#103 - possibly that slower than builds before?

[Polygonius]

I am not sure - but i have the FEELING that #103 is slower than the versions before?

I can not recognize any big difference with metal on or not. I have a group of 8 blurrs: With metal it takes round 12 seconds that the activity monitor is going to near zero, after "visible" this, without metal it takes round 15 seconds. I had not testest with former builds, but i MEAN all was much faster? Is that possible?

[Chris B]

Hey Polygonius,

I haven't noticed a difference. If you have a file and some specific tasks you'd like me to test I'm happy to do so. I have access to previous builds so this isn't a problem for me if you'd like me to look into it.

[Polygonius]

Thanks Chris.

 

File contains 2 groups: 1 group with 8 gaussian blur and another group with 8 bilateral blurr. Well, no real situation, but better for demonstration!

The 8 gaussians comes in 2,3 secondes COMPLETE (but will shrink the layer, with or without "take transparency"... (maybe another bug?), but the 8 bilateral blurr take a very long time to compute complete.

 

Also: The preview / current-view-size seems to be a VERY VERY important factor how long it takes, till the activity monitor come near zero. AND: if i activate/deactivate the group with a very small view and than eg. switch to c,d+4 it SEEMS to take LESS time than eg. in cmd+1 view from the beginning. Especially on my 27" retina.  Approximately you  have better tools for messure it! But keep this in mind - is it in the end not the algorithm/compute filesize/dpi itself, as just the "preview"????? 

TEST Performance.afphoto

[Chris B]

Hey Polygonius,

Thanks for the file.

I opened it in build 97, 98 and 103 and occasionally I was seeing the different builds give me different timings but nothing significant. The averages all worked out about the same.

The Bilateral blurs do seem to take a good 11 seconds or so on my 2017 MacBook Pro when I'm at 100%. The more I'm zoomed in, the longer it takes and I also noticed the Navigator Panel struggles to update.

I'm happy to poke this more and see what can be improved. Thank you  

[Andy Somerfield]

Hi,

Bilateral (and Median) are not GPU accelerated - this probably explains the difference in apparent performance.

Thanks,

A