helmut.bressler Posted November 8, 2021 Posted November 8, 2021 Hello, I'm using dx11. But its is even a bit more complicated - I have a lot of rendering logic in a custom plugin. I have configured the plugin so that rendering is off - but still it is loaded. So far I have only observed that behaviour on a single machine with a GTX 1080 card. The GPU temperature should be fine. I have spotted that issue in a standalone version first. After that I have started the project in the editor. Due to the long time it takes until this shows up, I have no nice test case I can handle over to you guys so far. I'm doing that now, but it will most likely take a couple of days, until I have some results. In particular I would like to exclude the custom plugin. I have the project running on another machine for a couple of days now, but it is running only at 30fps. If the issue depends on the frame number (the number of frames rendered), it will probably appear in around 3 to 4 days I would estimate. Thanks Helmut
silent Posted November 12, 2021 Posted November 12, 2021 Helmut, We successfully reproduced this behavior. Hope that would be fixed in 2.15 or right after in hotfix update. Thanks! How to submit a good bug report --- FTP server for test scenes and user uploads: ftp://files.unigine.com user: upload password: 6xYkd6vLYWjpW6SN
werner.poetzelberger Posted November 12, 2021 Posted November 12, 2021 (edited) From a more observational point of view, it seems, that some method stops working after a certain period/time/numbers of calculations. Could there be something like a noise, which runs out of precision and is not reset. Like a float which is simply 0 or infinite? oO Best. werner Edited November 12, 2021 by werner.poetzelberger
helmut.bressler Posted November 15, 2021 Posted November 15, 2021 Hello Silent, thats a great news! It would have taken weeks to create a nice test scene for you. Thanks Helmut
silent Posted November 15, 2021 Posted November 15, 2021 @helmut.bressler, actually fix is quite simple. In files: data/core/materials/base/objects/clouds/compositor/compositor.comp // replace this line: const uint frame_selector = uint(s_frame + min(ray_placement_noise * interleaved_frame_count, interleaved_frame_count - 1.0f)) % toUInt(interleaved_frame_count); // with this: const uint frame_selector = uint(s_frame + toInt(min(ray_placement_noise * interleaved_frame_count, interleaved_frame_count - 1.0f))) % toUInt(interleaved_frame_count); data/core/materials/base/objects/clouds/reprojection/reprojection.comp // replace this line: const uint frame_selector = uint(s_frame + min(ray_placement_noise * interleaved_frame_count, interleaved_frame_count - 1)) % toUInt(interleaved_frame_count); // with this: const uint frame_selector = uint(s_frame + toInt(min(ray_placement_noise * interleaved_frame_count, interleaved_frame_count - 1))) % toUInt(interleaved_frame_count); That should fix the issue (we hope). Thanks! How to submit a good bug report --- FTP server for test scenes and user uploads: ftp://files.unigine.com user: upload password: 6xYkd6vLYWjpW6SN
helmut.bressler Posted November 16, 2021 Posted November 16, 2021 Thank you Silent, I started now another test run with those modifications factored in. Looking at them, I would assume the issue is related to the framenumber. As I'm now testing this with the final software, I would expect it will take some time until I can confirm it is fixing the issue. (I had another machine here where it took 8 days). In case anyone else wants to apply the modifications to 2.15 alpha, its probably worth mentioning that the file locations are slightly different compared to what you have mentioned @silent :-) The lines which have to be modified are: data\core\shaders\clouds\compute\compositor.comp line 195 and data\core\shaders\clouds\compute\reprojection.comp line 40 additionally I believe toInt() should be replaced by to_int() Many thanks Helmut 1
silent Posted November 16, 2021 Posted November 16, 2021 Yes, that's correct. This fix was in the most recent engine version with modified file paths and math methods and I forgot to adjust them to the previous releases naming. How to submit a good bug report --- FTP server for test scenes and user uploads: ftp://files.unigine.com user: upload password: 6xYkd6vLYWjpW6SN
Recommended Posts