Mario, On Friday 27 June 2008 16:09, Mario Valle wrote: > > Ok, I see. You are running in gcc, i386 without any -march= compiler > > flag. So gcc assumes that you want to compile for a plain old i386 (read > > this does not even mean an i486). In this case gcc is limited to an > > instruction set where no such atomic operations are available. In fact > > you could not even programm a mutex with that i386 instruction set, as > > well as you could not build any SMP machine using an i386. > > Ahi Ahi Ahi!! > > That means that I'm working without this optimization since the beginning? > And nevertheless OSG is so fast? > > I specify in the advanced cmake options: > CMAKE_C_FLAGS -march=i486 -mtune=i486 > CMAKE_CXX_FLAGS -march=i486 -mtune=i486 > Then configure and generate, but the ./include/OpenThreads/Config file > continues with everything commented out. Have I missed something? Ok, then throw away your build to make sure that nothing si left.
I am not sure how many things are cached in what part of the cmake build system. I just want to be sure ... Then set again the cflags. And better set them to something like -march=pentiumpro. Also do not forget to include -O or -O3 or something like that. Without you do not get any optimizations I guess ... Then ensure that cmake does not see previous caches ... BTW: does anybody know a way to tell cmake that a variable should not be cached? > > In this case, when we cannot assume any atomic operations. So we need to > > use the previous mutex based behaviour. > > But for that, the old mechnisms are still there. You need to ensure that > > refcounting works in a threadsafe way using either the > > OSG_THREAD_SAFE_REF_UNREF environment variable or by calling > > OK, thanks. But for now I want only to contribute with my ignorance :-) to > polish this OSG functionality. > > > osg::Referenced::setThreadSafeReferenceCounting(true); > > > > before *any* concurrently used osg::Referenced object is created. > > Alternatively you can uncomment that ENFORCE_THREADSAFE define in > > src/osg/Referenced.cpp that does change the default behaviour of > > referenced. > > > >> As soon as the build finishes, I will report back to you. > > OK, now crashes in another place, see below. This run has been done before > changing march/mtune. > > > Are you running with the environment variable OSG_THREAD_SAFE_REF_UNREF > > set or unset? > > Unset Can you test with this variable set? I believe that these thread safety defaults in osg are not good in face of that multithreaded viewer stuff, but I just left them as is. Robert, Should we include that ENFORCE_THREADSAFE define in the cmake configuration? And in face of all this threading usage in osg, should the default change to on? And, yes, I should write a wiki page. I hope to find some time today ... Greetings Mathias -- Dr. Mathias Fröhlich, science + computing ag, Software Solutions Hagellocher Weg 71-75, D-72070 Tuebingen, Germany Phone: +49 7071 9457-268, Fax: +49 7071 9457-511 -- Vorstand/Board of Management: Dr. Bernd Finkbeiner, Dr. Florian Geyer, Dr. Roland Niemeier, Dr. Arno Steitz, Dr. Ingrid Zech Vorsitzender des Aufsichtsrats/ Chairman of the Supervisory Board: Prof. Dr. Hanns Ruder Sitz/Registered Office: Tuebingen Registergericht/Registration Court: Stuttgart Registernummer/Commercial Register No.: HRB 382196 _______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
