llama.cpp/ggml-alloc.h
slaren a113689571
ggml : add graph tensor allocator (#2411)
* ggml : add graph tensor allocator

* ggml : don't calculate data pointer of unallocated tensors when creating a view with an offset

* ggml : refactor ggml_view_Nd into ggml_view_tensor_offset
2023-07-30 15:58:01 +02:00

23 lines
653 B
C

#pragma once
#include "ggml.h"
#ifdef __cplusplus
extern "C" {
#endif
GGML_API struct ggml_allocr * ggml_allocr_new(void * data, size_t size, size_t alignment);
GGML_API struct ggml_allocr * ggml_allocr_new_measure(size_t alignment);
GGML_API void ggml_allocr_free(struct ggml_allocr * alloc);
GGML_API bool ggml_allocr_is_measure(struct ggml_allocr * alloc);
GGML_API void ggml_allocr_reset(struct ggml_allocr * alloc);
GGML_API void ggml_allocr_alloc(struct ggml_allocr * alloc, struct ggml_tensor * tensor);
GGML_API size_t ggml_allocr_alloc_graph(struct ggml_allocr * alloc, struct ggml_cgraph * graph);
#ifdef __cplusplus
}
#endif