Programming architecture

expression fast calculations: Programming Architecture

programming architecture

• All data is managed in arrays and data operations are vectorized.
• Large arrays are segmented (or tiled in case of 2d raster data) in order to parallel process data segments in separate threads, aka MT1
• Different independent operations are executed in parallel, aka MT2, Multi Threading,
• interest counting and a configurable CalcCache for storing intermediate results for as long as needed, but not longer.
• Use of C++ templates, STL, BOOST, GDAL.
• Management of run time properties of datasets to select the fastest algorithms, such as counting-sort and comparable methods for modus, join etc.
• Retain intermediate resources for reuse in repeatedly called functions, such as in: poly2grid.
• Usage of Intel’s Math Kernel Library for fast convolution, boost::uBlas for matrix operations.
• calculations can be segmented functionally and their results merged afterwards to run on multiple machines.

Under Study:

• C++ AMP for deploying GPU’s see http://wiki.objectvision.nl/index.php/Parallel_Processing_and_GPU_Acceleration and https://docs.microsoft.com/en-us/cpp/parallel/amp/cpp-amp-overview?view=vs-2019 GeoDMS is now compiled witt VS 2017.
• virtual data segments with segment level ownership and lifetime, aka MT3
• a RUST like unique write and shared read owners of intermediate results, value-based-calculating.