
|
||||
/**
|
||||
* @brief Checks if a given backend is a CANN backend.
|
||||
@ -57,7 +57,7 @@ GGML_API ggml_backend_t ggml_backend_cann_init(int32_t device);
|
||||
* @param backend The backend instance to check.
|
||||
* @return True if the backend is a CANN backend, false otherwise.
|
||||
*/
|
||||
GGML_API bool ggml_backend_is_cann(ggml_backend_t backend);
|
||||
GGML_BACKEND_API bool ggml_backend_is_cann(ggml_backend_t backend);
|
||||
|
||||
/**
|
||||
* @brief Retrieves the CANN buffer type for a specified device.
|
||||
@ -69,7 +69,7 @@ GGML_API bool ggml_backend_is_cann(ggml_backend_t backend);
|
||||
* @return A pointer to the buffer type interface for the specified device, or
|
||||
* nullptr if the device index is out of range.
|
||||
*/
|
||||
GGML_API ggml_backend_buffer_type_t
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t
|
||||
ggml_backend_cann_buffer_type(int32_t device);
|
||||
|
||||
/**
|
||||
@ -80,14 +80,14 @@ ggml_backend_cann_buffer_type(int32_t device);
|
||||
*
|
||||
* @return The number of CANN devices available.
|
||||
*/
|
||||
GGML_API int32_t ggml_backend_cann_get_device_count(void);
|
||||
GGML_BACKEND_API int32_t ggml_backend_cann_get_device_count(void);
|
||||
|
||||
/**
|
||||
* @brief pinned host buffer for use with the CPU backend for faster copies between CPU and NPU.
|
||||
*
|
||||
* @return A pointer to the host buffer type interface.
|
||||
*/
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_cann_host_buffer_type(void);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cann_host_buffer_type(void);
|
||||
|
||||
/**
|
||||
* @brief Retrieves the description of a specific CANN device.
|
||||
@ -99,7 +99,7 @@ GGML_API ggml_backend_buffer_type_t ggml_backend_cann_host_buffer_type(void);
|
||||
* @param description Pointer to a buffer where the description will be written.
|
||||
* @param description_size Size of the description buffer.
|
||||
*/
|
||||
GGML_API void ggml_backend_cann_get_device_description(
|
||||
GGML_BACKEND_API void ggml_backend_cann_get_device_description(
|
||||
int32_t device, char* description, size_t description_size);
|
||||
|
||||
/**
|
||||
@ -114,7 +114,7 @@ GGML_API void ggml_backend_cann_get_device_description(
|
||||
* @param total Pointer to a variable where the total memory size will be
|
||||
* stored.
|
||||
*/
|
||||
GGML_API void ggml_backend_cann_get_device_memory(int32_t device,
|
||||
GGML_BACKEND_API void ggml_backend_cann_get_device_memory(int32_t device,
|
||||
size_t* free,
|
||||
size_t* total);
|
||||
|
||||
|
||||
@ -54,54 +54,77 @@ extern "C" {
|
||||
GGML_NUMA_STRATEGY_COUNT
|
||||
};
|
||||
|
||||
GGML_API void ggml_numa_init(enum ggml_numa_strategy numa); // call once for better performance on NUMA systems
|
||||
GGML_API bool ggml_is_numa(void); // true if init detected that system has >1 NUMA node
|
||||
GGML_BACKEND_API void ggml_numa_init(enum ggml_numa_strategy numa); // call once for better performance on NUMA systems
|
||||
GGML_BACKEND_API bool ggml_is_numa(void); // true if init detected that system has >1 NUMA node
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value);
|
||||
GGML_API struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value);
|
||||
GGML_BACKEND_API struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value);
|
||||
GGML_BACKEND_API struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value);
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_set_i32 (struct ggml_tensor * tensor, int32_t value);
|
||||
GGML_API struct ggml_tensor * ggml_set_f32 (struct ggml_tensor * tensor, float value);
|
||||
GGML_BACKEND_API struct ggml_tensor * ggml_set_i32 (struct ggml_tensor * tensor, int32_t value);
|
||||
GGML_BACKEND_API struct ggml_tensor * ggml_set_f32 (struct ggml_tensor * tensor, float value);
|
||||
|
||||
GGML_API int32_t ggml_get_i32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_API void ggml_set_i32_1d(const struct ggml_tensor * tensor, int i, int32_t value);
|
