MantisBT - ParaView
View Issue Details
IDProjectCategoryView StatusDate SubmittedLast Update
0010828ParaViewBugpublic2010-06-14 13:282011-01-13 17:00
ReporterUtkarsh Ayachit 
Assigned ToUtkarsh Ayachit 
PrioritynormalSeverityminorReproducibilityhave not tried
StatusclosedResolutionfixed 
PlatformOSOS Version
Product Version 
Target Version3.8.1Fixed in Version3.8.1 
Project
Topic Name
Type
Summary0010828: Parallel Volume Rendering issues when geometry is preset
DescriptionFollow the steps:
* pvserver as 4 procs (cannot reproduce with num_procs < 4)
* Wavelet (Volume rendering)
* Contour (default)

Result is wrong (user provided image attached). If you apply "Process ID Scalars" to Contour, you can see that although the geometry is partitioned correctly, the process assignment is incorrect and hence the issues.

Steps To Reproduce
Additional Information
TagsNo tags attached.
Relationships
Attached Filespng depth-issue2.png (181,563) 2010-06-14 13:28
https://www.vtk.org/Bug/file/8183/depth-issue2.png
png

png bug-10828.png (328,377) 2010-07-21 21:56
https://www.vtk.org/Bug/file/8259/bug-10828.png
Issue History
Date ModifiedUsernameFieldChange
2010-06-14 13:28Utkarsh AyachitNew Issue
2010-06-14 13:28Utkarsh AyachitFile Added: depth-issue2.png
2010-06-14 13:28Utkarsh AyachitStatusbacklog => tabled
2010-06-14 13:28Utkarsh AyachitAssigned To => Utkarsh Ayachit
2010-06-14 13:36Utkarsh AyachitNote Added: 0021003
2010-06-14 13:36Utkarsh AyachitStatustabled => @80@
2010-06-14 13:36Utkarsh AyachitResolutionopen => fixed
2010-06-14 13:36Utkarsh AyachitNote Added: 0021004
2010-06-15 09:44Utkarsh AyachitTarget Version3.8.2 => 3.8.1
2010-07-07 09:55Utkarsh AyachitFixed in Version => Development
2010-07-08 10:15Utkarsh AyachitNote Edited: 0021003
2010-07-08 10:17Utkarsh AyachitNote Added: 0021313
2010-07-08 10:35Utkarsh AyachitNote Deleted: 0021313
2010-07-08 10:36Utkarsh AyachitNote Added: 0021320
2010-07-08 10:36Utkarsh AyachitFixed in VersionDevelopment => 3.8.1
2010-07-21 21:08Alan ScottNote Added: 0021452
2010-07-21 21:08Alan ScottStatus@80@ => closed
2010-07-21 21:55Alan ScottNote Added: 0021456
2010-07-21 21:55Alan ScottStatusclosed => @20@
2010-07-21 21:55Alan ScottResolutionfixed => reopened
2010-07-21 21:56Alan ScottFile Added: bug-10828.png
2010-07-28 16:54Utkarsh AyachitNote Added: 0021534
2010-07-28 16:54Utkarsh AyachitStatus@20@ => @80@
2010-07-28 16:54Utkarsh AyachitResolutionreopened => fixed
2010-07-29 23:34Alan ScottNote Added: 0021568
2010-07-29 23:34Alan ScottStatus@80@ => closed
2011-01-13 17:00Source_changeset_attached => VTK master a2bd8391
2011-01-13 17:00Source_changeset_attached => VTK master 020ef709

Notes
(0021003)
Utkarsh Ayachit   
2010-06-14 13:36   
(edited on: 2010-07-08 10:15)
commit 2a0b102b94e982fa3871f1309489774a54d59d66
Author: Utkarsh Ayachit <utkarsh.ayachit@kitware.com>
Date: Mon Jun 14 13:33:42 2010 -0400

    Fixed BUG 0010828. Incorrect KdTree used for partitioning geometry.
    
    vtkOrderedCompositeDistributor was not passing in the region assigments for the
    cuts to the distributor. This was resulting in D3 using it's own region
    assigments for the cuts which was resulting in geometry ending up on wrong
    processes. Fixed that.


=== VTK Commit ==
commit 5dca988dc39fba6f128d57a4ad906aa5aa63b0d8
Author: Utkarsh Ayachit <utkarsh.ayachit@kitware.com>
Date: Mon Jun 14 13:32:27 2010 -0400

    For BUG 0010828. Added API to set region assignments.
    
    vtkDistributedDataFilter had API to set user-defined cuts, but not the region
    assignments. This was needed to ensure that the cuts were assignmed to processes
    as expected when volume rendering image data. This commit exposes that API.
(0021004)
Utkarsh Ayachit   
2010-06-14 13:36   
Needs to go to 3.8 branch
(0021320)
Utkarsh Ayachit   
2010-07-08 10:36   
moved to release branch
(0021452)
Alan Scott   
2010-07-21 21:08   
Looks correct to me. Note I had to run clean to grid to get the wavelet to show up.

Tested 4 remote servers, trunk, linux.
(0021456)
Alan Scott   
2010-07-21 21:55   
Reopening. Utkarsh helped me run this properly. The secret is to change remote render threshold to 0.

It still fails. Attached is a screenshot.
(0021534)
Utkarsh Ayachit   
2010-07-28 16:54   
Could not reproduce. Alan can you verify that server-side windows are not overlapping.
(0021568)
Alan Scott   
2010-07-29 23:34   
This looks OK on the trunk, as of right now. I did see some stitching, but I think it may be the graphics card (blackrose). I did not see anything that looked like what this bug was written for. 8 processes.

Trunk, Linux, remote server.