FPC 33% slower than Delphi4 for 2D Gaussian Blur
Original Reporter info from Mantis: crlab @neurolabusc1
-
Reporter name: Chris Rorden
Original Reporter info from Mantis: crlab @neurolabusc1
- Reporter name: Chris Rorden
Description:
I am offering a bounty for anyone who can help speed up my image smoothing routines. My software (www.mricro.com) applies 3D smoothing, filtering and interpolation to MR images. I have found FPC/Lazarus consistently slower than Delphi.
I think it is fantastic that FPC supports lots of platforms. I am not sure if it is my particular coding style or that FPC is not optimized for random arrays. However, I am offering a reward for anyone who can help boost the speed of this code to within 10% of Delphi4:
http://www.sph.sc.edu/comd/rorden/mricron/bounty/
Additional information:
The slow code is in the routine
procedure SmoothInput (lFWHM: integer);
The procdure is supplied with the Full-Width-Half-Maximum of the smoothing kernel. A look-up-table is generated for the kernel and then applied sequentially horizontally and vertically.
Mantis conversion info:
- Mantis ID: 10275
- Version: 2.2.1
- Fixed in version: 2.4.0
- Monitored by: » crossbuilder (Burkhard Carstens), » etrusco (Flávio Etrusco), » @martin_frb (Martin Friebe), » scribly (Eric Heijnen), » @PascalDragon (Sven Barth)