diff options
Diffstat (limited to 'noao/digiphot/apphot/doc/ucache.hlp')
| -rw-r--r-- | noao/digiphot/apphot/doc/ucache.hlp | 15 |
1 files changed, 15 insertions, 0 deletions
diff --git a/noao/digiphot/apphot/doc/ucache.hlp b/noao/digiphot/apphot/doc/ucache.hlp new file mode 100644 index 00000000..34a091dd --- /dev/null +++ b/noao/digiphot/apphot/doc/ucache.hlp @@ -0,0 +1,15 @@ +If \fIcache\fR is yes and the host machine physical memory and working set size +are large enough, the input image pixels are cached in memory. If cacheing +is enabled and the task is run interactively the first measurement will appear +to take a long time as the entire image must be read in before the measurement +is actually made. All subsequent measurements will be very fast because the task +is accessing memory not disk. The point of cacheing is to speed up random +image access by making the internal image i/o buffers the same size as the +image itself. However if the input object lists are sorted in row order and +sparse cacheing may actually worsen not improve the execution time. Also at +present there is no point in enabling cacheing for images that are less than +or equal to 524288 bytes, i.e. the size of the test image dev$ypix, as the +default image i/o buffer is exactly that size. However if the size of dev$ypix +is doubled by converting it to a real image with the chpixtype task then the +effect of cacheing in interactive is can be quite noticeable if measurements +of objects in the top and bottom halfs of the image are alternated. |