||||
GGML_BACKEND_API int32_t ggml_get_i32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_BACKEND_API void ggml_set_i32_1d(const struct ggml_tensor * tensor, int i, int32_t value);
|
||||
|
||||
GGML_API int32_t ggml_get_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_API void ggml_set_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, int32_t value);
|
||||
GGML_BACKEND_API int32_t ggml_get_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_BACKEND_API void ggml_set_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, int32_t value);
|
||||
|
||||
GGML_API float ggml_get_f32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_API void ggml_set_f32_1d(const struct ggml_tensor * tensor, int i, float value);
|
||||
GGML_BACKEND_API float ggml_get_f32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_BACKEND_API void ggml_set_f32_1d(const struct ggml_tensor * tensor, int i, float value);
|
||||
|
||||
GGML_API float ggml_get_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_API void ggml_set_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, float value);
|
||||
GGML_BACKEND_API float ggml_get_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_BACKEND_API void ggml_set_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, float value);
|
||||
|
||||
GGML_API struct ggml_threadpool_params ggml_threadpool_params_default(int n_threads);
|
||||
GGML_API void ggml_threadpool_params_init (struct ggml_threadpool_params * p, int n_threads);
|
||||
GGML_API bool ggml_threadpool_params_match (const struct ggml_threadpool_params * p0, const struct ggml_threadpool_params * p1);
|
||||
GGML_API struct ggml_threadpool * ggml_threadpool_new (struct ggml_threadpool_params * params);
|
||||
GGML_API void ggml_threadpool_free (struct ggml_threadpool * threadpool);
|
||||
GGML_API int ggml_threadpool_get_n_threads(struct ggml_threadpool * threadpool);
|
||||
GGML_API void ggml_threadpool_pause (struct ggml_threadpool * threadpool);
|
||||
GGML_API void ggml_threadpool_resume (struct ggml_threadpool * threadpool);
|
||||
GGML_BACKEND_API struct ggml_threadpool_params ggml_threadpool_params_default(int n_threads);
|
||||
GGML_BACKEND_API void ggml_threadpool_params_init (struct ggml_threadpool_params * p, int n_threads);
|
||||
GGML_BACKEND_API bool ggml_threadpool_params_match (const struct ggml_threadpool_params * p0, const struct ggml_threadpool_params * p1);
|
||||
GGML_BACKEND_API struct ggml_threadpool * ggml_threadpool_new (struct ggml_threadpool_params * params);
|
||||
GGML_BACKEND_API void ggml_threadpool_free (struct ggml_threadpool * threadpool);
|
||||
GGML_BACKEND_API int ggml_threadpool_get_n_threads(struct ggml_threadpool * threadpool);
|
||||
GGML_BACKEND_API void ggml_threadpool_pause (struct ggml_threadpool * threadpool);
|
||||
GGML_BACKEND_API void ggml_threadpool_resume (struct ggml_threadpool * threadpool);
|
||||
|
||||
// ggml_graph_plan() has to be called before ggml_graph_compute()
|
||||
// when plan.work_size > 0, caller must allocate memory for plan.work_data
|
||||
GGML_API struct ggml_cplan ggml_graph_plan(
|
||||
GGML_BACKEND_API struct ggml_cplan ggml_graph_plan(
|
||||
const struct ggml_cgraph * cgraph,
|
||||
int n_threads, /* = GGML_DEFAULT_N_THREADS */
|
||||
struct ggml_threadpool * threadpool /* = NULL */ );
|
||||
GGML_API enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan);
|
||||
GGML_BACKEND_API enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan);
|
||||
|
||||
// same as ggml_graph_compute() but the work data is allocated as a part of the context
|
||||
// note: the drawback of this API is that you must have ensured that the context has enough memory for the work data
|
||||
GGML_API enum ggml_status ggml_graph_compute_with_ctx(struct ggml_context * ctx, struct ggml_cgraph * cgraph, int n_threads);
|
||||
GGML_BACKEND_API enum ggml_status ggml_graph_compute_with_ctx(struct ggml_context * ctx, struct ggml_cgraph * cgraph, int n_threads);
|
||||
|
||||
