Valgrind 内存检测

准备工作

1.valgrind (https://valgrind.org/)

2.rawlog, grid.gridmap, points.pointsmap

3.PC上的可执行文件test_rawlog, test_mclBestPose

只检测内存泄露，上述已足够。如果需要分析内存占用比例、接口调用热度，对于大型程序，需要KCachegrind可视化软件辅助。

4.Kubuntu 22.10（自带KCachegrind工具）

下载安装

在官网 (https://valgrind.org/)

下载最新源码：valgrind-3.20.0.tar.bz2

我安装了2次，第一次放在/opt目录，安装失败了，后来放在/home/tools目录下，安装成功了。可能是/opt目录需要权限。

安装命令：

tar -jxvf valgrind-3.12.0.tar.bz2 
cd valgrind-3.12.0
 ./configure 
make 
sudo make install

仿真代码

为了排除其他因素干扰，对test_rawlog和test_mclBestPose做了以下修改：

1.删除画图相关代码，保留纯算法接口（initialize和processxx）

2.代码中屏蔽SSE并行加速，实际跑机用不上

3.CMakeLists添加add_compile_options(-std=c++11 -g -O0 -Wall)，编译不优化

-g:调试信息

-O0:不优化

-Wall:全部警告

4.存储rawlog包，确认仿真代码可以正常使用

使用memcheck检测内存

1.按上述准备，编译仿真代码，生成可执行文件

2.检测内存泄露的命令 valgrind --tool=memcheck --leak-check=full ./test

3.测试结果

==17271== HEAP SUMMARY:
==17271==     in use at exit: 72,832 bytes in 5 blocks
==17271==   total heap usage: 3,627 allocs, 3,622 frees, 3,869,105 bytes allocated
==17271== 
==17271== LEAK SUMMARY:
==17271==    definitely lost: 0 bytes in 0 blocks
==17271==    indirectly lost: 0 bytes in 0 blocks
==17271==      possibly lost: 0 bytes in 0 blocks
==17271==    still reachable: 72,832 bytes in 5 blocks
==17271==         suppressed: 0 bytes in 0 blocks
==17271== Reachable blocks (those to which a pointer was found) are not shown.
==17271== To see them, rerun with: --leak-check=full --show-leak-kinds=all
==17271== 
==17271== For lists of detected and suppressed errors, rerun with: -s
==17271== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

4.分析结果：

HEAP SUMMARY：统计了全部申请的内存，释放的内存量。

LEAK SUMMARY：统计了三种类型的内存泄露definitely, indirectly, possibly，详细代码指向要看给出的报告内容。slam模块和mcl模块都没有内存泄露，无法提供参考。

ERROR SUMMARY：统计了代码中的错误。

内存优化

使用memcheck的xtree分析内存占用分布--slam

1.按上述准备，编译仿真代码，生成可执行文件

2.分析内存占用的命令：valgrind --xtree-memory=full ./test

3.测试结果

(1).修改前:20帧数据xtree

==18477== xtree memory report: /home/tb/sources/eva/build_pc/apps/test_pc/xtmemory.kcg.18477
==18477== HEAP SUMMARY:
==18477==     in use at exit: 72,832 bytes in 5 blocks
==18477==   total heap usage: 5,823 allocs, 5,818 frees, 7,517,316 bytes allocated
==18477== 
==18477== LEAK SUMMARY:
==18477==    definitely lost: 0 bytes in 0 blocks
==18477==    indirectly lost: 0 bytes in 0 blocks
==18477==      possibly lost: 0 bytes in 0 blocks
==18477==    still reachable: 72,832 bytes in 5 blocks
==18477==         suppressed: 0 bytes in 0 blocks
==18477== Rerun with --leak-check=full to see details of leaked memory
==18477== 
==18477== For lists of detected and suppressed errors, rerun with: -s
==18477== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

