0024832: Performance of new boolean operations has become worse

(0028840)
ifv (developer)
2014-04-15 13:22

Dear PKV,
please check this issue

(0028869)
pkv (developer)
2014-04-16 12:28
edited on: 2014-04-17 06:24

The performance is getting worse in the class Extrema_ExtCS.
The class is intensively used in computations of Edge/Face Interferences.
The measurement of the performance has be done for the OCC6.6.0 and DEV.
The results are the following:
DEV
  full elapsed time: 19.30 sec
  time spent by Extrema_ExtCS: 9.73 sec

OCC6.6.0
  full elapsed time: 12.01 sec
  time spent by Extrema_ExtCS: 0.59 sec

For the case there are toltal 163 pairs of E/F to treat in terms of bounding
boxes.Thus the method IntTools_BeanFaceIntersector::Perform() is invoked at
least 163 times.
Inside the method there is a lot of Extrema_ExtCS::Perform(...).

The attachment contains 2 examples of E/F intersections where the performance
is worse in >3 times (0.24sec in DEV vs. 0.07sec. in OCC6.6.0)

#-------------------------------------------------------------
#Please, use the following commands to reproduce the problem:
# case 1
restore ez82 b1
restore fz2 b2

bclearobjects; bcleartools;
baddobjects b1 b2

bfillds -t -s

# case 2
restore ez89 b1
restore fz2 b2

bclearobjects; bcleartools;
baddobjects b1 b2

bfillds -t -s
#----------------------------------------------------------
Unfortunately in OCC6.6.0 there is not "-t -s" options,
but it is possible to provide them using the source:
DEV/BOPTest_PartitionCommands.cxx as a sample.
The modified file for OCC6.6.0 is in the attachment.

(0029083)
abv (manager)
2014-04-25 19:14

The performance has downgraded due to fix made for #23995: new calculations have been added for additional check of points of discontinuities on C0 curves.

(0029147)
dbp (developer)
2014-04-29 19:22

Dear kgv,

please review the patch in branch CR24832_1.

(0029148)
dbp (developer)
2014-04-29 19:30

Performance was slightly improved by implementing a few optimizations.

On attached test case the results are the following:
Master: 19.5 sec
Branch: 15.5 sec (- 20%)

The main question is about processing of "sharp points" (Extrema_ExtCS::Perform). Exluding first and last sharp points allows to improve performance by additional 40%:

  ...
  for (i = 2; i < SharpPoints.Upper(); i++)
  {
    ...

Therefore it may be incorrect because some solutions are produced by these points.

(0029149)
kgv (developer)
2014-04-30 08:41
edited on: 2014-04-30 08:50

Dear dbp,

+       pointer PToExtPExtS to ExtPExtS from Extrema;
+       pointer PToExtPRevS to ExtPRevS from Extrema;

please change definition of these classes to inherit from Standard_Transient and use as handles. They used only within Extrema package and no existed public API methods/classes will be affected by this change.

>> Therefore it may be incorrect because some solutions are produced by these points.
please apply this change as discussed with Andrey.

(0029156)
dbp (developer)
2014-04-30 11:30

Dear kgv,

please review the patch in branch CR24832_2.

(0029157)
dbp (developer)
2014-04-30 11:34

On attached test case the results are the following:
Master: 19.5 sec
Branch: 10.2 sec (-20% (optimizations) - 28% (sharp points) = -48%)

(0029159)
kgv (developer)
2014-04-30 12:01

The same issue is still here for two other fields:

+    myPExtS    : SurfacePtr        from Adaptor3d;
+    myPRevS    : SurfacePtr        from Adaptor3d;

Please consider changing their definition as NCollection_Handle (using imported) - in Extrema_ExtPExtS/Extrema_ExtPRevS and remove redundant fields from Extrema_ExtPS.

(0029280)
dbp (developer)
2014-05-12 16:01

Dear kgv,

please review modifications in branch CR24832_2.

(0029290)
kgv (developer)
2014-05-13 08:47

Please test the patch in branch CR24832_3.

