neon优化-02关于例子和调试方法

最新推荐文章于 2023-06-27 11:26:24 发布

阳光小郭

最新推荐文章于 2023-06-27 11:26:24 发布

阅读量617

点赞数

分类专栏：性能优化文章标签： android simd optimization

本文链接：https://blog.csdn.net/GuoHuWei1989/article/details/119548899

版权

性能优化专栏收录该内容

1 篇文章 0 订阅

订阅专栏

1.优化例子

这里才从github上找了个开源的repo，只需要修改几个地方就可以配置arm64-v8的环境了
neon优化的链接

注意这里的repo含有子模块，git clone代码以后需要执行以下命令才可以下载完整的代码

git submodule init
git submodule update

1.1 编译步骤

1.下载ndk最新版本
2.安装cmake
3.安装adb等工具
4.sh build_android.sh进行编译和单元测试

build_android的说明：

set -e
set -x
basepath=$(cd `dirname $0`/; pwd)
BUILD_DIR=${basepath}/build
BUILD_ANDROID_NDK_HOME=/home/ldpe2g/Android/android-ndk-r15c #下载好的ndk的路径
DEPLOY_DIR=/data/local/tmp/ldp #手机上的路径
CMAKE=/home/ldpe2g/Android/cmake-3.11.0-Linux-x86_64/bin/cmake #安装好的cmake的路径
rm -rf ${BUILD_DIR}
if [[ ! -d ${BUILD_DIR} ]]; then
    mkdir -p ${BUILD_DIR}
fi
cd ${BUILD_DIR}
$CMAKE \
-DCMAKE_TOOLCHAIN_FILE=${BUILD_ANDROID_NDK_HOME}/build/cmake/android.toolchain.cmake \
-DANDROID_NDK=${BUILD_ANDROID_NDK_HOME} \
-DANDROID_ABI=armeabi-v7a with NEON \ # 这里修改成 arm64-v8a
-DANDROID_NATIVE_API_LEVEL=9 \
-DANDROID_TOOLCHAIN=gcc \ #这里修改成clang
-DANDROID_ARM_NEON=TRUE \ #这一行可以删掉 因为64位的arm neon都是支持的
../
make all -j4
adb shell "mkdir -p ${DEPLOY_DIR}/lib"
adb push ${basepath}/build/unit_test/runUnitTests ${DEPLOY_DIR}
adb push ${basepath}/build/src/libboxfilter.so ${DEPLOY_DIR}/lib
adb shell "cd ${DEPLOY_DIR}; export LD_LIBRARY_PATH=./lib ; ./runUnitTests"

这个工程用的是googletest,很容易添加自己的功能和测试代码。。。

1.2 优化效果

[==========] Running 6 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 6 tests from netTest
[ RUN      ] netTest.org_boxfilter
          [0] BoxFilfer Cost time: 1679.91ms
          [1] BoxFilfer Cost time: 1669.2ms
          [2] BoxFilfer Cost time: 1668.06ms
          [3] BoxFilfer Cost time: 1669.61ms
          [4] BoxFilfer Cost time: 1667.7ms
          [5] BoxFilfer Cost time: 1667.84ms
          [6] BoxFilfer Cost time: 1677.63ms
          [7] BoxFilfer Cost time: 1670.87ms
          [8] BoxFilfer Cost time: 1667.56ms
          [9] BoxFilfer Cost time: 1671.46ms

          BoxFilfer Average Cost time: 1670.99ms
[       OK ] netTest.org_boxfilter (17257 ms)
[ RUN      ] netTest.fast_boxfilter
          [0] Fast BoxFilfer Cost time: 341.669ms
          [1] Fast BoxFilfer Cost time: 341.137ms
          [2] Fast BoxFilfer Cost time: 341.387ms
          [3] Fast BoxFilfer Cost time: 340.85ms
          [4] Fast BoxFilfer Cost time: 341.06ms
          [5] Fast BoxFilfer Cost time: 341.151ms
          [6] Fast BoxFilfer Cost time: 341.605ms
          [7] Fast BoxFilfer Cost time: 340.94ms
          [8] Fast BoxFilfer Cost time: 341.17ms
          [9] Fast BoxFilfer Cost time: 340.895ms

          Fast BoxFilfer Average Cost time: 341.186ms
[       OK ] netTest.fast_boxfilter (3958 ms)
[ RUN      ] netTest.fast_boxfilter_v2
          [0] Fast BoxFilfer V2 Cost time: 75.034ms
          [1] Fast BoxFilfer V2 Cost time: 75.52ms
          [2] Fast BoxFilfer V2 Cost time: 75.206ms
          [3] Fast BoxFilfer V2 Cost time: 75.028ms
          [4] Fast BoxFilfer V2 Cost time: 75.324ms
          [5] Fast BoxFilfer V2 Cost time: 74.995ms
          [6] Fast BoxFilfer V2 Cost time: 75.311ms
          [7] Fast BoxFilfer V2 Cost time: 75.237ms
          [8] Fast BoxFilfer V2 Cost time: 75.218ms
          [9] Fast BoxFilfer V2 Cost time: 75.44ms

