[CMake] Moving large MSVS projects to CUDA

[James Bigler]

On Sat, May 5, 2012 at 3:32 AM, Ming <ymyang at wisc.edu> wrote:

>   *From: **James Bigler* <jamesbigler at gmail.com>* *
>  *Sent: Saturday, May 05, 2012 1:51 AM*
> *To: **Ming* <ymyang at wisc.edu>* *
> *Cc: **cmake at cmake.org* <cmake at cmake.org>* *
> *Subject: Re: [CMake] Moving large MSVS projects to CUDA*
> **
> *On Fri, May 4, 2012 at 12:39 PM, Ming <ymyang at wisc.edu> wrote:
> *
>
>>   *Hello CMake community,*
>> **
>> *I have experience using CMake to generate sime cuda-enabled executables
>> (where the cpp’s reside in one folder, etc.).*
>> *I am attempting to move a large project that utilizes CMake, from just
>> c++ to CUDA enabled c++.  *
>> **
>> *It is a project that generates a large set of libraries (comprising a
>> toolkit), using various cmake scripts.  I have modified the “add_libraries”
>> call to be “cuda_add_libraries”, and changed all of the .cc files to .cu
>> (but with no additional CUDA code).  An approach I thought would be similar
>> to changing a simple program from add_executable to “cuda_add_executable”
>> and renaming the extension to “.cu”*
>> *My goal was first to have it compile properly in its original form post
>> migration, but am met with some very strange errors in visual studio.  *
>> **
>> *I am wondering if anyone else has undertaken such a migration before
>> (libraries or binaries), and if so, what types of general steps need to be
>> taken other than just telling CMake we’re adding “.cu” files?*
>> **
>> *Thanks all in advance,*
>> *Ming*
>>
>> *--
>>
>> Powered by **www.kitware.com* <http://www.kitware.com>
>>
>> *Visit other Kitware open-source projects at **
>> http://www.kitware.com/opensource/opensource.html*<http://www.kitware.com/opensource/opensource.html>
>>
>> *Please keep messages on-topic and check the CMake FAQ at: **
>> http://www.cmake.org/Wiki/CMake_FAQ*<http://www.cmake.org/Wiki/CMake_FAQ>
>>
>> *Follow this link to subscribe/unsubscribe:
>> **http://www.cmake.org/mailman/listinfo/cmake*<http://www.cmake.org/mailman/listinfo/cmake>
>>
>
>
> *What kind of strange errors?
>
> Typically when adding a new CUDA enabled library, you simply start adding
> the .cu files with the host/device code mixed to a cuda_add_* target.
>
> James
> *
> Thank you all for the reply,
>
>   -The code in question is the Geant4 (
> http://geant4.web.cern.ch/geant4/support/download.shtml) toolkit, which
> uses CMake to handle its cross-platform configurations. The end result of
> the toolkit is to compile a large set of libraries.
> -I’ve changed all of the source files from .cc to .cu, and changed all *“add_library”
> *terms to *“cuda_add_library”*
> -I’ve also updated all of the *“include directories”* to *
> “cuda_include_directories”*
> -The end result is that I am able to generate a build directory (without
> any errors) using cmake, for VS2010.
>
> However when I attempt to build the package in an analagous fashion, I get
> errors from visual studio informing me mid-build that it:
>
> *Cannot load macro project “ for the following reason:*
> *C:\...\Projects\VSMacros80\Samples\Samples.vsmacros’ is an invalid or
> inaccessible macro project file*
> **
> Once I acknowledge this message, it then says:
>
> *CMake has regenerated 3 .sln and/or .vcproj files.*
> *Click:*
> *--Yes to reload the whole solution...*
> *--No to reload the solution ...*
> *--Cancel to do nothing*
>
> This message shows up even though CMake was done generating files and was
> closed prior to the build.
> The same thing happens on my work computer, and I am able to compile other
> programs just fine, so I do not believe it to be a corrupt install of
> visual studio.  Any suggestions on why this would occur is appreciated!
>
> Thanks,
> Ming
>
> Uploaded screenshots of the error messages:
> https://mywebspace.wisc.edu/ymyang/web/macroerr1.png
> https://mywebspace.wisc.edu/ymyang/web/macroerr2.png
>

I believe the macro error is a known problem with the CMake VS macro in VS
2010 (MS broke it).  The regeneration of the project files is also a known
behavior of FindCUDA.  The dependencies are computed by nvcc during build
and saved in a file.  When the dependency file is regenerated CMake will
want to reconfigure to add those new dependencies to the project file.
This is the expected behavior.

My current recommendation for CUDA + VS 2010 is to specify
CMAKE_SUPPRESS_REGENERATION OFF in your project file's CMakeLists.txt:
option(CMAKE_SUPPRESS_REGENERATION "Suppress regeneration of build scripts
during build" ON)

This will disable the VS CMake plugin, requiring you to manually configure
with CMake when you change the build scripts.  I also recommend configuring
CMake after your first build to pick up the file level dependencies that
CUDA generates.

James
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.cmake.org/pipermail/cmake/attachments/20120507/af2302bd/attachment.htm>