--- Comment #29 from dominiq at lps dot ens dot fr 2009-12-04 14:24 ---
AFAICT the inner loop of PR42108 is still unrolled.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42131
--- Comment #28 from jb at gcc dot gnu dot org 2009-12-02 09:23 ---
Subject: Bug 42131
Author: jb
Date: Wed Dec 2 09:22:50 2009
New Revision: 154900
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=154900
Log:
Typo in ChangeLog entry for PR fortran/42131
Modified:
trunk/gcc/
--- Comment #27 from jb at gcc dot gnu dot org 2009-12-01 18:32 ---
Subject: Bug 42131
Author: jb
Date: Tue Dec 1 18:32:37 2009
New Revision: 154876
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=154876
Log:
PR fortran/42131 Sign test using ternary operator
Modified:
trunk
--- Comment #26 from rguenther at suse dot de 2009-12-01 09:42 ---
Subject: Re: Weird translation of DO loops
On Mon, 30 Nov 2009, tkoenig at gcc dot gnu dot org wrote:
>
>
> --- Comment #22 from tkoenig at gcc dot gnu dot org 2009-11-30 19:15
> ---
> (In reply to comment
--- Comment #25 from tkoenig at gcc dot gnu dot org 2009-11-30 21:01
---
Fixed on trunk. Closing.
--
tkoenig at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #24 from tkoenig at gcc dot gnu dot org 2009-11-30 20:35
---
Subject: Bug 42131
Author: tkoenig
Date: Mon Nov 30 20:35:41 2009
New Revision: 154839
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=154839
Log:
2009-11-30 Thomas Koenig
PR fortran/42131
--- Comment #23 from jvdelisle at gcc dot gnu dot org 2009-11-30 20:19
---
Thomas, Ido not have email access at the moment.
I reviewed your patch and it is approved for trunk.
Thanks for the work.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42131
--- Comment #22 from tkoenig at gcc dot gnu dot org 2009-11-30 19:15
---
(In reply to comment #21)
> the "sign" for unsigned steps is always 1, you don't seem to account
> for unsignedness?
(Un)fortunately, there are no unsigned varaibles in Fortran.
> Note that I believe generating
--- Comment #21 from rguenther at suse dot de 2009-11-30 10:10 ---
Subject: Re: Weird translation of DO loops
On Mon, 30 Nov 2009, tkoenig at gcc dot gnu dot org wrote:
> --- Comment #20 from tkoenig at gcc dot gnu dot org 2009-11-30 07:31
> ---
> Created an attachment (id=1
--- Comment #20 from tkoenig at gcc dot gnu dot org 2009-11-30 07:31
---
Created an attachment (id=19182)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19182&action=view)
Patch that works for unrolling
It also passes do_3.F90.
I'll submit just in time for meeting the phase 4 dea
--- Comment #19 from tkoenig at gcc dot gnu dot org 2009-11-28 15:16
---
(In reply to comment #18)
> Well, in that case you can as well rely on twos-complement
> arithmetic and avoid all the overflow issues?
This is difficult without if statements or MAX_EXPR and MIN_EXPR,
because I n
--- Comment #18 from rguenther at suse dot de 2009-11-27 09:48 ---
Subject: Re: Weird translation of DO loops
On Thu, 26 Nov 2009, tkoenig at gcc dot gnu dot org wrote:
> --- Comment #15 from tkoenig at gcc dot gnu dot org 2009-11-26 23:43
> ---
> (In reply to comment #14)
>
--- Comment #17 from rguenther at suse dot de 2009-11-27 09:47 ---
Subject: Re: Weird translation of DO loops
On Thu, 26 Nov 2009, tkoenig at gcc dot gnu dot org wrote:
> --- Comment #13 from tkoenig at gcc dot gnu dot org 2009-11-26 21:56
> ---
> Created an attachment (id=1
--- Comment #16 from burnus at gcc dot gnu dot org 2009-11-27 08:29 ---
(In reply to comment #12)
> > Any tricks I have missed?
> Yes - we could provide for loop versioning in the front end.
[...]
> ELSE
>ABORT "M3 MUST NOT BE ZERO"
> ENDIF
Just for completeness a zeroness check is
--- Comment #15 from tkoenig at gcc dot gnu dot org 2009-11-26 23:43
---
(In reply to comment #14)
> Looks much better than the current situation. Is there a valid
> reason for the character(kind=4) casts? I would have thought
> that this should be a integer(kind=4).
