Re: [Aften-devel] Aften/CMake status

[Prakash P. <pr...@pu...>]

Am Samstag 21 Oktober 2006 07:13 schrieb Justin Ruggles:

> > Known issues:
> > - You need to pass proper CFLAGS to cmake to get SSE(3) code compiled
> > - It seems if SSE(3) is enabled with gcc, it als produces SSE code in t=
he
> > portable part, thus the binary is won't work on non-SSE capable CPUs.
> >
> > Both issues will hopefully be resolved in the next few days.
> >
> > Cheers,
>
> FYI...I compiled/installed gcc 4.1.1 on my system (AMD K6-2 + Slackware
> Linux 10.0).  It creates a binary which runs about 8% faster than that
> made by gcc 3.4.4.  Also, it does compile the SSE code when I pass in
> -msse and -msse3, although I still get the illegal instruction error
> when trying to execute on my k6-2.

Could you try again (using both compiler, pass via CC=3D flag to cmake)? Fl=
ags=20
for gcc should now be auto-detected and used specifically for the sse(3)=20
files only.

I am not sure how to set flags for msvc and icc to build portable code with=
=20
sse. Perhaps they do by default?

A useful feature could perhaps be that the aften lib caller would be able t=
o=20
ask which SIMD optimized routines are to be used. Do you have a good idea f=
or=20
the API (taking into account: possible GUI use, various languages, future u=
se=20
ofSIMD in other parts beside mdct?) My simple idea would be passing of an=20
array of strcuts with element string of the routien name and enum value for=
=20
used SIMD instruction set and possible simd instruction sets. I think anoth=
er=20
configurable option should be, that the user should be able to select a=20
specific simd instruction set, if he doesn't want aften's autoselection.

Cheers,
=2D-=20
(=B0=3D                 =3D=B0)
//\ Prakash Punnoor /\\
V_/                 \_V