mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2025-01-11 03:01:45 +00:00
06b00827a0
* removed ggml_task_backend, infavour of ggml_task_profile.runner and newly added id and name. * extracted mul_mat blas codes into ggml_compute_forward_mul_mat_blas, thus align with CUDA/CL a bit more and make it easier to fix profile and run tune. * rewrote task profile and update/add some cuda/cl codes, finnaly made CL GPU offloading work. * misc minor fix/update to tune, the data format was changed.
40 lines
1.4 KiB
C
40 lines
1.4 KiB
C
#pragma once
|
|
|
|
#include "ggml.h"
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
#define GGML_CUDA_MAX_DEVICES 16
|
|
|
|
struct ggml_tensor_extra_gpu {
|
|
void * data_device[GGML_CUDA_MAX_DEVICES]; // 1 pointer for each device for split tensors
|
|
};
|
|
|
|
void ggml_init_cublas(void);
|
|
void ggml_cuda_set_tensor_split(const float * tensor_split);
|
|
|
|
void ggml_cuda_mul(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
|
|
bool ggml_cuda_is_gpu_offloading(const struct ggml_tensor * src0);
|
|
size_t ggml_cuda_mul_mat_get_wsize(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
|
|
void ggml_cuda_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst, void * wdata, size_t wsize);
|
|
|
|
// TODO: export these with GGML_API
|
|
void * ggml_cuda_host_malloc(size_t size);
|
|
void ggml_cuda_host_free(void * ptr);
|
|
|
|
void ggml_cuda_transform_tensor(void * data, struct ggml_tensor * tensor);
|
|
|
|
void ggml_cuda_free_data(struct ggml_tensor * tensor);
|
|
void ggml_cuda_assign_buffers(struct ggml_tensor * tensor);
|
|
void ggml_cuda_assign_buffers_no_scratch(struct ggml_tensor * tensor);
|
|
void ggml_cuda_set_main_device(int main_device);
|
|
void ggml_cuda_set_scratch_size(size_t scratch_size);
|
|
void ggml_cuda_free_scratch(void);
|
|
bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|