There is a danger of being slightly misleading by talking in such broad terms about Mesh Filters and OpenCL. Im finding it to be a lot better with some of the worst bus in these areas fixed. There may be a few specific mesh filter examples that dont work, I havent had much time to test, but if so that could possibly be down to the openCL code inside the patches needing to be updated to work since they fixe the underlying QC-OpenCL bugs. Im not well versed enough in writing OpenCL stuff to be able to tell. Either way I think we have to be very specific when talking about OpenCL bugs in order to create an accurate impression of 10.6.3.
Anyway apart from whatever the remaining mesh filter problems turn out to be, this is, from my experience so far, the best QC OpenCL we have had with Snow Leopard. We get the fix that 10.6.2 first offered for the bug that prevented QC OpenCL from falling back to CPU if no suitable GPU found, without having to suffer all of the new problems 10.6.2 introduced. I doubt its perfect yet but it marks the end of a sorry chapter where I pretty much abandoned QC whilst waiting to see if 10.6.3 undid some of the damage, and Im pleased to say it has and I look forward to reality moving further in the direction of our original expectations for QC in Snow Leopard. If outstanding snow leopard iterator performance issues can be fixed by 10.6.4 then I think we are just about there? Cheers Steve On 30 Mar 2010, at 14:04, George Toledo wrote: > Mesh filters/OpenCL is still broken, and other bugs that I'm aware of seem to > still be there. > > Haven't done a valid iterator test yet (like Alex Drinkwater's old iterator > test). The iterator in SL can work ok in certain scenarios, so it's vital to > do an iterator test that makes QC evaluate more than it "should". > > -George Toledo > > On Tue, Mar 30, 2010 at 5:37 AM, Juan Ayala <[email protected]> wrote: > Well.. I was just about to report that all problems I had with the framework > -until now- were solved with the latest update. I mean... the compositions > with iterators are working now with 10.6.3. I'm using now QC 4.0 (103.1) > Framework version 4.1 (156.13) and so far everything is working. > > I can see again 60fps with all compositions affected including some sent by > other users to the list. For instance : > > http://ibeatsonline.com/index.php%3Fref=project-quartz-jp-quicktime-26_Constructivism.html > > or my favorite victim the "carousel" by Memo Akten found here: > > http://bit.ly/cRm2b1 > > BUT.... the but is... the first time I open the carousel it shows the same > same slow performance ( not the case with zugakousaku's file) . To see it > working fast I need to close and reopen the viewer. After that all works just > fine. > > > The console reports this: > > > > > > > juan > > > > > > > > > > > > > On 29 Mar 2010, at 22:45, vade wrote: > > Returning to this, just upgraded to 10.6.3 > > Mac Book Pro 2.33 Ghz hits the first Checkers at 30fps and iteratorTest at > 2.4ish. > > Seems 10.6.3 brings no Iterator fix? Is there any word on fixes for QC ? > > > On Nov 4, 2009, at 2:48 PM, Christopher Wright wrote: > > > <puts on asbestos suit ;)> > > I'm finding a lot of cases where iterator performance in QC4 is > > dramatically worse than in QC3 (typically about half as fast). > > > > <Checkers.qtz> (this does a 50x50 grid, with half the sprites disabled, so > > it's 2500/2 = 1250 sprites per frame -- MacBook can render this at 60fps in > > Leopard, MacBook Pro is struggling to hit 30fps in Snow Leopard.) > > > > <iteratorTest.qtz> (this one's iteration count is cranked up really high, > > but it's still ~5fps on my MacBook 1.83GHzGMA950 running 10.5.8, and only > > 2.2-2.5fps on my MacBook Pro 2.4GHz 8600M GT running 10.6.x (the MacBook on > > Snow Leopard also hits about 2.5fps, so the same machine is half as fast in > > iterators). Turning down the iteration count makes it scale up, with 10.5 > > always running about 2x as fast as 10.6). > > > > Is this expected? Does anyone else experience this kind of performance? > > Are there cases where iterators are actually faster (or at least as fast) > > on Snow Leopard? I'm only seeing a few cases where having all consumer > > subpatches get disabled causes snow leopard's iterator to skip work (unlike > > leopard's). Should I file a bug for this? > > > > -- > > [ christopher wright ] > > [email protected] > > http://kineme.net/ > > > > _______________________________________________ > > Do not post admin requests to the list. They will be ignored. > > Quartzcomposer-dev mailing list ([email protected]) > > Help/Unsubscribe/Update your Subscription: > > http://lists.apple.com/mailman/options/quartzcomposer-dev/doktorp%40mac.com > > > > This email sent to [email protected] > > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Quartzcomposer-dev mailing list ([email protected]) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/quartzcomposer-dev/juanpa%40wp.pl > > This email sent to [email protected] > > > > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Quartzcomposer-dev mailing list ([email protected]) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/quartzcomposer-dev/gtoledo3%40gmail.com > > This email sent to [email protected] > > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Quartzcomposer-dev mailing list ([email protected]) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/quartzcomposer-dev/steve%40dvmachine.com > > This email sent to [email protected]
_______________________________________________ Do not post admin requests to the list. They will be ignored. Quartzcomposer-dev mailing list ([email protected]) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/quartzcomposer-dev/archive%40mail-archive.com This email sent to [email protected]
