Add AVX2 version of adler32

[brianpopow]

Prerequisites

• I have written a descriptive pull-request title
• I have verified that there are no overlapping pull-requests open
• I have verified that I am following the existing coding patterns and practice as demonstrated in the repository. These follow strict Stylecop rules 👮.
• I have provided test coverage for my change (where applicable)

Description

This PR adds a AVX2 version of adler method.

Benchmark results:

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.19043.1526 (21H1/May2021Update)
Intel Core i7-6700K CPU 4.00GHz (Skylake), 1 CPU, 8 logical and 4 physical cores
.NET SDK=6.0.200
  [Host]     : .NET 5.0.14 (5.0.1422.5710), X64 RyuJIT
  DefaultJob : .NET 5.0.14 (5.0.1422.5710), X64 RyuJIT


|      Method |       Mean |   Error |  StdDev |
|------------ |-----------:|--------:|--------:|
|    AdlerSse |   160.4 ns | 1.10 ns | 0.92 ns |
|    AdlerAvx |   109.6 ns | 0.52 ns | 0.49 ns |
| AdlerScalar | 1,557.8 ns | 4.71 ns | 4.40 ns |

[JimBobSquarePants]

I kept meaning to migrate this across.

https://github.com/SixLabors/ZlibStream/blob/main/src/ZlibStream/Adler32.cs

Compression is currently broken in that repo (to a point I’m going to have to start again from the beginning of my fork and re-optimize with better tests) but Decompression works well and benchmarked faster at times than the runtime implementation. There’s some fast code in there.

[brianpopow]

I kept meaning to migrate this across.

https://github.com/SixLabors/ZlibStream/blob/main/src/ZlibStream/Adler32.cs

Compression is currently broken in that repo (to a point I’m going to have to start again from the beginning of my fork and re-optimize with better tests) but Decompression works well and benchmarked faster at times than the runtime implementation. There’s some fast code in there.

Ok, I was not aware of this. Will close this PR then.

[JimBobSquarePants]

I’d still compare performance though. Your Adler might still be faster!

[brianpopow]

I’d still compare performance though. Your Adler might still be faster!

The implementation from SixlaborsZlib is faster:

|      Method |        Mean |    Error |   StdDev |
|------------ |------------:|---------:|---------:|
|   AdlerSse2 |   167.95 ns | 1.974 ns | 1.750 ns |
|  AdlerSsse3 |    87.87 ns | 1.716 ns | 3.051 ns |
|    AdlerAvx |   111.50 ns | 0.761 ns | 0.712 ns |
| AdlerScalar | 1,566.36 ns | 5.335 ns | 4.730 ns |

AdlerSsse3 is the SixLabors/ZlibStream version.

After sleeping one night over it, I thought it might be still worth considering merging the AVX version of this PR to ImageSharp until we can make use of https://github.com/SixLabors/ZlibStream/blob/main/src/ZlibStream/Adler32.cs. I think it will take some time until we can make SixLabors/ZlibStream production ready. @JimBobSquarePants what do you think?

[JimBobSquarePants]

I don’t know when I’ll get time to work exclusively on ZlibStream. It’s a minimum of a few weeks full time work to incrementally implement the optimisations again unless someone can spot the issue. So yeah, merge ahead if it’s easier to do so

[brianpopow]

Ok then I think reopening this again makes sense.

[JimBobSquarePants]

LGTM 👍

[brianpopow]

LGTM 👍

ok lets wait for this to be merged when 2.0.1 is done. I think only hotfixes should go into 2.0.1

[brianpopow]

@JimBobSquarePants since we are skipping 2.0.1, this can now also get merged into the main branch, right?

[JimBobSquarePants]

Yeah, merge away 👍

[brianpopow]

I kept meaning to migrate this across.

https://github.com/SixLabors/ZlibStream/blob/main/src/ZlibStream/Adler32.cs

Compression is currently broken in that repo ...

@JimBobSquarePants: The tests in this repo seem to pass and there are encode tests. How did you notice its not working as it should?

[JimBobSquarePants]

Let me check my local copy and push failing tests. It’s definitely broken though and the tests are not adequate. Not related to adler though!