/**
 * Copyright (c) 2025-2026 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 range_float.h
 * \brief
 */

#ifndef RANGE_FLOAT_H
#define RANGE_FLOAT_H

#include "range_base.h"
#include "op_kernel/math_util.h"

namespace Range {
using namespace AscendC;

template <typename T, typename S>
class RangeFloat : public RangeBase<T> {
public:
    __aicore__ inline RangeFloat(){};
    __aicore__ inline void Init(
        GM_ADDR start, GM_ADDR end, GM_ADDR step, GM_ADDR out, GM_ADDR workspace,
        const RangeTilingDataFloat* tilingData);
    __aicore__ inline void Process(const RangeTilingDataFloat* tilingData);

private:
    __aicore__ inline void CopyIn(int64_t loopIdx, int64_t dataCount);
    __aicore__ inline void Compute(int64_t loopIdx, int64_t dataCount);
    __aicore__ inline void CopyOut(int64_t loopIdx, int64_t dataCount);
    __aicore__ inline void GenSequence(int64_t dataCount);

private:
    TPipe pipe_;
    constexpr static int64_t DB_BUFFER = 2;
    TQue<QuePosition::VECIN, DB_BUFFER> calcuDataQueue_;
    TQue<QuePosition::VECOUT, DB_BUFFER> outDataQueue_;
    TBuf<QuePosition::VECCALC> sequenceBuf_;
    TBuf<QuePosition::VECCALC> startDataBuf_;

    GlobalTensor<S> outputGm_;

    int64_t curNumOfCore_{0};
    int64_t curPerOfCore_{0};
    int64_t curTailOfCore_{0};
    int64_t loopCount_{0};
    int64_t coreOffset_{0};

    T start_{0};
    T step_{0};
};

template <typename T, typename S>
__aicore__ inline void RangeFloat<T, S>::Init(
    GM_ADDR start, GM_ADDR end, GM_ADDR step, GM_ADDR out, GM_ADDR workspace, const RangeTilingDataFloat* tilingData)
{
    if (GetBlockIdx() + 1 == tilingData->usedCoreNum) {
        curNumOfCore_ = tilingData->numOfTailCore;
        curPerOfCore_ = tilingData->perOfTailCore;
        curTailOfCore_ = tilingData->tailOfTailCore;
        loopCount_ = tilingData->loopOfTailCore;
    } else {
        curNumOfCore_ = tilingData->numOfPerCore;
        curPerOfCore_ = tilingData->perOfPerCore;
        curTailOfCore_ = tilingData->tailOfPerCore;
        loopCount_ = tilingData->loopOfPerCore;
    }

    // SetBuffer
    coreOffset_ = GetBlockIdx() * tilingData->numOfPerCore;
    outputGm_.SetGlobalBuffer((__gm__ S*)out + coreOffset_, curNumOfCore_);

    // InitBuffer
    pipe_.InitBuffer(calcuDataQueue_, DB_BUFFER, Ops::Base::CeilAlign(curPerOfCore_, ONCE_ALGN_NUM_INT32) * sizeof(T));
    pipe_.InitBuffer(outDataQueue_, DB_BUFFER, Ops::Base::CeilAlign(curPerOfCore_, ONCE_ALGN_NUM_INT32) * sizeof(S));
    pipe_.InitBuffer(sequenceBuf_, Ops::Base::CeilAlign(curPerOfCore_, ONCE_ALGN_NUM_INT32) * sizeof(T));
    if constexpr (!IsSameType<T, S>::value) {
        pipe_.InitBuffer(startDataBuf_, Ops::Base::CeilAlign(curPerOfCore_, ONCE_ALGN_NUM_INT32) * sizeof(T));
    }

    // GetValue
    start_ = tilingData->start;
    step_ = tilingData->delta;
}

template <typename T, typename S>
__aicore__ inline void RangeFloat<T, S>::GenSequence(int64_t dataCount)
{
    LocalTensor<T> sequenceUb = sequenceBuf_.Get<T>();

