[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: execution speed in *oct files

From: John W. Eaton
Subject: Re: execution speed in *oct files
Date: Tue, 8 Jun 1999 10:10:58 -0500 (CDT)

On  8-Jun-1999, Michael Hanke <address@hidden> wrote:

| I am not very experienced with gcc compiler issues. I am merely using
| compilers. So my question may be stupid. I had some rumour that
| certain gcc versions for some hardware platforms may generate buggy
| code if this switch is *not* used. So I am using it everytimes with a
| possible loss of performance. So my question is: In which
| gcc/egcs/pgcc versions is strength-reduce safe?

Searching DejaNews for

  gcc strength-reduce fixed

turned up a lot of hits.  Among them, there are some messages from Joe
Buck (maintainer of the g++ FAQ and long-time subscriber to gcc and
egcs development mailing lists, I believe).  Here is one of them:

  Marcus Sundberg <address@hidden> wrote:
  >Could someone give me the status of the strength-reduce feature
  >of GCC on ix86 processors, or point me to this information.

  2.7.2 had a bug in strength reduction that on ix86 platforms.  This was
  fixed in

  >Ie, on which versions is -fno-strenght-reduce needed to produce
  >correct code?

  Even with 2.7.2, tweaking the bug wasn't *that* common, but with
  and later there is no reason to use -fno-strength-reduce.

  >On which versions is -fno-strength-reduce the default, and on
  >which is strenght-reduce disabled?

  No version disables the flag -- it is turned on when you turn on

  >Is any versions of egcs/pgcc affected?

  egcs: No. pgcc: possibly the versions from several years ago were affected.

  Turning off strength reduction may substantially slow down your code.
  If you've gotten into the habit of saying -fno-strength-reduce out of
  superstition, you can stop now. :-)

  -- Joe Buck
     work: address@hidden, otherwise address@hidden or address@hidden


Octave is freely available under the terms of the GNU GPL.  To ensure
that development continues, see
Instructions for unsubscribing:

reply via email to

[Prev in Thread] Current Thread [Next in Thread]