September 20, 2021, 18:03 (GMT) |
Fix swapped glossy direct/indirect passes |
September 20, 2021, 17:59 (GMT) |
Fix broken motion vector pass link in compositor |
September 20, 2021, 17:07 (GMT) |
Merge branch 'master' into cycles-x |
September 20, 2021, 17:07 (GMT) |
Fix crash in background render after tile highlight changes |
September 20, 2021, 16:26 (GMT) |
Merge branch 'master' into cycles-x |
September 20, 2021, 15:59 (GMT) |
Cycles X: Remove dead resumable render code It will be implemented differently by specifying an EXR file. Remove the command line arguments to avoid false perception that the resumable render still works. Differential Revision: https://developer.blender.org/D12571 |
September 20, 2021, 15:43 (GMT) |
WIP: Cycles X: Schedule work based on occupancy Sacrifice refresh interval and increase GPU occupancy, lowering the final render time. Lowers `Blender 2.80 - Spring` demo file form predicted 30min measured 3 min to render 1/10th of samples) to about 7.5min. It is still considerably higher than the master branch, which finishes the file in just below 3 min, but it is already a better results. The timing is from RTX 5000. The viewport and CPU rendering should stay unaffected by this change. Differential Revision: https://developer.blender.org/D12570 |
September 20, 2021, 15:35 (GMT) |
Cycles X: Re-implement tile highlight Since the GPUDisplay usage for offline rendering we've lost tile highlight functionality. This change brings it back. Introduced new RenderEngine API call to manipulate highlight state without acquiring RenderResult (so that we don't allocate passes). Not fully ideal since the last tile will be highlighted after render is done until first full-frame is done, but is a cosmetic thing we can improve in the future. Differential Revision: https://developer.blender.org/D12519 |
September 20, 2021, 15:03 (GMT) |
Cycles X: make progressive multi-jitter the default sampling pattern Now that adaptive sampling is enabled by default this was already the case in practice, but use it now also for the non-adaptive case so the sampling patterns stays consistent when switching that on and off. A reason to do this now is that we will need to update all the test reference images for the cycles-x merge, and those are all affected by the sampling pattern. |
September 20, 2021, 14:51 (GMT) |
Cycles X: Improve progress report while processing files from disk Differential Revision: https://developer.blender.org/D12542 |
September 20, 2021, 14:26 (GMT) |
Fix T91425: VDB shadow rendering incorrect in some cases |
September 20, 2021, 13:56 (GMT) |
Fix baking failing with auto tiles, disable since it's not supported yet |
September 20, 2021, 12:40 (GMT) |
Fix incorrect auto minimum samples for adaptive sampling The comment did not match the implementation. Differential Revision: https://developer.blender.org/D12561 |
September 20, 2021, 11:06 (GMT) |
Merge branch 'master' into cycles-x |
September 17, 2021, 17:22 (GMT) |
Fix T91489: Cycles X crash when running with MESA_NO_ERROR |
September 17, 2021, 13:19 (GMT) |
Fix crash in background rendering after recent clearing bugfix |
September 17, 2021, 12:51 (GMT) |
Cleanup: clarify comment |
September 17, 2021, 11:29 (GMT) |
Cycles X: Improve performance of transparent shadows with OptiX This changes the shadow record-all any-hit program to accept all hits (return without calling `optixIgnoreIntersection`) beyond the furthest distance recorded after the maximum number of hits that can be recorded was reached. OptiX will not call the any-hit program anymore for hits beyond the distance of the accepted hits and also reduces the current ray length behind the scenes. As a result performance improves drastically in scenes where shadow rays can hit a lot of transparent objects, like the "koro" benchmark scene. With this applied I now get similar performance with both CUDA and OptiX in "koro". Not quite perfect yet, but much better than before. Reviewed By: brecht Differential Revision: https://developer.blender.org/D12524 |
September 17, 2021, 09:51 (GMT) |
Fix Cycles X crash viewing render result prior rendering |
September 17, 2021, 09:18 (GMT) |
Cycles X: Real memory saving when using tiled rendering The general idea is to delay reading EXR files for until after all view layers are rendered. Once they are all rendered, BlenderSession frees up as much memory as possible, and initiates processing of the on-disk files. The processing includes reading the file, optional denoising of the full frame (if the denoising is enabled), and writing the result via the write callback. The processing is done as a state-machine which routes specific calls to a full-frame processing, which allows to re-use same tile write callback in the software integration. In order to be able to know which view layer and render view is being written the API has been extended to write layer and view names. Done via BufferParams, as layer and view concepts are quite typical for the EXR files. The BufferParams also contains all fields needed for buffers access outside of the scene graph. Differential Revision: https://developer.blender.org/D12503 |
|