llama.cpp/ggml-cuda.h

22 lines
701 B
C
Raw Normal View History

#include "ggml.h"
#ifdef __cplusplus
extern "C" {
#endif
void ggml_init_cublas(void);
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);
#ifdef __cplusplus
}
#endif