Re: fold() and target-specific macros

[Andrew Haley <aph at redhat dot com>]

Roger Sayle writes:
 > 
 > Generally, I'd prefer a new flag_target_vm than reuse flag_syntax_only for
 > a slightly different purpose.

No problem there, but flag_syntax_only really means "don't generate code".

 > And in preference to that I'd prefer moving such target-specific
 > optimizations to RTL expansion if possible.
 > 
 > Thoughts?

We have two goals here: long term and short term.  Clearly, moving
vast amounts of target-dependent code isn't possible right now.  So,
I'm going to propose a two-pronged approach.

1.  Define flag_target_vm, as you suggest.  Near-automatic approval
    for patches to disable machine-specific folders when
    flag_target_vm is set.  This has the advantage of allowing
    machine-specific folders to be marked and easily grepped for.

2.  Defer machine-specific folding until some time after
    gimplification.

There's some related discussion about language-specifice folding, and
this hasn't yet come to any conclusion.  However, as we have a very
specific requirement for 1, can we do that now, please?

In the longer term, there may be many operations that we do in fold()
that simply don't make any sense with the current gcc architecture:
maybe we're just moving stuff around to no great effect.

Andrew.