c++ cuda trouble I'm having trouble getting c++ libraries to work with cuda

supotuco · June 19, 2009, 2:18pm

I’m using visual studio 2005. My cuda parts compile fine when there isn’t any libraries like iostream or fstream but when I include them I get 26 errors like:

1>file.obj : warning LNK4075: ignoring ‘/EDITANDCONTINUE’ due to ‘/INCREMENTAL:NO’ specification
1>LIBCMTD.lib(dbgheap.obj) : error LNK2005: _malloc already defined in LIBCMT.lib(malloc.obj)
1>LIBCMTD.lib(dbgheap.obj) : error LNK2005: __heap_alloc already defined in LIBCMT.lib(malloc.obj)
1>LIBCMTD.lib(dbgheap.obj) : error LNK2005: _calloc already defined in LIBCMT.lib(calloc.obj)

I’m not using any precompiled headers.

If you know how to get all the C++ libraries to work in cuda please help

prodigyaj · June 19, 2009, 4:21pm

check your additonal dependencies for linking , have u added winmm.lib ?

it usually works once u done that !!

supotuco · June 22, 2009, 4:41pm

I added winmm.lib and it didn’t fix anything. Do you have any other suggestions?

tmurray · June 22, 2009, 4:47pm

sounds like you’re linking some things as /MT and some as /MTd. make sure they are the same…

supotuco · June 22, 2009, 4:59pm

where would I check?

or more helpfully is there a template that I can download to create cuda programs with c++ because I got one from somebody on the forums and it is not very good (no offense)

veltrop · June 30, 2009, 3:29am

I had the same problem,

Try adding LIBCMT to Properties:Linker:Input:Ignore Specific Library.

Btw, ignoring LIBCMTD doesn’t seem to help.

supotuco · June 30, 2009, 2:25pm

this didn’t work but thanks for trying

veltrop · July 1, 2009, 1:45am

Yeah, maybe your .cu files are being linked against debug stuff and your .cpp files aren’t, or vice versa?

kickstart · July 29, 2009, 4:38pm

I have the same problem. How can I specify which library to use for CUDA compiler?

veltrop · July 30, 2009, 1:25am

Always having the problem or just in device emulation mode?

kickstart · July 30, 2009, 10:19am

In device emulation mode only.

veltrop · July 30, 2009, 10:26am

Cuda 2.2 cant do c++ with device emulation mode, so you are SOL. Supposedly this has been fixed in 2.3.

kickstart · July 30, 2009, 10:47am

Interestingly enough this happens to my emulator debug build only, release build seems to be ok. To me this looks like the problem should be solvable, I just need to find out how. My current guess is that there should be a way to tell the CUDA compiler which library to compile for.

kickstart · July 30, 2009, 11:22am

Seems like the problem has been solved.

The solution is: If, for example, you specify the use of multi-threaded debug libraries (compiler switch /MTd) for your C++ code then in emulator builds for your *.cu files the switch needs to be propagated from nvcc down to the underlying cl compiler via “-Xcompiler /MTd” option.