Closed GoogleCodeExporter closed 8 years ago
Here are some sample files. (All of them are versions of a GUI graphics editor
I'm
working on, so they may look a little unusual.) Most of the images are
small-scale
PNGs, the largest about 500 x 500 px. The images in the image library to the
right
are the ones that vary.
Small PNGs in image library ("control"--smaller images, generally about 2500 px
each):
http://dl.dropbox.com/u/947038/Glimmr%20Canvas%20Editor.png
Large PNGs in image library (these are the screenshots of the editor itself;
there
are also a few smaller PNGs):
http://dl.dropbox.com/u/947038/Glimmr%20Editor_PNG.gblorb
Large JPEGs in image library (the same large images as in the previous, but
JPEGs):
http://dl.dropbox.com/u/947038/Glimmr%20Editor_JPEG.gblorb
The easiest way to compare speeds is to note the initial time to open the
editor
windows, and then click on the first image in the image library at the right
side.
The image will be copied to the central window. Clicking elsewhere in the
central
window will then move the image around, giving a sense of the redraw rate.
Original comment by Ek.Temple@gmail.com
on 4 Mar 2010 at 2:00
I can't seem to get either of those gblorbs to load. I tested them with
gblorb.ulx,
and it crashes while extracting the unknown resources:
PICT 337 JPEG 235088 PIC337
UNKN 338 UNKN 651251913 UNK338
UNKN 339 UNKN 867273337 UNK339
UNKN 340 UNKN -21076728 UNK340
PICT 341 JPEG 211280 PIC341
Original comment by bcressey@gmail.com
on 4 Mar 2010 at 9:17
Hm, neither of these give me any trouble at all--I can load them in Gargoyle
(running through Wine on OS X) with no complaints, as well as in all other
terps. I
can also load them and extract all resources using gblorb.ulx. Maybe they were
corrupted in the download? I've zipped them for added insurance:
http://dl.dropbox.com/u/947038/Glimmr%20Editor_JPEG.gblorb.zip
http://dl.dropbox.com/u/947038/Glimmr%20Editor_PNG.gblorb.zip
Original comment by Ek.Temple@gmail.com
on 4 Mar 2010 at 9:46
Looking at the imgload.c, it appears that images are not getting reliably
cached in
memory on use.
So the issue may not necessarily be the speed of the scaling code, but the cost
of
retrieving megapixel images from disk.
Original comment by bcressey@gmail.com
on 4 Mar 2010 at 10:24
The zipped files worked just fine, thanks!
Original comment by bcressey@gmail.com
on 4 Mar 2010 at 10:25
Seems like two optimizations are possible:
- Keep the initial image in memory, to avoid trips to disk
- Keep the most recent scaled version of the image in memory, to make moving the
image less costly.
The only advantage of the current approach is that the original image is always
the
source for resize operations, but we can keep that behavior without the
constant hit
to execution speed.
Original comment by bcressey@gmail.com
on 4 Mar 2010 at 10:37
Fixed in r328.
Original comment by bcressey@gmail.com
on 22 Mar 2010 at 6:26
Original issue reported on code.google.com by
Ek.Temple@gmail.com
on 3 Mar 2010 at 9:48