/**
* 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.
*/


/* !
 * \file init_global_memory.asc
 * \brief
 */

#include "acl/acl.h"
#include "data_utils.h"
#include "kernel_operator.h"
#include "tiling/tiling_api.h"

namespace TEST_CASE {
__aicore__ inline void UnPadTilingFunc(const AscendC::ShapeInfo srcShape, const uint32_t stackBufferSize,
                                       const uint32_t typeSize, UnPadTiling& tiling)
{
    // common
    uint32_t srcHeight = srcShape.shape[0];
    uint32_t srcWidth = srcShape.shape[1];

    tiling.srcHeight = srcHeight;
    tiling.srcWidth = srcWidth;

    uint32_t baseBlockLen = 16 * AscendC::ONE_BLK_SIZE;
    uint32_t baseBlockSize = baseBlockLen / typeSize;
    uint32_t tmpBuffer1BlockNum = stackBufferSize * sizeof(uint8_t) / typeSize / baseBlockSize / 2;
    uint32_t tmpBuffer2Offset = tmpBuffer1BlockNum * baseBlockSize;

    tiling.tmpBuffer1BlockNum = tmpBuffer1BlockNum;
    tiling.tmpBuffer1RowNum = 16 * tmpBuffer1BlockNum;
    tiling.tmpBuffer2Offset = tmpBuffer2Offset;

    uint32_t widthTiling = 16 * tmpBuffer1BlockNum;
    uint32_t widthFractal = srcWidth / widthTiling;
    uint32_t widthFractalTail = srcWidth % widthTiling;
    if (widthFractalTail) {
        widthFractal += 1;
    } else {
        widthFractalTail = widthTiling;
    }

    tiling.widthTiling = widthTiling;
    tiling.widthFractal = widthFractal;
    tiling.widthFractalTail = widthFractalTail;
}

template <typename T>
class UnPad {
public:
    __aicore__ inline UnPad() {}
    __aicore__ inline void Init(GM_ADDR dstGm, GM_ADDR srcGm, uint16_t heightIn, uint16_t widthIn, uint16_t oriWidthIn,
                                AscendC::UnPadParams& unpadParamsIn, AscendC::TPipe* pipeIn)
    {
        pipe = pipeIn;
        height = heightIn;
        width = widthIn;
        oriWidth = oriWidthIn;
        unpadParams = unpadParamsIn;
        srcGlobal.SetGlobalBuffer((__gm__ T*)srcGm);
        dstGlobal.SetGlobalBuffer((__gm__ T*)dstGm);

        pipe->InitBuffer(inQueueSrcVecIn, 1, height * width * sizeof(T));
        pipe->InitBuffer(inQueueSrcVecOut, 1, height * (width) * sizeof(T));
    }
    __aicore__ inline void Process()
    {
        CopyIn();
        Compute();
        CopyOut();
    }

private:
    __aicore__ inline void CopyIn()
    {
        AscendC::LocalTensor<T> srcLocal = inQueueSrcVecIn.AllocTensor<T>();
        DataCopy(srcLocal, srcGlobal, height * width);
        inQueueSrcVecIn.EnQue(srcLocal);
    }
    __aicore__ inline void Compute()
    {
        AscendC::LocalTensor<T> dstLocal = inQueueSrcVecIn.DeQue<T>();
        AscendC::LocalTensor<T> srcOutLocal = inQueueSrcVecOut.AllocTensor<T>();

        uint32_t shape[] = {height, width};
        uint32_t oriShape[] = {height, oriWidth};
        AscendC::ShapeInfo shapeInfo(2, shape, 2, oriShape, AscendC::DataFormat::ND);
        dstLocal.SetShapeInfo(shapeInfo);

        uint32_t shape1[] = {height, static_cast<uint32_t>(width - unpadParams.rightPad)};
        uint32_t oriShape1[] = {height, oriWidth};
        AscendC::ShapeInfo shapeInfo1(2, shape1, 2, oriShape1, AscendC::DataFormat::ND);
        srcOutLocal.SetShapeInfo(shapeInfo1);

        AscendC::LocalTensor<uint8_t> stackBuffer;
        AscendC::PopStackBuffer<uint8_t, AscendC::TPosition::LCM>(stackBuffer);
        uint32_t stackBufferSize = stackBuffer.GetSize();

        UnPadTiling tiling;
        UnPadTilingFunc(shapeInfo, stackBufferSize, sizeof(T), tiling);

