Line data    Source code

       1              : /**
       2              :  * Copyright (c) 2026 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              : 
      11              : #ifndef CCU_MICROCODE_OPT_INSTRUCTION_SCHEDULER_H
      12              : #define CCU_MICROCODE_OPT_INSTRUCTION_SCHEDULER_H
      13              : 
      14              : #include <cstdint>
      15              : #include <vector>
      16              : 
      17              : #include "ccu_instr_info_v1.h"
      18              : #include "extract_operands.h"
      19              : 
      20              : namespace hcomm {
      21              : namespace CcuOpt {
      22              : 
      23              :     struct SchedulerStats {
      24              :         uint32_t nopRemoved = 0;     // 输入序列里被剥离的 NOP 数 (CkeOnly 不剥离, 恒 0)
      25              :         uint32_t nopInserted = 0;    // 调度过程因 latency 阻塞而新插入的 NOP 数
      26              :         uint32_t instrReordered = 0; // 与输入顺序不同的真实指令数 (CkeOnly 不重排, 恒 0)
      27              :         uint32_t basicBlocks = 0;    // 切分得到的 BB 个数 (CkeOnly 不切分, 占位为 1)
      28              : 
      29              :         // originIndex[i] = 调度后第 i 条指令对应的输入本地下标; -1 表示新插入的填充 NOP.
      30              :         std::vector<int32_t> originIndex{};
      31              :         std::vector<uint16_t> strippedNopIndices{}; // CkeOnly 不剥离, 恒为空.
      32              :     };
      33              : 
      34              :     // 指令调度算法档位. 极简后端优化只提供 CkeOnly 一档:
      35              :     //  * CkeOnly: 默认档 (唯一档). 完全不重排 / 不剥离, 只识别 CKE 寄存器的写后读 (setcke ->
      36              :     //             waitcke/clearcke) 并按固定 cke latency 补 NOP; XN / MS 写后读交由硬件
      37              :     //             interlock 处理, 不补任何 NOP. 配合指令空间预留可保证优化后指令数不越界.
      38              :     enum class SchedLevel : uint8_t {
      39              :         CkeOnly = 0,
      40              :     };
      41              : 
      42              :     struct InstructionSchedulerOptions {
      43              :         SchedLevel level = SchedLevel::CkeOnly;
      44              :     };
      45              : 
      46              :     class InstructionScheduler {
      47              :     public:
      48           21 :         explicit InstructionScheduler(InstructionSchedulerOptions opts = {}) : opts_(opts) {}
      49              : 
      50              :         CcuRep::CcuInstrInfo Schedule(const CcuRep::CcuInstrInfo& input);
      51              : 
      52           24 :         const SchedulerStats& Stats() const { return stats_; }
      53              : 
      54              :     private:
      55              :         InstructionSchedulerOptions opts_;
      56              :         SchedulerStats stats_{};
      57              : 
      58              :         // CkeOnly (默认档) 实现: 仅顺序扫描 + 只对 CKE 写后读按固定 cke latency 补 NOP,
      59              :         // XN / MS 写后读不补 (硬件 interlock).
      60              :         CcuRep::CcuInstrInfo ScheduleCkeOnly(const CcuRep::CcuInstrInfo& input);
      61              :     };
      62              : 
      63              : } // namespace CcuOpt
      64              : } // namespace hcomm
      65              : 
      66              : #endif // CCU_MICROCODE_OPT_INSTRUCTION_SCHEDULER_H