Optimize opencv builds #790

fnoop · 2018-10-19T22:07:56Z

Bump to opencv 4 and optimise for x86, raspberry and tegra platforms

fnoop · 2018-10-23T07:45:56Z

Aruco compiler flags for optimised builds:
fnoop/aruco@db2feab#diff-d7ed762694f84a7b37a09f7dde5fa6aaR118

fnoop · 2018-10-23T07:49:31Z

Useful info from vision_landing
goodrobots/vision_landing#95

fnoop · 2018-10-23T08:52:24Z

https://raspberrypi.stackexchange.com/a/43234
lol

fnoop · 2018-10-23T10:13:10Z

https://www.theimpossiblecode.com/blog/build-faster-opencv-raspberry-pi3/
https://www.theimpossiblecode.com/blog/faster-opencv-smiles-tbb/

fnoop · 2018-10-23T10:13:17Z

Raspberry:

Look into VFPV4 and VFPV5 flags: -mfpu=neon-vfpv4
-D BUILD_TESTS=OFF -D BUILD_PERF_TESTS=OFF -D OPENCV_ENABLE_NONFREE=ON

Can we use -fast-math?

Note on this thread (note it's old version of gcc):
https://www.raspberrypi.org/forums/viewtopic.php?t=107203

And I complied your code with both gcc-5.0 and gcc-4.8 and use the flag: -mcpu=cortex-a7
-mfloat-abi=hard -mfpu=neon-vfpv4. I find the NEON version of the code runs slower on both gcc-5.0 and gcc-4.8. And the neon version on gcc-4.8 runs fast than gcc-5.0.
if I only use the flag -mfpu=neon-vfpv4,the code complied on gcc-4.8 is about 300% percent slower.

Good thread:
https://www.raspberrypi.org/forums/viewtopic.php?t=144115

fnoop · 2018-10-23T10:36:42Z

Look into opencl:
https://github.com/doe300/VC4CL

fnoop · 2018-10-23T11:05:06Z

https://github.com/opencv/opencv/wiki/CPU-optimizations-build-options

fnoop · 2019-05-02T10:45:17Z

https://github.com/thortex/rpi3-opencv/
doe300/VC4CL#29

fnoop · 2020-05-09T11:18:17Z

Linked to #963, which deals with optimizing for nvidia tegra platform

fnoop · 2020-05-14T07:05:35Z

Raspberry cmake:

