- User Since
- Aug 31 2008, 5:42 PM (468 w, 1 d)
Thu, Aug 3
Sounds reasonable. :)
Just some note about SSE2/3 in general, I think we can deprecate these soon (especially for 2.8x). Of course we still need a non SIMD Kernel for other architectures, but every computer that runs 2.8x should be modern enough to have SSE4 as well.
Don't think I will miss this, so fine with me. Just wondering if this (alongside your other UI / settings changes) is meant for 2.7x still or for 2.8x?
Jun 14 2017
Nice work Lukas!
May 19 2017
Don't think we need the extra _fac variants in constant fold. We can just use is_one and is_zero?
Apr 27 2017
Apr 26 2017
Looks good to me then. :) We could only get half textures working on Fermi, if we take some of the float4 slots and make them half4 ones, but I wouldn't bother.
In general looks fine. Should be a first step towards easier code in this area. Once we can drop Fermi fully (as soon as nvidia drops it with CUDA 9), we can simplify it even further and get rid of the special cases.
Please always assign entire Cycles project as Reviewers. I refactored most of the image code last year, so I'd like a chance to check on the proposed changes here too. Quite busy atm, but will check in the coming days.
Apr 3 2017
Thank you for changing it, I prefer code readability and the ability to quickly debug stuff there over some very small speedup. :)
I don't see why you changed the if/else to switch statements, you said that the matrix inlining is responsible for the slow compilation?
Mar 28 2017
I don't see an issue with this approach, fine with it.
Mar 27 2017
Mar 21 2017
Please write your report in English.
Read through the code quickly, looks good to me. Just found one little code style issue. :)
Mar 19 2017
This is not a valid bug report. Please ask in a Blender forum for user help!
Mar 1 2017
Thanks for fixing the comment.
Please be more precise when changing such things, how is that a cleanup? Also the comment still mentions ".active" which is now meaningless after the change.
Feb 18 2017
Doesn't look like a bug, just reflections from the lamp. Disable MIS for the lamp and the birght spots go away.
Correct, we only support HD 7000 cards and above with GCN architecture.
Feb 17 2017
Feb 4 2017
Jan 28 2017
We expect clear instructions and if possible a blend file showcasing the issue, as clearly outlined in our bug report guidelines.
Jan 20 2017
I don't see how this is an underlying issue with the layout engine. Using a .active or .disabled tag, assigned to a certain layout block will affect all of its assigned properties. So using two different UI blocks is a proper fix.
Jan 14 2017
I would not scatter Cycles development at this point, either we move all Cycles development to the 2.8 branch (which I wouldn't do just yet) or we stick to master. Putting some things into the branch while keeping maintenance and smaller changes in master will only ask for trouble later. Also we still have a few months until a 2.79 release, enough time to fix issues when they come up.
Jan 6 2017
I don't see this as a new texture or special case, more as a reasonable improvement to an already implemented procedural texture.
Jan 3 2017
I tested the patch quickly, works nicely. LGTM from my side. Thanks for the addition!
Jan 2 2017
Dec 27 2016
We explicitly white list certain OpenCL configurations, not sure we differentiate correctly on OSX though.
Intel GPUs are not officially supported yet for Cycles rendering. This might change in the future, with various split kernel improvements, but for now it's a known limitation.
Dec 26 2016
Do you have a Renderman addon enabled? Disable it and it should be fine. If you use a vanilla Blender build without addon it works fine.
Please edit the content of such reports to "SPAM" or whatever too, so the content disappears.
Dec 24 2016
Try lowering the sub samples (Diffuse, SSS). That might help.
Dec 23 2016
Dec 14 2016
Dec 8 2016
Idea and code looks good to me. :) Having this as a closure seems good, more intuitive than adding yet another UI checkbox to object/mesh data settings imo.
Easy to add now, much better than before my refactor, which made PathState available to SVM directly. :)
Decoupled Ray Marching for homogeneous volumes should work on GPU (when enabled via ifdefs of course), but when I tested that, it wasn't stable. For heterogenous volumes, the current code uses dynamic mem alloc, which is slow on GPU and was giving me mem errors. That was 1-2 years ago though, situation might have changed.
Nov 29 2016
We need a clear way to reproduce the issue, with an example blend file. Also, 2.77a is old, please try the 2.78a release. Also, are you on Windows, Linux, Mac?
Nov 26 2016
Seems you run an old and unsupported GPU/driver. There is nothing we can do to fix this.
Nov 14 2016
Nov 7 2016
Oct 20 2016
Thanks for submitting the patch. Haven't tested it yet, but some comment about code style. That should be fixed in all places.
Oct 17 2016
OpenCL split kernel doesn't support volume rendering yet.
Oct 14 2016
Haven't tested the patch but code looks ok to me. Nice work! Have you tested this with both Mega and Split OpenCL kernel?
Oct 13 2016
we don't support OSL metadata yet, that has never been implemented. It's a known (low priority) todo.
Oct 12 2016
I am against putting this into e.g. extern/. It's part of Cycles SVM/OSL shading system and we should not start outsourcing such things here imho. Then I'd rather have these files twice in the source tree (once Cycles, once GLSL).
Oct 10 2016
Blender 2.78 is out, please update to the latest version and try again.
Oct 8 2016
This bugtracker is for bugs only, and the above mentioned behavior has been a design decision, and therefore works as intended.
Oct 7 2016
My bad. Thanks Julian!
It will happen when some developer picks it up and implements it. :)
This is a known todo, and has been requested many times already in various places. :) We will tackle it at some point.
"Based on a (as simple as possible) attached .blend file with minimum amount of steps" is missing.
Oct 5 2016
Oct 2 2016
Opens fine here with a master build and 2.78 release on Linux.
The HDR texture is missing in the blend file.
Oct 1 2016
Do you get asked during installation if you want to create a Desktop Icon, did you check this?
Sep 30 2016
Hi Bob, these sound great. You can upload patches here https://developer.blender.org/differential/diff/create/ then we can check on the work and possibly include it.
Sep 25 2016
Sep 21 2016
That's not the case, at least not when we use Filter Glossy (see e.g. GlassBsdfNode::simplify_settings). So in case of a non sharp closure, in combination with Filter Glossy, the closure isn't changed.
Sep 19 2016
Happy to see ccl_fetch go away. Can run some tests on my Geforce 730 tomorrow! Nice work!
Sep 17 2016
Adding Julian (UI team) and Sergey (Cycles dev) for further feedback.
We can make this more clear maybe, but it's still not considered a bug.
Sep 16 2016
In order to hide this object in the viewport render, you have to toggle the eye icon in the outliner. And the "Only Render" setting you highlighted is for OpenGL only.
Sep 7 2016
We run different algorithms for Volume rendering on CPU and GPU, GPUs do not support the improved decoupled ray marching sampling. This is a known todo and not considered a bug atm.
But then, RC1 should work as it has this fix?
Please don't mix issues into one report, that makes it difficult to troubleshot.
Sep 4 2016
Please follow the bug submission guidelines and attach a blend file and the texture that you use here, so we can easily reproduce the issue.