public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug fortran/45676] New: Move array assignments out of loop
@ 2010-09-15 5:44 tkoenig at gcc dot gnu dot org
2010-09-15 6:40 ` [Bug fortran/45676] " ubizjak at gmail dot com
2010-09-15 9:47 ` rguenth at gcc dot gnu dot org
0 siblings, 2 replies; 3+ messages in thread
From: tkoenig at gcc dot gnu dot org @ 2010-09-15 5:44 UTC (permalink / raw)
To: gcc-bugs
Consider
program main
integer :: a(100), b(100), c(100)
a = 2
do i=1,10
b = a
c = 1/b
print *,c
end do
end program main
The statements
b = a
c = 1/b
could be moved out of the loop. (The second one still has the
problem that we currently don't know that print doesn't read
values). Currently, this doesn't happen, as a dump shows:
<bb 4>:
# ivtmp.13_40 = PHI <10(3), ivtmp.13_39(6)>
MEM[(c_char * {ref-all})&b] = MEM[(c_char * {ref-all})&a];
<bb 5>:
# ivtmp.24_8 = PHI <ivtmp.24_45(5), 0(4)>
D.1598_11 = MEM[symbol: b, index: ivtmp.24_8, offset: 0B];
D.1599_12 = 1 / D.1598_11;
MEM[symbol: c, index: ivtmp.24_8, offset: 0B] = D.1599_12;
ivtmp.24_45 = ivtmp.24_8 + 4;
if (ivtmp.24_45 == 400)
goto <bb 6>;
else
goto <bb 5>;
<bb 6>:
dt_parm.2.common.filename = &"foo.f90"[1]{lb: 1 sz: 1};
dt_parm.2.common.line = 7;
dt_parm.2.common.flags = 128;
dt_parm.2.common.unit = 6;
_gfortran_st_write (&dt_parm.2);
parm.3.dtype = 265;
parm.3.dim[0].lbound = 1;
parm.3.dim[0].ubound = 100;
parm.3.dim[0].stride = 1;
parm.3.data = &c[0];
parm.3.offset = -1;
_gfortran_transfer_array (&dt_parm.2, &parm.3, 4, 0);
_gfortran_st_write_done (&dt_parm.2);
ivtmp.13_39 = ivtmp.13_40 - 1;
if (ivtmp.13_39 == 0)
goto <bb 7>;
else
goto <bb 4>;
<bb 7>:
return;
Not sure if this is better handled in the front or middle end, though.
--
Summary: Move array assignments out of loop
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: enhancement
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: tkoenig at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45676
^ permalink raw reply [flat|nested] 3+ messages in thread
* [Bug fortran/45676] Move array assignments out of loop
2010-09-15 5:44 [Bug fortran/45676] New: Move array assignments out of loop tkoenig at gcc dot gnu dot org
@ 2010-09-15 6:40 ` ubizjak at gmail dot com
2010-09-15 9:47 ` rguenth at gcc dot gnu dot org
1 sibling, 0 replies; 3+ messages in thread
From: ubizjak at gmail dot com @ 2010-09-15 6:40 UTC (permalink / raw)
To: gcc-bugs
------- Comment #1 from ubizjak at gmail dot com 2010-09-15 06:40 -------
Related to PR42108 and PR45223.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45676
^ permalink raw reply [flat|nested] 3+ messages in thread
* [Bug fortran/45676] Move array assignments out of loop
2010-09-15 5:44 [Bug fortran/45676] New: Move array assignments out of loop tkoenig at gcc dot gnu dot org
2010-09-15 6:40 ` [Bug fortran/45676] " ubizjak at gmail dot com
@ 2010-09-15 9:47 ` rguenth at gcc dot gnu dot org
1 sibling, 0 replies; 3+ messages in thread
From: rguenth at gcc dot gnu dot org @ 2010-09-15 9:47 UTC (permalink / raw)
To: gcc-bugs
------- Comment #2 from rguenth at gcc dot gnu dot org 2010-09-15 09:47 -------
We can't hoist invariant control flow. Also print *,c is surely thought
to be an escape point for c and thus may clobber it.
I'd rate this impossible to do for the middle-end (and generally not worth
the hassle to implement). Better fix your sources ;)
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45676
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2010-09-15 9:47 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-09-15 5:44 [Bug fortran/45676] New: Move array assignments out of loop tkoenig at gcc dot gnu dot org
2010-09-15 6:40 ` [Bug fortran/45676] " ubizjak at gmail dot com
2010-09-15 9:47 ` rguenth at gcc dot gnu dot org
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).