Timeloop
Timeloop is an infrastructure for modeling, mapping and generating code for dense and sparse tensor-algebra workloads on a range of hardware accelerator architectures.
Timeloop combines two components: a fast analytical model that emulates a range of accelerator architecture designs to project performance and energy use, and a mapper that searches the space of possible mappings for a tensor-algebra problem on a given architecture, then generates code from the result.
Version 2.0, also called Sparseloop, adds stochastic modeling of compressed-sparse tensor algebra, and version 3.0, called Ruby, adds support for imperfectly factorized mappings, spatial skews and flattened mappings. The project depends on the isl and barvinok libraries, and its documentation site hosts installation instructions, usage guides and a hands-on tutorial recommended for new users before working with the tool.