Oops I said Mesh Bulge by accident, it was actually Mesh Bend that I was able to fix by tinkering with the OpenCL Kernel patch code.
The changes to DAE scale & rotation I noticed, a pain for sure, think it affects the model in the Apple Wind Tunnel example, the wind sim part of which still works for me. Certinly it would be clearer to people which bugs still remain if Apple gave their examples some love and updated the OpenCL ones that dont work properly anymore. What stability issues are you having? I have same GPUs as you and stability seems way better, I cant make OpenCL crash like I used to. But I know you are likely to have ushed QC further than me so so examples would be appreciated. Unless Im mistaken we dont actually know if there is much in the way of hardware-specific OpenCL problems with this release, certainly Ive seen at least one OpenCL example that acts differently on CPU than it does on GPU but this may be down to inherent limitations with some OpenCL functionality on CPU class devices as opposed to GPU, I dont know. (Apple snow example demonstrates this stuff) Although its annoying that Apple failed to update those couple of mesh filters to work with the new & improved OpenCL, its hard for me to think of it as a terrible problem because it seems to have been caused by improvements to OpenCL and so in some ways it gives me hope. Cheers Steve On 2 Apr 2010, at 07:55, George Toledo wrote: > Jiggle is purple because there is either an interaction or mouse patch hidden > in it, as it is a virtual patch (can't remember which, and I'm in 10.5.8 > right now so it's a pain to check).... the mouse/interaction makes it > externally driven. > > Also, on my system, Mesh Bulge has been working- there is no problem with it, > it's Mesh Bend that's been broken, along with Mesh Noise (since 10.6.2 or > some 10.6.1.whatever). > > Some DAE's don't render at the same scaling or rotations as they did on > original release either. It's good to know that for some, this version is > working better, but this definitely isn't as stable as the original release > of QC4, on my system (nVidia 9400/9600 Macbook Pro). > > I don't want to seem like I'm beating up on it, I just think the praise is > unrealistic, especially for anyone who has to consider something running on > machines of differing configurations. It's more realistic and sane to avoid > OpenCL in QC at this point, and work around other bugs that have been > introduced since QC3 as best possible. For someone to design an app around, > say, QC rendering DAE's, and have a simple OS update change all scaling and > rotations is pretty unacceptable, and that's square one. It's hard for anyone > to take OpenCL and the Mesh Kit seriously. For things like mesh filters to > have regressed is also unfortunate. > > That said, QC4 is definitely usable, and there are some great things about it. > > Best regards, > George Toledo > > On Fri, Apr 2, 2010 at 2:04 AM, George Toledo <[email protected]> wrote: > One million and two. > > Are you under the impression that Quartz Composer is for building games? > > -George Toledo > > > On Fri, Apr 2, 2010 at 1:55 AM, Letty Logan <[email protected]> wrote: > How many games has each of you built already, if I may asked ? > > iPhone Mail > > On Mar 31, 2010, at 10:26 AM, Steven Watkins <[email protected]> wrote: > > Hello, > > Ah good that sounds more promising. I was just wondering myself why the > Jiggle is Purple. > > Im also getting somewhere with fixing the broken mesh filters. > > They can be edited by finding them in /System/Library/Compositions > > Im just looking at the two OpenCL Kernel patches in Mesh Noise and they give > errors because attenuation is set to be a constant but one line is trying to > change attenuation by clamping it. Crudely working round this by removing > that line in both patches causes the Mesh Noise to jump into life. Someone > that has a clue about programming will have to decide whether this is a bug > with OpenCL or actually a bugfix which has caused poor code in these > particular patches to fail. > > Whilst writing this email I had a very quick look at Mesh Bulge and I think > its the same story - the OpenCL kernel patch now picks up on more attempts to > assign stuff to read-only variables and so the patch doesnt run. > > Cheers > > Steve Elbows > > On 31 Mar 2010, at 17:31, vade wrote: > > Nice catch > > That does indeed work. It may have been the specific filter I was testing. I > tend to ignore the templates. > > So, in short, > > Mac Book Pro 2.33 Ghz - ATI X1600 sans OpenCL capability: > > Works: > Mesh Template shows proper output and renders a sin-wave distorted sphere. > (the default when opening a new Mesh Filter Template) > > Mesh Twist Filter > Mesh Wave > Mesh Bulge > Mesh Jiggle (although it shows up as a Provider - ie Purple - ) and acts a > bit odd, but it does render. > > Fails: > Mesh Bend > Mesh Noise > > > Of course, the two that I tested earlier were Bend and Noise. I will update > my Bug Report with this info. Thanks for the prod. This is interesting and > good to know OpenCL works muuuccch better than 10.6.2 (those patches, all of > them, used to crash QC for me). > > > On Mar 31, 2010, at 12:18 PM, Steven Watkins wrote: > > Oh dear, that doesnt sound good. What GPU is in your machine? Perhaps there > are certain hardware combinations that CPU fallback still fails on, Ive been > testing non-GPU OpenCL on a 2008 Mac Pro with a Radeon HD 2600. > > So when you click on Mesh Filter in the template startup window, you dont > even get a preview that shows a mesh being wave deformed? Thats the most > basic test I can think of without posting a composition. > > Cheers > > Steve > > On 31 Mar 2010, at 15:07, vade wrote: > > Do we? I do not have that OpenCL support on my GPU, and mesh filters fail > miserably. Do you have a specific example? I am unsure what if anything > OpenCL should work on my hardware to be honest - I imagine OpenCL on CPU > should but at this point... ha :) > > On Mar 30, 2010, at 6:32 PM, Steven Watkins wrote: > > 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, > > > > _______________________________________________ > 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/lettylogan%40yahoo.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/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/archive%40mail-archive.com This email sent to [email protected]
