[XviD-devel] sad() preliminary results

[James Hauxwell]

OK,

After yesterdays discussion I have rewritten the sad() routines in
intrinsics.  My preliminary findings are that there is a 50% reduction
in instruction count over the NASM hand coded version.  Obviously this
won't lead to a 50% performance increase as I expect most of the time is
going in stall cycles to the memory system.

To complete my tests I need to validate the routine with some test data
and then replace in the .dll for a comparative benchmark.

Until tomorrow!

Jim