Triton

A language and compiler for writing highly efficient custom Deep-Learning primitives. The aim is to provide an open-source environment to write fast code at higher productivity than CUDA, but also with higher flexibility than other existing DSLs.