[Pixman] Performance of radial gradients

Jonathan Morton jonathan.morton at movial.com
Mon Feb 22 08:42:45 PST 2010

On Mon, 2010-02-22 at 18:39 +0200, Siarhei Siamashka wrote:

> The usage of sqrt is probably not a fatal performance problem.

Most desktop-style PowerPCs in the wild have a dedicated fsqrt{f}
instruction (with approximately the same expense as fdiv{f}), and
Altivec can efficiently use the Newton-Raphson method as well.

Even x87 has transcendentals covered rather well.  In fact it's one of
the few things it's genuinely good at, justifying the relatively high
expense of an FPU in the early days.

So provided there aren't too many of them, I would not consider sqrt a
huge problem - just like a divide really.

From: Jonathan Morton
      jonathan.morton at movial.com

More information about the Pixman mailing list