The casts are g
--- Comment #14 from kargl at gcc dot gnu dot org 2009-11-26 22:07 ---
(In reply to comment #13)
> }
> countm1.1 = (((character(kind=4)) D.1338 - (character(kind=4)) D.1337) *
> (character(kind=4)) step_sign.2) / (character(kind=4)) (step_sign.2 * D.1339);
>
> implementing the
--- Comment #13 from tkoenig at gcc dot gnu dot org 2009-11-26 21:56
---
Created an attachment (id=19159)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19159&action=view)
patch that implements the multiplication idea
This generates
if (D.1339 > 0)
{
if (D.1338 <
--- Comment #12 from toon at moene dot org 2009-11-24 18:03 ---
> Any tricks I have missed?
Yes - we could provide for loop versioning in the front end.
I.e., generate code like:
IF (M3 > 0) THEN
... compute loop count ...
... perform loop ...
ELSE IF (M3 < 0) THEN
... comput
--- Comment #11 from tkoenig at gcc dot gnu dot org 2009-11-23 21:48
---
Created an attachment (id=19104)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19104&action=view)
another proposed patch
Here's another proposed patch, but there is a problem with it.
If we calculate (m2 -
--- Comment #10 from burnus at gcc dot gnu dot org 2009-11-22 19:04 ---
"Do loop with HUGE stepsize":
http://groups.google.com/group/comp.lang.fortran/browse_thread/thread/63348a0461ccf3e0
> Current status is that g77 prints 0 and gfortran prints 1.
To add: NAG f95, g95, sunf95, and ifo
--- Comment #9 from toon at moene dot org 2009-11-22 10:20 ---
Richard wondered about this earlier:
>countm1.1 = (character(kind=4)) (D.1337 - D.1336) / (character(kind=4))
D.1338;
but perhaps it's better to asked explicitly:
Where does the "(character(kind=4))" comes from in this
--- Comment #8 from tkoenig at gcc dot gnu dot org 2009-11-21 23:42 ---
Subject: Re: Weird translation of DO loops
On Sat, 2009-11-21 at 23:23 +, rguenther at suse dot de wrote:
> That's better.
Not yet correct, though, this causes regressions for
program main
charac
--- Comment #7 from rguenther at suse dot de 2009-11-21 23:23 ---
Subject: Re: Weird translation of DO loops
On Sat, 21 Nov 2009, tkoenig at gcc dot gnu dot org wrote:
> --- Comment #6 from tkoenig at gcc dot gnu dot org 2009-11-21 23:07
> ---
> Created an attachment (id=190
--- Comment #6 from tkoenig at gcc dot gnu dot org 2009-11-21 23:07 ---
Created an attachment (id=19076)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19076&action=view)
proposed patch
This patch generates
D.1336 = m1;
D.1337 = m2;
D.1338 = m3;
i = D.1336;
if
--- Comment #5 from toon at moene dot org 2009-11-21 21:40 ---
> The middle-end prefers do { } while () loop style so it knows the loop is
> always executed.
And the Fortran Standard describes the loops being built (by compilers) just
so:
1. First you determine what is m1, m2, m3
2. T
--- Comment #4 from rguenth at gcc dot gnu dot org 2009-11-21 19:24 ---
The middle-end prefers do { } while () loop style so it knows the loop is
always executed. It even tries to transform other loop forms into this by
copying the loop header. So if the FE already can cheaply produce
--- Comment #3 from tkoenig at gcc dot gnu dot org 2009-11-21 18:31 ---
(In reply to comment #2)
> Sorry, Steve - my mistake.
>
> The original message should have been:
>
> To illustrate this with a simple example:
>
> DO I = M1, M2, M3
>B(I) = A(I)
> ENDDO
>
> would be most easi
--- Comment #2 from toon at moene dot org 2009-11-21 17:33 ---
Sorry, Steve - my mistake.
The original message should have been:
To illustrate this with a simple example:
DO I = M1, M2, M3
B(I) = A(I)
ENDDO
would be most easily, and straightforwardly, implemented as follows:
--- Comment #1 from kargl at gcc dot gnu dot org 2009-11-21 16:26 ---
(In reply to comment #0)
> To illustrate this with a simple example:
>
> DO I = M1, M2, M3
>B(I) = A(I)
> ENDDO
>
> would be most easily, and atraightforwardly, implemented as follows:
>
> IF (M3 > 0 .AND.
29 matches
Mail list logo
