http://blog.cberger.net/2008/09/02/opengtl-0-9-5-and-darkroom-1-3/ What I do, where I live, what I think. Wed, 16 Jun 2010 13:53:02 +0000 http://wordpress.org/?v=2.9.2 hourly 1 http://blog.cberger.net/2008/09/02/opengtl-0-9-5-and-darkroom-1-3/comment-page-1/#comment-170 Hans Tue, 02 Sep 2008 23:27:00 +0000 http://cyrilleberger.wordpress.com/2008/09/02/opengtl-0-9-5-and-darkroom-1-3#comment-170 Replace (...)/3.0 with (...)*(1.0/3.0), so the compiler can replace the division with a multiply.</></>A division has a latency of 24 cycles and - more importantly - isn't pipelined on most machines.</></>Stuff like this adds up.</></>Btw, the impact of extra memory loads is overrated. i386 only has 8 registers, so memory-access has evolved to being very good. Replace (…)/3.0 with (…)*(1.0/3.0), so the compiler can replace the division with a multiply.A division has a latency of 24 cycles and – more importantly – isn’t pipelined on most machines.Stuff like this adds up.Btw, the impact of extra memory loads is overrated. i386 only has 8 registers, so memory-access has evolved to being very good.

]]>