        AscendC::UnPad(srcOutLocal, dstLocal, unpadParams, tiling);
        inQueueSrcVecOut.EnQue(srcOutLocal);
        inQueueSrcVecIn.FreeTensor(dstLocal);
    }
    __aicore__ inline void CopyOut()
    {
        AscendC::LocalTensor<T> srcOutLocalDe = inQueueSrcVecOut.DeQue<T>();
        AscendC::DataCopy(dstGlobal, srcOutLocalDe, height * (width - unpadParams.leftPad - unpadParams.rightPad));
        inQueueSrcVecOut.FreeTensor(srcOutLocalDe);
    }

private:
    AscendC::TPipe* pipe;
    AscendC::TQue<AscendC::QuePosition::VECIN, 1> inQueueSrcVecIn;
    AscendC::TQue<AscendC::QuePosition::VECOUT, 1> inQueueSrcVecOut;
    AscendC::GlobalTensor<T> srcGlobal;
    AscendC::GlobalTensor<T> dstGlobal;

    uint16_t height;
    uint16_t width;
    uint16_t oriWidth;

    AscendC::UnPadParams unpadParams;
};

} // namespace TEST_CASE

extern "C" __global__ __aicore__ void unpad_custom(GM_ADDR src_gm, GM_ADDR dst_gm)
{
    KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
    AscendC::TPipe pipe;
    TEST_CASE::UnPad<float> op;
    AscendC::UnPadParams unpadParams{0, 1};
    op.Init(dst_gm, src_gm, 8, 8, 8, unpadParams, &pipe);
    op.Process();
}

static bool CompareResult(const void* outputData, uint32_t outSize)
{
    void* goldenData;
    aclrtMallocHost((void**)(&goldenData), outSize);
    size_t goldenSize = outSize;
    bool ret = ReadFile("./output/golden.bin", goldenSize, goldenData, goldenSize);
    if (ret) {
        printf("ReadFile golden.bin success!\n");
    } else {
        printf("test failed!\n");
        return false;
    }
    constexpr float EPS = 1e-4;
    int64_t wrongNum = 0;

    for (size_t i = 0; i < outSize / sizeof(float); i++) {
        float a = (reinterpret_cast<const float*>(outputData))[i];
        float b = (reinterpret_cast<const float*>(goldenData))[i];
        float ae = std::abs(a - b);
        float re = ae / std::abs(b);
        if (ae > EPS && re > EPS) {
            printf("CompareResult golden.bin failed output is %lf, golden is %lf\n", a, b);
            wrongNum++;
        }
    }
    aclrtFreeHost(goldenData);
    if (wrongNum != 0) {
        printf("wrongNum: %ld\n", wrongNum);
        return false;
    } else {
        printf("CompareResult golden.bin success!\n");
        return true;
    }
}

int32_t main(int32_t argc, char* argv[])
{
    size_t param1FileSize = 64 * sizeof(float);
    size_t param2FileSize = 56 * sizeof(float);
    uint32_t numBlocks = 1;

    aclInit(nullptr);
    aclrtContext context;
    int32_t deviceId = 0;
    aclrtSetDevice(deviceId);
    aclrtCreateContext(&context, deviceId);
    aclrtStream stream = nullptr;
    aclrtCreateStream(&stream);

    uint8_t* param1Host;
    uint8_t* param1Device;
    aclrtMallocHost((void**)(&param1Host), param1FileSize);
    aclrtMalloc((void**)&param1Device, param1FileSize, ACL_MEM_MALLOC_HUGE_FIRST);
    ReadFile("./input/input_x.bin", param1FileSize, param1Host, param1FileSize);
    aclrtMemcpy(param1Device, param1FileSize, param1Host, param1FileSize, ACL_MEMCPY_HOST_TO_DEVICE);

    uint8_t* param2Host;
    uint8_t* param2Device;
    aclrtMallocHost((void**)(&param2Host), param2FileSize);
    aclrtMalloc((void**)&param2Device, param2FileSize, ACL_MEM_MALLOC_HUGE_FIRST);

    unpad_custom<<<numBlocks, nullptr, stream>>>(param1Device, param2Device);

    aclrtSynchronizeStream(stream);

    aclrtFree(param1Device);
    aclrtFreeHost(param1Host);

    aclrtMemcpy(param2Host, param2FileSize, param2Device, param2FileSize, ACL_MEMCPY_DEVICE_TO_HOST);

    WriteFile("./output/output.bin", param2Host, param2FileSize);

    bool goldenResult = true;
    goldenResult = CompareResult(param2Host, param2FileSize);
    if (goldenResult) {
        printf("test pass!\n");
    } else {
        printf("test failed!\n");
    }

    aclrtFree(param2Device);
    aclrtFreeHost(param2Host);

    aclrtDestroyStream(stream);
    aclrtDestroyContext(context);
    aclrtResetDevice(deviceId);
    aclFinalize();

    return 0;
}