// TODO: move to backend interface
|
||||
GGML_API int ggml_cpu_has_neon (void);
|
||||
GGML_API int ggml_cpu_has_sve (void);
|
||||
GGML_API int ggml_cpu_has_matmul_int8(void);
|
||||
// get the sve vector length in bytes
|
||||
GGML_API int ggml_cpu_get_sve_cnt(void);
|
||||
//
|
||||
// system info
|
||||
//
|
||||
|
||||
// x86
|
||||
GGML_BACKEND_API int ggml_cpu_has_sse3 (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_ssse3 (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_avx (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_avx2 (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_f16c (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_fma (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_avx_vnni (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_avx512 (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_avx512_vbmi(void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_avx512_vnni(void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_avx512_bf16(void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_amx_int8 (void);
|
||||
// ARM
|
||||
GGML_BACKEND_API int ggml_cpu_has_neon (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_arm_fma (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_fp16_va (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_matmul_int8(void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_sve (void);
|
||||
GGML_BACKEND_API int ggml_cpu_get_sve_cnt (void); // sve vector length in bytes
|
||||
// other
|
||||
GGML_BACKEND_API int ggml_cpu_has_riscv_v (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_vsx (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_wasm_simd (void);
|
||||
GGML_BACKEND_API int ggml_cpu_has_llamafile (void);
|
||||
|
||||
// Internal types and functions exposed for tests and benchmarks
|
||||
|
||||
@ -115,6 +138,7 @@ extern "C" {
|
||||
const void * GGML_RESTRICT y, int nr, int nc);
|
||||
|
||||
struct ggml_type_traits_cpu {
|
||||
ggml_from_float_t from_float;
|
||||
ggml_from_float_to_mat_t from_float_to_mat;
|
||||
ggml_vec_dot_t vec_dot;
|
||||
enum ggml_type vec_dot_type;
|
||||
@ -124,25 +148,25 @@ extern "C" {
|
||||
ggml_gemm_t gemm;
|
||||
};
|
||||
|
||||
GGML_API const struct ggml_type_traits_cpu * ggml_get_type_traits_cpu(enum ggml_type type);
|
||||
GGML_BACKEND_API const struct ggml_type_traits_cpu * ggml_get_type_traits_cpu(enum ggml_type type);
|
||||
|
||||
GGML_API void ggml_cpu_init(void);
|
||||
GGML_BACKEND_API void ggml_cpu_init(void);
|
||||
|
||||
//
|
||||
// CPU backend
|
||||
//
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_cpu_init(void);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_cpu_init(void);
|
||||
|
||||
GGML_API 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_threadpool (ggml_backend_t backend_cpu, ggml_threadpool_t threadpool);
|
||||
GGML_API void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
|
||||
GGML_BACKEND_API bool ggml_backend_is_cpu (ggml_backend_t backend);
|
||||
GGML_BACKEND_API void ggml_backend_cpu_set_n_threads (ggml_backend_t backend_cpu, int n_threads);
|
||||
GGML_BACKEND_API void ggml_backend_cpu_set_threadpool (ggml_backend_t backend_cpu, ggml_threadpool_t threadpool);
|
||||
GGML_BACKEND_API void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_cpu_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_cpu_reg(void);
|
||||
|
||||
#ifdef GGML_USE_CPU_HBM
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
|
||||
#endif
|
||||
|
||||
#ifdef __cplusplus
|
||||
|
||||
@ -7,7 +7,7 @@
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#ifdef GGML_USE_HIPBLAS
|
||||
#ifdef GGML_USE_HIP
|
||||
#define GGML_CUDA_NAME "ROCm"
|
||||
#define GGML_CUBLAS_NAME "hipBLAS"
|
||||
#elif defined(GGML_USE_MUSA)
|
||||
@ -20,27 +20,27 @@ extern "C" {
|
||||
#define GGML_CUDA_MAX_DEVICES 16
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_cuda_init(int device);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_cuda_init(int device);
|
||||
|
