From: "Tim Prince" <tprince@computer.org>
To: "Jan Hubicka" <jh@suse.cz>, "Toon Moene" <toon@moene.indiv.nluug.nl>
Cc: "Joern Rennecke" <amylaar@redhat.com>, "Jan Hubicka" <jh@suse.cz>,
<gcc@gcc.gnu.org>
Subject: Re: What is acceptable for -ffast-math? (Was: associative law in combine)
Date: Sat, 28 Jul 2001 23:04:00 -0000 [thread overview]
Message-ID: <004501c117f4$84a48c80$9865fea9@timayum4srqln4> (raw)
In-Reply-To: <20010718104420.E29584@atrey.karlin.mff.cuni.cz>
----- Original Message -----
From: "Jan Hubicka" <jh@suse.cz>
To: "Toon Moene" <toon@moene.indiv.nluug.nl>
Cc: "Joern Rennecke" <amylaar@redhat.com>; "Jan Hubicka" <jh@suse.cz>;
<gcc@gcc.gnu.org>
Sent: Wednesday, July 18, 2001 1:44 AM
Subject: What is acceptable for -ffast-math? (Was: associative law in
combine)
> > Joern Rennecke wrote:
> >
> > > > > Jan Hubicka wrote:
> >
> > > > OK, so to speak loud - where is the proper place to convert
> > > > a/b/c to a/(b*c) at tree level. fold-const or some other?
> > >
> > > Only if b and c are constants, the operations are floating point,
and
> > > b can be multiplied with c without loss of precision or overflow.
> > > Or if b and/or c is a power of two, and b can be multiplied with c
without
> > > overflow.
> >
> > Joern is right, Jan. One can argue about the loss of precision
(under
> > unsafe math optimisations), but not the overflow. I overlooked that
> > issue in my reply to you. Because overflow can only be determined
at
> > compile time with constants, this conversion cannot be right for
> > variables.
> OK, I believe, that our concept of unsafe_math_optimizations allows
such
> transformation, but I see, that the line between acceptable
> unsafe_math_optimization and unacceptable one is pretty fragine.
>
> In case -ffast-math is not enought to ask for such transformation,
> we probably should invent switch for that. This change (plus the
> other changes to avoid divisions) seems to play important role
> on optimizing some FP software (due to extreme cost of fp division).
>
> Honza
This optimization is specifically permitted for appropriate data types
by the Fortran standard, with the reservation that parentheses employed
to prevent re-association must be observed. As gcc is unable to
distinguish between the expressions (a/b)/c and a/b/c, this optimization
would violate the standard in the former case, so is undesirable for
enabling under -ffast-math.
When Honza and I first discussed this, and also the transformation
for(i=0;i<n;++i)
a[i]=b[i]/c;
to
for(i=0;i<n;++i)
a[i]=b[i]*(1/c);
I mentioned that these would put too many risks in the -ffast-math
category. I would like to see -ffast-math permit only those
optimizations outside of IEEE compliance which are generally expected
(e.g. according to Fortran standard).
For gcc-3.1, I would like to see an additional switch to permit
optimizations such as the ones on floating point division mentioned
above. I don't care about the name, but it should be more mnemonic than
the ones used by commercial compilers; something like -ffast-div.
