Line data    Source code

       1              : /**
       2              :  * Copyright (c) 2025 Huawei Technologies Co., Ltd.
       3              :  * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
       4              :  * CANN Open Software License Agreement Version 2.0 (the "License").
       5              :  * Please refer to the License for details. You may not use this file except in compliance with the License.
       6              :  * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
       7              :  * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
       8              :  * See LICENSE in the root of the software repository for the full text of the License.
       9              :  */
      10              : #ifndef HCCL_CCU_SUPER_FAST_LOAD_H
      11              : #define HCCL_CCU_SUPER_FAST_LOAD_H
      12              : #include <cstdlib>
      13              : #include <array>
      14              : #include <vector>
      15              : #include <cstring>
      16              : #include <utility>
      17              : #include <cstddef>
      18              : 
      19              : #include "securec.h"
      20              : #include "string_util.h"
      21              : 
      22              : #include "types.h"
      23              : #include "op_type.h"
      24              : #include "orion_adapter_rts.h"
      25              : #include "ccu/ccu_task_param.h"
      26              : #include "task_param.h"
      27              : #include "ccu_ins.h"
      28              : #include "acl/acl_rt.h"
      29              : 
      30              : namespace Hccl {
      31              : using CcuParamsMappingKeyType = std::uint32_t;
      32              : constexpr std::size_t CCU_SFL_PARAM_KEY_LEN = 3;
      33              : using CcuSFLMappingKey = std::array<CcuParamsMappingKeyType, CCU_SFL_PARAM_KEY_LEN>;
      34              : struct CachedCCUParams {
      35              : public:
      36              :     rtCcuTaskInfo_t* ccuParams{nullptr};
      37              :     std::vector<std::size_t> count;
      38              :     std::vector<TaskParam> taskParams;
      39              :     u64 execId{0};
      40              :     std::size_t totalCounts{0};
      41              :     CcuInstType insType{};
      42              :     bool isSlave{false};
      43              : 
      44            0 :     CachedCCUParams() = default;
      45              :     explicit CachedCCUParams(
      46              :         std::vector<std::vector<Hccl::CcuTaskParam>>&& ccuInstruction,
      47              :         std::vector<std::vector<CcuProfilingInfo>>&& profilingInfo, std::size_t execId, CcuInstType insType,
      48              :         bool isSlave, void* comm);
      49              : 
      50              :     CachedCCUParams(const CachedCCUParams&) = delete;
      51              :     CachedCCUParams& operator=(const CachedCCUParams&) = delete;
      52              : 
      53              :     CachedCCUParams(CachedCCUParams&& other) noexcept;
      54              :     CachedCCUParams& operator=(CachedCCUParams&& other) noexcept;
      55              : 
      56              :     ~CachedCCUParams();
      57              : 
      58              : private:
      59           13 :     inline void* aligned_malloc(size_t align, size_t size) const
      60              :     {
      61           13 :         void* p = nullptr;
      62           13 :         if (posix_memalign(&p, align, size) != 0) {
      63            0 :             throw std::bad_alloc();
      64              :         }
      65           13 :         return p;
      66              :     }
      67              : 
      68           13 :     inline void aligned_free(void* ptr) const
      69              :     {
      70           13 :         if (ptr) {
      71           13 :             std::free(ptr);
      72           13 :             ptr = nullptr;
      73              :         }
      74           13 :     }
      75              : 
      76           13 :     inline bool is_power_of_2(std::size_t x) const { return static_cast<bool>(x) && !static_cast<bool>((x & (x - 1))); }
      77           13 :     inline std::size_t round_up_to(std::size_t size, std::size_t alignment) const
      78              :     {
      79           13 :         return alignment != 0 ? (size + alignment - 1) / alignment * alignment : size;
      80              :     }
      81           13 :     inline void* alloc_aligned_raw(std::size_t alignment, std::size_t size) const
      82              :     {
      83           13 :         if (alignment == 0) {
      84            0 :             alignment = alignof(std::max_align_t);
      85              :         }
      86           13 :         if (!is_power_of_2(alignment)) {
      87            0 :             return nullptr;
      88              :         };
      89           13 :         return aligned_malloc(alignment, round_up_to(size, alignment));
      90              :     }
      91              : 
      92              :     rtCcuTaskInfo_t*
      93              :     alloc_and_memcpy_aligned(const std::vector<std::vector<rtCcuTaskInfo_t>>& vecs, std::size_t alignment);
      94              : };
      95              : 
      96            9 : inline void SuperFastLoad(rtCcuTaskInfo_t* params, aclrtStream const streamPtr, int counts)
      97              : {
      98           27 :     for (int i = 0; i < counts; ++i) {
      99           18 :         HrtCcuLaunch(params[i], streamPtr);
     100              :     }
     101            9 : }
     102              : } // namespace Hccl
     103              : 
     104              : namespace std {
     105              : constexpr std::size_t NUMBER_SIX = 6;
     106              : constexpr std::size_t NUMBER_TWO = 2;
     107              : struct ArrayHasher {
     108              :     std::size_t
     109           28 :     operator()(const std::array<Hccl::CcuParamsMappingKeyType, Hccl::CCU_SFL_PARAM_KEY_LEN>& ccuParams) const noexcept
     110              :     {
     111           28 :         std::size_t hashVal = 0;
     112          112 :         for (auto ccuParam : ccuParams) {
     113           84 :             hashVal ^= std::hash<std::uint32_t>{}(ccuParam) + 0x9e3779b9 + (hashVal << NUMBER_SIX)
     114           84 :                        + (hashVal >> NUMBER_TWO);
     115              :         }
     116           28 :         return hashVal;
     117              :     }
     118              : };
     119              : 
     120              : template <>
     121              : struct hash<Hccl::OpType> {
     122              :     std::size_t operator()(const Hccl::OpType& type) const noexcept { return static_cast<std::size_t>(type); }
     123              : };
     124              : template <>
     125              : struct hash<const Hccl::OpType> {
     126           48 :     std::size_t operator()(const Hccl::OpType& type) const noexcept { return static_cast<std::size_t>(type); }
     127              : };
     128              : }; // namespace std
     129              : #endif // HCCL_CCU_SUPER_FAST_LOAD_H