This is the mail archive of the
mailing list for the GCC project.
Re: Reload patch v4
- To: Bill Currie <bcurrie at tssc dot co dot nz>
- Subject: Re: Reload patch v4
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Thu, 10 Sep 1998 15:24:49 -0600
- cc: Bernd Schmidt <crux at pool dot informatik dot rwth-aachen dot de>, egcs-patches at cygnus dot com
- Reply-To: law at cygnus dot com
In message <35F83D6A.66E@tssc.co.nz>you write:
> Wouldn't this help [compiler] performance a little as well? I imagine
> any improvements would depend on access paterns (ie, separate arrays
> accessed with similar indexes or each array being scanned separatly).
Accesses into the reg structure at least tend to be sequential by reg
or nearly random. From a locality of reference standpoint it isn't
likely to help much.
It does help simplify address calculations though, and on some ports
that can be important (particularly the PA where lots computations of
address in the static store tends to force %r1 to be spilled).
The global rtl structure may help locality some. I haven't actually
tried any experiments.
> Either way, this sort of thing would definity improve readability, which
> is much more important in a compiler than raw performance.
It improves readability and makes infrastructure changes easier. For
example converting the register array to virtual arrays was basically
straightforward after we had all the register info in a single