MantisBT - Community
View Issue Details
0023581Community[OCCT] OCCT:Meshpublic2012-11-25 20:312019-09-06 18:36
Roman Lygin 
abv 
normalminor 
feedbackopen 
ALL
[OCCT] 6.5.4 
[OCCT] 7.5.0* 
0023581: [Regression] Performance of BRepMesh 6.5.4 can be up to Nx worse than 6.3.1
(Like 0023580, inspired by this forum thread - http://www.opencascade.org/org/forum/thread_23944/?add_id=23956&forum=3 [^])

Performance of BRepMesh measured on a few medium complexity models can be up to 3x worse vs 6.3.1
see enclosed readme-performance.txt
No tags attached.
related to 0023105closed oan Exception during Meshing / Missing triangles 
txt readme-performance.txt (1,291) 2012-11-25 20:31
https://tracker.dev.opencascade.org/
png f1_hollow_631.png (461,410) 2012-11-29 11:23
https://tracker.dev.opencascade.org/
png f1_hollow_654.png (474,296) 2012-11-29 11:23
https://tracker.dev.opencascade.org/
png f1_shading_631.png (203,912) 2012-11-29 11:23
https://tracker.dev.opencascade.org/
png f1_shading_654.png (219,844) 2012-11-29 11:24
https://tracker.dev.opencascade.org/
txt readme-performance2.txt (2,020) 2012-11-29 23:11
https://tracker.dev.opencascade.org/
zip shaver.zip (785,709) 2012-11-29 23:11
https://tracker.dev.opencascade.org/
? f1.brep (3,719,786) 2019-03-12 18:43
https://tracker.dev.opencascade.org/
? cylinder_head.brep (4,330,831) 2019-03-12 18:43
https://tracker.dev.opencascade.org/
Issue History
2012-11-25 20:31Roman LyginNew Issue
2012-11-25 20:31Roman LyginAssigned To => oan
2012-11-25 20:31Roman LyginFile Added: readme-performance.txt
2012-11-29 11:23kgvFile Added: f1_hollow_631.png
2012-11-29 11:23kgvFile Added: f1_hollow_654.png
2012-11-29 11:23kgvFile Added: f1_shading_631.png
2012-11-29 11:24kgvFile Added: f1_shading_654.png
2012-11-29 11:45kgvNote Added: 0022438
2012-11-29 11:45kgvAssigned Tooan => Roman Lygin
2012-11-29 11:45kgvStatusnew => feedback
2012-11-29 23:11Roman LyginNote Added: 0022464
2012-11-29 23:11Roman LyginStatusfeedback => assigned
2012-11-29 23:11Roman LyginFile Added: readme-performance2.txt
2012-11-29 23:11Roman LyginFile Added: shaver.zip
2012-11-29 23:12Roman LyginAssigned ToRoman Lygin => kgv
2012-11-29 23:12Roman LyginStatusassigned => feedback
2012-12-02 17:54kgvAssigned Tokgv => oan
2012-12-02 17:54kgvStatusfeedback => assigned
2013-03-07 15:44oanRelationship addedrelated to 0023105
2019-03-12 18:43oanFile Added: f1.brep
2019-03-12 18:43oanFile Added: cylinder_head.brep
2019-03-12 18:55gitNote Added: 0082880
2019-03-12 18:56oanNote Added: 0082881
2019-03-12 18:58oanNote Edited: 0082881bug_revision_view_page.php?bugnote_id=82881#r20825
2019-03-12 19:00oanNote Edited: 0082881bug_revision_view_page.php?bugnote_id=82881#r20826
2019-03-12 19:10oanAssigned Tooan => abv
2019-03-12 19:10oanStatusassigned => feedback
2019-04-21 09:22abvTarget Version => 7.4.0
2019-09-06 18:36abvTarget Version7.4.0 => 7.5.0*

Notes
(0022438)
kgv   
2012-11-29 11:45   
Dear Roman,

performance regressions couldn't be analyzed in such form.

First at all bug description should provide complete information about used models (should be attached to bug, path in OCCT repository, publicity accessible URL, confidential shapes could be provided to bugmaster via e-mail) and test scenarios.
I found only one model "F1.brep" (http://www.opencascade.org/showroom/shapegallery/gal5/ [^] "The F1") which seems to be exactly corresponds to shape in report.

Second, this is incorrect to calculate time of vdisplay command to compare meshing algorithm. vdisplay calculates deflection parameters by it's own, may use different mesh plugins and involves 3D visualization tasks as well.
The more acceptable test case will be:

pload MODELING
restore F1.brep s
tclean s
chrono l reset
chrono l start
incmesh s 15
chrono l stop
chrono l show
trinfo s

pload VISUALIZATION
vinit
vsetdispmode 1
vdisplay s
vaxo
vfit
vzfit
trinfo s

Results for OCCT631:
> Elapsed time: 0.826 Second(s)
> CPU user time: 0.624004 seconds
> This shape contains 13982 triangles.
> 15723 nodes.
> Maximal deflection 14.332640235601703

Results for OCCT654:
> Elapsed time: 1.52602131781 Seconds
> CPU user time: 1.3260085 seconds
> This shape contains 25230 triangles.
> 20457 nodes.
> Maximal deflection 14.364915117921329

Concerning to F1.brep it is obviously noticeable on attached snapshots that OCCT631 significantly deviate from required deflection whilst OCCT654 produce more acceptable result (but require more triangles and time).
(0022464)
Roman Lygin   
2012-11-29 23:11   
Kirill, thanks for a fast turn-around.
OK to change the methodology (although the 'vdisplay' is closer to the overall end-user experience).
See the updated readme enclosed. The shaver.brep is enclosed, although you can rebuild from the web-site, following the URLs.
If you would like to follow up feel free to drop me a personal email. Discussion via Notes here can be not always efficient. We can put a summary afterwards here.
thanks.
Roman
(0082880)
git   
2019-03-12 18:55   
Branch CR23581 has been created by oan.

SHA-1: c62618cd602f954b4167ef6e09a16a00ed102cac


Detailed log of new commits:

Author: oan
Date: Mon Mar 4 09:55:25 2019 +0300

    0023581: [Regression] Performance of BRepMesh 6.5.4 can be up to Nx worse than 6.3.1
    
    Added test case.
(0082881)
oan   
2019-03-12 18:56   
(edited on: 2019-03-12 19:00)
Current master gives the following results in seconds comparing to 6.5.4 and 6.3.1:

6.3.1
f1: 0.3125
cylinder_head: 0.46875
shaver: 0.53125

6.5.4
f1: 1.1875
cylinder_head: 1.453125
shaver: 1.546875

7.4.0
f1: 0.53125
cylinder_head: 0.734375
shaver: 0.75

General point of performance degradation was change of triangulation algorithm done in version next to 6.3.1.

It is proposed to close the issue now.