MAPS is an open-source, header-only C++ CUDA template library for automatic multi-GPU programming and optimization of GPU kernels. The framework leverages memory access patterns to provide near-optimal performance on various architectures.
The Research Software Company working with Tal Ben Nun, of The Hebrew University of Jerusalem, wrapped the Memory Access Patterns open source C++ library in Python. The project made efficient GPU programming available to a significantly wider audience.
The project was implemented in Python by processing Python ASTs and generating C++/CUDA code that is automatically compiled with PyCUDA.
