Gfortran question

[David Lapointe]

Hi Ed,

The AMD tech said that the acml lib for gfortran was built with version 4.0.x 
and would not work for 4.1.x. I could use g77 but I would have the same 
question.

The one main reason I use acml is for R.  There is a quite a speedup over the 
default blas library. I suppose I could try the Goto BLAS library. It has 
source available.

Errors:
I am trying to compile R (2.3.0) with acml3.1.0 on SuSE 10.1 ( 
gfortran) on an opteron system. 

I get reference errors from libacml compiling the ACML example programs as 
well as with R.

/usr/lib64/gcc/x86_64-suse-linux/4.1.0/../../../../lib64/libacml.so: 
undefined reference to `_gfortran_filename' 
/usr/lib64/gcc/x86_64-suse-linux/4.1.0/../../../../lib64/libacml.so: 
undefined reference to `_gfortran_ioparm' 
/usr/lib64/gcc/x86_64-suse-linux/4.1.0/../../../../lib64/libacml.so: 
undefined reference to `_gfortran_line' 

I did try this with gfortran on a different system ( SuSE 10.0, x86_64, 
gfortran 4.0.x) and it does work. Maybe if I can convince AMD to compile ACML 
with 4.1.x!!

David
On Tuesday 04 July 2006 2:36 pm, you wrote:
> On Tue, 2006-07-04 at 12:16 -0400, David Lapointe wrote:
> > I am using SuSE 10.1 which has gcc/gfortran 4.1.0.  I have some
> > applications which won't compile with the AMD ACML math library (
> > actually the configure step won't allow). Tech @AMD says don't use
> > gfortran 4.1, but gfortran 4.0 will work.
> >
> > My question is how to back down to gfortran 4.0 w/o havoc.
>
> Hi David,
>
> The gfortran shipped in GCC 4.0.x is *quite* buggy and is, officially,
> just a "technology preview" not meant for production code.  The gfortran
> in GCC 4.1.x and later has *many* bug fixes and is much more stable and
> capable.  So if there is any way you can avoid 4.0.x then by all means
> do so.
>
> BTW -- what errors or problems are you seeing with 4.1.x ?
>
> Ed

-- 
 .david
 David Lapointe
"What we obtain too cheap, we esteem too lightly." - T.Paine