Applying FUNCTION_ARG_PADDING to register arguments

[Joern Rennecke]

> The first problem is that MUST_PASS_IN_STACK macro in expr.h forces BLKmode
> arguments to be passed in the stack if the padding is opposite the natural
> direction.  This is wrong for irix6.  In general, MUST_PASS_IN_STACK should
> not be overriding FUNCTION_ARG if FUNCTION_ARG says that something is only
> passed in registers.  This seems to be a relic left over from GCC 1.x.  I
> suspect that other ports may also be generating incorrect (non ABI compliant)
> code because of this.  For instance, I noticed that the standard MIPS ABI
> support suffers from this bug also.  However, at this point, it seems unsafe to
> just delete this macro, because that will introduce unexpected and unnecessary
> incompatibilites for targets where gcc is the only compiler.  Providing a
> way for tm.h files to override MUST_PASS_IN_STACK seems the safest way for
> now.  For now I have just modified expr.h but I don't think this is safe.

Actually this modification would be sufficient to solve the problem I've
originally set out to fix.

Thus, I think the way to go is to change MUST_PASS_IN_STACK in expr.h not
to take padding into account (like in your patch), and make the definition
conditional on #ifndef MUST_PASS_IN_STACK.  The targets that actually want
this double argument passing can then define MUST_PASS_IN_STACK to be
identical to the old definition.

I have gone through the 35 configurations in egcs, and found that
for most of them, this change would have no effect, due to their
endianness / definition of FUNCTION_ARG_PADDING, definition of FUNCTION_ARG,
or in the case of dsp16xx, BITS_PER_UNIT being ideintical to PARM_BOUNDARY.
For the SH, the proposed change would be a bugfix.  So that leaves
nine targets where I'm not sure if we'd like to define MUST_PASS_IN_STACK
in the old way:

arc
arm
h8300 (problem only for TARGET_QUICKCALL, but still...)
m32r
mips
pa
pyr
romp
sparc

If we can't resolve the issue for one or more of these targets in a timely
manner, we can just put in the old definition of MUST_PASS_IN_STACK into
the respective tm.h file, with a comment that it is unknown if that is
actually intentional, but kept for backward compatibility.