Subject: Re: gcc optimization with strength-reduce
To: None <michaelv@HeadCandy.com>
From: Niklas Hallqvist <email@example.com>
Date: 04/03/1996 17:58:51
>>>>> "Michael" == Michael L VanLoon <-- HeadCandy.com" <michaelv@HeadCandy.com>> writes:
Michael> Was the bug ever fixed that caused gcc 2.7.2 to lose when
Michael> building with -O2 or higher and not using
Michael> -fno-strength-reduce. I. e. can I safely remove
Michael> -fno-strength-reduce from my makefiles if I'm still using -O6
Michael> (not sure if anything even happens above 2 or 3... 6 is just
Michael> the traditional "turbo-build" option :-).
There has been a great debate of this over in GCC developer-land
whether or not a GCC 2.7.3 should be released fixing this. The
consensus seemed to be that it should be done, but it was not clear
how. There is a patch that apparently fixes the problem for the known
cases, but Kenner seemed to think it was not "small enough" to be safe
for a bug-patch release. Actually this bug has been in GCC for about
five years so it's not like it's common that it actually shows up.
However both our kernel and the Linux kernel triggers it so it's
severe enough to warrant a new release. This was a fact that wasn't
very clear to Kenner, the main maintainer of GCC, as bug reports from
these environments gets filtered via the compiler providers, and
doesn't hit gcc-bug directly. As this bug never got represented in
many messages there, the impression he got was that it wasn't very
important to fix.
Anyway, unless you want a patch for your local gcc to try out, you'll
have to await 2.7.3. I guess I can hunt down the proposed patch, if
you want it.
Niklas Hallqvist Phone: +46-(0)31-40 75 00 Home: +46-(0)31-41 93 95
Applitron Datasystem Fax: +46-(0)31-83 39 50 Home: +46-(0)31-41 93 96
Molndalsvagen 95 Email: firstname.lastname@example.org GSM: +46-(0)70-714 10 35
S-412 63 GOTEBORG WWW: Here
Sweden IRC: niklas (#NetBSD)