Re: Installing cuda-4.0
Solved except for a side topic. See my closing question to this regard. Installation of cuda 4.0 in amd64 wheezy succeeds from Debian packages, and there is no conflict with the installation of the nvidia driver the Debian way. No need to go to the nvidia web, so that both the nvidia driver and cuda 4.0 place the files according to Debian rules. Following installation of the nvidia drives, it suffices to install: nvidia-cuda-dev libnvidia-compiler libcupti-dev libcupti4 libcupti-doc (not sure whether these libcupti are really needed) The cuda headers are under /usr/include, so that, in configuring the computational package to compile, the flag --cuda-prefix /usr (not /usr/include, in order that all CUDA bin, lib, and include are found. I understand that I am discovering the hot water. However, as I had no reply to my question QUESTION: What failed is a special (minor) feature of the computational code. To get that, it was required to point to the location of the lapack and c standard libraries. Which I tried to satisfy by pointing to /usr/lib/lapack/liblapack.so.3gf.0 /usr/lib/x86_64-linux-gnu/libstdc++.so.6. Which prevented the compilation to occur. Is this link to the libraries correct for Debian amd64 wheezy? Thanks francesco pietra On Mon, Mar 11, 2013 at 6:40 PM, Francesco Pietra chiendar...@gmail.com wrote: Hello: With amd64 wheezy, intel i7-3930K and two GTX-680, so far I got correctly the X server by installing build-essential nvidia-kernel-dkms and nvidia-xconfig. I could also run successfully binary CUDA-enabled codes (NAMD molecular dynamics, which contains the runtime libcudart.so.4) Now, in order to include special plugins, I am trying to compile NAMD from source. The plugins were correctly patched, charm++ was correctly compiled for multicore, however, final make was vainly looking for the cuda/include directory at /usr/local/encap/cuda-4.0/include. Non existing. I have now installed nvidia-cuda-dev libnvidia-compiler, whereby which nvcc reports /usr/bin/nvcc so that there is now the nvidia compiler. Now, francesco@:~$ dpkg -l | grep nvidia ii glx-alternative-nvidia 0.2.2 amd64allows the selection of NVIDIA as GLX provider ii libgl1-nvidia-alternatives 304.64-4 amd64transition libGL.so* diversions to glx-alternative-nvidia ii libgl1-nvidia-glx:amd64304.48-1 amd64NVIDIA binary OpenGL libraries ii libglx-nvidia-alternatives 304.64-4 amd64transition libgl.so diversions to glx-alternative-nvidia ii libnvidia-compiler:amd64 304.64-4 amd64NVIDIA runtime compiler library ii libnvidia-ml1:amd64304.48-1 amd64NVIDIA management library (NVML) runtime library ii nvidia-alternative 304.48-1 amd64allows the selection of NVIDIA as GLX provider ii nvidia-cuda-dev4.2.9-2 amd64NVIDIA CUDA development files ii nvidia-cuda-doc4.2.9-2 all NVIDIA CUDA and OpenCL documentation ii nvidia-cuda-gdb4.2.9-2 amd64NVIDIA CUDA GDB ii nvidia-cuda-toolkit4.2.9-2 amd64NVIDIA CUDA toolkit ii nvidia-glx 304.48-1 amd64NVIDIA metapackage ii nvidia-installer-cleanup 20120630+3 amd64Cleanup after driver installation with the nvidia-installer ii nvidia-kernel-common 20120630+3 amd64NVIDIA binary kernel module support files ii nvidia-kernel-dkms 304.48-1 amd64NVIDIA binary kernel module DKMS source ii nvidia-libopencl1:amd64304.64-4 amd64NVIDIA OpenCL library ii nvidia-opencl-dev 4.2.9-2 amd64NVIDIA OpenCL development files ii nvidia-settings304.64-1 amd64Tool for configuring the NVIDIA graphics driver ii nvidia-smi 304.48-1 amd64NVIDIA System Management Interface ii nvidia-support 20120630+3 amd64NVIDIA binary graphics driver support files ii nvidia-vdpau-driver:amd64 304.48-1 amd64NVIDIA vdpau driver ii nvidia-visual-profiler 4.2.9-2 amd64NVIDIA Visual Profiler ii nvidia-xconfig 304.48-1 amd64X configuration tool for non-free NVIDIA drivers ii xserver-xorg-video-nvidia 304.48-1 amd64NVIDIA binary Xorg driver francesco@gig64:~$ Still missing is /encap/cuda-4.0/include. To avoid messing the system, may I ask how to correctly get /cuda-.40/include (in order to specify --cuda-prefix... in the
