iraf-v216 · Code · Issues (50) · Pull requests (81)
iraf.net Issue #109
[2.16.1][linux] convolve does not work correctly
open olebole opened this issue on 2017-10-20 · 1 comments
olebole commented on 2017-10-20
When running the original binary linux (32 bit) distribution of 2.16.1, the convolve
task does not work properly:
vocl> convolve dev$pix pix.cnv "1. 1. 1.; 1. 0. 1.; 1. 1. 1." bound=wrap
ERROR: floating point invalid operation
With a self-compiled version (8c330867617f79c8e2790ef0c988324ef8760bc1) the convolve
task finishes:
vocl> convolve dev$pix pix.cnv "1. 1. 1.; 1. 0. 1.; 1. 1. 1." bound=wrap
vocl> imstat pix.cnv
# IMAGE NPIX MEAN STDDEV MIN MAX
pix.cnv 262144 858.6 842.7 -32768. 32055.
However this result is still different from what one gets on 64 bit:
vocl> convolve dev$pix pix.cnv "1. 1. 1.; 1. 0. 1.; 1. 1. 1." bound=wrap
vocl> imstat pix.cnv
# IMAGE NPIX MEAN STDDEV MIN MAX
pix.cnv 262144 860.5 804.5 -30822. 32055.
olebole commented on 2017-10-21
On MacOSX, all tests show the result of the linux64 version, independently of
- 2.16.1 binary download or own compilation
- 32 or 64 bit
- Yosemite, El Capitan or Sierra
This suggests, that the problem is Linux (32 bit) specific – and the problem exists already in the current 2.16.1 binary. I can reproduce it on Ubuntu 12.04, Ubuntu 14.04 and on Debian testing (others, I didn’t check yet).
Last updated on 2017-10-21