    // 生成vci连续序列
    const T firstValue = static_cast<T>(0);
    uint32_t calCount = static_cast<uint32_t>(dataCount);
    CreateVecIndex(sequenceUb, firstValue, calCount);
}

template <typename T, typename S>
__aicore__ inline void RangeFloat<T, S>::CopyIn(int64_t loopIdx, int64_t dataCount)
{
    LocalTensor<T> xCalcuUb = calcuDataQueue_.AllocTensor<T>();
    LocalTensor<T> sequenceUb = sequenceBuf_.Get<T>();
    int32_t calCount = static_cast<int32_t>(dataCount);

    // 随机数序列深拷贝，并计算[总offset] = [核间offset] + [核内loop的offset]
    auto curOffset = coreOffset_ + loopIdx * curPerOfCore_;
    const T offsetScalarValue = static_cast<T>(curOffset);
    Adds(xCalcuUb, sequenceUb, offsetScalarValue, calCount);

    calcuDataQueue_.EnQue(xCalcuUb);
}

template <typename T, typename S>
__aicore__ inline void RangeFloat<T, S>::Compute(int64_t loopIdx, int64_t dataCount)
{
    LocalTensor<S> outDataUb = outDataQueue_.AllocTensor<S>();
    LocalTensor<T> xCalcuUb = calcuDataQueue_.DeQue<T>();
    int32_t calCount = static_cast<int32_t>(dataCount);

    // 等差数列的基本公式，a_n = a_1 + (n - 1) * d, 实现乘加操作
    const T startScalarValue = static_cast<T>(start_);
    const T stepScalarValue = static_cast<T>(step_);

    if constexpr (IsSameType<T, S>::value) {
        Duplicate(outDataUb, startScalarValue, calCount);
        Axpy(outDataUb, xCalcuUb, stepScalarValue, calCount);
    } else {
        LocalTensor<T> xStartUb = startDataBuf_.Get<T>();
        Duplicate(xStartUb, startScalarValue, calCount);
        Axpy(xStartUb, xCalcuUb, stepScalarValue, calCount);
        AscendC::Cast(outDataUb, xStartUb, RoundMode::CAST_RINT, calCount);
    }

    outDataQueue_.EnQue<S>(outDataUb);
    calcuDataQueue_.FreeTensor(xCalcuUb);
}

template <typename T, typename S>
__aicore__ inline void RangeFloat<T, S>::CopyOut(int64_t loopIdx, int64_t dataCount)
{
    LocalTensor<S> outDataUb = outDataQueue_.DeQue<S>();
    DataCopyExtParams copyParamsYOut{
        static_cast<uint16_t>(1), static_cast<uint32_t>(dataCount * sizeof(S)), static_cast<uint32_t>(0),
        static_cast<uint32_t>(0), static_cast<uint32_t>(0)};
    DataCopyPad(outputGm_[loopIdx * curPerOfCore_], outDataUb, copyParamsYOut);
    outDataQueue_.FreeTensor(outDataUb);
}

template <typename T, typename S>
__aicore__ inline void RangeFloat<T, S>::Process(const RangeTilingDataFloat* tilingData)
{
    if (GetBlockIdx() >= tilingData->usedCoreNum) {
        return;
    }

    GenSequence(curPerOfCore_);
    for (int64_t n = 0; n < loopCount_ - 1; n++) {
        CopyIn(n, curPerOfCore_);
        Compute(n, curPerOfCore_);
        CopyOut(n, curPerOfCore_);
    }
    {
        CopyIn(loopCount_ - 1, curTailOfCore_);
        Compute(loopCount_ - 1, curTailOfCore_);
        CopyOut(loopCount_ - 1, curTailOfCore_);
    }
}

} // namespace Range

#endif // RANGE_FLOAT_H