YOUR ACCOUNT

Login or Register to post new topics or replies
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
The attached file shows that a scale to a blend can increase the time by 10x smile:?:

Scale Bug.ffxml
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
Didn't have time to diagnose the problem the other day, but it is not a scale issue, but an image issue.

Curious why a flip operation will cause this increase in time with an image?
Or rotate .25,.5,.75.

Image Bug.ffxml
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
Am I misunderstanding image handling or is this an issue?
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
GMM
Moderator
Filter Forge, Inc
Posts: 3491
Can anyone with a Mac confirm this? I'm afraid out testers are too busy with more critical bugs.
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
It may not be a mac only issue and the title here should probably be changed to 'Image processing'.

Also it is not version specific, I tested it on FF 2.012, and there is the same problem:

Image Bug v2.ffxml
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
So here are some screen captures and more details.

Regarding the 1st and the 6th Selector position ( in Image Bug.ffxml ), the Scale 1 and Flip XY:

_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
Selector 1 ( Scale 1 ) timing is:

_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
Selector 6 ( Flip XY ) timing is:

_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
When the Noise / Image switch is flipped to Noise, there is no appreciable timing difference between Selector 1 and 6.

I would almost understand this if it was bitmap based and not sample based architecture, but for the mean time, another...

smile:?:
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
Indigo Ray
Adam

Posts: 1442
Filters: 82
I'm using FF4.006 on Windows 7 and I do not confirm this. "Flip" barely increases render time. So it could be a Mac bug.
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
Thanks for checking this. I came up with the following table of times on 3 machines using Image Bug.ffxml in FF4.006:


Windows XP 2.6 Ghz single core

Image, Selector 1: 4.58
Image, Selector 6: 9.68


Mac G4 10.4 1 Ghz single core

Image, Selector 1: 9.42
Image, Selector 6: 11.35


Mac Intel 10.6 2.8 Ghz quad core

Image, Selector 1: 0.65
Image, Selector 6: 8.22



Might it be the parallel processing somehow to blame in the last case? But still, seeing roughly double for Win XP?
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
SpaceRay
SpaceRay

Posts: 12298
Filters: 35
I have tested this and I have the following result

Using Windows 8 64 bit (just installed with a total fresh start) and the latest FF 4.0.6 with the silent update using Intel Quad Core i7 2600K 3.4GHz

Using Image Bug filter with Noise/Image checkbox enabled

Image, Selector 1: 0,37 seconds

Image, Selector 6: 1,15 seconds

Using Image Bug V2 filter

600x600 with image noise checkbox disabled

Fast/10x Slow checkbox enabled 1,42 seconds

Fast/10x Slow checkbox disabled 0,38 seconds

---------------------------------------
600x600 with image noise checkbox enabled

Fast/10x Slow checkbox enabled 0,52 seconds

Fast/10x Slow checkbox disabled 0,47 seconds

---------------------------------------------
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
Roughly 2x again, thanks for confirming.

Quote
testers are too busy with more critical bugs.


Not that this is a huge issue, but it would be helpful to understand why this is. A tiler I'm working on suffers. smile:|
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
SpaceRay
SpaceRay

Posts: 12298
Filters: 35
Quote
xirja Roughly 2x again


0,37 x 3 = 1,11 so I think is more than 3x instead of roughly 2x smile;)
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Indeed, any Transform component with non-zero sample coordinate distortion adds extra rendering time (about a double on my Windows 7 machine).

My bet is on sample cache. Having any distortion invalidates the sample cache, so your image-providing subtree gets sampled twice, instead of being sampled once with the second sample from Blend being retrieved from the sample cache.

Try to replace Blend with a switch or anything else that fires one sample instead of two, and you'll see what I mean.
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
I think I almost understand, especially if the difference is double the time, or the sum of the two independent processes, but still here FF 4.008 on a Mac Intel 10.6 2.8 Ghz quad core:

Image, Selector 1: 0.65 seconds
Image, Selector 6: 8.22 seconds = 13 x

and replacing the blend with a 50% threshold at 0 smoothing :

Image, Selector 1: 0.55 seconds
Image, Selector 6: 3.03 seconds = 6 x
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
So, 13 or 6 times more time depending.

Quote
GMM

Can anyone with a Mac confirm this?
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
My apologies for continuing this gripe, but the inexplicable is tormenting me!

Again, for a "Flip" transform it takes greater than 8 x the render time WHEN THE SOURCE IS AN IMAGE on a mac multi-core!!!???.

PREPOSTEROUS! What is this?

I mean for god's sake, replacing the Flip with a Blur (Gaussian) at 100 takes almost half the time???!!!

So what is sooo peculiar about the Transform components and Images here?

Image Bug Simplified.ffxml
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail
xirja
Idididoll Forcabbage

Posts: 1698
Filters: 8
Aha, it seems the blur set to .000001 is the work around!

But still perplexed!

Image Bug Solved_.ffxml
_____________________________________________________

http://web.archive.org/web/2021062908...rjadesign/
_____________________________________________________
  Details E-Mail

Join Our Community!

Filter Forge has a thriving, vibrant, knowledgeable user community. Feel free to join us and have fun!

33,711 Registered Users
+18 new in 30 days!

153,531 Posts
+39 new in 30 days!

15,347 Topics
+72 new in year!

Create an Account

Online Users Last minute:

22 unregistered users.