---------- Forwarded message --------- From: Jakub Golinowski <jakub.golinow...@gmail.com> Date: Sun, Jun 10, 2018, 10:47 AM Subject: [STE||AR GSoC-students] HPX Backend for OpenCV - First Phase Report To: <gsoc-stude...@stellar-group.org>
Dear Ste||ar GSoC community, I would like to submit short report of my progress during the first phase of GSoC. The main achievement of this moth is a working HPX backend for OpenCV. During the development it was discovered that the parallel_for_ loop which is the base for all parallel execution of OpenCV's jobs is not only exposed to the user but is also used indirectly by some OpenCV filters. In the latter case there is a workload based partitioning introduced based on the domain-specific knowledge of the particular computer vision operation. Because of that the simple backend version with partitioning done fully by HPX was extended to the version supporting partitioning propsed by OpenCV. What is more there is a question of when to start the HPX runtime. For now there are to versions developed: - simple backend that assumes that HPX runtime is on, - start-stop backend that is starting and stopping the runtime for each parallel_for call. After the benchmarking experiments it is decided that the final version of the backend will be a "smart start-stop". It will perform all the necessary checks to determine if the backend is running and then based on that either start the runtime (and of course stop it later) or directly proceed to hpx calls. The backend itself is developed in two branches of my fork of OpenCV repository (hpx_backend <https://github.com/Jakub-Golinowski/opencv/tree/hpx_backend> and hpx_backend_eval <https://github.com/Jakub-Golinowski/opencv/tree/hpx_backend_eval>). The eval branch is more up-to-date and has additional cmake options which enable to choose between versions of the backend: - with or without support for OpenCV partitioning, - simple mode (assume HPX runs) or start-stop mode. Small example applications using HPX, application used for benchmarking (opencv_mandelbrot) and necessary bash and python scripts can be found in this repository: https://github.com/Jakub-Golinowski/opencv_hpx_backend In the next 2 months the backend will be developed in its final version (most probably smart start-stop) and the efforts to merge it to OpenCV repository will be started. Apart from that, the example application combining OpenCV with freshly developed HPX backend, access to webcam and QT GUI will be developed. Best regards, Jakub Golinowski _______________________________________________ gsoc-students mailing list gsoc-stude...@stellar-group.org https://mail.cct.lsu.edu/mailman/listinfo/gsoc-students
_______________________________________________ hpx-users mailing list hpx-users@stellar.cct.lsu.edu https://mail.cct.lsu.edu/mailman/listinfo/hpx-users
