---
title: 使用指南
description: "KML_CONV只提供filter2D的核心算子实现，用户需按使用指南将核心算子注册进开源OpenCV，并调用开源OpenCV filter2d接口完成滤波计算。"
url: https://www.hikunpeng.com/document/detail/zh/kunpenghpcs/hpckit/devg/kunpengaccel_kml_0711.html
sourcePath: /source/zh/kunpenghpcs/hpckit/devg/kunpengaccel_kml_0711.html
indexId: dd5f49fd7d8aec00470664564dfb1d6d29b7d0d273d3ddde12fef7bd4ee68c5d61
---
# 使用指南

KML_CONV只提供filter2D的核心算子实现，用户需按使用指南将核心算子注册进开源OpenCV，并调用开源OpenCV filter2d接口完成滤波计算。

#### 算子注册

1. 获取OpenCV源码。
  下载地址：https://github.com/opencv/opencv/archive/refs/tags/4.9.0.tar.gz(https://github.com/opencv/opencv/archive/refs/tags/4.9.0.tar.gz)

2. 解压OpenCV安装包。

```
tar -xvf opencv-4.9.0.tar.gz
```


3. 注册kml_conv核心算子。

  a. opencv/3rdparty 目录下新增名为“kconv”的文件夹。

```
cd your_path_to_opencv/opencv-4.9.0/3rdparty
mkdir -p kconv
```


  b. 拷贝头文件及相应依赖库至“opencv/3rdparty/kconv”文件夹下。

    - 拷贝conv头文件。

```
cp cv.h your_path_to_opencv/opencv-4.9.0/3rdparty/kconv
```


    - 拷贝CONV、BLAS、FFT依赖库。

```
cp libkconv_ext.so your_path_to_opencv/opencv-4.9.0/3rdparty/Kunpeng
cp libkblas.so your_path_to_opencv/opencv-4.9.0/3rdparty/Kunpeng
cp libkfftf.so your_path_to_opencv/opencv-4.9.0/3rdparty/Kunpeng
cp libkffth.so your_path_to_opencv/opencv-4.9.0/3rdparty/Kunpeng
```


  c. “opencv/3rdparty/Kunpeng”文件夹下新增头文件与源文件。

    - 新增头文件kml_cv.hpp。

```
#ifndef KML_CV_HPP
#define KML_CV_HPP
#include <opencv2/core/base.hpp>
#include "cv.h"
#endif
```


    - 新增源文件kml_cv.cpp。

```
#include "kml_cv.hpp"
```


  d. “opencv/3rdparty/kconv”文件夹下新增CMakeLists.txt。

```
project(kml_conv_ext CXX)
add_library(kml_conv_ext STATIC kml_cv.cpp)
target_link_libraries(kml_conv_ext ${CMAKE_CURRENT_SOURCE_DIR}/libkconv.so ${CMAKE_CURRENT_SOURCE_DIR}/libkblas.so ${CMAKE_CURRENT_SOURCE_DIR}/libkfftf.so ${CMAKE_CURRENT_SOURCE_DIR}/libkffth.so)
set(KML_HAL_VERSION "0.0.1" PARENT_SCOPE)
set(KML_HAL_LIBRARIES "kml_conv_ext" PARENT_SCOPE)
set(KML_HAL_HEADERS "kml_conv_ext.hpp" PARENT_SCOPE)
set(KML_HAL_INCLUDE_DIRS "${CMAKE_BINARY_DIR}" PARENT_SCOPE)
set_target_properties(kml_conv_ext PROPERTIES LINKER_LANGUAGE CXX)
set_target_properties(kml_conv_ext PROPERTIES ARCHIVE_OUTPUT_DIRECTORY ${3P_LIBRARY_OUTPUT_PATH})
if(NOT BUILD_SHARED_LIBS)
ocv_install_target(kail_dnn_ext EXPORT OpenCVModules ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT dev)
endif()
set(OPENCV_SRC_DIR "${CMAKE_SOURCE_DIR}")
target_include_directories(kml_conv_ext PRIVATE ${CMAKE_CURRENT_SOURCE_DIR} ${OPENCV_SRC_DIR}/modules/core/include)
configure_file("${CMAKE_CURRENT_SOURCE_DIR}/kml_cv.hpp" "${CMAKE_BINARY_DIR}/kml_cv.hpp" COPYONLY)
configure_file("${CMAKE_CURRENT_SOURCE_DIR}/cv.h" "${CMAKE_BINARY_DIR}/cv.h" COPYONLY)
set_target_properties(${KML_LIBRARIES} PROPERTIES
OUTPUT_NAME ${KML_LIBRARIES}
DEBUG_POSTFIX "${OPENCV_DEBUG_POSTFIX}"
COMPILE_PDB_NAME ${KML_LIBRARIES}
COMPILE_PDB_NAME_DEBUG "${KML_LIBRARIES}${OPENCV_DEBUG_POSTFIX}"
ARCHIVE_OUTPUT_DIRECTORY ${3P_LIBRARY_OUTPUT_PATH}
)
```


