[Ffmpeg-devel] [RFC] wmadec init float to fixedpoint conversion

[Rich Felker]

On Fri, Mar 03, 2006 at 01:59:39PM +0100, Michael Niedermayer wrote:
> Hi
> 
> On Fri, Mar 03, 2006 at 12:36:28PM +0000, Benjamin Larsson wrote:
> > Hi, attached is a WIP for removing all the floats in the init routines
> > for wma. What I'm aiming for is that the code maintains it's readable
> > way and that all floats are removed. Currently I'm keeping an option to
> > still use floats through a define. Is that needed ?
> 
> for the init code? if your integer code is accurate then no
> for the actual wma decode? certainly, many modern CPUs have very fast
> FPUs which are much faster then integer instructions and all the shifts
> and other extra maintaince code you need for integers

I doubt the performance matters for init code, unless there's a CPU
fault for every single floating point operation in which case the
float version would be very slow. I'm in favor of removing as much
float code as possible, but that's just MHO.

Rich