* 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 kernel_operator_conv2d_intf_impl.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message("impl/basic_api/kernel_operator_conv2d_intf_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 \"basic_api/kernel_operator_conv2d_intf.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_KERNEL_OPERATOR_CONV2D_INTF_IMPL_H__
#endif
#ifndef ASCENDC_MODULE_OPERATOR_CONV2D_INTERFACE_IMPL_H
#define ASCENDC_MODULE_OPERATOR_CONV2D_INTERFACE_IMPL_H
#include "kernel_tensor.h"
#include "kernel_operator_conv2d_base_impl.h"
#include "kernel_operator_data_copy_intf_impl.h"
#include "kernel_struct_data_copy.h"
namespace AscendC {
template <typename T>
[[deprecated("NOTICE: GetConv2dTiling has been deprecated and will be removed in the next version. "
"Please do not use it!")]]
__aicore__ inline Conv2dTilling GetConv2dTiling(Conv2dParams& conv2dParams)
{
Conv2dTilling tilling;
GetTypeforC0<T>(conv2dParams, tilling);
tilling.loopMode = LoopMode::MODE_NM;
tilling.strideH = conv2dParams.stride[0];
tilling.strideW = conv2dParams.stride[1];
tilling.dilationH = conv2dParams.dilation[0];
tilling.dilationW = conv2dParams.dilation[1];
tilling.hi = conv2dParams.imgShape[0];
tilling.wi = conv2dParams.imgShape[1];
tilling.ho = (tilling.hi + conv2dParams.padList[2] + conv2dParams.padList[3] -
tilling.dilationH * (conv2dParams.kernelShape[0] - 1) - 1) /
tilling.strideH +
1;
tilling.wo = (tilling.wi + conv2dParams.padList[0] + conv2dParams.padList[1] -
tilling.dilationW * (conv2dParams.kernelShape[1] - 1) - 1) /
tilling.strideW +
1;
tilling.height = conv2dParams.kernelShape[0];
tilling.width = conv2dParams.kernelShape[1];
tilling.howo = tilling.ho * tilling.wo;
tilling.mNum = tilling.howo;
tilling.nNum = conv2dParams.cout;
tilling.kNum = conv2dParams.cin * tilling.height * tilling.width;
CalculateConv2dTiling(tilling);
return tilling;
}
* @ingroup：Conv2D
* @brief：Given an input tensor and a weight tensor, perform the convolution 2-D operation, and output the
* result tensor
* @param [out] dst output LocalTensor
* @param [in] bias input LocalTensor
* @param [in] featureMap input LocalTensor
* @param [in] weight input LocalTensor
* @param [in] intriParams.imgShape Shape of "featureMap"
* @param [in] intriParams.kernel Shape shape of "weight"
* @param [in] intriParams.stride Stride of convolution
* @param [in] intriParams.cin Fractal layout parameter,cin = c1 * c0
* @param [in] intriParams.cout Fractal layout parameter
* @param [in] intriParams.padList Padding rows/columns
* @param [in] intriParams.dilation Void convolution parameter
* @param [in] intriParams.initY dst initialization parameters
* @param [in] intriParams.partialSum Judge whether the result is moved out
*/
template <typename T, typename U>
[[deprecated("NOTICE: Conv2D has been deprecated and will be removed in the next version. "
"Please do not use it!")]]
__aicore__ inline __in_pipe__(MTE2)
__out_pipe__(MTE3) void Conv2D(const LocalTensor<T> &dst, const LocalTensor<U> &featureMap,
const LocalTensor<U> &weight, Conv2dParams &conv2dParams, Conv2dTilling &tilling)
{
if (conv2dParams.initY == 2) {
return;
}
Conv2D(dst, dst, featureMap, weight, conv2dParams, tilling);
}
template <typename T, typename U>
[[deprecated("NOTICE: Conv2D has been deprecated and will be removed in the next version. "
"Please do not use it!")]]
__aicore__ inline __in_pipe__(MTE2)__out_pipe__(MTE3) void Conv2D(const LocalTensor<T> &dst,
const LocalTensor<T> &bias, const LocalTensor<U> &featureMap, const LocalTensor<U> &weight,
Conv2dParams &conv2dParams, Conv2dTilling &tilling)
{
#if ASCENDC_CPU_DEBUG
if (!CheckConv2DParams(dst, bias, featureMap, weight, conv2dParams, tilling)) {
return;
}
#endif
#if (__NPU_ARCH__ == 1001) || (__NPU_ARCH__ == 2002)
DataCopyParams dataCopyParams;
dataCopyParams.blockCount = 1;
DataCopyEnhancedParams enhancedParams;
enhancedParams.blockMode = BlockMode::BLOCK_MODE_MATRIX;
#endif
const Hardware dstScope = GetPhyType((TPosition)dst.GetPosition());
LocalTensor<T> l0c;
if (dstScope == Hardware::L0C) {
l0c = dst[0];
} else {
#if (__NPU_ARCH__ == 1001) || (__NPU_ARCH__ == 2002)
TBuffAddr tbufc;
tbufc.logicPos = static_cast<uint8_t>(TPosition::C2);
l0c.SetAddr(tbufc);
l0c.InitBuffer(0, TOTAL_L0C_SIZE / sizeof(PrimT<T>));
dataCopyParams.blockLen = dst.GetSize() * sizeof(PrimT<T>) / 1024;
DataCopy(l0c, dst, dataCopyParams, enhancedParams);
#endif
}
if (tilling.loopMode == LoopMode::MODE_NM) {
Conv2DExecNm(l0c, bias, featureMap, weight, conv2dParams, tilling);
} else if (tilling.loopMode == LoopMode::MODE_MN) {
Conv2DExecMn(l0c, bias, featureMap, weight, conv2dParams, tilling);
} else {
}
#if (__NPU_ARCH__ == 1001) || (__NPU_ARCH__ == 2002)
if ((!conv2dParams.partialSum) && (dstScope == Hardware::UB)) {
pipe_barrier(PIPE_ALL);
dataCopyParams.blockLen = tilling.roundM * tilling.roundN * sizeof(PrimT<T>) / 1024;
DataCopy(dst, l0c, dataCopyParams, enhancedParams);
}
#endif
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_KERNEL_OPERATOR_CONV2D_INTF_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_KERNEL_OPERATOR_CONV2D_INTF_IMPL_H__
#endif
