AW: NAMD 2.9 CUDA compilation

From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Tue Mar 12 2013 - 02:53:06 CDT

Hi,

then I guess the cuda prefix should be just "/usr" then, the folders lib and include are expected within there.

Norman Geist.

> -----Ursprüngliche Nachricht-----
> Von: Francesco Pietra [mailto:chiendarret_at_gmail.com]
> Gesendet: Dienstag, 12. März 2013 08:48
> An: Norman Geist
> Betreff: Re: namd-l: NAMD 2.9 CUDA compilation
>
> Hi Norman:
> In the meantime I made some progress. With the installation of
> "nvidia-cuda-dev" and "libnvidia-compiler", I got the needed
> /usr/bin/nvcc
>
> I had already installed all the packages provided by Debian
> nvidia-cuda-toolkit, to which I added libcupti-dev libcupti4 and
> libcupti-doc.
>
> Now, as far as I can understand, the required headers are under
> /usr/include, such as, fpr example,
>
> /usr/include/cuda_runtime.h
> /usr/include/cuda.h
> etc
>
> I wonder whether now the correct configuration can be obtained with
>
> ./config Linux-x86_64-g++ --with-cuda --cuda-prefix /usr/include
>
>
> (not yet tried, just because I am considering whether there can be any
> conflict that could mess the X-server. I got the nvidia driver the
> Debian way, i.e., by installing "build-essential" "nvidia-kernel-dkms"
> "nvidia-xconfig" This is very convenient, as the nvidia driver is
> updated automatically at "apt-get update/upgrade"
>
>
> On previous attempts, with merely
>
> ./config Linux-x86_64-g++ --with-cuda
>
> the system was looking for /usr/local/encap/cuda4.0/include
>
> Thanks
>
> francesco
>
> On Tue, Mar 12, 2013 at 7:57 AM, Norman Geist
> <norman.geist_at_uni-greifswald.de> wrote:
> > Hi Francesco,
> >
> > the usual path for the cuda runtime is "/usr/local/cuda" I guess. If
> your installation does not have this path, you have to use the
> configure option to specify the location on you system. Also IMHO it is
> not enough to just include the the folder to your includes, as it will
> also need nvcc and other stuff maybe. I guess the best to try is to
> supply the right location to you cuda installation at configure stage.
> >
> > Norman Geist.
> >
> >> -----Ursprüngliche Nachricht-----
> >> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im
> >> Auftrag von Francesco Pietra
> >> Gesendet: Montag, 11. März 2013 16:57
> >> An: NAMD
> >> Betreff: Fwd: namd-l: NAMD 2.9 CUDA compilation
> >>
> >> I recognize now that I was previously using namd 2.9 binary, which
> was
> >> not activated during the compilation below. Therefore ,
> >>
> >> -DNAMD_FFTW -DNAMD_CUDA -I. -I/usr/local/encap/cuda-4.0/include
> >>
> >>
> >> will not find the runtime. What should be installed for the runtime?
> >>
> >> From "apt-cache search cuda";
> >>
> >> libcublas4 - NVIDIA CUDA BLAS runtime library
> >> libcudart4 - NVIDIA CUDA runtime library
> >> libcufft4 - NVIDIA CUDA FFT runtime library
> >> libcuinj4 - NVIDIA CUDA INJ runtime library
> >> libcupti-dev - NVIDIA CUDA Profiler Tools Interface development
> files
> >> libcupti-doc - NVIDIA CUDA Profiler Tools Interface documentation
> >> libcupti4 - NVIDIA CUDA Profiler Tools Interface runtime library
> >> libcurand4 - NVIDIA CUDA Random Numbers Generation runtime library
> >> libcusparse4 - NVIDIA CUDA Sparse Matrix runtime library
> >> libnpp4 - NVIDIA Performance Primitives runtime library
> >> nvidia-cuda-dev - NVIDIA CUDA development files
> >> nvidia-cuda-doc - NVIDIA CUDA and OpenCL documentation
> >> nvidia-cuda-gdb - NVIDIA CUDA GDB
> >> nvidia-cuda-toolkit - NVIDIA CUDA toolkit
> >> nvidia-visual-profiler - NVIDIA Visual Profiler
> >> libcuda1 - NVIDIA CUDA runtime library
> >> libcuda1-ia32 - please switch to multiarch libcuda1:i386
> >> libnvcuvid1 - NVIDIA CUDA nvcuvid runtime library
> >> libnvidia-compiler - NVIDIA runtime compiler library
> >>
> >>
> >> Thanks
> >>
> >> francesco
> >>
> >> ---------- Forwarded message ----------
> >> From: Francesco Pietra <chiendarret_at_gmail.com>
> >> Date: Mon, Mar 11, 2013 at 4:40 PM
> >> Subject: namd-l: NAMD 2.9 CUDA compilation
> >> To: NAMD <namd-l_at_ks.uiuc.edu>
> >>
> >>
> >> Hello:
> >> I am attempting to compile namd 2.9 with cuda on Debian GNU/Linux
> >> amd64. Configured;
> >>
> >> ./config Linux-x86_64-g++ --with-cuda
> >>
> >>
> >> The output of "make depends", followed by "make"
> >>
> >> ended with error:
> >>
> >> g++ -m64 -O3 -I.rootdir/charm-6.4.0/multicore-linux64/include
> >> -DCMK_OPTIMIZE=1 -Isrc -Iinc -Iplugins/include -DSTATIC_PLUGIN
> >> -I.rootdir/tcl-threaded/include -DNAMD_TCL -I.rootdir/fftw/include
> >> -DNAMD_FFTW -DNAMD_CUDA -I. -I/usr/local/encap/cuda-4.0/include
> >> -DNAMD_VERSION=\"2.9\" -DNAMD_PLATFORM=\"Linux-x86_64-multicore-
> CUDA\"
> >> -DNAMD -DREMOVE_PROXYRESULTMSG_EXTRACOPY
> >> -DNODEAWARE_PROXY_SPANNINGTREE -DUSE_NODEPATCHMGR
> >> -fexpensive-optimizations -ffast-math -o obj/ComputeNonbondedCUDA.o
> >> -c src/ComputeNonbondedCUDA.C
> >> src/ComputeNonbondedCUDA.C:6:26: fatal error: cuda_runtime.h: No
> such
> >> file or directory
> >> compilation terminated.
> >> make: *** [obj/ComputeNonbondedCUDA.o] Error 1
> >>
> >> Actually, /src does not include cuda-related files. Was the
> >> configuration command malformed?
> >>
> >> Thanks. Please inform me if more information is needed. I tried to
> be
> >> concise.
> >>
> >> francesco pietra
> >

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:23:03 CST