Here is the test: <https://github.com/altavir/kmat...

# mathematics

[altavir]

Here is the test: https://github.com/altavir/kmath/blob/dev/benchmarks/src/main/kotlin/scientifik/kmath/structures/BufferNDStructureBenchmark.kt Did not wrap it in JMH yet since JMH gradle plugin is not very friendly with rapid changes in kotlin code. For now for 1000x1000 real matrix kotlin works abut 10% faster than numpy cell. Inlined, but not specialized implementation is a little bit slower and boxing implementation is 2x slower (which is also not very much since in real life). There is also a lazy implementation on coroutines which will dramatically increase performance when multiple nd-operations are performed subsequently and/or you do not need all of the elements of resultion matrix, but it requires further work and some testing.