  e. 注册KML_CONV高性能filter2D核心算子。
    修改“opencv-4.9.0/CMakeLists.txt”
    第965-995行。

```
set(OpenCV_HAL "kml_conv_ext")
foreach(hal ${OpenCV_HAL})
if(hal STREQUAL "carotene")
elseif(hal STREQUAL "openvx")
add_subdirectory(3rdparty/openvx)
ocv_hal_register(OPENVX_HAL_LIBRARIES OPENVX_HAL_HEADERS OPENVX_HAL_INCLUDE_DIRS)
list(APPEND OpenCV_USED_HAL "openvx (ver ${OPENVX_HAL_VERSION})")
elseif(hal STREQUAL "kml_conv_ext")
add_subdirectory(3rdparty/kconv)
ocv_hal_register(KML_HAL_LIBRARIES KML_HEADERS KML_INCLUDE_DIRS)
list(APPEND OpenCV_USED_HAL "kml_conv_ext (ver ${KML_VERSION})")
else()
ocv_debug_message(STATUS "OpenCV HAL: ${hal} ...")
ocv_clear_vars(OpenCV_HAL_LIBRARIES OpenCV_HAL_HEADERS OpenCV_HAL_INCLUDE_DIRS)
find_package(${hal} NO_MODULE QUIET)
if(${hal}_FOUND)
ocv_hal_register(OpenCV_HAL_LIBRARIES OpenCV_HAL_HEADERS OpenCV_HAL_INCLUDE_DIRS)
list(APPEND OpenCV_USED_HAL "${hal} (ver ${${hal}_VERSION})")
endif()
endif()
```


4. 编译安装OpenCV。

```
cd your_path_to_opencv/opencv-4.9.0
mkdir -p build
cd build
cmake .. -DWITH_ADE=OFF -DCMAKE_INSTALL_PREFIX=your_path_to_install_opencv # cmake到build文件夹下
make -j
make install # 安装到your_path_to_install_opencv中
```


#### OpenCV filter2D接口定义

完成上述KML_CONV filter2D算子注册后，用户可直接调用OpenCV filter2D接口完成滤波计算。

OpenCV filter2D使用指南：

https://docs.opencv.org/4.9.0/d4/d86/group__imgproc__filter.html(https://docs.opencv.org/4.9.0/d4/d86/group__imgproc__filter.html)

C++ interface：

void cv::filter2D( InputArray src, OutputArray dst, int ddepth, InputArray kernel, Point anchor = Point(-1,-1), double delta = 0, int borderType = BORDER_DEFAULT)

参数列表：

| 参数名 | 类型 | 描述 | 输入/输出 |
| --- | --- | --- | --- |
| src | InputArray | 输入图像 | 输入 |
| dst | OutputArray | 输出图像 | 输出 |
| int | ddepth | 输出图像的深度 | 输入 |
| kernel | InputArray | 卷积核 | 输入 |
| anchor | Point | 卷积核锚点，表示滤波起始点在卷积核中的相对位置 默认值为为Point(\-1,\-1),表示滤波起始点位于卷积核中心 | 输入 |
| delta | 0 | 偏置值，默认值为0 | 输入 |
| int | borderType | 边界填充类型，支持以下入参: BORDER\_CONSTANT BORDER\_REPLICATE BORDER\_REFLECT BORDER\_REFLECT\_101 BORDER\_REFLECT101 BORDER\_DEFAULT（默认值） BORDER\_ISOLATED | 输入 |


完成上述KML_CONV算子注册后的OpenCV 将新增对CV_16F图像深度filter2D计算的功能支持。下表是OpenCV filter2D接口所支持的输入图像深度、输出图像深度组合。

| 输入图像深度 | 输出图像深度 |
| --- | --- |
| CV\_8U | \-1/CV\_16S/CV\_32F/CV\_64F （输出图像深度值为\-1时，表示输出图像深度与输入图像深度相同） |
| CV\_16U/CV\_16S | \-1/CV\_32F/CV\_64F |
| CV\_32F | \-1/CV\_32F |
| CV\_64F | \-1/CV\_64F |
| CV\_16F | \-1/CV\_16F |


当输入图像深度与输出图像深度均为CV_32F或CV_16F时，将调用KML_CONV高性能算子实现完成滤波计算。其他情况下，将调用OpenCV自身算子实现完成滤波计算。


#### 依赖

#include <opencv2/imgproc/imgproc.hpp>


#### 示例

C++ interface：

```
cv::Mat src = (cv::Mat_<float>(6, 6) << 1, 2, 3, 4, 5, 6,
7, 8, 9, 10, 11, 12,
13, 14, 15, 16, 17, 18,
19, 20, 21, 22, 23, 24,
25, 26, 27, 28, 29, 30,
31, 32, 33, 34, 35, 36);
cv::Mat kernel = (cv::Mat_<float>(3, 3) << 0, -1, 0,
-1, 5, -1,
0, -1, 0);
cv::Mat dst;
cv::filter2D(src, dst, -1, kernel, cv::Point(-1, -1), 0, cv::BORDER_CONSTANT);
/*
* dst = [-4, -2, 0, 2, 4, 13;
*        13, 8, 9, 10, 11, 25;
*        25, 14, 15, 16, 17, 37;
*        37, 20, 21, 22, 23, 49;
*        49, 26, 27, 28, 29, 61;
*        98, 70, 72, 74, 76, 115]
*/
```