From mboxrd@z Thu Jan  1 00:00:00 1970
From: Jeffrey A Law <law@cygnus.com>
To: Joern Rennecke <amylaar@cygnus.co.uk>
Cc: rth@cygnus.com, gcc@gcc.gnu.org
Subject: Re: Your recent reload changes 
Date: Tue, 22 Aug 2000 14:31:00 -0000
Message-id: <12581.966979793@upchuck>
References: <200008222125.WAA31259@phal.cygnus.co.uk>
X-SW-Source: 2000-08/msg00455.html

  In message < 200008222125.WAA31259@phal.cygnus.co.uk >you write:
  > > Reload 3: R1_REGS, RELOAD_FOR_INPUT_ADDRESS (opnum = 1), can't combine, 
  > > secondar
  > > y_reload_p
  > >         reload_reg_rtx: (reg:SI 1 %r1)
  > 
  > Looking at the reload_insi pattern of the pa, I see that it applies to all
  > destinations of all register classes, and is used with any source that is
  > not a hard register.  Couldn't you make this pattern restrictive, so that
  > it only triggers when you genuinely need a scratch register?
  > Reload is perfectly capable of dealing with intermediary registers on its
  > own.
Sometimes we need an intermediate, sometimes we need a scratch.  Find a way
to rewrite it to handle both with a more restrictive pattern and I'll happily
use it.

Not surprisingly it's been tried multiple times, and every time we've come
back to the code you see in pa.md/pa.c.  Good luck.


jeff