split global and reload

[Zack Weinberg]

On Tue, Apr 11, 2000 at 04:05:39PM -0700, Richard Henderson wrote:
> On Sun, Apr 09, 2000 at 05:09:40PM -0700, Zack Weinberg wrote:
> > This patch separates global register allocation from reload.  That
> > means they are reported separately in the -Q timing summary, and get
> > separate RTL dump files.  No code generation should change; it simply
> > makes it easier to determine which of the two is misbehaving.
> 
> This doesn't make sense at all.  We call back into global
> during reload.  See retry_global_alloc.

Yes, I know.  That's why the patch keeps global's data structures
(reg_allocno[], etc.) around until reload is done with them.

The callbacks don't do nearly as much work as global itself.  It
seems reasonable to me to treat those operations as part of reload.

zw