------- Comment #1 from burnus at gcc dot gnu dot org 2008-03-25 23:52 -------
Thanks for testing/reducing/reporting this.
If one looks at the bounds, one sees that for negative lower bounds the output
is wrong. NAG correctly has:
2 1 0 1 -2 -3 -4
3 4 5 0 7 8 9
while gfortran wrongly has:
2 1 0 1 1 1 1
3 4 5 0 0 0 0
ila1[S.2 + -1] = lla.dim[S.2 + -1].stride >= 0 && lla.dim[S.2 + -1].ubound >=
lla.dim[S.2 + -1].lbound || lla.dim[S.2 + -1].stride < 0 && lla.dim[S.2 +
-1].ubound <= lla.dim[S.2 + -1].lbound ? (integer(kind=4)) lla.dim[S.2 +
-1].lbound : 1;
This looks ok. However, I do not understand the stride calculation:
lla.dim[3].lbound = (integer(kind=8)) nf10;
lla.dim[3].ubound = 1;
lla.dim[3].stride = 48;
so far so good, but why the following?
D.950 = (2 - (integer(kind=8)) nf10) * 48;
lla.dim[4].lbound = -2;
lla.dim[4].ubound = 7;
lla.dim[4].stride = D.950;
D.950 is (2-10)*48 = -384 ?!?!?
--
burnus at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
OtherBugsDependingO| |32834
nThis| |
Status|UNCONFIRMED |NEW
Ever Confirmed|0 |1
Keywords| |wrong-code
Last reconfirmed|0000-00-00 00:00:00 |2008-03-25 23:52:07
date| |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35698
