Altivec, -fsigned-char CFLAGS

[e-d0uble]

I'm just curious:

Isn't it foolish to add `-maltivec -mabi=altivec' to global CFLAGS on G4 ppc's (uging gcc>3.1) ?!? Only applications that are coded to use altivec in the first place will utilize them... Correct? I don't think gcc is generating these instructions from scratch? What software (GNU or otherwise) has altivec instructions in it?? I have seen none so far.. So isn't this kind of useless? Or am I completely wrong here?

Also, does `fsigned-char' always need to be added?? AFAIK, certain programs need it:

#include <stdio.h>
int main() {
char a = -3;
if(a > 0)
printf("Ooops!\n");
return 0;
}

will say Ooops! Unless compiled with -fsigned-char, but I'm certain there are cases where declaring char as signed is a performance hit.

what do you think?

Ethan

[Stummel]

Well,

there is altivec enabled software: Ogle e.g. and i think mol can use it to. Ogle runs a lot faster using altivec.
I think, but I'm not shure, that you have to compile your base system with altivec support, to enable it. But I might be wrong with that.
My first gentoo was compiled with 2.95 and O3 optimisation - and nautilus started quite fast. But now, on my gcc 3.2 system, with altivec and O3 optimisation, nautilus is a lot faster than the old one. Maybe, thats because of gcc 3.2. I don't know.

There is a benchmark programm arround - hint I think - there have also a altivec optimized versition. Try it out.

CU Stummel

[Gerk]

there are quite a few emerging pieces of software using altivec code.

mplayer, xine, lame, mol just to name a few off the top of my head

gcc also does use some small internal altive references when using these options at compile time. Setting -fsigned-char should be harmless and in fact save cycles for apps that constantly try to evaluate if its needed or not (and sadly there are a few)