This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: [FORTRAN] Improved expansion of simple WHERE statements
On Sat, Feb 04, 2006 at 07:02:37PM -0800, Steve Kargl wrote:
> On Sat, Feb 04, 2006 at 03:43:25PM -0800, Steve Kargl wrote:
> > On Sat, Feb 04, 2006 at 03:35:53PM -0800, Jerry DeLisle wrote:
> >
> > I'm reading the patch now. It should have no effect and show
> > no bugs from NIST or LAPACK. NIST and LAPACK are Fortran 77
> > codes that pre-date WHERE (and TRANSPOSE and FORALL).
> >
> > I'll note that I changed one of my codes because the use of a
> > simple WHERE caused memory usage to jump from 1GB to 2GB.
> >
>
> Well, I may have spoke too soon. I'm seeing a regression with
> FM013.f. A reduced test cast is
>
> PROGRAM FM013
> IF (ICZERO) 31270, 1270, 31270
> 1270 CONTINUE
> 1272 ASSIGN 1273 TO J
> 1273 ASSIGN 1274 TO J
> 1274 ASSIGN 1275 TO J
> GOTO 1276
> 1275 continue
> 1276 GOTO J, ( 1272, 1273, 1274, 1275 )
> 31270 IVDELE = IVDELE + 1
> END
>
> kargl[203] gfc4x -o z -w fm013.f
> kargl[204] gfc4x -o z -w -O fm013.f
> /var/tmp/ccMIfiYq.o(.text+0x8): In function `MAIN__':
> : undefined reference to `.L5'
> /var/tmp/ccMIfiYq.o(.text+0x14): In function `MAIN__':
> : undefined reference to `.L7'
> collect2: ld returned 1 exit status
>
> I think this is unrelated to Roger's patch, but I've reverted
> to a clean tree to see what happens.
>
Roger's patch appears to be ok. This regression is in a clean
trunk. I wish people committing changes to the middle and backend
would do regression tests with gfortran.
--
Steve