Use of TBB, IPP and OpenMP on OSX or Linux

advanced
#1

Hello All,

Is it worth to use TBB, Intel IPP on OSX 10.6 or Linux?

(Especially with opencv)

I already tested OpenMP on Linux (4 cores)
It is now used at 400% instead of 100%.
But there is no much more performance result than before. :-/
The detection time is the same.

Does someone already tested one of those implementation.

If yes could you provide example?

Many thanks in advance,

#2

You can build OpenCV with IPP for a performance boost (how much I’m not sure) with the instructions here:

http://opencv.willowgarage.com/wiki/InstallGuide

it requires that you build your own OpenCV instead of using the .a included in ofxOpenCV.

#3

Thanks for the answer,

But unfortunately the result is the same.
The detection time is more than 600ms when I try the facedetect.cpp example provided by Opencv 2.1.

The following procedure has been followed (works also for ubuntu 10.10):
http://www.w-bremer.de/en/howtos/72-opencv-ubuntu-ipp

Here is the cmake options and result:

root@desktop:/var/tmp/opencv.build# cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D BUILD_PYTHON_SUPPORT=ON OPENCV_BUILD_3RDPARTY_LIBS=ON -D USE_SSE=ON -D USE_SSE2=ON -D USE_SSE3=ON -D USE_IPP=ON -D IPP_PATH=/opt/intel/ipp/6.1.2.051/em64t/sharedlib -D WITH_TBB=YES -D WITH_UNICAP=YES …/OpenCV-2.1.0
– Detected version of GNU GCC: 44 (404)
– checking for module ‘libunicap’
– found libunicap, version 0.9.5
– checking for module ‘libucil’
– found libucil, version 0.9.5
– checking for module ‘libdc1394-2’
– package ‘libdc1394-2’ not found
– checking for module ‘libdc1394’
– package ‘libdc1394’ not found
– Use INCLUDE: /usr/lib/python2.6/dist-packages/numpy/core/include
– TBB detected:
CMake Warning (dev) at CMakeLists.txt:513 (link_directories):
This command specifies the relative path

as a link directory.

Policy CMP0015 is not set: link_directories() treats paths relative to the
source dir. Run “cmake --help-policy CMP0015” for policy details. Use the
cmake_policy command to set the policy and suppress this warning.
This warning is for project developers. Use -Wno-dev to suppress it.

– IPP detected: TRUE
– Parsing ‘cvconfig.h.cmake’
running mkdir -p “/var/tmp/opencv.build/unix-install/” 2>&1

– General configuration for opencv 2.1.0 =====================================

– Built as dynamic libs?: ON
– Compiler:
– C++ flags (Release): -Wall -Wno-long-long -pthread -ffunction-sections -O3 -DNDEBUG -fomit-frame-pointer -O3 -ffast-math -msse -msse2 -DNDEBUG
– C++ flags (Debug): -Wall -Wno-long-long -pthread -ffunction-sections -g -O0 -DDEBUG -D_DEBUG
– Linker flags (Release):
– Linker flags (Debug):

– GUI:
– GTK+ 2.x: 1
– GThread: 1

– Image I/O:
– JPEG: TRUE
– PNG: TRUE
– TIFF: TRUE
– JPEG 2000: TRUE

– Video I/O:
– DC1394 1.x: 0
– DC1394 2.x: 0
– FFMPEG: 1
– codec: 1
– format: 1
– util: 1
– swscale: 1
– gentoo-style: 1
– GStreamer: 1
– UniCap: 1
– PvAPI:
– V4L/V4L2: Using libv4l
– Xine: FALSE

– Interfaces:
– Old Python: 0
– Python: ON
– Python interpreter: /usr/bin/python2.6
– Python numpy: 1
– Use IPP: /opt/intel/ipp/6.1.2.051/em64t/sharedlib
– Use TBB: YES
– Build Documentation 0

– Install path: /usr/local

– cvconfig.h is in: /var/tmp/opencv.build

–
– Configuring done
– Generating done
– Build files have been written to: /var/tmp/opencv.build

On mac and win32 it works well with nearly same hardware.

Did I miss a library or anything else??

Any clue will be appreciated…