|
|
Up |
|
|
  |
Author: michael.caracotsiosmichael.caracotsios Date: Sep 12, 2008 12:25
It appears that if I compile LAPACK with CVF under the option: Full
Optimizations I have trouble computing the eigenvalues and
eigenvectors of the transpose of this matrix:
A=RESHAPE( (/ -8, -9, -7, 6, 3, 7, -3, 4,
-6, -9, 2, 9, 9, -2, 7, -5, -10, 3,
-2, 4, 9, -3, 3, 6, 5/),SHAPE(A))
The error is with the sign function. (it throws an exception)
However, if I compile with the option Optimize for Speed (O3) there is
no problem. Anyone has observed similar problems with the optimization
flags for LAPACK?
Michael
|
| |
|
| | 3 Comments |
|
  |
Author: glen herrmannsfeldtglen herrmannsfeldt Date: Sep 12, 2008 14:29
> The error is with the sign function. (it throws an exception)
An exception for the SIGN function? Pretty funny.
In earlier versions of Fortran, SIGN was undefined when
the second argument was zero. Some implementations used
it as a way to test the sign of REAL variables with the
value zero. As of at least Fortran 2003:
"Case (iv): If B is of type real and is zero, then
(1) If the processor cannot distinguish between positive
and negative real zero, the value of the result is |A|.
(2) If B is positive real zero, the value of the result is |A|.
(3) If B is negative real zero, the value of the result is -|A|."
On many systems, SIGN can be implemented by replacing the sign
bit of the first argument with the sign bit of the second argument
and returning the result. For implementations that allow for
a floating point negative zero, it can be used to test for that.
(Fortran 2003 specifically excludes this ability for INTEGER
zero. The return value is positive for integer zero.)
|
| Show full article (1.40Kb) |
|
| | no comments |
|
  |
Author: Tim PrinceTim Prince Date: Sep 12, 2008 15:56
glen herrmannsfeldt wrote:
>
> It looks like it calls an actual function for one case, but
> inlines a bit move for the other. Hopefully they will update
> for Fortran 2003 soon.
>
> -- glen
>
No, there's no likelihood of an F2003 update of Compaq Fortran. That's
still owned by HP. There are several currently maintained Fortran
compilers.
Incidentally, for the f95/f2003 version of SIGN(...
|
| Show full article (1.03Kb) |
| no comments |
|
  |
|
|
  |
Author: Steve LionelSteve Lionel Date: Sep 15, 2008 05:12
On Sat, 13 Sep 2008 15:37:51 -0700, nospam@see.signature (Richard Maine)
wrote:
>glen herrmannsfeldt ugcs.caltech.edu> wrote:
>
>> I thought he read my answer as expecting all of F2003, when I
>> meant just that one feature. Even more so if it is in F95.
>
>Regardless of what feature you are talking about, the pretty clear
>answer is that ain't *NOTHING* going to happen to CVF. Hasn't in some
>time and has pretty much zero chance of happening in the future.
Richard is correct. HP has nobody working on CVF. I cannot imagine any
circumstances that would allow for a revival and future updates of that
product.
--
Steve Lionel
Developer Products Division
Intel Corporation
Nashua, NH
For email address, replace "invalid" with "com"
|
| Show full article (0.98Kb) |
| no comments |
|
RELATED THREADS |
  |
|
|
|