On Sun, 10 Nov 2002 11:26:25 -0800 (PST)
Linus Torvalds <[EMAIL PROTECTED]> wrote:
>
> On Sun, 10 Nov 2002, Felix K�hling wrote:
> >
> > not too long ago I read on this list that you have to be extra careful
> > about using floating point in the kernel. Is the same true for MMX? If
> > so, would it be dangerous to compile a kernel module with -march=athlon?
> > AFAIK gcc-3.2 generates MMX instructions if the target CPU supports
> > them.
>
> If so, the kernel will crash, eventually. However, I thought that gcc
> wouldn't use MMX on its own (ie that you have to use the MMX data types to
> get MMX operations). Not true?
I saw this in the context of a gcc-3.2 bug where it mixed FPU and MMX
instructions. It used "movd" to move some value temporarily to %mm0.
Adding "-mno-mmx -mno-3dnow" to the commandline disabled the use of
movd.
I thought the gcc bug was that it mixed MMX with FPU instructions. But
it seems the bug was that it used MMX instructions at all. In the 2.4.19
kernel sources I added "-S" to the CFLAGS in arch/i386/Makefile and ran
"make clean" and "make -k bzImage modules". Then I grepped for "%mm" in
all .o files. The only one I found was arch/i386/lib/mmx.o (explicit MMX
inline assembly).
> Many other architectures use a compiler flag like "-msoft-float" or
> "-mno-fpu" to explicitly tell gcc that it cannot use the FP unit. If gcc
> starts using MMX on its own, we'll have to find the switch that tells it
> not to do that. You might want to ask some gcc people about it..
>
> Linus
Felix
__\|/__ ___ ___ ___
__Tsch��_______\_6 6_/___/__ \___/__ \___/___\___You can do anything,___
_____Felix_______\�/\ \_____\ \_____\ \______U___just not everything____
[EMAIL PROTECTED] >o<__/ \___/ \___/ at the same time!
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Dri-devel mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/dri-devel
