This repository provides a basic starting point for a Feel++ application including:
✓ Feel++ applications in C++ to use Feel++ and Feel++ toolboxes in src
✓ documentation using asciidoc and antora
✓ python Feel++ notebooks that can be downloaded from the documentation
✓ continuous integration including tests for the C++ applications
✓ docker image generation for the project
The documentation for solar-shading is available at here and you can build on it for your project by enabling the github pages for your repository.
Updating the solar-shading version
The version of the project is defined in the files CMakeLists.txt, docs/antora.yml and docs/package.json.
You need to update with the same version in all files.
Release process
✓ update the version in CMakeLists.txt
✓ update the version in docs/antora.yml
✓ commit the changes with the message "Release vx.y.z". At this point the CI will generate the docker image and push it to docker hub
The current version of solar shading algorithm will not scale and should be fully review.
The bad things that I see are :
Too many submeshes. Here we create a submesh for each building and a submesh for each face of building. : In fact no submesh are required
A BVH is built for each building and then a ray intersection is done by a loop for each building (very bad complexity)
The save on disk is done for each face of building in csv file and apply inside the algo will slow down the execution.
The idea is to build only one BVH containing all faces of the building. Then we do only on loop for each faces and run intersect function of the BVH. If an intersection, we have access to mesh face and so we can have easily the building id or face id (by created an map between : face -> (building id, faceid). Then, we need to reduce the amount of data generated and speed up this write exported these quantity with HDF5 for example. We can build a large vector at the begining and then using EigenMap we can have an access to each matrix without any more memory. The write on disk should be done at the end (only once).
The perf will be significantly improved in sequential. Then we can go the parallelism with MPI and thread.
Some refactoring is required for the SolarMask computation in order to separate task creation to actual task orchestration and enable choosing different frameworks