> My suggestion would be to file a bug. The Velocity Engine Team is very
> actively pursuing optimization on these frameworks and would be interested
> in knowing if there's anything that they may have overlooked.
Since you brought this up ;-)
Is it expected that tiny differences in destination image size can result in
a 4x difference in elapsed time for vImageScale_Planar8? For example,
starting with an image 1728x2180 (1728 rowbytes), scaling to 863x1089 (864
rowbytes) takes 150ms but scaling to 864x1090 (864 rowbytes) takes 650ms.
Not all images are slow when scaled down exactly 1/2, and the test image
mentioned above is slow at many other scales, and other test images the
difference is 2x (I think I've seen 3x as well). I think I'm setting up
rowbytes correctly. The timings are consistent within ~10ms across multiple
runs. Here's a partial output of timings of that test image:
856 x 1080 - ratio: 2.01869 - 0.149367 secs
857 x 1081 - ratio: 2.01634 - 0.623081 secs
858 x 1082 - ratio: 2.01399 - 0.625644 secs
859 x 1084 - ratio: 2.01164 - 0.163413 secs
860 x 1085 - ratio: 2.0093 - 0.150244 secs
861 x 1086 - ratio: 2.00697 - 0.644116 secs
862 x 1087 - ratio: 2.00464 - 0.64057 secs
863 x 1089 - ratio: 2.00232 - 0.147098 secs
864 x 1090 - ratio: 2 - 0.650623 secs
865 x 1091 - ratio: 1.99769 - 0.670462 secs
866 x 1093 - ratio: 1.99538 - 0.15652 secs
867 x 1094 - ratio: 1.99308 - 0.155308 secs
868 x 1095 - ratio: 1.99078 - 0.642182 secs
869 x 1096 - ratio: 1.98849 - 0.640839 secs
870 x 1098 - ratio: 1.98621 - 0.144765 secs
871 x 1099 - ratio: 1.98393 - 0.144555 secs
872 x 1100 - ratio: 1.98165 - 0.6453 secs
What gives? And can I avoid this? And yeah, those timings are from a test
project I built to submit. Should it go as a bug report or DTS incident or
both?
--
Scott Ribe
email@hidden
http://www.killerbytes.com/
(303) 665-7007 voice
_______________________________________________
Do not post admin requests to the list. They will be ignored.
PerfOptimization-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/perfoptimization-dev/email@hidden
This email sent to email@hidden
