/**
 * 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 round_v200_impl.h
 * \brief
 */

#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
    "impl/adv_api/detail/math/round/round_v200_impl.h is an internal header file and must not be used directly. Functions or variables defined in this file may be removed in the future. Please use \"#include \"adv_api/math/round.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_ROUND_ROUND_V200_IMPL_H__
#endif
#ifndef IMPL_MATH_ROUND_ROUND_V200_IMPL_H
#define IMPL_MATH_ROUND_ROUND_V200_IMPL_H
#include "kernel_basic_intf.h"
#include "kernel_tensor.h"
#include "kernel_tiling/kernel_tiling.h"
#include "../../common/check.h"

#if defined(__NPU_ARCH__) && __NPU_ARCH__ == 2002
namespace AscendC {
constexpr uint32_t ROUND_HALF_CALC_FAC_200 = 2;
constexpr uint32_t STRIDE_OF_DIFFERENT_DIGITS = 2;

template <typename T, bool isReuseSource = false>
__aicore__ inline void RoundComputeCount(
    const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor, const LocalTensor<uint8_t>& sharedTmpBuffer,
    const uint32_t calCount)
{}

template <>
__aicore__ inline void RoundComputeCount<float, false>(
    const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor,
    const LocalTensor<uint8_t>& sharedTmpBuffer, const uint32_t calCount)
{
    // Calculate the amount of data that can be stored in the temporary space and split the data into the entire block
    // and tail block.
    uint32_t sharedTmpBufferSize = sharedTmpBuffer.GetSize();
    uint32_t splitCount = sharedTmpBufferSize / sizeof(float) / ONE_BLK_SIZE * ONE_BLK_SIZE;
    CheckTmpBufferSize(splitCount, 0, sharedTmpBufferSize);

    uint32_t loopCount = calCount / splitCount;
    uint32_t calcTail = calCount % splitCount;

    const LocalTensor<int32_t>& tmpTensor = sharedTmpBuffer.ReinterpretCast<int32_t>();
    SetVectorMask<float, MaskMode::COUNTER>(0, splitCount);

    // In the case of the float data type, there is no direct instruction for the round operation. Therefore, multiple
    // conversions are required.
    for (uint32_t i = 0; i < loopCount; ++i) {
        Cast<int32_t, float, false>(
            tmpTensor, srcTensor[i * splitCount], RoundMode::CAST_RINT, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
        Cast<float, int32_t, false>(
            dstTensor[i * splitCount], tmpTensor, RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
    }
    if (calcTail > 0) {
        SetVectorMask<float, MaskMode::COUNTER>(0, calcTail);
        Cast<int32_t, float, false>(
            tmpTensor, srcTensor[loopCount * splitCount], RoundMode::CAST_RINT, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
        Cast<float, int32_t, false>(
            dstTensor[loopCount * splitCount], tmpTensor, RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
    }
}

template <>
__aicore__ inline void RoundComputeCount<half, false>(
    const LocalTensor<half>& dstTensor, const LocalTensor<half>& srcTensor, const LocalTensor<uint8_t>& sharedTmpBuffer,
    const uint32_t calCount)
{
    // Calculate the amount of data that can be stored in the temporary space and split the data into the entire block
    // and tail block.
    uint32_t sharedTmpBufferSize = sharedTmpBuffer.GetSize();
    uint32_t splitCount = sharedTmpBufferSize / sizeof(float) / ROUND_HALF_CALC_FAC_200 / ONE_BLK_SIZE * ONE_BLK_SIZE;
    CheckTmpBufferSize(splitCount, 0, sharedTmpBufferSize);

    uint32_t loopCount = calCount / splitCount;
    uint32_t calcTail = calCount % splitCount;

    const LocalTensor<float>& tmpTensor = sharedTmpBuffer.ReinterpretCast<float>();
    const LocalTensor<int32_t>& tmpTensor1 = sharedTmpBuffer.ReinterpretCast<int32_t>();

    SetMaskCount();
    SetVectorMask<half, MaskMode::COUNTER>(0, splitCount);
    // In the case of the half data type, there is no direct instruction for the round operation. Therefore, multiple
    // conversions are required.
    for (uint32_t i = 0; i < loopCount; ++i) {
        Cast<float, half, false>(
            tmpTensor, srcTensor[i * splitCount], RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE / STRIDE_OF_DIFFERENT_DIGITS});
        PipeBarrier<PIPE_V>();

        Cast<int32_t, float, false>(
            tmpTensor1, tmpTensor, RoundMode::CAST_RINT, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
        Cast<float, int32_t, false>(
            tmpTensor, tmpTensor1, RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
        Cast<half, float, false>(
            dstTensor[i * splitCount], tmpTensor, RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE / STRIDE_OF_DIFFERENT_DIGITS, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
    }
    if (calcTail > 0) {
        SetVectorMask<half, MaskMode::COUNTER>(0, calcTail);
        Cast<float, half, false>(
            tmpTensor, srcTensor[loopCount * splitCount], RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE / STRIDE_OF_DIFFERENT_DIGITS});
        PipeBarrier<PIPE_V>();
        Cast<int32_t, float, false>(
            tmpTensor1, tmpTensor, RoundMode::CAST_RINT, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
        Cast<float, int32_t, false>(
            tmpTensor, tmpTensor1, RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
        Cast<half, float, false>(
            dstTensor[loopCount * splitCount], tmpTensor, RoundMode::CAST_NONE, MASK_PLACEHOLDER, (uint8_t)1,
            {1, 1, DEFAULT_REPEAT_STRIDE / STRIDE_OF_DIFFERENT_DIGITS, DEFAULT_REPEAT_STRIDE});
        PipeBarrier<PIPE_V>();
    }
    SetMaskNorm();
    ResetMask();
}
} // namespace AscendC
#endif

#endif // IMPL_MATH_ROUND_ROUND_V200_IMPL_H

#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_ROUND_ROUND_V200_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_ROUND_ROUND_V200_IMPL_H__
#endif