| Anonymous | Login | Signup for a new account | 2013-05-24 21:42 PDT | ![]() |
| Main | My View | View Issues | Change Log | Roadmap |
| View Issue Details [ Jump to Notes ] | [ Issue History ] [ Print ] | ||||||||||||
| ID | Project | Category | View Status | Date Submitted | Last Update | ||||||||
| 0001233 | LuxRender | Core | public | 2012-05-14 21:15 | 2013-05-21 14:33 | ||||||||
| Reporter | AGriffiths | ||||||||||||
| Assigned To | jeanphi | ||||||||||||
| Priority | high | Severity | crash | Reproducibility | always | ||||||||
| Status | feedback | Resolution | open | ||||||||||
| Platform | PC | OS | Windows x64 | OS Version | 7 | ||||||||
| Product Version | 1.0RC1 | ||||||||||||
| Target Version | 1.3 | Fixed in Version | |||||||||||
| Summary | 0001233: Unable to process a scene using DAZ 3D Millenium Dragon 2 | ||||||||||||
| Description | Two issues, related to using DAZ 3D Millenium Dragon 2. Rendering from DAZ Studio 4 Pro x64 via Reality 2.2 plugin to LuxRender x64 1.0RC1 (2nd release of RC1). Issue 1 - I have an existing scene with 2 DAZ's Millenium Dragon 2 and a whole bunch of other stuff. Send it to Lux via your Reality Plugin. Lux Opens up and starts to load the scene. It gets stuck in Lux when 'Applying 1 levels of loop subdivision ...' for the shape associated with the 'SPIKES' parameter. I have narrowed down the line of code in Lux to line 301 of loopsubdiv.cpp (luxlib\core) and more specifically the call to gamma(v[i]->valence() when i = 25,575 (always the same index). It goes into the do {} while loop at line 217 of loopsubdiv.h abd never comes out because f is never equal to startFace. However, if I set the SPIKES parameter to any non-zero value (e.g. 0.001) then the issue does not occur. Even more bizarre is that this happens with only 1 of the two instances of MilDrag 2 in the scene. Issue 2 - Again using DAZ's Millenium Dragon 2 - in a brand new DAZ scene (I did this to hopefully eliminate the possibility of my original scene from issue one being the cause). Render to Lux as usual, Lux loads, opens scene successfully this time, however Lux gets to the point of preprocessing and crashes. This time the location is at line 222 in qbvhaccel.cpp (luxrays\src\accelerators) because binId is -2147483648 (0x80000000) - obviously out of range and thus causes an Access Violation. In this case i is always 8138 and primsCentroids[primIndex][axis] has NaN values for x, y, & z. Regards, Andy | ||||||||||||
| Steps To Reproduce | Issue 1 - Load scene rs-22-1.lxs in LuxRender. Scene will load an then endlessly loop when the log says 'Shape 0000005 (plymesh): Applying 1 levels of loop subdivision to 28776 triangles' LuxIssue1.zip Issue 2 - Load scene rs-22-2.lxs in LuxRender. LuxRender will load the scene and crash whe it gets to the rendering stage. LuxIssue2.zip - Trying to upload, but 5MB limit causing issues. | ||||||||||||
| Additional Information | Running Windows 7 x64 DAZ Studio 4.0.3.47 x64 Pro Reality 2.20.1.117 x64 LuxRender 1.0RC1 x64 dated 20th April 2012 - The second build of RC1 - Used the 'lux-d48df2ee119a' source code to build debug version. AMD Phenom II X6 1055T 2.80 GHz 6 core CPU AMD Radeon HD 6990 Dual GPU, 4GB GDDR5 Video Memory 16GB RAM | ||||||||||||
| Tags | No tags attached. | ||||||||||||
| Mercurial Changeset # | |||||||||||||
| Requires Documentation Update | No | ||||||||||||
| Requires Exporter Update | |||||||||||||
| Attached Files | |||||||||||||
Notes |
|
|
(0003412) Lord Crc (administrator) 2012-05-15 15:17 |
