AW: compilation with CUDA error

From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Tue Oct 30 2012 - 02:19:22 CDT

If I remember correctly and if it hasn't changed yet, namd is not using CUDA
features of charm++, so you should try to build charm++ without cuda
support.

 

Norman Geist.

 

Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag
von Aron Broom
Gesendet: Montag, 29. Oktober 2012 18:44
An: Thomas Evangelidis
Cc: namd-l
Betreff: Re: namd-l: compilation with CUDA error

 

it doesn't say anywhere in the instructions which version of cuda it is
intended for? My impression was that it was 4.0, but that's a big guess.

On Mon, Oct 29, 2012 at 9:44 AM, Thomas Evangelidis <tevang3_at_gmail.com>
wrote:

Greetings,

I am trying to compile NAMD CVS-2012-09-22_Linux-x86_64-multicore-CUDA using
Cuda 4.2 with the following command lines:

CUDA_DIR=/usr/local/cuda MAKE=gmake ./build charm++ net-linux-x86_64 icc
ifort smp cuda -j8 --with-production -O3 -axSSE4.2 -unroll -DCMK_OPTIMIZE=1
./config Linux-x86_64-icc --charm-arch net-linux-x86_64-cuda-ifort-smp-icc
--with-fftw3 --fftw-prefix
/home/thomas/Programs/fftw-3.3.2/single_precision_Intel_build --cuda-prefix
/usr/local/cuda
cd Linux-x86_64-icc
CUDA_DIR=/usr/local/cuda gmake -j8

but I get the following error:

icpc: command line warning #10121: overriding '-shared-intel' with
'-i-static'
icpc: warning #10237: -lcilkrts linked in dynamically, static library not
available
.rootdir/charm-6.4.0/net-linux-x86_64-cuda-ifort-smp-icc/bin/../lib/libcudah
ybridapi.a(hybridapi.o): In function `gpuProgressFn':
tmpxft_00003316_00000000-1_cuda-hybrid-api.cudafe1.cpp:(.text+0xb26):
undefined reference to `kernelSelect(workRequest*)'
tmpxft_00003316_00000000-1_cuda-hybrid-api.cudafe1.cpp:(.text+0xbe1):
undefined reference to `kernelSelect(workRequest*)'
/usr/local/cuda/lib64/libcudart.so: undefined reference to
`std::_List_node_base::_M_unhook()@GLIBCXX_3.4.14'
/usr/local/cuda/lib64/libcudart.so: undefined reference to
`std::_List_node_base::_M_hook(std::_List_node_base*)@GLIBCXX_3.4.14'
Fatal Error by charmc in directory
/home/thomas/Programs/NAMD_CVS-2012-09-22_Source/Linux-x86_64-icc

I also get the same error with Cuda 5.0. I then tried to replace the
libcudart.so (and libcudart.so.4) with the one that is shipped with NAMD but
got:

icpc: command line warning #10121: overriding '-shared-intel' with
'-i-static'
icpc: warning #10237: -lcilkrts linked in dynamically, static library not
available
.rootdir/charm-6.4.0/net-linux-x86_64-cuda-ifort-smp-icc/bin/../lib/libcudah
ybridapi.a(hybridapi.o): In function `gpuProgressFn':
tmpxft_00003316_00000000-1_cuda-hybrid-api.cudafe1.cpp:(.text+0xb26):
undefined reference to `kernelSelect(workRequest*)'
tmpxft_00003316_00000000-1_cuda-hybrid-api.cudafe1.cpp:(.text+0xbe1):
undefined reference to `kernelSelect(workRequest*)'
Fatal Error by charmc in directory
/home/thomas/Programs/NAMD_CVS-2012-09-22_Source/Linux-x86_64-icc

Does anyone know how to fix this? Do I need to install an older version of
CUDA?

thanks,
Thomas

-- 
======================================================================
Thomas Evangelidis
PhD student
University of Athens
Faculty of Pharmacy
Department of Pharmaceutical Chemistry
Panepistimioupoli-Zografou
157 71 Athens
GREECE
email: tevang_at_pharm.uoa.gr
          tevang3_at_gmail.com
website: https://sites.google.com/site/thomasevangelidishomepage/
 
 
-- 
Aron Broom M.Sc
PhD Student
Department of Chemistry
University of Waterloo

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:22:12 CST