* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
#ifndef BUILT_IN_LLM_COMMON_CACHE_MANAGER_H_
#define BUILT_IN_LLM_COMMON_CACHE_MANAGER_H_
#include <mutex>
#include "flow_func/meta_multi_func.h"
#include "flow_func/meta_params.h"
#include "flow_func/meta_run_context.h"
#include "flow_func/tensor_data_type.h"
#include "llm_common/llm_common.h"
#include "llm_common/cluster_manager.h"
namespace FlowFunc {
using CacheIndex = std::pair<uint64_t, uint64_t>;
class CacheManager {
public:
static CacheManager &GetInstance();
FsmStatus AllocateCache(const std::shared_ptr<MetaRunContext> &run_context,
const AllocateCacheReqInfo &req_info,
uint64_t *data_addresses);
FsmStatus DeallocateCache(int64_t cache_id);
FsmStatus RemoveCacheIndex(const CacheIndex &cache_index,
bool is_prefix,
uint32_t num_tensor_indices = 0,
const TransferInfo *tensor_indices = nullptr);
FsmStatus CopyBlockCache(const CopyCacheReqInfo &req_info) const;
FsmStatus CopyCache(const CopyCacheReqInfo &req_info) const;
FsmStatus PullCache(PullKvReqInfo *req_info, ClusterManager &cluster_manager, uint64_t start_tick) const;
FsmStatus TransferCache(TransferKvReqInfo *req_info, ClusterManager &cluster_manager, uint64_t start_tick) const;
bool GetCacheEntry(const CacheIndex &cache_index, CacheEntry &cache_entry, bool is_prefix) const;
bool GetCacheEntry(int64_t cache_id, CacheEntry &cache_entry) const;
bool GetCacheIndex(const std::pair<int64_t, uint32_t> &cache_id_and_batch_index, CacheIndex &cache_index) const;
void Enable();
bool IsEnabled() const;
FsmStatus CheckAddr(uint64_t addr, uint64_t check_size);
private:
void RemoveAddrToSize(const CacheEntry &cache_entry);
static FsmStatus CheckCopyCacheParam(const CacheEntry &src_cache_entry, const CacheEntry &dst_cache_entry,
const CopyCacheReqInfo &req_info, uint64_t ©_size);
void AddCacheIndices(CacheEntry &cache_entry, const AllocateCacheReqInfo &req_info,
const std::vector<CacheIndex> &cache_indices);
void AddCacheIndicesForBlocks(const AllocateCacheReqInfo &req_info,
const std::vector<CacheIndex> &cache_indices);
FsmStatus CheckAndCreateCacheIndices(const AllocateCacheReqInfo &req_info,
std::vector<CacheIndex> &cache_indices) const;
const CacheEntry *DoGetCacheEntry(int64_t cache_id) const;
static FsmStatus CheckTensorIndices(const PullKvReqInfo &req_info,
const CacheEntry &cache_entry,
bool enable_page_attention);
static FsmStatus CheckReqInfo(const CacheEntry &cache_entry, PullKvReqInfo *req_info, bool &enable_page_attention);
mutable std::mutex mu_;
bool enabled_ = false;
std::map<int64_t, CacheEntry> cache_id_to_entry_;
std::map<CacheIndex, int64_t> cache_index_to_id_;
std::map<CacheIndex, int64_t> prefix_index_to_id_;
std::map<std::pair<int64_t, uint32_t>, CacheIndex> cache_id_and_batch_id_to_cache_index_;
std::map<uint64_t, uint64_t> addr_to_size_;
};
}
#endif