GF(p^m) array structure

[mhostetter]

Consider making N-D GF(p^m) arrays actually (N + 1)-D arrays over GF(p) under the hood. This would require intercepting indexing calls. This would prevent the continual int_to_poly() and poly_to_int() calls in the arithmetic functions, which I believe take the most time.

Intercepting broadcasting routines will probably be needed as well.

[mhostetter]

Potentially create a custom dtype for finite field types. For GF(p^m) it could be a subarray dtype with m items from GF(p)?

https://numpy.org/devdocs/reference/generated/numpy.dtype.subdtype.html#

[mhostetter]

This would cause too many issues. Extension fields are fast enough.