/**
 * 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.
 */

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_dequant_rope_quant_kvcache.h"

#define CHECK_RET(cond, return_expr) \
  do {                               \
    if (!(cond)) {                   \
      return_expr;                   \
    }                                \
  } while (0)

#define LOG_PRINT(message, ...)     \
  do {                              \
    printf(message, ##__VA_ARGS__); \
  } while (0)

int64_t GetShapeSize(const std::vector<int64_t>& shape) {
  int64_t shapeSize = 1;
  for (auto i : shape) {
    shapeSize *= i;
  }
  return shapeSize;
}

void PrintOutResult(std::vector<int64_t> &shape, void** deviceAddr) {
  auto size = GetShapeSize(shape);
  std::vector<int8_t> resultData(size, 0);
  auto ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]),
                         *deviceAddr, size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return);
  for (int64_t i = 0; i < size; i++) {
    LOG_PRINT("mean result[%ld] is: %d\n", i, resultData[i]);
  }
}

int Init(int32_t deviceId, aclrtStream* stream) {
  // 固定写法，资源初始化
  auto ret = aclInit(nullptr);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret);
  ret = aclrtSetDevice(deviceId);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
  ret = aclrtCreateStream(stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret);
  return 0;
}

template <typename T>
int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
                    aclDataType dataType, aclTensor** tensor) {
  auto size = GetShapeSize(shape) * sizeof(T);
  // 调用aclrtMalloc申请device侧内存
  auto ret = aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMalloc failed. ERROR: %d\n", ret); return ret);
  // 调用aclrtMemcpy将host侧数据拷贝到device侧内存上
  ret = aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMemcpy failed. ERROR: %d\n", ret); return ret);

  // 计算连续tensor的strides
  std::vector<int64_t> strides(shape.size(), 1);
  for (int64_t i = shape.size() - 2; i >= 0; i--) {
    strides[i] = shape[i + 1] * strides[i + 1];
  }

  // 调用aclCreateTensor接口创建aclTensor
  *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
                            shape.data(), shape.size(), *deviceAddr);
  return 0;
}

int main() {
  // 1. 固定写法，device/stream初始化, 参考AscendCL对外接口列表
  // 根据自己的实际device填写deviceId
  int32_t deviceId = 0;
  aclrtStream stream;
  auto ret = Init(deviceId, &stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);

// 2. 构造输入与输出，需要根据API的接口定义构造
  int64_t shapeB = 1;
  int64_t shapeS = 1;
  int64_t shapeNq = 2;
  int64_t shapeNkv = 1;
  int64_t shapeD = 32;
  int64_t shapeH = shapeD * (shapeNq + shapeNkv + shapeNkv);
  std::vector<int64_t> inputShape = {shapeB, shapeS, shapeH};
  std::vector<int64_t> cosShape = {shapeB, shapeS, 1, shapeD};
  std::vector<int64_t> sinShape = {shapeB, shapeS, 1, shapeD};
  std::vector<int64_t> kcacheShape = {shapeB, shapeH, 1, shapeD};
  std::vector<int64_t> vcacheShape = {shapeB, shapeH, 1, shapeD};
  std::vector<int64_t> indicesShape = {shapeB};
  std::vector<int64_t> kscaleShape = {shapeD};
  std::vector<int64_t> vscaleShape = {shapeD};
  std::vector<int64_t> koffsetShape = {shapeD};
  std::vector<int64_t> voffsetShape = {shapeD};

  std::vector<int64_t> weightShape = {shapeH};
  std::vector<int64_t> activationShape = {shapeB};
  std::vector<int64_t> biasShape = {shapeH};

  std::vector<int16_t> inputHostData(shapeB * shapeS * shapeH, 1);
  std::vector<int16_t> cosHostData(shapeB * shapeS * shapeD, 1);
  std::vector<int16_t> sinHostData(shapeB * shapeS * shapeD, 1);
  std::vector<int8_t> kcacheHostData(shapeB * shapeH * shapeD, 6);
  std::vector<int8_t> vcacheHostData(shapeB * shapeH * shapeD, 6);
  std::vector<int32_t> indicesHostData(shapeB, 0);
  std::vector<int32_t> kscaleHostData(shapeD, 2);
  std::vector<int32_t> vscaleHostData(shapeD, 2);
  std::vector<int32_t> koffsetHostData(shapeD, 2);
  std::vector<int32_t> voffsetHostData(shapeD, 2);

  std::vector<int32_t> weightHostData(shapeH, 2);
  std::vector<int32_t> activationHostData(shapeB, 2);
  std::vector<int32_t> biasHostData(shapeH, 2);

  void* inputDeviceAddr = nullptr;
  void* cosDeviceAddr = nullptr;
  void* sinDeviceAddr = nullptr;
  void* kcacheDeviceAddr = nullptr;
  void* vcacheDeviceAddr = nullptr;
  void* indicesDeviceAddr = nullptr;
  void* kscaleDeviceAddr = nullptr;
  void* vscaleDeviceAddr = nullptr;
  void* koffsetDeviceAddr = nullptr;
  void* voffsetDeviceAddr = nullptr;

