einsum-derive
Proc-macro based einsum implementation for ndarray crate
```rust
use ndarray::array;
use einsum_derive::einsum;
let a = array![
[1.0, 2.0],
[3.0, 4.0]
];
let b = array![
[1.0, 2.0],
[3.0, 4.0]
];
let c = einsum!("ij,jk->ik", a, b);
assert_eq!(c, array![
[6.0, 8.0],
[12.0, 16.0]
]);
```
This proc-macro wil compile the input subscripts "ij,jk->ik"
to generate Rust code executing corresponding operation.
Status / Roadmap
- [x] [Optimal contraction by memorizing partial summation to reduce computation order.](https://github.com/termoshtt/einsum-derive/pull/18)
- For example, three matrix multiplication
ij,jk,kl->il is factorized into
two successive einsum ij,jk->ik and ik,kl->il.
- [ ] [Call BLAS routines if possible](https://github.com/termoshtt/einsum-derive/issues/22)
- [ ] [Ellipsis
... support](https://github.com/termoshtt/einsum-derive/issues/7)
Architecture
- einsum-derive
crate is proc-macro crate to provide above
einsum! macro.
- einsum-codegen
crate implements parser for the einsum subscripts like
ij,jk->ik and generates Rust code.
Links
- numpy.einsum is well-known einsum implementation in Python.
- opt_einsum is an implementation for optimizing einsum computation for NumPy and other linear algebra packages.
- oracleofnj/einsum is a runtime-based implementation of einsum for rust-ndarray
License
© 2022 Toshiki Teramura (@termoshtt)
This project is licensed under either of
- Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)
at your option.