||||
GGML_API bool ggml_backend_is_cuda(ggml_backend_t backend);
|
||||
GGML_BACKEND_API bool ggml_backend_is_cuda(ggml_backend_t backend);
|
||||
|
||||
// device buffer
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_cuda_buffer_type(int device);
|
||||
GGML_BACKEND_API 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_backend_buffer_type_t ggml_backend_cuda_split_buffer_type(int main_device, const float * tensor_split);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cuda_split_buffer_type(int main_device, 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_cuda_host_buffer_type(void);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cuda_host_buffer_type(void);
|
||||
|
||||
GGML_API int ggml_backend_cuda_get_device_count(void);
|
||||
GGML_API void ggml_backend_cuda_get_device_description(int device, char * description, size_t description_size);
|
||||
GGML_API void ggml_backend_cuda_get_device_memory(int device, size_t * free, size_t * total);
|
||||
GGML_BACKEND_API int ggml_backend_cuda_get_device_count(void);
|
||||
GGML_BACKEND_API void ggml_backend_cuda_get_device_description(int device, char * description, size_t description_size);
|
||||
GGML_BACKEND_API void ggml_backend_cuda_get_device_memory(int device, size_t * free, size_t * total);
|
||||
|
||||
GGML_API bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size);
|
||||
GGML_API void ggml_backend_cuda_unregister_host_buffer(void * buffer);
|
||||
GGML_BACKEND_API bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size);
|
||||
GGML_BACKEND_API void ggml_backend_cuda_unregister_host_buffer(void * buffer);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_cuda_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_cuda_reg(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
|
||||
@ -37,13 +37,13 @@ struct ggml_vk_device ggml_vk_current_device(void);
|
||||
// forward declaration
|
||||
typedef struct ggml_backend * ggml_backend_t;
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_kompute_init(int device);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_kompute_init(int device);
|
||||
|
||||
GGML_API bool ggml_backend_is_kompute(ggml_backend_t backend);
|
||||
GGML_BACKEND_API bool ggml_backend_is_kompute(ggml_backend_t backend);
|
||||
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_kompute_buffer_type(int device);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_kompute_buffer_type(int device);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_kompute_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_kompute_reg(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
|
||||
@ -39,27 +39,27 @@ extern "C" {
|
||||
// user-code should use only these functions
|
||||
//
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_metal_init(void);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_metal_init(void);
|
||||
|
||||
GGML_API bool ggml_backend_is_metal(ggml_backend_t backend);
|
||||
GGML_BACKEND_API bool ggml_backend_is_metal(ggml_backend_t backend);
|
||||
|
||||
GGML_DEPRECATED(
|
||||
GGML_API ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size),
|
||||
GGML_BACKEND_API ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size),
|
||||
"obsoleted by the new device interface - https://github.com/ggerganov/llama.cpp/pull/9713");
|
||||
|
||||
GGML_API void ggml_backend_metal_set_abort_callback(ggml_backend_t backend, ggml_abort_callback abort_callback, void * user_data);
|
||||
GGML_BACKEND_API void ggml_backend_metal_set_abort_callback(ggml_backend_t backend, ggml_abort_callback abort_callback, void * user_data);
|
||||
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void);
|
||||
GGML_BACKEND_API 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);
|
||||
GGML_BACKEND_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);
|
||||
GGML_BACKEND_API void ggml_backend_metal_capture_next_compute(ggml_backend_t backend);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_metal_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_metal_reg(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
