|
From: Christoph B. <bar...@or...> - 2007-07-27 10:10:13
Attachments:
m_execontext.c.bz2
m_execontext.c.linear.bz2
|
Hi, I have tested linear probing for m_execontext.c, but it was far slower than chaining. There were long chains of filled buckets and the termination condition did not help very much. Maybe you find an error if you look at m_execontext.c.linear. However I am now advocating the use of chaining in m_execontext.c. I've implemented resizing of the hash table. Such that one does not have to choose a big initial size. Chaining uses more memory than linear probing but is far more efficient. Resizing of the hashtable costs some performance but for my testprogramms the gains are far bigger. The implementation is in m_execontext.c. Currently old hashtables are not freed because there is no valgrind partner of the following method: VG_(am_munmap_client) Are you going to implement VG_(am_munmap_valgrind)? Currently I do not see more improvements for m_execontext.c Greetings Christoph Bartoschek |