Build problems with 0.0.20

[gdabram]

Is this stuff still active? I saw it mentioned in HPCwire
http://www.hpcwire.com/hpcwire/2011-06-09/top_10_objections_to_gpu_computing_reconsidered.html
and decided to take a look. Ran into a few problems compiling it on a
RHEL5 system using g++:

src/gmac/pthreads/pthread.cpp line 21 wouldn't compile. I changed it
from:
.... ((destructor())) ....
to
.... ((destructor)) ....

src/util/posix/Private-impl.h line 12 is initializing the key_ value
inside an assert. In non-debuggable code, the assert statement goes
away and the variable isn't initialized causing a segfault. Ought to
be:

pthread_key_create(&var.key_, NULL);
assert(var.get() == 0);

Also the install requires the presence of src/include/gmac/opencl.h
and src/include/gmac/opencl_types.h. Works if you touch them.

[Isaac Gelado]

Hi,
GMAC is still alive :-). We have been working during the last months
to improve the stability and performance of the code. What started as
minor modifications became a complete refactoring of the GMAC code.
Good news is that next week we will be updating GMAC to version 1.0,
which will include some new things like GPU-to-GPU copies, using the
write-combining buffer.. and other stuff.

Thanks for the bug reports; actually we do not have any RH machine to
test the code, so we appreciate your feedback. The issue with
pthread_create_key() (i.e. executing necessary code inside an
ASSERTION) is also present in a couple places more in the code...
hopefully in next week release we will have all of them fixed.

Cheers,
Isaac

--
Isaac Gelado

 'As gold which he cannot spend will make no man rich
    so knowledge which he cannot apply will make no man wise'