|
From: Christoph B. <bar...@or...> - 2007-08-16 08:41:34
|
Hi, I have bundled all changes to the current trunk I made in the attached patch. The patch allows you to run valgrind on big cpu-bound programs that do lots of allocations and deallocatoins and normally run one hour or more. My experience shows that for such programs valgrinds memcheck needs more than a week of runtime to complete. The patch helps you to reduce the runtime to one or two days. When should you try this patch? If the ratio between normal runtime and valgrind runtime gets worse than 40 this patch might help. If OProfile shows that less than 40% of the total runtime is spent in code generated by valgrind this patch might help. If you have millions of allocations and deallocations with millions of different tracebacks, this patch helps to reduce memory consumption and decreases runtime. What are the drawbacks? The patch is only tested by three users so far. And there are some regressions in the leak-check tool. I guess this is due to the different way memory allocations are handled. I do not see any regressions in normal memcheck oprations. Any comments are appreciated Christoph Bartoschek |