4 ms·
The xtensor vision: C++ for R, Python and Julia
- oscar6echo 8y agovery VERY sound and compelling vision !
- mattip 8y agoWhile I like the intention of “write once use anywhere”, converting code from python, r, julia to c++ is not an easy task. I would like to see more tracing JITs that could capture the essence of the operations, store the result, and somehow convert it to a more permanent representation that could be converted between languages. PyPy and Julia have such a JIT compiler, but work needs to be done to enable persistance
- droelf 8y agoThere is definitely an intention to write some sort of JIT-to-xtensor compiler at some point. I've personally worked on a "simple" (1000 lines of Python) NumPy-to-Armadillo (C++) JIT compiler before, and we're in close contact with the author of Pythran, which also compiles to C++. So hopefully at some point you'll be able to go from NumPy to all the languages, and NumPy becomes a high-level array operation description language with multiple backends.