[polyml] Benchmarks: Poly/ML 5.6.1 vs 5.6.0

On 13/02/2017 11:37, Kostirya wrote:

Many thanks! Poly/ML 5.6.1 FFI is faster than in 5.6.0!

That's good.

So my server (redis-sharding) fasted on 30-80% on Poly/ML 5.6.1. I am also thinking to add kevent and epoll calls into libpolyml :-)

But I also noticed that Int32 is very slower in 5.6.1. Int32 is slower than LargeInt. But I don't understand why is it. :-(

Int32 is implemented in terms of LargeInt and simply adds a test to each operation that could return a result that can't be represented in 32-bits. On a 32-bit machine the test will always involve a call into the run-time system. It needs better implementation but it's rather low on the list of priorities at the moment.

David