Committed, MMIX: minor cleanups. Switch to TARGET_CPU_CPP_BUILTINS.

[Hans-Peter Nilsson]

On Mon, 10 Jun 2002, Richard Henderson wrote:
> On Mon, Jun 10, 2002 at 06:27:18PM -0400, Hans-Peter Nilsson wrote:
> > Minor cleanup; remove code that was never executed or was
> > duplicated.  Also switches to TARGET_CPU_CPP_BUILTINS, though
> > unfortunately it can't be used to detect contradicting machine
> > options.  I see other ports drop such detection, so I figured oh
> > well.
>
> If they're truely contradictory, rather than opposites being
> overridden, then detect this in override_options.

I used contradictory as a synonym to opposite here, but I don't
think they should override each other.  Should -mbig-endian
override -mlittle-endian, or should it be complained about?  I
think only -mfoo and -mno-foo should be allowed to silently
override each other.

There are two ABI:s for MMIX: one derived from Knuth's MMIX
texts and one "classic" I dreamt up, avoiding register stack
registers for parameters; a few percent faster.  Specifying both
on a command line would be contradictory.  They use the same
target_flags bit; perhaps I should now instead use different
target_flags bits to detect when both are specified.  Better
grab those bits while they're plenty. :-)  However, it's not
exactly something to get sleepless over.

>  Otherwise
> to a large extent we're fixing bugs.

I saw TARGET_CPU_CPP_BUILTINS migration where contradicting
-m-combinations were previously detected using CPP_SPEC, and
thought the detection and %e message was just dropped.  Maybe
all those tests *are* migrated to the respective
OVERRRIDE_OPTIONS?  Ah, no; for example it seems there's no
"-mbig-endian and -mlittle-endian may not be used together"
after arm/arm.h:1.146, I don't see the corresponding tweak in
arm.c.

brgds, H-P