35 lines
1.1 KiB
C
35 lines
1.1 KiB
C
#pragma once
|
|
|
|
#include "ggml.h"
|
|
#include "ggml-backend.h"
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
#define GGML_VK_NAME "Vulkan"
|
|
|
|
GGML_API void ggml_vk_init(void);
|
|
|
|
GGML_API void ggml_vk_preallocate_buffers_graph(struct ggml_tensor * node);
|
|
GGML_API void ggml_vk_preallocate_buffers(void);
|
|
GGML_API void ggml_vk_build_graph(struct ggml_tensor * node, bool last_node);
|
|
GGML_API bool ggml_vk_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor);
|
|
#ifdef GGML_VULKAN_CHECK_RESULTS
|
|
void ggml_vk_check_results_1(struct ggml_compute_params * params, struct ggml_tensor * tensor);
|
|
#endif
|
|
GGML_API void ggml_vk_graph_cleanup(void);
|
|
|
|
// backend API
|
|
GGML_API GGML_CALL ggml_backend_t ggml_backend_vk_init(void);
|
|
|
|
GGML_API GGML_CALL bool ggml_backend_is_vk(ggml_backend_t backend);
|
|
|
|
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_vk_buffer_type(void);
|
|
// 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_vk_host_buffer_type(void);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|