PDF version of this entire document
To more effectively debug those pointer issues, valgrind might have to be installed on the 64-bit machine (I asked Yaron to use the admin account to achieve it yesterday afternoon). MEX has some really horrible debugging facilities and it leaves the user stranded in the land of GDB (based on the advice from MathWorks' site) or just trial-and-error. Debugging segmentation faults is hard without advanced debuggers or valgrind. Meanwhile, new workaround code was written to reject improper allocations of memory (initalisation of FMM leaving pointers that are not accessible to the marching operations). The original code seems to have been written for 32-bit Windows, whereas the servers use swap and highmem/Physical Address Extension (PAE), so we have added a constraint for it never to use the swap rather than make a more permanent fix. This at least should facilitate crashless experiments and development.
