C[omp]ute

Welcome to my blog, which was once a mailing list of the same name and is still generated by mail. Please reply via the "comment" links.

Always interested in offers/projects/new ideas. Eclectic experience in fields like: numerical computing; Python web; Java enterprise; functional languages; GPGPU; SQL databases; etc. Based in Santiago, Chile; telecommute worldwide. CV; email.

Personal Projects

Lepl parser for Python.

Colorless Green.

Photography around Santiago.

SVG experiment.

Professional Portfolio

Calibration of seismometers.

Data access via web services.

Cache rewrite.

Extending OpenSSH.

Last 100 entries

Re: Python's sad, unimaginative Enum; Some explanation; Printing binary trees sideways; About "Python's sad, unimaginative Enum"; Atoms in python; Some good feedback here; Frustration Understood; I agree with you #nt; What would be imaginative?; Re: Enum; this is fucking useless; Enum; Python's sad, unimaginative Enum; Possible Fix; Work, Exhaustion, Vacation; VirtualBox with Centos 6.3 to 6.4, client; Matasano - Programming Lessons Learned; PDF to HTML; Alternate Substitution; Why RSA Works; Trigger; Dreaming of Death; Example: Tracing; Using Coroutines In Protocol Simulations; Python 3.3 Only; Pure Python SHA1 and MD4 Implementations; Ubuntu on VirtualBox; Starting TOR as a service on OpenSuse 12.3; 1001 Albums; Using fail2ban on OpenSuse 12.3; PPPoE on OpenSuse 12.3; Good Article on Unified Physics; It's Police (Carabineros); Linux Software for Listening to and Exploring Music; Android is Pretty Bad; Lucky Number; 3D Printing for Casting; Cover Art for MPDroid; Who'd a thought the French were so bigoted?; PS Input Signal; Small Problem with Roksan K2 Amp; Roksan K2 Amp + ATC SCM7 Speakers; Do What Makes Sense; Re: Arguing About Tests, Still; Arguing About Tests, Still; Images; Good Article on NY Drummers; Related Bug Report; Getting Python 3.3 and Virtualenv Working in OpenSuse 12.3; How I Am; Awesome video about digital audio; The Difference Between Dimensional and Normalized Databases; The rise of the new Chinese bogeyman; Updated Syntax; Very First Steps to C-ORM; The Ideal User Interface For Music Exploration; Can The Republicans Be Saved?; Rate Limiting Calls to EchoNest; Mods to Cache; Comparing UYKFG and UYKFD/E/F; Someone Else is Concerned; EchoNest-based Playlist Generator for MPD; Example Voting Results; A Heavyweight Python Cache; Identifying Artists with EchoNest; Notes on Pregalex / Pregabalina / Lyrica; The Neil Cowley Trio; Drake - Make for Data; A Reliable Python Web Service; Useful Python Date/Time Library?; Need to Sleep, But this is Good; Command Line Set Difference; Little Details...; Linux Command Line Tricks; AutoTools Tutorial; Hangman Tactics; A Tor Proxy Embedded In A Web Page; Tree (Nested Dicts) in Python; Sleeping at Parties; I Know Someone Who Hurts Other People; Light and Tea; Description of the LCS35 Time Capsule Crypto-Puzzle; Re: I can relate to that ...; I can relate to that ...; Re: It's 2012 Why Does My IDE Suck?; My Own Alternative Medicine; Nice explanation of SVM; Why and How Writing Crypto is Hard; Re: It's 2012 Why Does My IDE Suck?; Incremental Regular Expressions; BBC Map Confused at Pole; Social Media: Ground Zero in the Culture War; My Visit to the Psycho Doc; Learning Modern 3D Graphics Programming; Hope you got some crackers to go with the cheese; Re: But how easy would it be ...; But how easy would it be ...; Powerline Freq Fingerprinting of Audio; The Folly of Scientism; Cheese - Because You're Going to Die Anyway; Another GPU Success - PyCUDA, Cross-Correlations

Calling OpenCL from Octave / Matlab

From: andrew cooke <andrew@...>

Date: Wed, 9 Dec 2009 11:42:22 -0300

[I am optimising some Matlab code but, because I don't have a Matlab
licence on my local dev machine, am using Octave.  So what follows is
a bit odd, because I am using Octave, but via the Matlab-compatible
tools.  That doesn't affect the main points, though...]

Calling OpenCL from Octave / Matlab turns out to be (at least in a
simple case) very similar to calling C.  The main issues are (1) C/C++
interop and (2) libraries.


Currently the Octave code calls a C routine which, in turn, calls the
OpenCL C++ code.  The OpenCL C++ code is compiled into a library
(called "support" below).  I used this approach because I already knew
Octave (+ Matlab) can call C, and wasn't sure about C++.  So this
moves worries about C/C++ interop to the C level (where I know how to
deal with them) rather than the Octave/Matlab level (where live might
be more complicated).  In other words (to labour the point) I am using
C as the main interface to Octave / Matlab, not C++.