From what I can gather, issue 1 is due to numerical issues. It seems some of the vertices are just too close to each other. In at least one case a subdivided vertex, which should be between the two source vertices, actually ends up being located at the exact position of one of the source vertices. This creates a degenerate triangle which the code fails to identify and handle. I haven't looked into issue 2 yet. |
|
(0003413) AGriffiths (reporter) 2012-05-15 23:27 |
Thanks for the feedback. That makes sense - The SPIKES are set to a value of 0 which morphs them to a position where the are not seen. So by setting the parameter to 0.001 was just enough to avoid that issue. I guess that the rotation/scale of the character must have played a big factor in causing this as well. That would be called Murphy's law. Thanks, Andy |
|
(0003440) AGriffiths (reporter) 2012-06-23 10:42 edited on: 2012-06-23 10:43 |
Not wishing to be a pain, but has there been any progress on these issues? Lookng forward to trying this scene out with Lux, but need the fixes. Cheers, Andy |
|
(0003441) Lord Crc (administrator) 2012-06-27 02:35 edited on: 2012-06-27 02:48 |
Unfortunately this is a tricky issue. I managed to change some code so that it handles the numerical precision issues better, however this is just a bandaid fix. I would like to make the code properly handle degenerate edges, so that it will never crash for geometry like yours. I'm still working on that. The fix for issue 1 may or may not solve issue 2, so I will have to wait with the investigation into that until 1 is resolved. |
|
(0003442) AGriffiths (reporter) 2012-06-27 08:37 |
Much appreciated - thanks for the update. Cheers, Andy |
|
(0003458) Lord Crc (administrator) 2012-07-05 02:56 |
I think I'll solve issue 1 by detecting the erroneous condition and simply prevent further subdivision. We may later make the code more robust to these conditions, however for now this should prevent the infinite loops. I've not yet figured out exactly how to handle issue 2. I know what the problem is, so hopefully I can figure out something soon. |
|
(0003459) AGriffiths (reporter) 2012-07-05 19:44 |
Thanks for the update - much appreciated. |
|
(0003941) jeanphi (administrator) 2013-05-21 14:33 |
Issue 1 should now be fixed (I can render the scene here in GPU mode. Issue 2 seems to be fixed too, Lux now aborts with a nice error message when doing hybrid rendering, and renders ok on CPU. |
Issue History |
|||
| Date Modified | Username | Field | Change |
| 2012-05-14 21:15 | AGriffiths | New Issue | |
| 2012-05-14 21:15 | AGriffiths | File Added: LuxIssue1.zip | |
| 2012-05-14 21:17 | AGriffiths | File Added: LuxIssue2.zip | |
| 2012-05-14 22:43 | Lord Crc | Assigned To | => Lord Crc |
| 2012-05-14 22:43 | Lord Crc | Status | new => assigned |
| 2012-05-15 15:17 | Lord Crc | Note Added: 0003412 | |
| 2012-05-15 23:27 | AGriffiths | Note Added: 0003413 | |
| 2012-06-23 10:42 | AGriffiths | Note Added: 0003440 | |
| 2012-06-23 10:43 | AGriffiths | Note Edited: 0003440 | View Revisions |
| 2012-06-27 02:35 | Lord Crc | Note Added: 0003441 | |
| 2012-06-27 02:48 | Lord Crc | Note Edited: 0003441 | View Revisions |
| 2012-06-27 08:37 | AGriffiths | Note Added: 0003442 | |
| 2012-07-05 02:56 | Lord Crc | Note Added: 0003458 | |
| 2012-07-05 19:44 | AGriffiths | Note Added: 0003459 | |
| 2012-08-21 07:12 | jeanphi | Target Version | => 1.0 |
| 2012-09-11 06:07 | jeanphi | Target Version | 1.0 => 1.1 |
| 2012-10-08 00:19 | jeanphi | Target Version | 1.1 => 1.2 |
| 2013-02-25 05:11 | jeanphi | Target Version | 1.2 => 1.3 |
| 2013-05-21 14:31 | jeanphi | Assigned To | Lord Crc => jeanphi |
| 2013-05-21 14:33 | jeanphi | Note Added: 0003941 | |
| 2013-05-21 14:33 | jeanphi | Status | assigned => feedback |
| Copyright © 2000 - 2012 MantisBT Group |