[Wien] Fortran 90 bug (?) causes Segmentation fault in DSTART (doreallocate)

[Oleg Rubel]

Dear All,

thank you very much for the comments and suggestions.

I tried to benchmark Laurence's suggestion

     do j = 1, min2
         do i = 1 , min1
         !do j = 1 , min2
             hilfsfeld(i,j)=tf(i,j)
         end do
     end do

     $ x dstart -c
     DSTART ENDS
     63.335u 11.743s 1:15.12 99.9%	0+0k 0+0io 0pf+0w

vs

     do i = 1, min1
         do j = 1 , min2
             hilfsfeld(i,j)=tf(i,j)
         end do
     end do

     $ x dstart -c
     DSTART ENDS
     62.678u 11.837s 1:14.59 99.8%	0+0k 0+0io 0pf+0w

It seems that there is no noticeable difference in timing, but thanks anyway.

Gerhard - I did not try full static linking. Switching the  
optimization levels to -O0 does not help with the original code. I  
cannot step down to an older version of the compiler. Regarding size  
of 'tr' and 'hilfsfeld'. It seems that bounds are OK, in the sense  
that 1:min1 and 1:min2 do not go out of the bounds of 'tr' and  
'hilfsfeld'. Usually size(tr,1)=size(hilfsfeld,1)=3 and  
size(tr,2)=>size(hilfsfeld,2). Just to remind, it works fine for  
smaller size(hilfsfeld,2). It is also not an issue of the memory  
shortage. (16 GB free is more than enough).

So it looks like it is necessary to change the code. Since DSTART is  
not a bottle neck in terms of timing, I would not wary much. Important  
is that it works and probably worth changing in the release version.

Sincerely,

Oleg

-- 
Oleg Rubel, Ph.D.
Thunder Bay Regional Research Institute
290 Munro St
Thunder Bay, Ontario
P7A 7T1, CANADA
E-mail: Oleg.Rubel at physik.uni-marburg.de
Homepage: http://www.staff.uni-marburg.de/~rubel/