User:Barre/ITK Registration Optimization

ITK Registration Optimization (BW NAC) Project

My (Sebastien Barre) notes so far. Once the dust settles, the relevant sections will be moved to the project pages listed below.

Project

The ultimate specific goal is B-Spline registration optimization for linux and windows on multi-core and multi-processor, shared memory machines. [...] Also, setup tools and a reporting mechanism for ITK speed to be monitored and reported by us and others. BWH is the driving force behind this work.

• Non Rigid Registration at NA-MIC
• ITK Registration Optimization at NA-MIC
• Slicer3:Performance Analysis at NA-MIC
• User:Barre/ITK Registration Optimization at NA-MIC (this page)

Contacts

• Stephen Aylward
• Julien Jommier
• Sebastien Barre
• Lydia Ng

Quick Links

• Google Performance Tools
• kcachegrind, The KCachegrind Handbook, Running Kcachegrind on Mac OSX 10.4
• Also check my BW-NAC Google Notebook for fresher links.

Source Code

 cvs -d :pserver:<login>@public.kitware.com:/cvsroot/BWHITKOptimization login

Enter your VTK <login>, and password, then:

 cvs -d :pserver:<login>@public.kitware.com:/cvsroot/BWHITKOptimization co BWHITKOptimization

to checkout the code.

Testing Data

• NAMIC: Deformable registration speed optimization (DSpace @ Insight-Journal)

Suggested Benchmarks

All tests should cout two values

• the time required
• an measure of the error (0 = no error; 1 = 100% error)

Tests to be developed and suggested parameter settings

• LinearInterpTest <numThreads> <dimSize> <factor> [<outputImage>]
  • NumThreads = 1, 2, 4, and #OfCoresIf>4 (for every platform)
  • DimSize = 100, 200 (meaning: 100^3 and 200^3 images)
  • Factor = 1.5, 2, 3 (thereby producing up to 600^3 images)
  • = 24 tests (approx time on dual-core for all tests = 1.5 minutes)
• BSplineInterpTest <numThreads> <dimSize> <factor> [<outputImage>]
  • NumThreads = 1, 2, 4, and #OfCoresIf>4 (for every platform)
  • DimSize = 100, 200 (meaning: 100^3 and 200^3 images)
  • Factor = 1.5, 2, 3 (thereby producing up to 600^3 images)
  • = 24 tests (approx time on dual-core for all tests = ??)
• SincInterpTest <numThreads> <dimSize> <factor> [<outputImage>]
• BSplineTransformLinearInterpTest <numThreads> <dimSize> <nodes> [<outputImage>]
• MeanReciprocalSquaredDifferenceMetricTest
• MeanSquaresMetricTest
• NormalizedCorreltationMetricTest
• GradientDifferentMetricTest
• MattesMutualInformationMetricTest
• MutualInformationMetricTest
• NormalizedMutualInformationMetricTest
• MutualInformationHistogramMetricTest
• NormaalizedMutualInformationHistogramMetricTest

Potential Issues with Timing

• Repository was updated so that it can compile on Unix.

__rtdsc()

CallMonWin includes <intrin.h> to call __rdtsc(), a header that does not exist in Microsoft compilers prior to Visual Studio 8/2005. It seems however that one can call __rdtsc() directly from assembly:

• GameDev.net forum, see third comment by Evil Steve (!)
• CPUID for x64 Platforms and Microsoft Visual Studio* .NET 2005
• High Resolution Elapsed Timer

A few articles advises against the use of __rdtsc(), especially in a multicore/multithread context:

• GameDev.net forum
• High Performance Timing under Windows

The suggested alternative is to use Performance Counters. Hardware counters are actually not an OS feature per se, but a CPU feature that has been around for some time. They provide high-resolution timers that can be used to monitor a wide range of resources:

• Accessing Nanosecond Timers While Performance Tuning on Windows Servers
• High Resolution Timing

The issue remains on how to access those counters in a cross-platform way:

• PAPI: "The Performance API (PAPI) project specifies a standard API for accessing hardware performance counters".
  Stephen/Christian reported that Dual Core CPUs were not supported, but it seems from the release notes for PAPI 3.5 (2006-11-09) that both Intel Core2Duo and Pentium D (i.e. dual core) are indeed supported.

Process Priority

Whatever our choices, several articles also suggest to bump the application's priority to real-time before performing testing to make sure the wall-clock() results are as realistic as possible. It is however very important to set it back to normal (see last paragraph. Example (win32):

	DWORD dwPriorityClass = GetPriorityClass(GetCurrentProcess());
	int nPriority = GetThreadPriority(GetCurrentThread());
	SetPriorityClass(GetCurrentProcess(),REALTIME_PRIORITY_CLASS);
	SetThreadPriority(GetCurrentThread(),THREAD_PRIORITY_TIME_CRITICAL);
[...]g
        SetThreadPriority(GetCurrentThread(),nPriority);
	SetPriorityClass(GetCurrentProcess(),dwPriorityClass);

Thread Affinity

We should consider using SetThreadAffinityMask to make sure that the starting time is recorded on the same thread as the ending time (Win32). Will that constrain the rest of the program to run on a single thread, very good question. Also check Sleep(0), reported in a few discussions, including this long one.

Test Platforms

The primary target platform at the 8, 16, and 32 processor machines at BWH. However, preliminary tests have been performed on KHQ computers.

KHQ

A full software stack was compiled on several machines at Kitware. Each component was build in two flavors, both shared/debug and static/release:

• Tcl/Tk 8.4
• VTK (cvs)
• ITK (cvs)
• ITK Applications (cvs)
• FLTK (1.1 svn)
• BWHItkOptimization (cvs)

All platforms are so far described in the BWHItkOptimization/Results directory:

Host	#CPU	CPU	Freq	RAM	Arch	OS	Login
amber2	2	Pentium Xeon	2.8 GHz	4 GB	64 bits	Linux 2.6 (Red Hat Enterprise 4)	kitware (ssh, vnc; cd ~/barre)
fury	1	Pentium 4 (hyperthread)	2.8 GHz	1 GB	32 bits	Linux 2.6 (Fedora Core 4)	barre, jjomier, aylward
panzer	1	Intel Core Duo (dual core)	1.66 GHz	1 GB	32 bits	Mac OS X 10.4.8	barre, jjomier, aylward
sanakhan	1	Pentium M	1.8 GHz	1 GB	32 bits	Windows XP SP2	barre
tetsuo	1	Pentium D (dual core)	3.2 GHz	2 GB	32 bits	Windows XP SP2	barre

Tests

• LinearInterp: (to describe)

BWH

I have not yet access to the machines at BWH. Stephen has, and will let me know.

Status

• kcachegrind and timing are being performed on amber2. Stay tuned.
• valgrind is not supported on x86_64 architecture :( Now using fury instead of amber2.
• RegTests/RunLinearInterpTest.sh.in is configured automatically to run and times LinearInterp with various combinations of threads, size and factor parameters.
  • It was run on fury (release static): Results/fury.kitware.timings-rel.txt
  • It was run on fury (debug): Results/fury.kitware.timings-dbg.txt
  • It was run on amber2 (release static): Results/amber2.kitware.timings-rel.txt