Windows setting and profiling

[meiqua]

Motivation

Many friends run this repo on Windows but face some problems often. Though you can find answers in issues, we summarize some tips here to make things easier.

Dependency

This repo only depends on Opencv3 (4 is fine too but not tested). You can follow Installation by Using the Pre-built Libraries to install Opencv 3.4.11.
After downloading this repo, use cmake-gui to config, or add files manually. We need c++11 here so it's better to use >= Visual Studio 2017.

Visual Studio Config

1. Build in release mode, set /o2
2. enable SIMD and openMP options
3. You may want to use custom openmp reduction. If you are lazy, just delete the 3 lines that the compiler complains

All Configs:

/MP /GS- /Qpar /GL /W0 /Gy /Zc:wchar_t- /I"C:\Users\meiqua\software\shape_based_matching\MIPP" /I"C:\Users\meiqua\library\opencv\build\include" /I"C:\Users\meiqua\library\opencv\build\include\opencv" /Gm- /O2 /Ob2 /sdl- /Fd"shape_based_matching_test.dir\Release\vc142.pdb" /Zc:inline /fp:fast /D "CMAKE_INTDIR=\"Release\"" /fp:except- /errorReport:prompt /GF /WX- /Zc:forScope- /GR- /arch:AVX2 /Gr /Oy /Oi /MT /openmp /std:c++14 /Fa"shape_based_matching_test.dir\Release\" /nologo /Fo"shape_based_matching_test.dir\Release\" /Ot /Fp"shape_based_matching_test.dir\Release\shape_based_matching_test.pch" /diagnostics:column 

Configs captured:

Profiling

In a i5-10400 PC, using fusion branch, it costs 70ms for 5,000,000 pixels image under Ubuntu16.04.
For windows Visual Studio 2019 with opencv 3.4.11, it cost 70ms too (with custom openmp reduction).