Zero-extend int

[buybackoff]

Following Mendel's link to "Peanut butter" https://devblogs.microsoft.com/dotnet/performance-improvements-in-net-6/#peanut-butter.

This gives 2-3% on my machine on the same HT core. The bench with different physical cores became very unstable on my machine: jumping from 1xx to 280 (compared to my previous PR, which works as it did on the same machine). This is not related to this change, it jumps before the change as well.

Adding ExtraWork method, which was removed for some reason, every time is tedious so I have not, but it makes batching behavior more stable. So these numbers need to be double checked on different machines.

[MendelMonteiro]

Yes, well spotted :)

The associated PR is dotnet/runtime#51190 for reference's sake.

[buybackoff]

We could avoid zero extend at all by storing ArrayDataOffset as nint

[MendelMonteiro]

It would make the code easier to understand as well and is exactly what nint was intended for

[ocoanet]

Thank you for spotting this.

|  Method |      Mean |     Error |    StdDev |
|-------- |----------:|----------:|----------:|
|   Read1 | 0.8264 ns | 0.0014 ns | 0.0013 ns |
|   Read2 | 0.7641 ns | 0.0013 ns | 0.0012 ns |

Not a huge win but this method method was already quite optimized.
I take every perf improvement that has a reasonable impact on complexity 🙂

I will add back ExtraWork to the perf test if it helps stabilizing results on your machine.
However, for such low level improvements, benchmarks might be more appropriate.

[buybackoff]

Adding ExtraWork is extra work 😄
I did not need it that much actually this time. It helps control batching, I could add it next time.

[ocoanet]

I get the same perf improvements using nint (or nuint) for ArrayDataOffset and index. For some reason the Conv_U for ArrayDataOffset does not change anything, only the one for index is required, but it feels more coherent to apply the conversion for both.

I will accept the PR and then apply the same trick to other InternalUtil, unless you want to change them too?

[buybackoff]

The change after adding the second Conv_U for ArrayDataOffset was quite small, close to noise, but after around 5 runs with and without it the result looked better with it. Maybe just noise. For JIT it a constant, probably it knows what to do itself.

apply the same trick to other InternalUtil, unless you want to change them too?

Go ahead!

[ocoanet]

Just in time for V5.0!