next prev parent reply other threads:[~2001-07-28 23:04 UTC|newest]
Thread overview: 119+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <20010716224423.F29145@atrey.karlin.mff.cuni.cz>
[not found] ` <3B547DF4.7F39BF61@moene.indiv.nluug.nl>
[not found] ` <20010717200629.E5166@atrey.karlin.mff.cuni.cz>
[not found] ` <3B5489F1.54DE59CE@moene.indiv.nluug.nl>
[not found] ` <20010717205837.I5166@atrey.karlin.mff.cuni.cz>
[not found] ` <3B548CBF.DDD59F97@moene.indiv.nluug.nl>
[not found] ` <20010717211516.N5166@atrey.karlin.mff.cuni.cz>
[not found] ` <3B548FAA.E174E765@moene.indiv.nluug.nl>
2001-07-17 12:23 ` associative law in combine Jan Hubicka
2001-07-17 15:41 ` Dima Volodin
2001-07-17 15:59 ` Joern Rennecke
2001-07-18 1:01 ` Toon Moene
2001-07-18 1:47 ` What is acceptable for -ffast-math? (Was: associative law in combine) Jan Hubicka
2001-07-28 23:04 ` Tim Prince [this message]
2001-07-29 6:33 ` Jan Hubicka
2001-07-29 10:18 ` Tim Prince
2001-07-29 10:26 ` Jan Hubicka
2001-07-29 12:11 ` Tim Prince
2001-07-29 12:17 ` Jan Hubicka
2001-07-29 10:50 ` Linus Torvalds
2001-07-18 11:30 ` associative law in combine Joern Rennecke
2001-07-29 12:52 * Re: What is acceptable for -ffast-math? (Was: associative lawin combine) Linus Torvalds
2001-07-29 14:03 ` What is acceptable for -ffast-math? (Was: associative law incombine) Stephen L Moshier
2001-07-29 21:17 ` What is acceptable for -ffast-math? (Was: associative law in combine) Fergus Henderson
2001-07-30 0:23 ` Gabriel Dos Reis
2001-07-29 14:22 dewar
2001-07-29 21:33 What is acceptable for -ffast-math? (Was: associative law incombine) Linus Torvalds
2001-07-30 14:43 ` What is acceptable for -ffast-math? (Was: associative law in combine) Alexandre Oliva
2001-07-30 15:45 ` Neil Booth
2001-07-30 16:03 ` Alexandre Oliva
2001-07-30 16:11 ` Neil Booth
2001-07-30 16:28 ` Alexandre Oliva
2001-07-30 19:08 ` Joern Rennecke
2001-07-30 19:22 ` Alexandre Oliva
2001-07-30 19:29 ` Gabriel Dos Reis
2001-07-30 19:34 ` Alexandre Oliva
2001-07-30 19:54 ` Gabriel Dos Reis
2001-07-30 19:27 ` Gabriel Dos Reis
2001-07-30 5:57 dewar
2001-07-30 6:00 dewar
2001-07-30 13:08 ` Toon Moene
2001-07-30 6:01 dewar
2001-07-30 6:53 ` Tim Hollebeek
2001-07-30 6:14 dewar
2001-07-30 8:30 ` Kevin Handy
2001-07-30 8:59 mike stump
2001-07-30 11:37 What is acceptable for -ffast-math? (Was: associative law incombine) Linus Torvalds
2001-07-30 11:53 ` What is acceptable for -ffast-math? (Was: associative law in combine) Gabriel Dos Reis
2001-07-30 18:40 ` Olivier Galibert
2001-07-30 19:06 ` Gabriel Dos Reis
2001-07-31 1:35 ` Linus Torvalds
2001-07-31 2:04 ` Gabriel Dos Reis
2001-07-31 2:35 ` Olivier Galibert
2001-07-31 2:58 ` Gabriel Dos Reis
2001-07-31 18:10 ` Linus Torvalds
2001-07-30 11:52 dewar
2001-07-30 12:26 dewar
2001-07-30 13:10 dewar
2001-07-30 15:29 dewar
2001-07-30 15:39 ` Toon Moene
2001-07-30 16:11 dewar
2001-07-30 16:29 ` Alexandre Oliva
2001-07-31 8:13 ` Kevin Handy