So the script to build the code looks like this:

g++ -fpic -Wall
-I/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/OpenCL/common/inc
-I/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/shared/inc -o support.o -c
support.cpp
gcc -shared -Wl,-soname,libsupport.so -olibsupport.so support.o

gcc -o main main.c -L. -lsupport -L/usr/lib64/
-L/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/OpenCL/common//lib
-L/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/shared/lib/
-L/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/shared/lib/linux -lOpenCL
-loclUtil -lshrutil -lstdc++

mkoctfile -v --mex singlec5.c -L. -lsupport -L/usr/lib64/
-L/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/OpenCL/common//lib
-L/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/shared/lib/
-L/home/andrew/pkg/NVIDIA_GPU_Computing_SDK/shared/lib/linux -lOpenCL
-loclUtil -lshrutil -lstdc++

(note the use of fpic - more on that below; also my toolkit is
installed in a non-standard location).  The "main" above is a separate
C program that calls the OpenCL routines.  Getting this working first
allowed me to fix basic errors without worrying about Octave.

The file singlec5.c is the "mex" file that handles the call from
Octave, unpacks/packs data structures and calls the OpenCl code in the
support library.


All the above is pretty normal, I think.  However, it won't work "out
of the box" because it requires that the support libraries (provided
by NVidia and used in their demos, which I based my code on) are
compiled only as ".a" (archive) libraries and not ".so" (shared
object) (this is on x86-64 Linux (OpenSuse 11.1)).

To build the libraries as shared objects, modify the following two
files so that "-fpic" is included in CXX and CC as well as LINK:

NVIDIA_GPU_Computing_SDK/OpenCL/common/common_opencl.mk
NVIDIA_GPU_Computing_SDK/shared/common.mk

Then rebuild the libraries:

cd NVIDIA_GPU_Computing_SDK/OpenCL/common/obj/release
gcc -shared -Wl,-soname,liboclUtil.so -oliboclUtil.so oclUtils.cpp.o
mv liboclUtil.so ../../../common/lib/

cd NVIDIA_GPU_Computing_SDK/shared/obj/release
gcc -shared -Wl,-soname,libshrUtil.so -olibshrUtil.so shrUtils.cpp.o
mv libshrUtil.so ../../lib/

Finally, the LD_LIBRARY_PATH should include

NVIDIA_GPU_Computing_SDK/shared/lib
NVIDIA_GPU_Computing_SDK/OpenCL/common/lib

(Paths will need to be changed as appropriate - I hope that's obvious....!)

Andrew

Extern C

From: andrew cooke <andrew@...>

Date: Wed, 9 Dec 2009 11:53:38 -0300

I should also have added, although I guess it's obvious, that the
header file for support.cpp, support.h, includes

  extern "C"

before each function declaration.  This lets C code link to the C++
routines (it removes the C++ name mangling, and may do other things I
do not understand).

Andrew

Or Simply Don't Use The Libs

From: andrew cooke <andrew@...>

Date: Wed, 16 Dec 2009 11:53:00 -0300

Hmmm.  On another machine, working with Matlab rather than Octave, and
with CentOS rather than OpenSuse, the above failed to fix things.

So I revised my code.  For now I have changed the logging to use
printf().  Apart from that, the only critical support code used was
oclLoadProgSource.  That just reads the source into memory.  Reading
files in C is tedious, but not impossible; here's a replacement:

    FILE *file;
    size_t blockAvailable = 512;
    size_t blockRead;
    char *kernelSource = NULL;
    size_t kernelLength = 0;
    size_t currentSize = 0;

    printf("Loading %s\n", kernelPath);
    file = fopen(kernelPath, "r");
    if (! file) {
        printf("Could not find kernel\n");
        err = 1;
        goto cleanup;
    }

    blockRead = blockAvailable;
    while (blockRead == blockAvailable) {
        blockAvailable *= 2;
        currentSize += blockAvailable;
        kernelSource = (char *)realloc(kernelSource, currentSize);
        blockRead = fread(kernelSource + kernelLength, sizeof(char),
                          blockAvailable, file);
        kernelLength += blockRead;
    }
    *(kernelSource + kernelLength + 1) = 0;
    printf(kernelSource);

You could add more checks (eg for realloc and fread) to handle more
errors, but the above works for now (this is very much proof of
concept code).

Andrew

More On OpenCL and Matlab Here

From: andrew cooke <andrew@...>

Date: Thu, 17 Dec 2009 19:47:47 -0300

http://www.acooke.org/cute/Experience0.html

Andrew

Comment on this post