mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-10-30 08:42:00 +00:00
a0b3ac8c48
This change makes it possible to build ggml-cuda.cu and ggml-metal.m as independent dynamic shared objects, that may be conditionally linked at runtime in a multiplatform binary. It introduces a GGML_CALL annotation that documents which functions have a cyclic call relationship, between the application code and GPU modules. This change does nothing, unless the build defines -DGGML_MULTIPLATFORM which causes back-references and function pointers to conform to MS ABI which is supported by NVCC, ROCm, XCode, GCC and Clang across platforms
53 lines
2.1 KiB
C
53 lines
2.1 KiB
C
#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
|
|
|
|
// Always success. To check if CUDA is actually loaded, use `ggml_cublas_loaded`.
|
|
GGML_API GGML_CALL void ggml_init_cublas(void);
|
|
|
|
// Returns `true` if there are available CUDA devices and cublas loads successfully; otherwise, it returns `false`.
|
|
GGML_API GGML_CALL bool ggml_cublas_loaded(void);
|
|
|
|
GGML_API GGML_CALL void * ggml_cuda_host_malloc(size_t size);
|
|
GGML_API GGML_CALL void ggml_cuda_host_free(void * ptr);
|
|
|
|
GGML_API GGML_CALL bool ggml_cuda_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
|
|
GGML_API GGML_CALL bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor);
|
|
|
|
GGML_API GGML_CALL int ggml_cuda_get_device_count(void);
|
|
GGML_API GGML_CALL void ggml_cuda_get_device_description(int device, char * description, size_t description_size);
|
|
|
|
// 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);
|
|
|
|
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);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|