--- Comment #12 from tkoenig at gcc dot gnu dot org 2008-06-29 19:06
---
Subject: Bug 36341
Author: tkoenig
Date: Sun Jun 29 19:06:06 2008
New Revision: 137255
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=137255
Log:
2008-06-29 Thomas Koenig <[EMAIL PROTECTED]>
PR
--- Comment #11 from tkoenig at gcc dot gnu dot org 2008-06-28 20:37
---
(In reply to comment #9)
> Should not
>
> + mpz_init_set (f->shape[0], a->shape[1]);
>
> be
>
> + mpz_init_set (f->shape[0], a->shape[0]);
>
Yes, it should be that.
--
http://gcc.gnu.org/b
--- Comment #10 from dominiq at lps dot ens dot fr 2008-06-28 17:30 ---
The failures reported in comment #8 disappear with the fix of comment #9.
However I have gotten an Heisenbug for 32 bit mode:
FAIL: gfortran.dg/bounds_check_11.f90 -Os (test for warnings, line 18)
FAIL: gfortran
--- Comment #9 from dominiq at lps dot ens dot fr 2008-06-28 14:38 ---
Should not
+ mpz_init_set (f->shape[0], a->shape[1]);
be
+ mpz_init_set (f->shape[0], a->shape[0]);
?
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36341
--- Comment #8 from dominiq at lps dot ens dot fr 2008-06-28 12:51 ---
With the patch in comment #7 on i686-apple-darwin9, the test
gcc/testsuite/gfortran.fortran-torture/execute/intrinsic_matmul.f90
fails with
v(1:2) = matmul(a, y)
1
Error: Different shape for array assignment
--- Comment #7 from tkoenig at gcc dot gnu dot org 2008-06-28 07:54 ---
Created an attachment (id=15822)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15822&action=view)
compile-time patch
This patch works for compile-time, and passes
all *matmul* regression tests.
--
tkoenig
--- Comment #6 from tkoenig at gcc dot gnu dot org 2008-06-23 21:28 ---
We should probably set the shape for
arguments of known shape in gfc_resolve_matmul.
--
tkoenig at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #5 from tkoenig at gcc dot gnu dot org 2008-06-12 18:20 ---
This is an instance of PR 34670.
--
tkoenig at gcc dot gnu dot org changed:
What|Removed |Added
Oth
--- Comment #4 from terry at chem dot gu dot se 2008-05-28 02:00 ---
Somewhat reduced testcase that exhibits the same behaviour:
program distgeom
implicit none
real(kind=8),dimension(7,12)::B
real(kind=8),dimension(7,7)::U
real(kind=8),dimension(6,12)::dzeta
call random_number(U)
call r
--- Comment #3 from burnus at gcc dot gnu dot org 2008-05-27 16:34 ---
gfortran should do better. NAG f95 detects at compile time:
dzeta=matmul(U,B)
Error: Array shapes differ in dimension 1 (extent 6 and 7)
Here, dzeta is (6,12) and matmul(U,B) is (7,12).
gfortran does not even de
10 matches
Mail list logo