2001-07-30 18:00 dewar
2001-07-30 18:25 ` Joe Buck
2001-07-30 18:02 dewar
2001-07-30 18:08 dewar
2001-07-30 18:38 dewar
2001-07-30 18:39 dewar
2001-07-30 19:46 dewar
2001-07-30 20:00 ` Gabriel Dos Reis
2001-07-30 20:20 ` Alexandre Oliva
2001-07-30 20:25 ` Gabriel Dos Reis
2001-07-30 20:54 dewar
2001-07-30 21:11 ` Gabriel Dos Reis
2001-07-30 21:39 ` Joern Rennecke
2001-07-30 21:13 dewar
2001-07-30 21:34 ` Gabriel Dos Reis
2001-07-30 21:43 ` Joern Rennecke
2001-07-30 21:53 ` Gabriel Dos Reis
2001-08-03 7:12 ` Nick Ing-Simmons
2001-08-01 8:55 ` Theodore Papadopoulo
2001-08-01 9:15 ` Gabriel Dos Reis
2001-08-01 11:21 ` Theodore Papadopoulo
2001-08-01 11:44 ` Gabriel Dos Reis
2001-08-01 9:24 ` Tim Hollebeek
2001-08-01 9:54 ` What is acceptable for -ffast-math? (Was: associative law incombine) Linus Torvalds
2001-08-01 10:26 ` Gabriel Dos_Reis
2001-08-01 11:13 ` What is acceptable for -ffast-math? (Was: associative law in combine) Alexandre Oliva
2001-08-01 11:36 ` Gabriel Dos Reis
2001-08-01 12:07 ` Alexandre Oliva
2001-08-01 13:21 ` Gabriel Dos_Reis
2001-08-01 14:20 ` Toon Moene
2001-07-31 7:26 dewar
2001-07-31 15:57 ` Toon Moene
2001-07-31 21:55 ` Tim Prince
2001-08-03 6:12 ` Per Abrahamsen
2001-07-31 7:59 mike stump
2001-07-31 8:19 mike stump
2001-07-31 8:35 dewar
2001-07-31 8:36 mike stump
2001-07-31 8:36 dewar
2001-07-31 8:37 dewar
2001-07-31 9:22 mike stump
2001-07-31 16:38 dewar
2001-07-31 17:37 dewar
2001-07-31 18:12 What is acceptable for -ffast-math? (Was: associative law incombine) Linus Torvalds
2001-07-31 20:55 ` What is acceptable for -ffast-math? (Was: associative law in combine) Gabriel Dos Reis
2001-07-31 18:15 dewar
2001-07-31 18:20 dewar
2001-07-31 18:50 ` Joern Rennecke
2001-07-31 21:27 ` Tim Prince
2001-07-31 18:23 dewar
2001-07-31 19:10 dewar
2001-08-01 3:02 Vincent Penquerc'h
2001-08-01 6:04 dewar
2001-08-01 6:48 ` Vincent Penquerc'h
2001-08-03 0:46 ` Eric W. Biederman
2001-08-01 6:52 dewar
2001-08-01 9:58 What is acceptable for -ffast-math? (Was: associative law incombine) Gabriel Dos_Reis
2001-08-01 10:08 ` Wolfgang Bangerth
2001-08-01 11:12 ` Gabriel Dos_Reis
2001-08-01 11:27 ` What is acceptable for -ffast-math? (Was: associative law in combine) Theodore Papadopoulo
2001-08-01 11:47 ` Gabriel Dos_Reis
2001-08-03 7:32 ` Nick Ing-Simmons
2001-08-03 6:01 ` Per Abrahamsen
2001-08-01 9:59 dewar
2001-08-01 10:04 dewar
2001-08-01 10:28 ` Gabriel Dos_Reis
2001-08-01 10:05 dewar
2001-08-01 10:13 dewar
2001-08-01 10:38 dewar
2001-08-01 10:39 dewar
2001-08-01 12:06 Phil Edwards
2001-08-01 19:04 Carlo Wood
2001-08-02 3:37 Vincent Penquerc'h
2001-08-03 14:48 dewar
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='004501c117f4$84a48c80$9865fea9@timayum4srqln4' \
--to=tprince@computer.org \
--cc=amylaar@redhat.com \
--cc=gcc@gcc.gnu.org \
--cc=jh@suse.cz \
--cc=toon@moene.indiv.nluug.nl \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).