Libraries
- Full Unicode support, including Codon-native SIMD Unicode primitives and
Unicode-specific compiler optimizations. Learn more in our blog post:
exaloop.io/blog/codon-unicode - Other complementary additions including
bytesandbytearray. - Ordered dictionary type to match CPython (old unordered implementation is
still usable via-unordered-dictflag).
NumPy
- Uses ILP64 BLAS (both OpenBLAS and Accelerate) now by default.
- Numerous other performance improvements including better fusion optimization.
Type system
- Unified many core numeric types (e.g.
intandInt[64]are now the same
type,byteandUInt[8]are now the same type, etc.). - Many types can now be used together without explicit casting (e.g.
u8(1) + i16(2)now works and casts to the wider type automatically).
GPU
- Support for
float16andbfloat16. - Support for more math functions (e.g.
log1p,log1pf). - Improved GPU optimization pipeline.