Current tally of high-performance, deep-learning-oriented DSLs/IRs/compilers, in no particular order:<p>- TensorComprehensions (Facebook): <a href="https://github.com/facebookresearch/TensorComprehensions" rel="nofollow">https://github.com/facebookresearch/TensorComprehensions</a><p>- XLA (Google): <a href="https://www.tensorflow.org/performance/xla/" rel="nofollow">https://www.tensorflow.org/performance/xla/</a><p>- taco (MIT): <a href="http://tensor-compiler.org/" rel="nofollow">http://tensor-compiler.org/</a><p>- DLVM (UIUC): <a href="http://dlvm.org/" rel="nofollow">http://dlvm.org/</a><p>- nGraph (Intel): <a href="http://ngraph.nervanasys.com/docs/cpp/" rel="nofollow">http://ngraph.nervanasys.com/docs/cpp/</a><p>- TVM (DMLC): <a href="https://github.com/dmlc/tvm" rel="nofollow">https://github.com/dmlc/tvm</a><p>Honorable mention to Julia (<a href="http://julialang.org" rel="nofollow">http://julialang.org</a>) as well.