C[omp]ute

From: andrew cooke <andrew@...>

Date: Fri, 16 Oct 2009 16:58:11 -0300

Some notes about getting OpenCL working on OpenSuse (I am using 11.1).

1 - Install the Open CL driver.  You *must* also uninstall and
previously installed, "standard" NVidia driver by deselecting it in
Yast (if you don't, you'll get an error about the kernel module
version not matching the driver version number).

2 - Install the CUDA 2.3 toolkit (and add the "bin" directory to your path)

3 - Install the OpenCL toolkit (see below)

As with the early CUDA releases, the NVidia installer runs a perl
script that is buffered, which makes running the installer confusing.

The solution is to do the following:


gpucomputingsdk_2.3b_linux.run --noexec --target opencl-2.3b
cd opencl-2.3b

sed -i '/#!/ a\
select((select(STDOUT), $|=1)[0]);' install-sdk-linux.pl

./install-sdk-linux.pl


(the sed command is on two lines, the first line ends with a
backslash; alternatviely you can edit install-sdk-linux.sh and add
"select((select(STDOUT), $|=1)[0]);" after the first line)

4 - You also need an OpenCL implementation.  I am using PyOpenCL -
http://mathema.tician.de/software/pyopencl

From Yast install numpy, libboost_python and libboost_thread, then:


./configure.py --boost-python-libname=boost_python
--boost-thread-libname=boost_thread
python2.6 setup.py build
sudo python2.6 setup.py install

cd examples
python2.6 benchmark-all.py
 Execution time of test without OpenCL:  7.82306098938 s
 ===============================================================
 Platform name: NVIDIA
 Platform profile: FULL_PROFILE
 Platform vendor: NVIDIA Corporation
 Platform version: OpenCL 1.0
 ---------------------------------------------------------------
 Device name: GeForce 8400 GS
 Device type: GPU
 Device memory:  511 MB
 Device max clock speed: 1400 MHz
 Device compute units: 1
 Execution time of test: 0.247648 s
 Results OK


Andrew