opencv neon 최적화

엥? 2.1에서 단순하게 neon 활성화 하는데 이렇게 오른다고? ㄷㄷㄷ

OpenCv 2.1

– cvMatchTemplate() algorithm as the test case

cvMatchTemplate( img, tpl, res, CV_TM_CCORR_NORMED );

Flags:

CFLAGS += -pipe -O3

• Result: ~19.35sec @800Mhz

Flags:

CFLAGS += -pipe -O3 -mfpu=neon -ftree-vectorize -mfloat-abi=softfp

• Result: ~4.91sec @800Mhz

~75% increase in performance

[링크 : http://elinux.org/images/1/1c/Optimizing_the_Embedded_Platform_Using_OpenCV.pdf]

[링크 : http://answers.opencv.org/question/33940/are-these-functions-accelerated-by-arm-neon/]

    [링크 : http://answers.opencv.org/question/17845/open-source-neon-optimizations/]

NEON intrinsics 노가다..

[링크 : http://web.stanford.edu/class/cs231m/lectures/lecture-4-opencv.pdf]

+

g++ -lopencv_core -lopencv_video -lopencv_highgui -lopencv_imgproc -mfpu=neon -ftree-vectorize -mfloat-abi=hard MatchTemplate_Demo.cpp -o b.out

./b.out 1.jpg 2.jpg

음.. 차이가 없네...

openCV 3.0 부터 제한적으로 NEON을 옵션으로 제공하는 듯...

[링크 : http://answers.opencv.org/question/35233/neon-support-in-opencv/]

[링크 : http://opencv.org/opencv-3-0.html]