forked from ggerganov/whisper.cpp
/
ggml-cuda.h
23 lines (17 loc) · 906 Bytes
/
ggml-cuda.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
#include "ggml.h"
#ifdef __cplusplus
extern "C" {
#endif
void ggml_init_cublas(void);
void ggml_cuda_mul(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
bool ggml_cuda_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
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(struct ggml_tensor * tensor);
void ggml_cuda_load_data(const char * fname, struct ggml_tensor * tensors, size_t offset);
#ifdef __cplusplus
}
#endif