-- General configuration for OpenCV 4.3.0 =====================================
--   Version control:               4.3.0
--
--   Extra modules:
--     Location (extra):            /srv/maverick/var/build/opencv_contrib/modules
--     Version control (extra):     4.3.0
--
--   Platform:
--     Timestamp:                   2020-05-13T23:04:46Z
--     Host:                        Linux 4.19.97-v7l+ armv7l
--     CMake:                       3.13.4
--     CMake generator:             Unix Makefiles
--     CMake build tool:            /usr/bin/make
--     Configuration:               Release
--
--   CPU/HW features:
--     Baseline:                    VFPV3 NEON
--       requested:                 DETECT
--       required:                  VFPV3 NEON
--   C/C++:
--     Built as dynamic libs?:      YES
--     C++ standard:                11
--     C++ Compiler:                /usr/bin/c++  (ver 8.3.0)
--     C++ flags (Release):         -fsigned-char -W -Wall -Werror=return-type -Werror=non-virtual-dtor -Werror=address -Werror=sequence-point -Wformat -Werror=format-security -Wmissing-declarations -Wundef -Wini
t-self -Wpointer-arith -Wshadow -Wsign-promo -Wuninitialized -Winit-self -Wsuggest-override -Wno-delete-non-virtual-dtor -Wno-comment -Wimplicit-fallthrough=3 -Wno-strict-overflow -fdiagnostics-show-option -pthre
ad -fomit-frame-pointer -ffunction-sections -fdata-sections  -mfpu=neon -fvisibility=hidden -fvisibility-inlines-hidden -O3 -DNDEBUG  -DNDEBUG
--     C++ flags (Debug):           -fsigned-char -W -Wall -Werror=return-type -Werror=non-virtual-dtor -Werror=address -Werror=sequence-point -Wformat -Werror=format-security -Wmissing-declarations -Wundef -Wini
t-self -Wpointer-arith -Wshadow -Wsign-promo -Wuninitialized -Winit-self -Wsuggest-override -Wno-delete-non-virtual-dtor -Wno-comment -Wimplicit-fallthrough=3 -Wno-strict-overflow -fdiagnostics-show-option -pthre
ad -fomit-frame-pointer -ffunction-sections -fdata-sections  -mfpu=neon -fvisibility=hidden -fvisibility-inlines-hidden -g  -O0 -DDEBUG -D_DEBUG
--     C Compiler:                  /usr/bin/cc
--     C flags (Release):           -fsigned-char -W -Wall -Werror=return-type -Werror=non-virtual-dtor -Werror=address -Werror=sequence-point -Wformat -Werror=format-security -Wmissing-declarations -Wmissing-pro
totypes -Wstrict-prototypes -Wundef -Winit-self -Wpointer-arith -Wshadow -Wuninitialized -Winit-self -Wno-comment -Wimplicit-fallthrough=3 -Wno-strict-overflow -fdiagnostics-show-option -pthread -fomit-frame-poin
ter -ffunction-sections -fdata-sections  -mfpu=neon -fvisibility=hidden -O3 -DNDEBUG  -DNDEBUG
--     C flags (Debug):             -fsigned-char -W -Wall -Werror=return-type -Werror=non-virtual-dtor -Werror=address -Werror=sequence-point -Wformat -Werror=format-security -Wmissing-declarations -Wmissing-pro
totypes -Wstrict-prototypes -Wundef -Winit-self -Wpointer-arith -Wshadow -Wuninitialized -Winit-self -Wno-comment -Wimplicit-fallthrough=3 -Wno-strict-overflow -fdiagnostics-show-option -pthread -fomit-frame-poin
ter -ffunction-sections -fdata-sections  -mfpu=neon -fvisibility=hidden -g  -O0 -DDEBUG -D_DEBUG
--     Linker flags (Release):      -Wl,--gc-sections -Wl,--as-needed
--     Linker flags (Debug):        -Wl,--gc-sections -Wl,--as-needed
--     ccache:                      NO
--     Precompiled headers:         NO
--     Extra dependencies:          dl m pthread rt

--   OpenCV modules:
--     To be built:                 alphamat aruco bgsegm bioinspired calib3d ccalib core datasets dnn dnn_objdetect dnn_superres dpm face features2d flann freetype fuzzy gapi hdf hfs highgui img_hash imgcodecs i
mgproc intensity_transform line_descriptor ml objdetect optflow phase_unwrapping photo plot python2 python3 quality rapid reg rgbd saliency shape stereo stitching structured_light superres surface_matching text t
racking ts video videoio videostab xfeatures2d ximgproc xobjdetect xphoto
--     Disabled:                    world
--     Disabled by dependency:      -
--     Unavailable:                 cnn_3dobj cudaarithm cudabgsegm cudacodec cudafeatures2d cudafilters cudaimgproc cudalegacy cudaobjdetect cudaoptflow cudastereo cudawarping cudev cvv java js matlab ovis sfm v
iz
--     Applications:                tests perf_tests examples apps
--     Documentation:               NO
--     Non-free algorithms:         YES
--   Video I/O:
--     DC1394:                      YES (2.2.5)
--     FFMPEG:                      YES
--       avcodec:                   YES (58.35.100)
--       avformat:                  YES (58.20.100)
--       avutil:                    YES (56.22.100)
--       swscale:                   YES (5.3.100)
--       avresample:                YES (4.0.0)
--     GStreamer:                   YES (1.14.4)
--     OpenNI2:                     NO
--     v4l/v4l2:                    YES (linux/videodev2.h)
--
--   Parallel framework:            TBB (ver 2020.0 interface 11100)
--
--   Trace:                         YES (with Intel ITT)
--
--   Other third-party libraries:
--     Lapack:                      YES (/srv/maverick/software/openblas/lib/libopenblas.so)
--     Eigen:                       YES (ver 3.3.7)
--     Custom HAL:                  YES (carotene (ver 0.0.1))
--     Protobuf:                    build (3.5.1)
--
--   OpenCL:                        YES (no extra features)
--     Include path:                /srv/maverick/var/build/opencv/3rdparty/include/opencl/1.2
--     Link libraries:              Dynamic load

fnoop · 2020-05-14T09:33:15Z

On raspbian buster, gcc (gcc version 8.3.0 (Raspbian 8.3.0-6+rpi1)) seems to choose good optimisations:

