Paweł Bylica

Skip nop instructions

7

This was originally included in #622. I can't get good measurement of the performance on Haswell (I will try other architectures later). But this is rather no-brainer as we simply...

Invoke refactoring

5

Depends on #602.

Optimize internal calls by passing only stack end for arguments

1

Instead of passing the reference to the call arguments (a span on the top of the stack) just pass the pointer to the stack end. Also, don't get the result...

Depends on #638.

With this change the `sp` is kept in a register, instead on the stack as previously. The code gets ~25% smaller. But still execution time is the same.

Add PGO builds

2

The PGO builds give good boost to the performance, but we need to select out training set. ``` Comparing o/master to ../pgouse/o/master Benchmark Time CPU Time Old Time New CPU...

https://github.com/cyb70289/utf8