Re: Installing cuda-4.0
Hi Francesco, your issue may not be a CUDA issue, but rather a NAMD. CUDA debian packages are fine. You may look in this directory, in particular for installing NAMD success story on Debian and for some deb ball written by some folks. In general the main issue with these kinds of source is that the Linux customary rules are absolutely not followed, namely, they have their own way to build and install (and most of the time it looks very much that those rules were not written by linux folks). Jerome On 11/03/13 18:40, Francesco Pietra wrote: Hello: With amd64 wheezy, intel i7-3930K and two GTX-680, so far I got correctly the X server by installing build-essential nvidia-kernel-dkms and nvidia-xconfig. I could also run successfully binary CUDA-enabled codes (NAMD molecular dynamics, which contains the runtime libcudart.so.4 this sounds as a good start for a mess: cuda stuff may be distributed by cuda ball, point. ) Now, in order to include special plugins, I am trying to compile NAMD from source. The plugins were correctly patched, charm++ was correctly compiled for multicore, however, final make was vainly looking for the cuda/include directory at /usr/local/encap/cuda-4.0/include. your building configuration is incorrect: /usr/local tree is meant to be managed by the local superuser, not by the Debian machinery. Non existing. I have now installed nvidia-cuda-dev libnvidia-compiler, whereby which nvcc reports /usr/bin/nvcc so that there is now the nvidia compiler. You must configure prperly the NAMD stuff. Now, francesco@:~$ dpkg -l | grep nvidia ii glx-alternative-nvidia 0.2.2 amd64allows the selection of NVIDIA as GLX provider ii libgl1-nvidia-alternatives 304.64-4 amd64transition libGL.so* diversions to glx-alternative-nvidia ii libgl1-nvidia-glx:amd64304.48-1 amd64NVIDIA binary OpenGL libraries ii libglx-nvidia-alternatives 304.64-4 amd64transition libgl.so diversions to glx-alternative-nvidia ii libnvidia-compiler:amd64 304.64-4 amd64NVIDIA runtime compiler library ii libnvidia-ml1:amd64304.48-1 amd64NVIDIA management library (NVML) runtime library ii nvidia-alternative 304.48-1 amd64allows the selection of NVIDIA as GLX provider ii nvidia-cuda-dev4.2.9-2 amd64NVIDIA CUDA development files ii nvidia-cuda-doc4.2.9-2 all NVIDIA CUDA and OpenCL documentation ii nvidia-cuda-gdb4.2.9-2 amd64NVIDIA CUDA GDB ii nvidia-cuda-toolkit4.2.9-2 amd64NVIDIA CUDA toolkit ii nvidia-glx 304.48-1 amd64NVIDIA metapackage ii nvidia-installer-cleanup 20120630+3 amd64Cleanup after driver installation with the nvidia-installer ii nvidia-kernel-common 20120630+3 amd64NVIDIA binary kernel module support files ii nvidia-kernel-dkms 304.48-1 amd64NVIDIA binary kernel module DKMS source ii nvidia-libopencl1:amd64304.64-4 amd64NVIDIA OpenCL library ii nvidia-opencl-dev 4.2.9-2 amd64NVIDIA OpenCL development files ii nvidia-settings304.64-1 amd64Tool for configuring the NVIDIA graphics driver ii nvidia-smi 304.48-1 amd64NVIDIA System Management Interface ii nvidia-support 20120630+3 amd64NVIDIA binary graphics driver support files ii nvidia-vdpau-driver:amd64 304.48-1 amd64NVIDIA vdpau driver ii nvidia-visual-profiler 4.2.9-2 amd64NVIDIA Visual Profiler ii nvidia-xconfig 304.48-1 amd64X configuration tool for non-free NVIDIA drivers ii xserver-xorg-video-nvidia 304.48-1 amd64NVIDIA binary Xorg driver francesco@gig64:~$ Still missing is /encap/cuda-4.0/include. To avoid messing the system, may I ask how to correctly get /cuda-.40/include (in order to specify --cuda-prefix... in the configuration of the code)? As far as I know, it should be located in the cuda SDK directory structure. Thanks francesco pietra -- To UNSUBSCRIBE, email to debian-amd64-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/5140249d.10...@rezozer.net
