Forums

fmult in G.726

Started by mayank agarwal December 19, 2005
Hi all,
In the G.726 standard fmult function is taking around
24 percent mips and in this multiplication is done in
floating point format.
how can it be done in more effective way to achieve
the efficient mips on C64x simulator.
If anyone has done this before he can help me.
Regards,
Mayank

__________________________________________________



Mayank,

Are you using the c64x 'fast floating point library' -
'fastrts'??

mikedunn

--- mayank agarwal <may_ank77@may_...> wrote:

> Hi all,
> In the G.726 standard fmult function is taking
> around
> 24 percent mips and in this multiplication is done
> in
> floating point format.
> how can it be done in more effective way to achieve
> the efficient mips on C64x simulator.
> If anyone has done this before he can help me.
> Regards,
> Mayank
>
> __________________________________________________
> >
>
> c6x-unsubscribe@c6x-...




Mike-

> Are you using the c64x 'fast floating point library' -
> 'fastrts'??

What does fastrts do? Replace some floating-point operations with 32-bit fixed-point
operations?

-Jeff

> --- mayank agarwal <may_ank77@may_...> wrote:
>
> > Hi all,
> > In the G.726 standard fmult function is taking
> > around
> > 24 percent mips and in this multiplication is done
> > in
> > floating point format.
> > how can it be done in more effective way to achieve
> > the efficient mips on C64x simulator.
> > If anyone has done this before he can help me.
> > Regards,
> > Mayank



Jeff,

Go to TI's web site and search 'fastrts' for the
complete info from TI [the lib is a free download].
As I remember, the 'std rts lib' is compiled from a
common C lib for all DSPs. The c6x 'fastrts' includes
some hand-optimized asm code. You can replace the std
rts lib with fastrts OR you can use them together
since fastrts functions can be called by the 'standard
name' or the 'new fastrts name'.

I am not sure about the 'fine print'/downside to
fastrts. Some functions will show very noticeable
improvement.

mikedunn

--- Jeff Brower <jbrower@jbro...> wrote:

> Mike-
>
> > Are you using the c64x 'fast floating point
> library' -
> > 'fastrts'??
>
> What does fastrts do? Replace some floating-point
> operations with 32-bit fixed-point
> operations?
>
> -Jeff
>
> > --- mayank agarwal <may_ank77@may_...> wrote:
> >
> > > Hi all,
> > > In the G.726 standard fmult function is taking
> > > around
> > > 24 percent mips and in this multiplication is
> done
> > > in
> > > floating point format.
> > > how can it be done in more effective way to
> achieve
> > > the efficient mips on C64x simulator.
> > > If anyone has done this before he can help
> me.
> > > Regards,
> > > Mayank
>