I adjusted some parameters, and got 20% better utilization of the
pipeline. The speed is now 470-480 chains a second. The code for this
has been committed to svn 

http://reflextor.com/trac/a51/browser/ati_shared_lib

The code will only work on a full pipeline ( > 147456 chains at 288
stream cores )- and will idle if it runs out of data.

The workload is divided between CPU / GPU, the GPU only does the A5/1
rounds, while the CPU switches the round functions for the GPU when
chains reach the hard coded distinguishing point (15 bits)

 Frank




_______________________________________________
A51 mailing list
[email protected]
http://lists.lists.reflextor.com/cgi-bin/mailman/listinfo/a51

Reply via email to