(2).修改后:20帧数据xtree

==25649== xtree memory report: /home/tb/sources/eva/build_pc/apps/test_pc/xtmemory.kcg.25649
==25649== HEAP SUMMARY:
==25649==     in use at exit: 72,832 bytes in 5 blocks
==25649==   total heap usage: 3,608 allocs, 3,603 frees, 3,826,308 bytes allocated
==25649== 
==25649== LEAK SUMMARY:
==25649==    definitely lost: 0 bytes in 0 blocks
==25649==    indirectly lost: 0 bytes in 0 blocks
==25649==      possibly lost: 0 bytes in 0 blocks
==25649==    still reachable: 72,832 bytes in 5 blocks
==25649==         suppressed: 0 bytes in 0 blocks
==25649== Rerun with --leak-check=full to see details of leaked memory
==25649== 
==25649== For lists of detected and suppressed errors, rerun with: -s
==25649== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

结论：内存占用降低50%（7,517,316 bytes-->3,826,308 bytes）

4.分析结果

用KCachegrind可视软件打开xtmemory.kcg.xxx文件，View→Primary Event Type→total allocated Bytes，得到如下图表：

修改前：进入processActionObservation()后，内存分配最多的接口是CICP::Align()，里面内存分配最多的接口是determineMatching2D()，里面内存分配最多的地方在vector。

（图略）

修改后：占用的比例发生变化，CICP::Align()从55%->31%。有所改善。

（图略）

一般把占用内存最多的地方优化了，就能达到明显的效果。其他不太需要考虑。

文件生成了仿真代码的内存占用树状图，从图中能看到哪个接口占用内存比例较高，对占用比例较高的代码进行优化，能高效降低内存占用

使用memcheck的xtree分析内存占用分布—mcl

1.按上述准备，编译仿真代码，生成可执行文件

2.分析内存占用的命令：valgrind --xtree-memory=full ./test

3.测试结果

(1).修改前：5帧数据xtree

==18531== 
==18531== xtree memory report: /home/tb/sources/eva/build_pc/apps/test_pc/xtmemory.kcg.18531
==18531== HEAP SUMMARY:
==18531==     in use at exit: 72,832 bytes in 5 blocks
==18531==   total heap usage: 7,384 allocs, 7,379 frees, 14,494,661 bytes allocated
==18531== 
==18531== LEAK SUMMARY:
==18531==    definitely lost: 0 bytes in 0 blocks
==18531==    indirectly lost: 0 bytes in 0 blocks
==18531==      possibly lost: 0 bytes in 0 blocks
==18531==    still reachable: 72,832 bytes in 5 blocks
==18531==         suppressed: 0 bytes in 0 blocks
==18531== Rerun with --leak-check=full to see details of leaked memory
==18531== 
==18531== Use --track-origins=yes to see where uninitialised values come from
==18531== For lists of detected and suppressed errors, rerun with: -s
==18531== ERROR SUMMARY: 138 errors from 17 contexts (suppressed: 0 from 0)

(2).修改后:5帧数据xtree

==10032== xtree memory report: /home/tb/sources/eva/build_pc/apps/test_pc/xtmemory.kcg.10032
==10032== HEAP SUMMARY:
==10032==     in use at exit: 72,832 bytes in 5 blocks
==10032==   total heap usage: 3,028 allocs, 3,023 frees, 3,846,855 bytes allocated
==10032== 
==10032== LEAK SUMMARY:
==10032==    definitely lost: 0 bytes in 0 blocks
==10032==    indirectly lost: 0 bytes in 0 blocks
==10032==      possibly lost: 0 bytes in 0 blocks
==10032==    still reachable: 72,832 bytes in 5 blocks
==10032==         suppressed: 0 bytes in 0 blocks
==10032== Rerun with --leak-check=full to see details of leaked memory
==10032== 
==10032== For lists of detected and suppressed errors, rerun with: -s
==10032== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

结论：内存占用降低75%（14,494,661 bytes-->3,846,855 bytes）。且修改前的代码存在ERROR，未初始化的变量，修改后ERROR为0.

用KCachegrind可视软件打开xtmemory.kcg.xxx文件，View→Primary Event Type→total allocated Bytes，得到如下图表：

修改前：进入test()后，内存分配最多的接口是CRelocalization()构造，里面分配了20000个pose。其余接口内存分配总占用58%。

（图略）

修改后：CRelocalization()构造占用内存比例增加，其余接口比例下降，优化在了其他接口中。实际上构造的20000个pose占用是不变的，没有修改这一部分。目前还没有好的方式对此进行优化。

（图略）