Re: Installing cuda-4.0
On 13/03/13 08:02, Jerome BENOIT wrote: Hi Francesco, your issue may not be a CUDA issue, but rather a NAMD. CUDA debian packages are fine. You may look in this directory, in particular for installing NAMD success story on Debian and for some deb ball written by some folks. In general the main issue with these kinds of source is that the Linux customary rules are absolutely not followed, namely, they have their own way to build and install (and most of the time it looks very much that those rules were not written by NON linux folks). ^ Sorry for the typo, Jerome Jerome On 11/03/13 18:40, Francesco Pietra wrote: Hello: With amd64 wheezy, intel i7-3930K and two GTX-680, so far I got correctly the X server by installing build-essential nvidia-kernel-dkms and nvidia-xconfig. I could also run successfully binary CUDA-enabled codes (NAMD molecular dynamics, which contains the runtime libcudart.so.4 this sounds as a good start for a mess: cuda stuff may be distributed by cuda ball, point. ) Now, in order to include special plugins, I am trying to compile NAMD from source. The plugins were correctly patched, charm++ was correctly compiled for multicore, however, final make was vainly looking for the cuda/include directory at /usr/local/encap/cuda-4.0/include. your building configuration is incorrect: /usr/local tree is meant to be managed by the local superuser, not by the Debian machinery. Non existing. I have now installed nvidia-cuda-dev libnvidia-compiler, whereby which nvcc reports /usr/bin/nvcc so that there is now the nvidia compiler. You must configure prperly the NAMD stuff. Now, francesco@:~$ dpkg -l | grep nvidia ii glx-alternative-nvidia 0.2.2 amd64 allows the selection of NVIDIA as GLX provider ii libgl1-nvidia-alternatives 304.64-4 amd64 transition libGL.so* diversions to glx-alternative-nvidia ii libgl1-nvidia-glx:amd64 304.48-1 amd64 NVIDIA binary OpenGL libraries ii libglx-nvidia-alternatives 304.64-4 amd64 transition libgl.so diversions to glx-alternative-nvidia ii libnvidia-compiler:amd64 304.64-4 amd64 NVIDIA runtime compiler library ii libnvidia-ml1:amd64 304.48-1 amd64 NVIDIA management library (NVML) runtime library ii nvidia-alternative 304.48-1 amd64 allows the selection of NVIDIA as GLX provider ii nvidia-cuda-dev 4.2.9-2 amd64 NVIDIA CUDA development files ii nvidia-cuda-doc 4.2.9-2 all NVIDIA CUDA and OpenCL documentation ii nvidia-cuda-gdb 4.2.9-2 amd64 NVIDIA CUDA GDB ii nvidia-cuda-toolkit 4.2.9-2 amd64 NVIDIA CUDA toolkit ii nvidia-glx 304.48-1 amd64 NVIDIA metapackage ii nvidia-installer-cleanup 20120630+3 amd64 Cleanup after driver installation with the nvidia-installer ii nvidia-kernel-common 20120630+3 amd64 NVIDIA binary kernel module support files ii nvidia-kernel-dkms 304.48-1 amd64 NVIDIA binary kernel module DKMS source ii nvidia-libopencl1:amd64 304.64-4 amd64 NVIDIA OpenCL library ii nvidia-opencl-dev 4.2.9-2 amd64 NVIDIA OpenCL development files ii nvidia-settings 304.64-1 amd64 Tool for configuring the NVIDIA graphics driver ii nvidia-smi 304.48-1 amd64 NVIDIA System Management Interface ii nvidia-support 20120630+3 amd64 NVIDIA binary graphics driver support files ii nvidia-vdpau-driver:amd64 304.48-1 amd64 NVIDIA vdpau driver ii nvidia-visual-profiler 4.2.9-2 amd64 NVIDIA Visual Profiler ii nvidia-xconfig 304.48-1 amd64 X configuration tool for non-free NVIDIA drivers ii xserver-xorg-video-nvidia 304.48-1 amd64 NVIDIA binary Xorg driver francesco@gig64:~$ Still missing is /encap/cuda-4.0/include. To avoid messing the system, may I ask how to correctly get /cuda-.40/include (in order to specify --cuda-prefix... in the configuration of the code)? As far as I know, it should be located in the cuda SDK directory structure. Thanks francesco pietra -- To UNSUBSCRIBE, email to debian-amd64-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/514029b8.7070...@rezozer.net