[Debian-coldfire-devel] Progress?

[Wouter Verhelst]

On Sat, Jun 03, 2006 at 08:17:44AM +1000, Adam Conrad wrote:
> Wouter Verhelst wrote:
> > Yes, but none of those are of great importance, and can be worked
> > around. Apart from the FF1 opcode ("Find First One", an opcode to find
> > the first 1-bit in an operator), they all belong to things such as the
> > MAC/EMAC ((Enhanced) Multiply ACcumulator, for hardware-accelerated
> > vector math) or the crypto accelerator units.
> 
> These should be handled via glibc's hwcap interface (which m68k has no,
> as of yet, put to use, but will certainly be doing in the future), much
> in the same was as i386 has /usr/lib/i686/cmov, /usr/lib/i686,
> /usr/lib/i586, etc.

Sure, I absolutely agree.

> It's worth masking off these new options as "spiffy high performance
> hwcap options", because they're just that.

Well, yeah, in an m68k context at least :-)

> In the other direction, however, it's more worth having the kernel
> emulate older m68k opcodes that the ColdFire can't support, since the
> ColdFire is clearly fast enough to pull that off.

I still think that producing code which needs no emulation on either
hardware is a better way to go. Remember that apart from the above, the
coldfire ISA is a subset of the classic m68k one.

Perhaps that's something we can only figure out through benchmarking?

> (We certainly wouldn't want a 68040 trying to emulate the new whizbang
> vector math stuff, though... Eek)

How about an 030? ;-)

-- 
Fun will now commence
  -- Seven Of Nine, "Ashes to Ashes", stardate 53679.4