[flight] [mav@maverick-raspberry ~/software/opencv]$ gcc -mcpu=native -march=native -Q --help=target
The following options are target specific:
  -mabi=                      		aapcs-linux
  -march=                     		armv8-a+crc+simd
  -mcpu=                      		cortex-a72
  -mfloat-abi=                		hard
  -mfpu=                      		vfp

-mfpu perhaps could be changed. Available fpu options:
Known ARM FPUs (for use with the -mfpu= option): auto crypto-neon-fp-armv8 fp-armv8 fpv4-sp-d16 fpv5-d16 fpv5-sp-d16 neon neon-fp-armv8 neon-fp16 neon-vfpv3 neon-vfpv4 vfp vfp3 vfpv2 vfpv3 vfpv3-d16 vfpv3-d16-fp16 vfpv3-fp16 vfpv3xd vfpv3xd-fp16 vfpv4 vfpv4-d16
Important to note that NEON is not an fpu replacement but a SIMD implementation, and it costs to transfer data back and forth to NEON. -mfpu=neon-vfpv4 might be a good option.

This suggests -mfpu=neon-fp-armv8:
https://www.raspberrypi.org/forums/viewtopic.php?t=244095
and this:
https://forums.gentoo.org/viewtopic-t-1098908-start-0.html
and:
https://github.com/thortex/rpi3-opencv

Excellent reference:
https://gist.github.com/fm4dd/c663217935dc17f0fc73c9c81b0aa845

fnoop · 2020-05-14T10:11:43Z

Note OpenCV Release build sets -O3 which automatically turns on -ftree-vectorize

fnoop · 2020-05-14T12:08:04Z

https://answers.opencv.org/question/230199/correct-method-to-specify-mfpu-compile-flag/
opencv/opencv#17292

fnoop added a commit that referenced this issue Oct 22, 2018

Move to opencv 4, optimise builds, add openvino #790 #791

2445c66

fnoop added this to the 2.0 milestone Aug 10, 2019

fnoop modified the milestones: 2.0, 1.2 May 9, 2020

fnoop added a commit that referenced this issue May 16, 2020

opencv compile optimize for rpi #790

7935a85

fnoop added the Documentation label May 17, 2020

fnoop added a commit that referenced this issue Jul 23, 2020

opencv compile optimize for rpi #790

c6b8416

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Optimize opencv builds #790

Optimize opencv builds #790

fnoop commented Oct 19, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018 •

edited

Loading

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented May 2, 2019

fnoop commented May 9, 2020 •

edited

Loading

fnoop commented May 14, 2020 •

edited

Loading

fnoop commented May 14, 2020 •

edited

Loading

fnoop commented May 14, 2020

fnoop commented May 14, 2020 •

edited

Loading

Optimize opencv builds #790

Optimize opencv builds #790

Comments

fnoop commented Oct 19, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018 • edited Loading

fnoop commented Oct 23, 2018

fnoop commented Oct 23, 2018

fnoop commented May 2, 2019

fnoop commented May 9, 2020 • edited Loading

fnoop commented May 14, 2020 • edited Loading

fnoop commented May 14, 2020 • edited Loading

fnoop commented May 14, 2020

fnoop commented May 14, 2020 • edited Loading

fnoop commented Oct 23, 2018 •

edited

Loading

fnoop commented May 9, 2020 •

edited

Loading

fnoop commented May 14, 2020 •

edited

Loading

fnoop commented May 14, 2020 •

edited

Loading

fnoop commented May 14, 2020 •

edited

Loading