[patch] track register arguments in call frame info

[Andrew Pinski]

On Mar 28, 2005, at 11:36 AM, Michael Matz wrote:

>
> With active -fregparam there are two regressions on i686 which are not
> trivially fixable (having to do with the fact that regparam.c doesn't
> reset args_size at barrier like dwarf2out does, because it's CFG based 
> and
> hence doesn't see those barriers) visible only with nested functions
> without a frame pointer and --enable-checking.  But on i686 this flag
> doesn't make sense anyway (no register passing in general), so IMHO 
> that's
> not as important.  There are no other regressions.  I should mention 
> that
> this testing was done on mainline from 2005-02-11 .  The patch below is
> for current mainline and I've up to now only tested bootstrap on 
> x86-64.
> I'm going to retest it again on current mainline on the above seven
> architectures.

Actually the point about no register passing on x86 is not fully true
any more and has not been true since 3.4.0 since unit-at-a-time can
change the ABI for static functions and can get even worse with
whole program mode which can turn non static functions to use register
for passing arguments.

-- Pinski