Hmmm, OK, it looks like 'ended' is set in the code after all in
pthread_create_wrapper, after the thread function exits. I'll keep
investigating this.
--
Monitor your physical, virtual and cloud infrastructure from a singl
Hi, time to look into this issue again.
I built a debug version of winpthreads but as you can understand
debugging this is a PITA.
Hang seems to happen around line 1539 of src/thread.c of winpthreads.
This reads like:
if (tv->ended == 0)
WaitForSingleObject (tv->h, INFINITE);
Trying to fig
2012/8/25 Antony Riakiotakis
> It is reproducable with just one thread too. This should be the simpler
> case:
>
> #include
> #include
> #include
>
> void *do_thread(void *) {
> int niterations = 200;
> int i;
>
> #pragma omp parallel for
> for (i = 0; i < niter
Hi niXman, no it won't cause a link error because the #pragma just
gets ignored so essentially there is no parallelism, hence, no hang.
It is interesting though that when using -fopenmp, the -lpthread flag
is unnecessary. Maybe when using openmp links some other version of
pthread somehow?
Just fo
2012/9/12 Antony Riakiotakis:
> Hi, any news on this issue? Can I help in some way?
Hi Antony,
I want to pay your attention to the fact that if I compile the code
without using the '-fopenmp' option, then the bug does not occur.
Although, far as I remember in this case must occur a link error. Bu
Hi, any news on this issue? Can I help in some way?
--
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discu
It is reproducable with just one thread too. This should be the simpler case:
#include
#include
#include
void *do_thread(void *) {
int niterations = 200;
int i;
#pragma omp parallel for
for (i = 0; i < niterations; i++)
{
printf (
Hi, I think I have figured it out:
The issue may be that internally openmp may be allocating thread
identifiers that may already be used by pthread so pthread_join
somehow ends up waiting for the parent thread to finish, deadlocking
itself.
Try this:
#include
#include
#include
void *do_threa
Indeed. I do not have much experience with thread programmng unfortunately. :/
In any case trying to debug this from within blender is quite hard
because I can't get a stacktrace. A disassembly shows that the problem
occurs within pthread_create_wrapper. Here is a small disassembly
around the cras
This test program is invalid. Your application tries to create an
thread and afterwards immediately terminates (by returning from main).
By add the line 'pthread_join (id, NULL);' after creation of the
thread, you will see the message "I am a thread!" as expected.
Regards,
Kai
-
Just as a note, removing the 'private' openmp directive also crashes.
--
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers
Hi, I am looking into openmp-related crashes that we've had into
blender. I see that the issue is known (
http://www.mail-archive.com/mingw-w64-public@lists.sourceforge.net/msg03982.html
) and that people from this mailing list have requested test files to
debug so I am sending a small program that
12 matches
Mail list logo
