Discussion:
[PyCUDA] 2nd try: 2014.1 git build, without siteconf.py
Bruce Labitt
2014-08-30 23:05:04 UTC
Permalink
I am trying to install PyCuda from git. I have Ubuntu 14.04, and CUDA6.5
installed. (Driver is 340.19 from Nvidia, CUDA from Nvidia) CUDA examples
seem to work, at least the ones supported my my hardware. (3.0) Partially
stuck on setting up siteconf.py correctly. Perhaps someone could help me
getting the right configuration.

1. Have libboost 1.54 installed from the repo. How do I find which version
of gcc was used to compile libboost?
2. What does CUDADRV_LIB_DIR mean? the directory where *what* file is
located? $ locate cannot find any file named cudadrv
3. For CUDART_LIB_DIR, can I use ${CUDA_ROOT}/lib, or do I need to use the
full path ${CUDA_ROOT}/lib/targets/x86_64-linux/lib ?
4. CXXFLAGS recommended? LDFLAGS?
5.Software updater is saying time for an updated kernel. If I update the
kernel, and its sources, do I need to reinstall the nvidia driver? Cuda6.5
required a new video driver to work.
6. When I did sudo make install, the installation complained it could not
find nvcc. This was because root did not have the correct PATH and
LD_LIBRARY_PATH. What is the best way to fix this? For myself, I just
edited my .bashrc file. Is this ok to do for root?

Thanks for any help!
Bruce
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.tiker.net/pipermail/pycuda/attachments/20140830/f38d3e80/attachment.html>
Andreas Kloeckner
2014-08-30 23:29:27 UTC
Permalink
Post by Bruce Labitt
I am trying to install PyCuda from git. I have Ubuntu 14.04, and CUDA6.5
installed. (Driver is 340.19 from Nvidia, CUDA from Nvidia) CUDA examples
seem to work, at least the ones supported my my hardware. (3.0) Partially
stuck on setting up siteconf.py correctly. Perhaps someone could help me
getting the right configuration.
1. Have libboost 1.54 installed from the repo. How do I find which version
of gcc was used to compile libboost?
To be clear, you don't need boost installed to compile PyCUDA, unless
you like to use the extra-difficult build. Just set "USE_SHIPPED_BOOST =
True" and relax.
Post by Bruce Labitt
2. What does CUDADRV_LIB_DIR mean? the directory where *what* file is
located? $ locate cannot find any file named cudadrv
libcuda.so
Post by Bruce Labitt
3. For CUDART_LIB_DIR, can I use ${CUDA_ROOT}/lib, or do I need to use the
full path ${CUDA_ROOT}/lib/targets/x86_64-linux/lib ?
Yes.
Post by Bruce Labitt
4. CXXFLAGS recommended? LDFLAGS?
Nope. Don't set those in siteconf.py to retain the defaults unless you
have a good reason.
Post by Bruce Labitt
5.Software updater is saying time for an updated kernel. If I update the
kernel, and its sources, do I need to reinstall the nvidia driver? Cuda6.5
required a new video driver to work.
Yes.
Post by Bruce Labitt
6. When I did sudo make install, the installation complained it could not
find nvcc. This was because root did not have the correct PATH and
LD_LIBRARY_PATH. What is the best way to fix this? For myself, I just
edited my .bashrc file. Is this ok to do for root?
Editing bashrc is fine. PyCUDA generally assumes that nvcc is somewhere
on the path, so you'll have to do that for the user that you'll run as,
too. Alternative: edit /etc/bashrc, or just say

export PATH=$PATH:.../cuda/bin

every time before you want to run PyCUDA code.

HTH,
Andreas
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 810 bytes
Desc: not available
URL: <http://lists.tiker.net/pipermail/pycuda/attachments/20140830/146778c6/attachment.sig>
Tomasz Rybak
2014-08-31 10:49:26 UTC
Permalink
You can try to look into Debian package.
I've changed it so the latest version (2014.1-1) should
build without any changes on both Debian and Ubuntu.

If you want to have packages (*.deb), you can clone
my git repository with packaging files:
http://anonscm.debian.org/gitweb/?p=collab-maint/python-pycuda.git
then call
debian/rules get-orig-source
unpack pycuda_*.tar.xz, copy debian/ directory to
pycuda-* directory, ensure that you have all build
dependencies and call
dpkg-buildpackage -rfakeroot
I'll create few *.deb files which you can install on your system.

Hope it helps.
--
Tomasz Rybak GPG/PGP key ID: 2AD5 9860
Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860
http://member.acm.org/~tomaszrybak

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part
URL: <http://lists.tiker.net/pipermail/pycuda/attachments/20140831/cce3759b/attachment.sig>
Loading...