|
[Sponsors] |
September 13, 2002, 05:23 |
F90 Performance, Pointer
|
#1 |
Guest
Posts: n/a
|
Hey everybody, I'm currently porting a pure F77 code to F90 by using derived data types, pointers and all that modern stuff. Unfortunately, I do loose significantly performance on nearly all platforms (HP;SGI;DEC;IBM) eventhough I do not use the pointers within computationally intensive subroutines. In those routines the data is seen as pure arrays. Using the pointer data directly in strong do-loops (something I really would like to do) kills the perfomance completely. Does somebody experience something similar in the past and does somebody now about Compiler Options which might help? I'm pretty much convinced that aliasing is the problem I'm struggling with.
Thanks in advance -Rolf |
|
September 13, 2002, 05:45 |
Re: F90 Performance, Pointer
|
#2 |
Guest
Posts: n/a
|
Hi Rolf,
once I have encountered a similar drop in performance using F90 pointers. I my case, the compiler obviously generated local (hidden) copies of the fields before they were passed to the subroutine. Unfortunately I have found no remedy for it. Best regards Markus |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
parallel performance on BX900 | uzawa | OpenFOAM Installation | 3 | September 5, 2011 15:52 |
Linux kernel performance enhancement | ConnClark | OpenFOAM | 0 | January 17, 2011 18:18 |
Hexadecimal and F90 | ganesh | Main CFD Forum | 0 | August 4, 2010 09:24 |
Makefile for f90 under unix | Jinwon | Main CFD Forum | 1 | September 22, 2007 05:10 |
ANSYS CFX 10.0 Parallel Performance for Windows XP | Saturn | CFX | 4 | August 13, 2006 12:27 |