- User Since
- Jan 3 2003, 6:23 PM (733 w, 1 d)
Started happening after rB8f0dc3cef6c3: Fix T50052: bpy.utils.unregister_module doesn't unregister classes of….
Could we auto reload when changing the scripts path?
It's great to see this finally tackled, and nice to see so much red in the diff.
What is the status here?
Some minor comments, overall seems fine.
Wed, Dec 28
imb_ext_image_qt and imb_ext_movie_qt can be removed as well.
Tue, Dec 27
Looks fine generally.
There is not enough information here for us to investigate. Please provide a .blend and exact steps to redo the problem using it, otherwise there's nothing we can do.
Lowering Diffuse/SSS/.. samples and increasing AA samples does not lower quality, it's only about performance and noise. Using 8x8 = 64 SSS samples per AA sample is usually the wrong balance, I'd advise to use much lower values.
Dec 19 2016
Seems like a good approach. The downside of this patch is that the simplification is no longer multithreaded. It should be possible to multithreaded BlenderSync::sync_materials though, so that it's even faster than before due to multithreaded Blender material export.
Dec 14 2016
No time to review the patch now, but it's good functionality I think.
Seems reasonable, users can indeed always add one extra bounce.
Dec 13 2016
I think doing this using objects is the right solution, even if textures let you do some similar things. The purpose here is to be able to apply the cutout on an entire scene without having to go and modify the materials for each object in the scene.
I have no time to review this code now, but great that you got the performance so close.
There are two tricks that I know of to solve the aliasing issue:
- Render at higher resolution, tonemap, and scale down. Either manually or as some kind of automatic thing where subpixel samples are preserved. Maybe even using something like OpenDCX deep EXR files, but this is of course quite complicated and affects the entire pipeline.
- Do a remapping before and after pixel filtering, scaling down high values and scaling them up again. This is biased but still preserves things a bit better than clamping. See e.g. info on the remapping option in RIS here and here.
Dec 6 2016
Nice work, always cool to see Cycles used in different ways.
Adding LABEL_SINGULAR seems fine to me for BSDFs with zero roughness. But that PATH_RAY_MIS_SKIP test I think is wrong.
Dec 5 2016
This patch also seems to be for an older version of the code, in latest master there have been a bunch of changes to the device code so this will need to be updated for that.
Dec 4 2016
Dec 3 2016
Nov 30 2016
The HSV value edited in the color picker is in display space (sRGB by default). The values in the API are in scene linear space.
Nov 29 2016
That seems quite acceptable, especially if solving this correlation helps with convergence in some cases.
I don't know if we want to do this, but perhaps we should actually be using shadow_blocked() here to get partial transparency like AO? It's more expensive if there is transparency, but also useful for volumes and cutout transparency.
Nov 27 2016
There are multiple ways to do HSV to RGB conversion, and it also depends on the color space (sRGB, linear, ..).
It uses the distance to the bounding box of the object, not to the center of the object.
Nov 26 2016
Perhaps when we have adaptive sampling we can do some kind of brightness estimation and adaptively change the threshold. Currently I suggest to lower the default threshold to 0.01.
Nov 23 2016
Thanks, should be fixed in rB411836d97c03: Fix Cycles device backwards compatibility error if device type is unavailable..
Thanks, fixed in rB57141ea30e9e: Fix spelling in Cycles distance culling description..
Nov 21 2016
Nov 20 2016
If the problem here is indeed C++ symbol collisions, couldn't we just hide all those symbols? Adding *__Z* in osx_locals.map should hide all Blender C++ symbols. We already hide Boost symbols but maybe there are some that we missed.
That error indeed indicates it's the same issue that was already fixed, and if there's another issue it's probably one of the multiple other OpenCL bugs open.
Committing this first as experimental seems ok.
Nov 19 2016
I can't spot any bugs, just minor nitpicks. A box filter is quite ok for downscaling I think.
Here's what I'd like to commit:
- Backwards compatibility only for compute_device_type, the device ID is too complicated to map and only affects multi GPU users.
- Add userpref.blend version and check that, it's not the same as regular .blend.
- Rename old property to legacy_compute_device_type, and hide in the UI.
- Only apply backwards compatibility for users that haven't already resaved their user prefs (<= 2.78.1).
- Always show device enum in render properties, greyed out if not usable so that users can look at the tooltip to guide them to the user prefs.
If it crashes without texture, or with only a few objects, etc, then you could share a smaller file. If it needs to be a big file to repro that's ok though.
We really need a .blend file and exact steps to reproduce the problem, I have not seen this issue reported before, so there must be some specific things that trigger it, which we can't guess well.
Nov 18 2016
Nov 17 2016
I fixed the auto tile size addon in T50001.
@Sergey Sharybin (sergey), the heuristic looks ok, I can't find anything in the Windows API docs to solve this completely.
Nov 13 2016
There is not enough information here for us to investigate, we need a .blend file, sound file and exact steps to reproduce the problem.
I guess this is one for @Joshua Leung (aligorith).
No reply since a week, we really need to know if this is still an issue in the latest build, and if so what the exact error message is. So according to tracker policy I'll close this report, but can reopen is more information is provided.
If you try the latest build from https://builder.blender.org/download/ it should be fixed there, though you may need to reconfigure your GPU user preferences since that changed as well.
@Bastien Montagne (mont29), it can be fixed like this, the zip was not including files recursively. It seems the API docs haven't been updated in a few weeks though, so maybe something else is broken as well.
No, I was just waiting to see if there was any feedback from Sergey, but if there is none I'll commit this soon.
This is due to rB93ace71bd7cd: Cycles: Only use new light sample threshold for new files. Initially we had the light sampling threshold optimization enabled in all existing files, but now it's done only in new files for backwards compatibility.
Nov 12 2016
Marking as incomplete for now.
This was broken in rBe9bcdcdbbd91: UI: Make eyedropper shortcut configurable.
This is by design currently, refraction is not transparency.
Assigning to @Gaia Clary (gaiaclary).
Fixed by D2343.
Spot lights don't work well with non-zero size, not just in Cycles but the light geometry as defined by Blender seems problematic.
To me this seems like the right direction, but I don't yet understand all the code or implications well enough. Is there a high level design document somewhere with the plan?