Ondřej Čertík

Implement all NumPy intrinsics via ASR

6

With high performance implementation both at compile time and at runtime.

Implement numpy.size() and numpy.shape()

4

And use the ArraySize intrinsic.

Vectorization

2

We need to implement ASR->ASR passes that implement vectorization. It seems we write a pragma (comment) which will direct the optimizer to apply a given transformation and specify exactly which...

Which modules to implement from the Python standard library

15

The Python standard library has a lot of modules: https://docs.python.org/3/library/index.html, but many of them are rarely used. First we should concentrate on the most commonly used modules, probably: * `sys`...

Example: ```python from numpy import empty, int16 from lpython import i16, i32, Const def spot_print_lpython_array(a: i16[:]) -> None: print(a) def main() -> i32: n : Const[i32] = 15 m :...

Some high priority ideas for GSoC are those that will bring LPython closer to production: * https://github.com/lcompilers/lpython/issues/2470 * https://github.com/lcompilers/lpython/issues/2409 * https://github.com/lcompilers/lpython/issues/2474 * https://github.com/lcompilers/lpython/issues/2471 * https://github.com/lcompilers/lpython/issues/2473 * Get ASR->Fortran finished *...

Improve dictionaries, sets, tuples, lists for all backends

1

We have initial implementation working, we now have to extend it to work for all backends well, support printing, etc.

Ensure more types and cases are handled, that it works with all backends (C and LLVM), that it works in "JIT" mode using the lpython decorator from CPython.

For the next deliverable: * [ ] fastGPT implementation in LPython (https://github.com/lcompilers/lpython/issues/2409) * [ ] Make sure LPython is at least as fast as C++ for every simple benchmark (currently...

Ensure the `@cpythoncall` works with the LLVM backend by refactoring it into an ASR->ASR pass, similar to how the new symbolic ASR->ASR pass works