This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Use of register 17 on x86
- To: Richard Kenner <kenner at vlsi1 dot ultra dot nyu dot edu>
- Subject: Re: Use of register 17 on x86
- From: Richard Henderson <rth at redhat dot com>
- Date: Sun, 18 Feb 2001 14:28:38 -0800
- Cc: gcc at gcc dot gnu dot org
- References: <10102181259.AA11911@vlsi1.ultra.nyu.edu>
On Sun, Feb 18, 2001 at 07:59:29AM -0500, Richard Kenner wrote:
> There is a choice of using a SCRATCH with that register class or the
> explicit register 17. If the former, gcse can move such computation, but
> we may try to move things using that class out of loops and have to do
> expensive moves in and out of CC.
>
> Has any thought been given to this tradeoff?
Yes. Moves into and out of flags are expensive enough that we
would prefer to avoid them at almost any cost.
> Also, why isn't x86 using CC0 anymore? That would seem to be the most
> efficient, both in terms of generated code and compilation speed.
Because you can't get good representation of the architecture
that way. It's is impossible to represent a comparison followed
by multiple cmove instructions. Given the improvements in code
generation with the new backend, I think that gives lie to CC0
being more efficient in terms of generated code.
r~