(0029360)
apn (administrator)
2014-05-15 20:41

Dear BugMaster,

Branch CR24832_3 (and products from GIT master) was compiled on Linux, Windows and MacOS platforms and tested.
SHA-1: e6b0d9ec4736dc4c7665609f61f23fdc8a338703

Number of compiler warnings:

occt component :
Linux: 17 (17 on master)
Windows: 0 (0 on master)
MacOS: 205 (212 on master)

products component :
Linux: 11 (11 on master)
Windows: 2 (2 on master)

Regressions/Differences:
No regressions

Testing cases:
Not needed

Testing on Linux:
Total MEMORY difference: 355757448 / 355968116
Total CPU difference: 51924.55999999994 / 51289.87999999977

Testing on Windows:
Total MEMORY difference: 379613708 / 379904880
Total CPU difference: 42886.953125 / 35684.234375

There are no differences in images found by testdiff.

(0029361)
abv (manager)
2014-05-15 20:50

Andrey, please provide results of testing performance confirming that the original problem is fixed

(0029460)
apn (administrator)
2014-05-22 12:37

LINUX:
bug24823_1 (ez82 fz2)
master: Tps: 0.35
fix: Tps: 0.02

bug24823_2 (ez89 fz2)
master: Tps: 0.35
fix: Tps: 0.02

bug24823_3 (test in description)
master: Elapsed time: 21.889494 Seconds
fix: Elapsed time: 9.117681 Seconds

http://occt-tests/24832_branch_linux/summary.html [^]
http://occt-tests/24832_master_linux/summary.html [^]

WINDOWS:
bug24823_1 (ez82 fz2)
master: Tps: 0.42
fix: Tps: 0.03

bug24823_2 (ez89 fz2)
master: Tps: 0.43
fix: Tps: 0.03

bug24823_3 (test in description)
master: Elapsed time: 28.3160944599 Seconds
fix: Elapsed time: 10.0781670224 Seconds

http://occt-tests/24832_branch_windows/summary.html [^]
http://occt-tests/24832_master_windows/summary.html [^]

Related Changesets
occt: master 0734c53d Timestamp: 2014-05-22 13:52:20 Author: dbp Committer: apn [ Details ] [ Diff ]	0024832: Performance of new boolean operations has become worse Replace C pointers with handles. Encapsulate handles in Extrema_ExtPExtS and Extrema_ExtPRevS classes.
	mod - src/Extrema/Extrema_ExtCS.cxx	[ Diff ] [ File ]
	mod - src/Extrema/Extrema_ExtPExtS.cdl	[ Diff ] [ File ]
	mod - src/Extrema/Extrema_ExtPExtS.cxx	[ Diff ] [ File ]
	mod - src/Extrema/Extrema_ExtPRevS.cdl	[ Diff ] [ File ]
	mod - src/Extrema/Extrema_ExtPRevS.cxx	[ Diff ] [ File ]
	mod - src/Extrema/Extrema_ExtPS.cdl	[ Diff ] [ File ]
	mod - src/Extrema/Extrema_ExtPS.cxx	[ Diff ] [ File ]