  void* weightDeviceAddr = nullptr;
  void* activationDeviceAddr = nullptr;
  void* biasDeviceAddr = nullptr;

  aclTensor* input = nullptr;
  aclTensor* cos = nullptr;
  aclTensor* sin = nullptr;
  aclTensor* kcache = nullptr;
  aclTensor* vcache = nullptr;
  aclTensor* indices = nullptr;
  aclTensor* kscale = nullptr;
  aclTensor* vscale = nullptr;
  aclTensor* koffset = nullptr;
  aclTensor* voffset = nullptr;
  aclTensor* weight = nullptr;
  aclTensor* activation = nullptr;
  aclTensor* bias = nullptr;

  ret = CreateAclTensor(inputHostData, inputShape, &inputDeviceAddr, aclDataType::ACL_INT32, &input);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(cosHostData, cosShape, &cosDeviceAddr, aclDataType::ACL_FLOAT16, &cos);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(sinHostData, sinShape, &sinDeviceAddr, aclDataType::ACL_FLOAT16, &sin);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(kcacheHostData, kcacheShape, &kcacheDeviceAddr, aclDataType::ACL_INT8, &kcache);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(vcacheHostData, vcacheShape, &vcacheDeviceAddr, aclDataType::ACL_INT8, &vcache);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(indicesHostData, indicesShape, &indicesDeviceAddr, aclDataType::ACL_INT32, &indices);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(kscaleHostData, kscaleShape, &kscaleDeviceAddr, aclDataType::ACL_FLOAT, &kscale);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(vscaleHostData, vscaleShape, &vscaleDeviceAddr, aclDataType::ACL_FLOAT, &vscale);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(koffsetHostData, koffsetShape, &koffsetDeviceAddr, aclDataType::ACL_FLOAT, &koffset);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(voffsetHostData, voffsetShape, &voffsetDeviceAddr, aclDataType::ACL_FLOAT, &voffset);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  ret = CreateAclTensor(weightHostData, weightShape, &weightDeviceAddr, aclDataType::ACL_FLOAT, &weight);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(activationHostData, activationShape, &activationDeviceAddr, aclDataType::ACL_FLOAT, &activation);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(biasHostData, biasShape, &biasDeviceAddr, aclDataType::ACL_FLOAT, &bias);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  std::vector<int64_t> qShape = {shapeB, shapeS, shapeNq, shapeD};
  std::vector<int16_t> qHostData(shapeB * shapeS * shapeNq * shapeD, 9);
  aclTensor *q = nullptr;
  void *qDeviceAddr = nullptr;

  std::vector<int64_t> kShape = {shapeB, shapeS, shapeNkv, shapeD};
  std::vector<int16_t> kHostData(shapeB * shapeS * shapeNkv * shapeD, 10);
  aclTensor *k = nullptr;
  void *kDeviceAddr = nullptr;
  std::vector<int64_t> vShape = {shapeB, shapeS, shapeNkv, shapeD};
  std::vector<int16_t> vHostData(shapeB * shapeS * shapeNkv * shapeD, 10);
  aclTensor* v = nullptr;
  void* vDeviceAddr = nullptr;

  ret = CreateAclTensor(qHostData, qShape, &qDeviceAddr, aclDataType::ACL_FLOAT16, &q);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(kHostData, kShape, &kDeviceAddr, aclDataType::ACL_FLOAT16, &k);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(vHostData, vShape, &vDeviceAddr, aclDataType::ACL_FLOAT16, &v);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  std::vector<int64_t> splitData = {shapeNq * shapeD, shapeNkv * shapeD, shapeNkv * shapeD};
  aclIntArray *sizeSplits = aclCreateIntArray(splitData.data(), splitData.size());

  // 3. 调用CANN算子库API，需要修改为具体的API
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;

  // 调用aclnnDequantRopeQuantKvcache第一段接口
  ret = aclnnDequantRopeQuantKvcacheGetWorkspaceSize(input, cos, sin, kcache, vcache, indices, kscale, vscale, koffset,
                                                     voffset, weight, activation, bias,sizeSplits, "static", "BSND", true,
                                                     "contiguous", q, k, v, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnDequantRopeQuantKvcacheGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);

  // 根据第一段接口计算出的workspaceSize申请device内存
  void* workspaceAddr = nullptr;
  if (workspaceSize > 0) {
    ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
  }

  // 调用aclnnDequantRopeQuantKvcache第二段接口
  ret = aclnnDequantRopeQuantKvcache(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnDequantRopeQuantKvcache failed. ERROR: %d\n", ret); return ret);

  // 4. 固定写法，同步等待任务执行结束
  ret = aclrtSynchronizeStream(stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);

  // 5. 获取输出的值，将device侧内存上的结果拷贝至host侧，需要根据具体API的接口定义修改
  PrintOutResult(kcacheShape, &kcacheDeviceAddr);
  PrintOutResult(vcacheShape, &vcacheDeviceAddr);

  // 6. 释放aclTensor和aclScalar，需要根据具体API的接口定义修改
  aclDestroyTensor(input);
  aclDestroyTensor(q);

  // 7. 释放device 资源
  aclrtFree(inputDeviceAddr);
  aclrtFree(qDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}