How would you benchmark for "performance regressions" exactly? One library is 
written in Java, other in Nim, you can't compare apples to oranges. I did 
although 
[benchmarked](https://github.com/b3liever/manu/blob/master/experiments/mult_bench.nim)
 different data representations with matrix dot products and "surprise" packed 
storage is faster.

Reply via email to