|
||||
@ -10,18 +10,18 @@ extern "C" {
|
||||
#define GGML_RPC_MAX_SERVERS 16
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_rpc_init(const char * endpoint);
|
||||
GGML_API bool ggml_backend_is_rpc(ggml_backend_t backend);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_rpc_init(const char * endpoint);
|
||||
GGML_BACKEND_API bool ggml_backend_is_rpc(ggml_backend_t backend);
|
||||
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint);
|
||||
|
||||
GGML_API void ggml_backend_rpc_get_device_memory(const char * endpoint, size_t * free, size_t * total);
|
||||
GGML_BACKEND_API void ggml_backend_rpc_get_device_memory(const char * endpoint, size_t * free, size_t * total);
|
||||
|
||||
GGML_API void ggml_backend_rpc_start_server(ggml_backend_t backend, const char * endpoint, size_t free_mem, size_t total_mem);
|
||||
GGML_BACKEND_API void ggml_backend_rpc_start_server(ggml_backend_t backend, const char * endpoint, size_t free_mem, size_t total_mem);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_rpc_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_rpc_reg(void);
|
||||
|
||||
GGML_API ggml_backend_dev_t ggml_backend_rpc_add_device(const char * endpoint);
|
||||
GGML_BACKEND_API ggml_backend_dev_t ggml_backend_rpc_add_device(const char * endpoint);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
|
||||
@ -17,32 +17,32 @@ extern "C" {
|
||||
#endif
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_sycl_init(int device);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_sycl_init(int device);
|
||||
|
||||
GGML_API bool ggml_backend_is_sycl(ggml_backend_t backend);
|
||||
GGML_BACKEND_API bool ggml_backend_is_sycl(ggml_backend_t backend);
|
||||
|
||||
// devide buffer
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device);
|
||||
GGML_BACKEND_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_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split);
|
||||
GGML_BACKEND_API 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_BACKEND_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 void ggml_backend_sycl_get_gpu_list(int *id_list, int max_len);
|
||||
GGML_API void ggml_backend_sycl_get_device_description(int device,
|
||||
GGML_BACKEND_API void ggml_backend_sycl_print_sycl_devices(void);
|
||||
GGML_BACKEND_API void ggml_backend_sycl_get_gpu_list(int *id_list, int max_len);
|
||||
GGML_BACKEND_API void ggml_backend_sycl_get_device_description(int device,
|
||||
char *description,
|
||||
size_t description_size);
|
||||
GGML_API int ggml_backend_sycl_get_device_count();
|
||||
GGML_API void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
|
||||
GGML_BACKEND_API int ggml_backend_sycl_get_device_count();
|
||||
GGML_BACKEND_API 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 bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size);
|
||||
// GGML_API void ggml_backend_sycl_unregister_host_buffer(void * buffer);
|
||||
// GGML_BACKEND_API bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size);
|
||||
// GGML_BACKEND_API void ggml_backend_sycl_unregister_host_buffer(void * buffer);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_sycl_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_sycl_reg(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
|
||||
@ -10,21 +10,21 @@ extern "C" {
|
||||
#define GGML_VK_NAME "Vulkan"
|
||||
#define GGML_VK_MAX_DEVICES 16
|
||||
|
||||
GGML_API void ggml_vk_instance_init(void);
|
||||
GGML_BACKEND_API void ggml_vk_instance_init(void);
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_vk_init(size_t dev_num);
|
||||
GGML_BACKEND_API ggml_backend_t ggml_backend_vk_init(size_t dev_num);
|
||||
|
||||
GGML_API bool ggml_backend_is_vk(ggml_backend_t backend);
|
||||
GGML_API int ggml_backend_vk_get_device_count(void);
|
||||
GGML_API void ggml_backend_vk_get_device_description(int device, char * description, size_t description_size);
