This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: regclass.c: scan_one_insn special casing
On Mon, 2004-03-29 at 07:14, David Edelsohn wrote:
> >>>>> Michael Matz writes:
>
> Michael> If anything the place of that code seems to be a bit strange, at least
> Michael> from current perspective. After these many years we might have a better
> Michael> place. OTOH it has to examine the insn constraints and the earliest point
> Michael> GCC is doing this is in regclass. Clearly it would be even better if
> Michael> reload wouldn't do the stupid thing which this code is supposed to help.
>
> This discussion thread appears to be mixing two issues: the code
> appearing in a common part of the compiler and these types of special
> cases scattered around the compiler. The diffusion of this type of code
> is a maintenance problem, but a single implementation should be a
> benefit.
Right. I thought it seemed more of a waste in scan_one_insn than
anything else. Maybe an "optimize address" type of function would be
best. Especially since this optimization isn't dependent on "optimize"
at all.
Though, there's the "don't split _this_ insn, we do that somewhere else"
weird effect, but the comment is likely different than the reason we're
splitting it in the backend.
-eric
--
Eric Christopher <echristo@redhat.com>