Matlab编译安装MatConvnet流程及问题解决

最近使用Matlab跑深度学习的项目，需要安装MatConvnet，在这个过程中遇到了一些问题，成功解决后特在此总结如下。

一、安装及编译流程

1. MatConvNet介绍： Installing - MatConvNet
2. 配置编译器

> mex -setup

3. 编译用于CPU的库

> cd <MatConvNet>
> addpath matlab
> vl_compilenn

4. 编译用于GPU的库

> vl_compilenn('enableGpu',true,'cudaRoot','C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0','cudaMethod' ,'nvcc','enableCudnn','true','cudnnRoot','C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA')

上述4步执行完如无报错则说明安装编译成功。

下面介绍安装编译过程中遇到的问题及成功解决方法。

二、问题及解决方法(注意本文中所涉及的文件路径请根据自身实际情况就行修改)

1. 报错1

Warning: CL.EXE not found in PATH. Trying to guess out of mex setup. 
> In vl_compilenn>check_clpath (line 650)
  In vl_compilenn (line 426) 
'cl.exe' is not recognized as an internal or external command, 
operable program or batch file. 
Error using vl_compilenn>check_clpath (line 656)
Unable to find cl.exe

Error in vl_compilenn (line 426)
    cl_path = fileparts(check_clpath()); % check whether cl.exe in path

解决方法：把 C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.16.27023\bin\Hostx64\x64 下的cl.exe复制到 D:\Matlab\matconvnet-1.0-beta25\matconvnet-1.0-beta25\ 下。

2. 报错2

Error using vl_compilenn>nvcc_compile (line 615)
Command "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\nvcc" -c -o
"D:\Matlab\matconvnet-1.0-beta25\matconvnet-1.0-beta25\matlab\mex\.build\bits\data.obj"
"D:\Matlab\matconvnet-1.0-beta25\matconvnet-1.0-beta25\matlab\src\bits\data.cu" -DENABLE_GPU -DENABLE_DOUBLE
-DENABLE_CUDNN -I".local\cudnn\include" -O -DNDEBUG -D_FORCE_INLINES --std=c++11
-I"D:\Matlab\extern\include" -I"D:\Matlab\toolbox\distcomp\gpu\extern\include"
-gencode=arch=compute_52,code=\"sm_52,compute_52\"  --compiler-options=/MD --compiler-bindir="C:\Program
Files (x86)\Microsoft Visual Studio\2017\Community\VC\bin"  failed.

Error in vl_compilenn (line 487)
      nvcc_compile(opts, srcs{i}, objfile, flags) ;

解决方法：在 C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC 下创建bin文件夹。

3. 报错3

c:\program files\nvidia gpu computing toolkit\cuda\v9.0\include\crt/host_config.h(133): fatal error C1189: #error:  -- unsupported Microsoft Visual Studio version! Only the versions 2012, 2013, 2015 and 2017 are supported! 
nvcc warning : The -std=c++11 flag is not supported with the configured host compiler. Flag will be ignored. 
data.cu 
Error using vl_compilenn>nvcc_compile (line 615)
Command "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\nvcc" -c -o
"D:\Matlab\matconvnet-1.0-beta25\matconvnet-1.0-beta25\matlab\mex\.build\bits\data.obj"
"D:\Matlab\matconvnet-1.0-beta25\matconvnet-1.0-beta25\matlab\src\bits\data.cu" -DENABLE_GPU -DENABLE_DOUBLE
-DENABLE_CUDNN -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\include" -O -DNDEBUG -D_FORCE_INLINES
--std=c++11 -I"D:\Matlab\extern\include" -I"D:\Matlab\toolbox\distcomp\gpu\extern\include"
-gencode=arch=compute_52,code=\"sm_52,compute_52\"  --compiler-options=/MD --compiler-bindir="C:\Program
Files (x86)\Microsoft Visual Studio\2017\Community\VC\bin"  failed.

Error in vl_compilenn (line 487)
      nvcc_compile(opts, srcs{i}, objfile, flags) ;

错误原因：CUDA和VS版本不匹配。
解决方法：打开C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\include\crt\host_config.h，把
#if _MSC_VER < 1600 || _MSC_VER > 1911
改为：
#if _MSC_VER < 1600 || _MSC_VER > 1920 // 只要版本号够高就行，随便挑个数字
即可解决。

4. 报错4

0x00007FF64188ADD0 (0x0000000000000000 0x000001E56C1C7F18 0x000066BA00000001 0x00000004000304ED) 
0x00007FF641886F3D (0x0000009EBEFFE798 0x0000000000000000 0x0000000000000000 0x000001E56C1DCE20) 
0x00007FF641888713 (0xnvcc warning : The -std=c++11 flag is not supported with the configured host compiler. Flag will be ignored. 
data.cu 
nvcc error   : 'cicc' died with status 0xC0000005 (ACCESS_VIOLATION) 
Error using vl_compilenn>nvcc_compile (line 615)
Command "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\nvcc" -c -o
"D:\Matlab\matconvnet-1.0-beta25\matconvnet-1.0-beta25\matlab\mex\.build\bits\data.obj"
"D:\Matlab\matconvnet-1.0-beta25\matconvnet-1.0-beta25\matlab\src\bits\data.cu" -DENABLE_GPU -DENABLE_DOUBLE
-DENABLE_CUDNN -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\include" -O -DNDEBUG -D_FORCE_INLINES
--std=c++11 -I"D:\Matlab\extern\include" -I"D:\Matlab\toolbox\distcomp\gpu\extern\include"
-gencode=arch=compute_52,code=\"sm_52,compute_52\"  --compiler-options=/MD --compiler-bindir="C:\Program
Files (x86)\Microsoft Visual Studio\2017\Community\VC\bin"  failed.

Error in vl_compilenn (line 487)
      nvcc_compile(opts, srcs{i}, objfile, flags) ;

错误原因：使用CUDA9.0可能会导致上述错误。
解决方法：当使用Visual Studio 2017 Community时，CUDA使用 10.0版本。即搭配为Compatible: Visual Studio 2017 | Cuda 10.0 | Matlab R2018a。重新编译即可解决。

> vl_compilenn('enableGpu', true, ...
  'cudaRoot', 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0', ...
  'cudaMethod', 'nvcc', ...
  'enableCudnn', true, ...
  'cudnnRoot', 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0');

5. 报错5

MatConvNet compiled with '-R2018a' and linked with '-R2017b'

解决方法：把 {MatConvNet路径}/matlab/vl_compilenn.m 第620行附近改成：

args = horzcat({'-outdir', mex_dir}, ...
flags.base, flags.mexlink, ...
'-R2018a',...//新增
{['LDFLAGS=$LDFLAGS ' strjoin(flags.mexlink_ldflags)]}, ...
{['LDOPTIMFLAGS=$LDOPTIMFLAGS ' strjoin(flags.mexlink_ldoptimflags)]}, ...
{['LINKLIBS=' strjoin(flags.mexlink_linklibs) ' $LINKLIBS']}, ...
objs) ;

同时把第359行附近改成：

flags.mexlink = {'-lmwblas'};

即可解决。

至此，使用Matlab编译安装MatConvnet的流程及在这个过程中遇到的问题和解决方法总结如上所示，请各位小伙伴认真对照修改，一定可以解决！如有问题请在评论区留言，我会及时回复！