On Sun, 15 Aug 2010, Kai-Uwe Behrmann wrote: > > Valgrind keeps completly silent except for some uninitialised values. > However filling them does not help. I tried on two multi core intel > machines on Linux.
Did you use 'valgrind --tool=helgrind'? A very recent valgrind is recommended. I use the SVN version 3.6. Unfortunately, this requires that you replace libgomp.so with an alternate one which uses POSIX threads rather than Linux proprietary features. I accomplished this by building the system version of the GCC compiler from source (built using special instructions at the valgrind site), and then replacing libgomp with my own build. Valgrind's helgrind only sees issues which actually occur during the run, but it is helpful. It found a number of issues in GraphicsMagick for me to fix. The latest Sun/Oracle Studio compiler includes a mode by which it reports thread race conditions during compilation. It uses the compiler's optimization smarts to do so. This seems much more complete than 'helgrind' to me but some reported race conditions may never be observed in practice on normal hardware. My port of the GraphicsMagick CMS code to lcms 2.0 still does not try to use cmsCreateProofingTransformTHR() in multiple threads. Bob -- Bob Friesenhahn bfrie...@simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer, http://www.GraphicsMagick.org/ ------------------------------------------------------------------------------ This SF.net email is sponsored by Make an app they can't live without Enter the BlackBerry Developer Challenge http://p.sf.net/sfu/RIM-dev2dev _______________________________________________ Lcms-user mailing list Lcms-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lcms-user
