LSB1.1: /proc/cpuinfo

[Stuart Anderson]

On Fri, 4 Jan 2002, Alan Cox wrote:

> > > What do apps actually care about
> > > 	-	MMX
> > > 	-	Athlon MMX2
> > > 	-	SSE
> > > 	-	SSE2
> >
> > Image manipulation programs need a way to determine the most efficient
> > algorithm/instructions to use.
>
> Which is down to MMX, MMX2, SSE, SSE2. Take a look at real existing apps
> like mplayer, mp1e and xine.

Are you saying that these are all that is needed, and that this problem has
already been solved?

I've seen comments that indicate that depending on SIGILL and SIGFPE to
determine that something isn't present may not be fool proof. Also, it takes
as much or more code to set up and test this as it would to read /proc/cpuinfo.

> > Probably not as much here, except as how this info may be used to identify
> > erratta on a feature that is important to the program.
>
> The x86 CPU errata are kernel handled so Im not sure that is actually an
> issue.

In setup.c there are several place where feature bits are turned on or off
based model and errata info. Why should this kind of handlig be duplicated?
Also, what about furture situations? There is no gurantee that a future bug
can be handled transparently to the app, and at no performance cost.


                                Stuart

Stuart R. Anderson                               anderson at metrolink.com

Metro Link Incorporated                          South Carolina Office
5807 North Andrews Way                           129 Secret Cove Drive
Fort Lauderdale, Florida 33309                   Lexington, SC 29072
voice: 954.660.2500                              voice: 803.951.3630
http://www.metrolink.com/                        XFree86 Core Team