[ILUG] EU Rejects Patent Software Law - For Now

[Ronan Cunniffe]

On Fri, 18 Feb 2005, Brian Foster wrote:

>  p.s.  someone is running a poll about having designed code for
>       using SIMD features: Yes, I have.  but not for Intel-ish
>       SIMD the poll is asking about.  but in fairness, nothing
>       significant; just some test code for the RTOS to confirm
>       applications which did use the SIMD facilities (some of
>       which was semi-autonomous) would be correctly handled.
>       hence this probably doesn't count....?   (we _did_ have
>       an application, an MPEG decoder as it happens, that heavily
>       used the SIMD facilities (much of it hand-written in
>       assembler!), but I was not involved.)

yay!  That's *one* person so far....

It drives me bananas that there's such a divide between the latest hyped
chip technology and what run-of-the-mill code actually does.  It's like
running 16-bit code on a mainframe, doing integer anti-aliasing or
computing normals by hand when there's a quarter *billion* transistors of
T & L goodness a few inches/nanoseconds away and just a few billion years
of coding evolution...

Actually, somebody posted that most recent Linux distros were compiled for
686.  This jarred but I couldn't put my finger on it.

Is SSE state saved during function calls?  If yes, it slows down all
applications on that chip, and if no, you could easily have two threads
in a single program, where both make calls to SSE-enabled libraries and
trample on each other's data.

I could be wrong, but I think the current approach is: don't save/restore
all the exotica, don't touch it from libraries, and if the application
wants to use it, that's the application programmer's problem.

Bah.
Ronan