Issue History
Date Modified	Username	Field	Change
2014-04-15 12:53	Timo	New Issue
2014-04-15 12:53	Timo	Assigned To	=> ifv
2014-04-15 12:53	Timo	File Added: Comp3.zip
2014-04-15 12:55	Timo	Relationship added	related to 0023807
2014-04-15 12:56	Timo	Description Updated	View Revisions
2014-04-15 12:59	Timo	OS	=> Windows
2014-04-15 12:59	Timo	OS Version	=> 7 (64 bit)
2014-04-15 13:22	ifv	Note Added: 0028840
2014-04-15 13:22	ifv	Assigned To	ifv => pkv
2014-04-15 13:22	ifv	Status	new => assigned
2014-04-16 12:28	pkv	Note Added: 0028869
2014-04-16 12:29	pkv	File Added: 0024832.zip
2014-04-16 12:29	pkv	Assigned To	pkv => ifv
2014-04-17 06:21	pkv	File Added: BOPTest_PartitionCommands.cxx
2014-04-17 06:24	pkv	Note Edited: 0028869	View Revisions
2014-04-25 18:34	abv	Relationship added	child of 0023995
2014-04-25 19:09	abv	Assigned To	ifv => dbp
2014-04-25 19:14	abv	Note Added: 0029083
2014-04-25 19:14	abv	Target Version	6.7.1 => 6.8.0
2014-04-29 19:22	dbp	Note Added: 0029147
2014-04-29 19:22	dbp	Assigned To	dbp => kgv
2014-04-29 19:22	dbp	Status	assigned => resolved
2014-04-29 19:30	dbp	Note Added: 0029148
2014-04-30 08:41	kgv	Note Added: 0029149
2014-04-30 08:41	kgv	Assigned To	kgv => dbp
2014-04-30 08:41	kgv	Status	resolved => assigned
2014-04-30 08:42	kgv	Note Edited: 0029149	View Revisions
2014-04-30 08:50	kgv	Note Edited: 0029149	View Revisions
2014-04-30 08:50	kgv	Note Edited: 0029149	View Revisions
2014-04-30 11:30	dbp	Note Added: 0029156
2014-04-30 11:30	dbp	Assigned To	dbp => kgv
2014-04-30 11:30	dbp	Status	assigned => resolved
2014-04-30 11:34	dbp	Note Added: 0029157
2014-04-30 12:01	kgv	Note Added: 0029159
2014-04-30 12:01	kgv	Assigned To	kgv => dbp
2014-04-30 12:01	kgv	Status	resolved => assigned
2014-05-12 16:01	dbp	Note Added: 0029280
2014-05-12 16:01	dbp	Assigned To	dbp => kgv
2014-05-12 16:01	dbp	Status	assigned => resolved
2014-05-13 08:47	kgv	Note Added: 0029290
2014-05-13 08:47	kgv	Assigned To	kgv => bugmaster
2014-05-13 08:47	kgv	Status	resolved => reviewed
2014-05-14 16:32	apn	Assigned To	bugmaster => apn
2014-05-15 20:41	apn	Note Added: 0029360
2014-05-15 20:42	apn	Test case number	=> Not needed
2014-05-15 20:42	apn	Assigned To	apn => bugmaster
2014-05-15 20:42	apn	Status	reviewed => tested
2014-05-15 20:50	abv	Note Added: 0029361
2014-05-22 12:37	apn	Note Added: 0029460
2014-05-23 14:25	apn	Changeset attached	=> occt master 0734c53d
2014-05-23 14:25	apn	Assigned To	bugmaster => apn
2014-05-23 14:25	apn	Status	tested => verified
2014-05-23 14:25	apn	Resolution	open => fixed
2014-06-16 13:18	abv	Relationship added	related to 0024979
2014-11-11 12:46	aiv	Fixed in Version	=> 6.8.0
2014-11-11 12:52	aiv	Status	verified => closed

Notes
(0028840) ifv (developer) 2014-04-15 13:22	Dear PKV, please check this issue

(0028869) pkv (developer) 2014-04-16 12:28 edited on: 2014-04-17 06:24	The performance is getting worse in the class Extrema_ExtCS. The class is intensively used in computations of Edge/Face Interferences. The measurement of the performance has be done for the OCC6.6.0 and DEV. The results are the following: DEV full elapsed time: 19.30 sec time spent by Extrema_ExtCS: 9.73 sec OCC6.6.0 full elapsed time: 12.01 sec time spent by Extrema_ExtCS: 0.59 sec For the case there are toltal 163 pairs of E/F to treat in terms of bounding boxes.Thus the method IntTools_BeanFaceIntersector::Perform() is invoked at least 163 times. Inside the method there is a lot of Extrema_ExtCS::Perform(...). The attachment contains 2 examples of E/F intersections where the performance is worse in >3 times (0.24sec in DEV vs. 0.07sec. in OCC6.6.0) #------------------------------------------------------------- #Please, use the following commands to reproduce the problem: # case 1 restore ez82 b1 restore fz2 b2 bclearobjects; bcleartools; baddobjects b1 b2 bfillds -t -s # case 2 restore ez89 b1 restore fz2 b2 bclearobjects; bcleartools; baddobjects b1 b2 bfillds -t -s #---------------------------------------------------------- Unfortunately in OCC6.6.0 there is not "-t -s" options, but it is possible to provide them using the source: DEV/BOPTest_PartitionCommands.cxx as a sample. The modified file for OCC6.6.0 is in the attachment.