          Fast BoxFilfer V2 Average Cost time: 75.2313ms
[       OK ] netTest.fast_boxfilter_v2 (1308 ms)
[ RUN      ] netTest.fast_boxfilter_v2_neon_intrinsics
          [0] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.076ms
          [1] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.52ms
          [2] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.959ms
          [3] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.4ms
          [4] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.465ms
          [5] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.367ms
          [6] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.753ms
          [7] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.778ms
          [8] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.297ms
          [9] Fast BoxFilfer V2 Neon Intrinsics Cost time: 72.758ms

          Fast BoxFilfer V2 Neon Intrinsics Average Cost time: 72.5373ms
[       OK ] netTest.fast_boxfilter_v2_neon_intrinsics (1272 ms)
[ RUN      ] netTest.fast_boxfilter_v2_neon_asm
          [0] Fast BoxFilfer V2 Neon Asm Cost time: 73.163ms
          [1] Fast BoxFilfer V2 Neon Asm Cost time: 72.803ms
          [2] Fast BoxFilfer V2 Neon Asm Cost time: 72.829ms
          [3] Fast BoxFilfer V2 Neon Asm Cost time: 73.206ms
          [4] Fast BoxFilfer V2 Neon Asm Cost time: 72.73ms
          [5] Fast BoxFilfer V2 Neon Asm Cost time: 73.08ms
          [6] Fast BoxFilfer V2 Neon Asm Cost time: 72.65ms
          [7] Fast BoxFilfer V2 Neon Asm Cost time: 72.883ms
          [8] Fast BoxFilfer V2 Neon Asm Cost time: 72.85ms
          [9] Fast BoxFilfer V2 Neon Asm Cost time: 72.598ms

          Fast BoxFilfer V2 Neon Asm Average Cost time: 72.8792ms
[       OK ] netTest.fast_boxfilter_v2_neon_asm (1275 ms)
[ RUN      ] netTest.fast_boxfilter_v2_neon_asm_v2
          [0] Fast BoxFilfer V2 Neon AsmV2 Cost time: 62.454ms
          [1] Fast BoxFilfer V2 Neon AsmV2 Cost time: 63.037ms
          [2] Fast BoxFilfer V2 Neon AsmV2 Cost time: 62.932ms
          [3] Fast BoxFilfer V2 Neon AsmV2 Cost time: 63.445ms
          [4] Fast BoxFilfer V2 Neon AsmV2 Cost time: 62.876ms
          [5] Fast BoxFilfer V2 Neon AsmV2 Cost time: 63.137ms
          [6] Fast BoxFilfer V2 Neon AsmV2 Cost time: 63.33ms
          [7] Fast BoxFilfer V2 Neon AsmV2 Cost time: 63.589ms
          [8] Fast BoxFilfer V2 Neon AsmV2 Cost time: 63.07ms
          [9] Fast BoxFilfer V2 Neon AsmV2 Cost time: 63.237ms

          Fast BoxFilfer V2 Neon AsmV2 Average Cost time: 63.1107ms
[       OK ] netTest.fast_boxfilter_v2_neon_asm_v2 (1177 ms)
[----------] 6 tests from netTest (26247 ms total)

[----------] Global test environment tear-down
[==========] 6 tests from 1 test suite ran. (26248 ms total)
[  PASSED  ] 6 tests.

2. 通过adb shell 进行gdb调试的方法

在terminal中执行下面的脚本就可以在手机端开启一个服务，port=5039

DEPLOY_DIR=/TEST_ALGO #手机上要存放的测试数据
HOST_GDBSERVER=${BUILD_ANDROID_NDK_HOME}/prebuilt/android-arm64/gdbserver/gdbserver
HOST_GDB=${BUILD_ANDROID_NDK_HOME}/prebuilt/linux-x86_64/bin/gdb
HOST_APP=./algo/sample_test #编译的本地可执行sample

TARGET_APP=${DEPLOY_DIR}/sample_test
TARGET_GDBSERVER=/data/app/gdbserver
PORT=5039
adb forward tcp:${PORT} tcp:${PORT}
adb push ${HOST_GDBSERVER} ${TARGET_GDBSERVER}
adb shell ${TARGET_GDBSERVER} :${PORT} ${TARGET_APP}

在电脑上另开一个terminal输入下面的命令，就可以开始调试了

${BUILD_ANDROID_NDK_HOME}/prebuilt/linux-x86_64/bin/gdb ./sample_test #进入gdb调试交互界面
target remote:5039 # 在gdb 交互界面中输入，可以调试服务端的程序

由于理解的不全面，所以上面的描述也许不准确，但是按照上面的步骤是可以正确在adb 中进行gdb调试native c/c++的

阳光小郭

关注

0
点赞
踩
3

收藏

觉得还不错? 一键收藏
0
评论
neon优化-02关于例子和调试方法

优化例子这里才从github上找了个开源的repo，只需要修改几个地方就可以配置arm64-v8的环境了neon优化的链接注意这里的repo含有子模块，git clone代码以后需要执行以下命令才可以下载完整的代码git submodule initgit submodule update编译步骤1.下载ndk最新版本安装cmake安装adb等工具sh build_android,sh进行编译和单元测试build_android的说明：set -ese
复制链接

扫一扫