|
||||
GGML_API void ggml_backend_vk_get_device_memory(int device, size_t * free, size_t * total);
|
||||
GGML_BACKEND_API bool ggml_backend_is_vk(ggml_backend_t backend);
|
||||
GGML_BACKEND_API int ggml_backend_vk_get_device_count(void);
|
||||
GGML_BACKEND_API void ggml_backend_vk_get_device_description(int device, char * description, size_t description_size);
|
||||
GGML_BACKEND_API void ggml_backend_vk_get_device_memory(int device, size_t * free, size_t * total);
|
||||
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_vk_buffer_type(size_t dev_num);
|
||||
GGML_BACKEND_API 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_backend_buffer_type_t ggml_backend_vk_host_buffer_type(void);
|
||||
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_vk_host_buffer_type(void);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_vk_reg(void);
|
||||
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_vk_reg(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
|
||||
@ -176,15 +176,15 @@
|
||||
#ifdef GGML_SHARED
|
||||
# if defined(_WIN32) && !defined(__MINGW32__)
|
||||
# ifdef GGML_BUILD
|
||||
# define GGML_API __declspec(dllexport)
|
||||
# define GGML_API __declspec(dllexport) extern
|
||||
# else
|
||||
# define GGML_API __declspec(dllimport)
|
||||
# define GGML_API __declspec(dllimport) extern
|
||||
# endif
|
||||
# else
|
||||
# define GGML_API __attribute__ ((visibility ("default")))
|
||||
# define GGML_API __attribute__ ((visibility ("default"))) extern
|
||||
# endif
|
||||
#else
|
||||
# define GGML_API
|
||||
# define GGML_API extern
|
||||
#endif
|
||||
|
||||
// TODO: support for clang
|
||||
@ -1490,7 +1490,7 @@ extern "C" {
|
||||
"use ggml_rope_ext_inplace instead");
|
||||
|
||||
// compute correction dims for YaRN RoPE scaling
|
||||
void ggml_rope_yarn_corr_dims(
|
||||
GGML_API void ggml_rope_yarn_corr_dims(
|
||||
int n_dims, int n_ctx_orig, float freq_base, float beta_fast, float beta_slow, float dims[2]);
|
||||
|
||||
// rotary position embedding backward, i.e compute dx from dy
|
||||
@ -2384,38 +2384,6 @@ extern "C" {
|
||||
GGML_API size_t gguf_get_meta_size(const struct gguf_context * ctx);
|
||||
GGML_API void gguf_get_meta_data(const struct gguf_context * ctx, void * data);
|
||||
|
||||
//
|
||||
// system info
|
||||
//
|
||||
|
||||
GGML_API int ggml_cpu_has_avx (void);
|
||||
GGML_API int ggml_cpu_has_avx_vnni (void);
|
||||
GGML_API int ggml_cpu_has_avx2 (void);
|
||||
GGML_API int ggml_cpu_has_avx512 (void);
|
||||
GGML_API int ggml_cpu_has_avx512_vbmi(void);
|
||||
GGML_API int ggml_cpu_has_avx512_vnni(void);
|
||||
GGML_API int ggml_cpu_has_avx512_bf16(void);
|
||||
GGML_API int ggml_cpu_has_amx_int8 (void);
|
||||
GGML_API int ggml_cpu_has_fma (void);
|
||||
GGML_API int ggml_cpu_has_arm_fma (void);
|
||||
GGML_API int ggml_cpu_has_metal (void);
|
||||
GGML_API int ggml_cpu_has_f16c (void);
|
||||
GGML_API int ggml_cpu_has_fp16_va (void);
|
||||
GGML_API int ggml_cpu_has_wasm_simd (void);
|
||||
GGML_API int ggml_cpu_has_blas (void);
|
||||
GGML_API int ggml_cpu_has_cuda (void);
|
||||
GGML_API int ggml_cpu_has_vulkan (void);
|
||||
GGML_API int ggml_cpu_has_kompute (void);
|
||||
GGML_API int ggml_cpu_has_gpublas (void);
|
||||
GGML_API int ggml_cpu_has_sse3 (void);
|
||||
GGML_API int ggml_cpu_has_ssse3 (void);
|
||||
GGML_API int ggml_cpu_has_riscv_v (void);
|
||||
GGML_API int ggml_cpu_has_sycl (void);
|
||||
GGML_API int ggml_cpu_has_rpc (void);
|
||||
GGML_API int ggml_cpu_has_vsx (void);
|
||||
GGML_API int ggml_cpu_has_cann (void);
|
||||
GGML_API int ggml_cpu_has_llamafile (void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
// restrict not standard in C++
|
||||
#define GGML_RESTRICT
|
||||
@ -2432,7 +2400,6 @@ extern "C" {
|
||||
size_t type_size;
|
||||
bool is_quantized;
|
||||
ggml_to_float_t to_float;
|
||||
ggml_from_float_t from_float;
|
||||
ggml_from_float_t from_float_ref;
|
||||
};
|
||||
|
||||
|
||||
Reference in New Issue
Block a user