(0029083) abv (manager) 2014-04-25 19:14	The performance has downgraded due to fix made for #23995: new calculations have been added for additional check of points of discontinuities on C0 curves.

(0029147) dbp (developer) 2014-04-29 19:22	Dear kgv, please review the patch in branch CR24832_1.

(0029148) dbp (developer) 2014-04-29 19:30	Performance was slightly improved by implementing a few optimizations. On attached test case the results are the following: Master: 19.5 sec Branch: 15.5 sec (- 20%) The main question is about processing of "sharp points" (Extrema_ExtCS::Perform). Exluding first and last sharp points allows to improve performance by additional 40%: ... for (i = 2; i < SharpPoints.Upper(); i++) { ... Therefore it may be incorrect because some solutions are produced by these points.

(0029149) kgv (developer) 2014-04-30 08:41 edited on: 2014-04-30 08:50	Dear dbp, + pointer PToExtPExtS to ExtPExtS from Extrema; + pointer PToExtPRevS to ExtPRevS from Extrema; please change definition of these classes to inherit from Standard_Transient and use as handles. They used only within Extrema package and no existed public API methods/classes will be affected by this change. >> Therefore it may be incorrect because some solutions are produced by these points. please apply this change as discussed with Andrey.

(0029156) dbp (developer) 2014-04-30 11:30	Dear kgv, please review the patch in branch CR24832_2.

(0029157) dbp (developer) 2014-04-30 11:34	On attached test case the results are the following: Master: 19.5 sec Branch: 10.2 sec (-20% (optimizations) - 28% (sharp points) = -48%)

(0029159) kgv (developer) 2014-04-30 12:01	The same issue is still here for two other fields: + myPExtS : SurfacePtr from Adaptor3d; + myPRevS : SurfacePtr from Adaptor3d; Please consider changing their definition as NCollection_Handle (using imported) - in Extrema_ExtPExtS/Extrema_ExtPRevS and remove redundant fields from Extrema_ExtPS.

(0029280) dbp (developer) 2014-05-12 16:01	Dear kgv, please review modifications in branch CR24832_2.

(0029290) kgv (developer) 2014-05-13 08:47	Please test the patch in branch CR24832_3.

(0029360) apn (administrator) 2014-05-15 20:41	Dear BugMaster, Branch CR24832_3 (and products from GIT master) was compiled on Linux, Windows and MacOS platforms and tested. SHA-1: e6b0d9ec4736dc4c7665609f61f23fdc8a338703 Number of compiler warnings: occt component : Linux: 17 (17 on master) Windows: 0 (0 on master) MacOS: 205 (212 on master) products component : Linux: 11 (11 on master) Windows: 2 (2 on master) Regressions/Differences: No regressions Testing cases: Not needed Testing on Linux: Total MEMORY difference: 355757448 / 355968116 Total CPU difference: 51924.55999999994 / 51289.87999999977 Testing on Windows: Total MEMORY difference: 379613708 / 379904880 Total CPU difference: 42886.953125 / 35684.234375 There are no differences in images found by testdiff.

(0029361) abv (manager) 2014-05-15 20:50	Andrey, please provide results of testing performance confirming that the original problem is fixed

(0029460) apn (administrator) 2014-05-22 12:37	LINUX: bug24823_1 (ez82 fz2) master: Tps: 0.35 fix: Tps: 0.02 bug24823_2 (ez89 fz2) master: Tps: 0.35 fix: Tps: 0.02 bug24823_3 (test in description) master: Elapsed time: 21.889494 Seconds fix: Elapsed time: 9.117681 Seconds http://occt-tests/24832_branch_linux/summary.html [^] http://occt-tests/24832_master_linux/summary.html [^] WINDOWS: bug24823_1 (ez82 fz2) master: Tps: 0.42 fix: Tps: 0.03 bug24823_2 (ez89 fz2) master: Tps: 0.43 fix: Tps: 0.03 bug24823_3 (test in description) master: Elapsed time: 28.3160944599 Seconds fix: Elapsed time: 10.0781670224 Seconds http://occt-tests/24832_branch_windows/summary.html [^] http://occt-tests/24832_master_windows/summary.html [^]