Integrate SimdUnicode for AVX-512

[EgorBo]

Contributes to #103781, only for AVX-512, other ISAs can be added if/once this is approved/merged.

I did some clean up, like replacing some SIMD apis with cross-platform ones/operators. Btw, I don't believe that ISimdVector can be used here. Also, I removed the initial "skip ASCII data" part since we already have a work horse for that.

cc @lemire, @Nick-Nuon let me know if you want to change something (including credits in THIRD-PARTY-NOTICES.TXT)

TODO: do some ad-hoc testing, make sure test coverage is good

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-text-encoding
See info in area-owners.md if you want to be subscribed.

[EgorBo]

@EgorBot -intel -amd

using System.Text;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;
using System.Text.Unicode;

BenchmarkRunner.Run<Bench>(args: args);

public class Bench
{
    public static IEnumerable<byte[]> GetUtf8BytesData()
    {
        // Chinese "Lorem Ipsum"
        var utf8 = "唐聞球方五保査禁答近確掲著協世好知長。育乗江校上価話戒宏口自森特室堂討。陸迎奔必秋最量注好枚挑周。間父癒曲在近真権幕覧超持樹件芸保展島船点。齢度約治末価埼坂内辞千故資接藤雨約宿県。定戻業担伸立発告敗家響意球禎。呼真局験善体続得新税知群孫大場。変省創与毎容開拡作北経眺間。樹野市現館開分供同南費海。投以画露両装知全茨済力上速田弘変掲材保内。王野嗅結択芸合験覧託委致就近資。励意親者著識連愚戦親能精球信相準大避一。民覧過走最国転開社加砲者度座図。提著学月牟止百県意能宝質約投分記加。中長塚相選暇版経田経問下訟全報府。要事集細両体要特義点必周優載治山集摘。手機掛果題銀料新政庁分堀住画禁信。味表柄読必望著後入協攻末源安 案志検江水口宿言京並属需就一生断導。通崎楽大最放新属健戦維議本金部兜素定市船"u8.ToArray();
        yield return utf8.AsSpan(0, 1000).ToArray();
        yield return utf8.AsSpan(0, 500).ToArray();
        yield return utf8.AsSpan(0, 250).ToArray();
        yield return utf8.AsSpan(0, 100).ToArray();
    }

    [Benchmark]
    [ArgumentsSource(nameof(GetUtf8BytesData))]
    public int GetUtf8Bytes(byte[] str) => Encoding.UTF8.GetCharCount(str);

    public static IEnumerable<byte[]> ValidateUtf8Data()
    {
        // ru-RU "Lorem Ipsum"
        var utf8 = "Лорем ипсум долор сит амет, хас тале феугаит ех, мел дицит сонет сцрипта ид? Еррорибус темпорибус адверсариум про те, видит ностер хас не, яуод феугаит цу ест. Но дицунт рецусабо диссентиас цум, оптион евертитур ан вих. Но мел антиопам молестиае, продессет абхорреант витуператорибус ат сит, дицант глориатур персецути при еу. При еяуидем пхаедрум рецусабо ех, не вим ерант вертерем Ехерци семпер те нец. Ид нолуиссе детерруиссет нам, яуо ан адхуц дицит пертинациа, мел тота цлита цомпрехенсам ид? Ид аугуе граецис еффициенди вис, ат анимал фиерент инструцтиор пер, не виде еффициенди при!"u8.ToArray();
        yield return utf8.AsSpan(0, 1000).ToArray();
        yield return utf8.AsSpan(0, 500).ToArray();
        yield return utf8.AsSpan(0, 250).ToArray();
        yield return utf8.AsSpan(0, 100).ToArray();
    }

    [Benchmark]
    [ArgumentsSource(nameof(ValidateUtf8Data))]
    public bool ValidateUtf8(byte[] str) => Utf8.IsValid(str);
}

[EgorBot]

Benchmark results on Intel

BenchmarkDotNet v0.13.12, Ubuntu 22.04.4 LTS (Jammy Jellyfish)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 16 logical and 8 physical cores
  Job-ARQQWD : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-JZIDFW : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

Method	Toolchain	str	Mean	Error	Ratio
GetUtf8Bytes	Main	Byte[1000]	238.02 ns	1.036 ns	1.00
GetUtf8Bytes	PR	Byte[1000]	117.88 ns	0.013 ns	0.50
GetUtf8Bytes	Main	Byte[100]	69.25 ns	0.287 ns	0.29
GetUtf8Bytes	PR	Byte[100]	61.40 ns	0.437 ns	0.26
GetUtf8Bytes	Main	Byte[250]	108.00 ns	0.512 ns	0.45
GetUtf8Bytes	PR	Byte[250]	121.58 ns	0.433 ns	0.51
GetUtf8Bytes	Main	Byte[500]	160.09 ns	0.493 ns	0.67
GetUtf8Bytes	PR	Byte[500]	125.87 ns	0.509 ns	0.53
ValidateUtf8	Main	Byte[1000]	342.05 ns	0.111 ns	1.00
ValidateUtf8	PR	Byte[1000]	114.51 ns	0.011 ns	0.33
ValidateUtf8	Main	Byte[100]	35.04 ns	0.007 ns	0.10
ValidateUtf8	PR	Byte[100]	23.51 ns	0.008 ns	0.07
ValidateUtf8	Main	Byte[250]	76.36 ns	0.105 ns	0.22
ValidateUtf8	PR	Byte[250]	73.53 ns	0.051 ns	0.21
ValidateUtf8	Main	Byte[500]	169.19 ns	0.008 ns	0.49
ValidateUtf8	PR	Byte[500]	84.19 ns	0.015 ns	0.25

BDN_Artifacts.zip

[EgorBot]

Benchmark results on Amd

BenchmarkDotNet v0.13.12, Ubuntu 22.04.4 LTS (Jammy Jellyfish)
AMD EPYC 7763, 1 CPU, 16 logical and 8 physical cores
  Job-GWHBIB : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX2
  Job-LYEYRH : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX2

Method	Toolchain	str	Mean	Error	Ratio
GetUtf8Bytes	Main	Byte[1000]	211.16 ns	0.075 ns	1.00
GetUtf8Bytes	PR	Byte[1000]	100.20 ns	0.025 ns	0.47
GetUtf8Bytes	Main	Byte[100]	67.82 ns	0.137 ns	0.32
GetUtf8Bytes	PR	Byte[100]	61.96 ns	0.298 ns	0.29
GetUtf8Bytes	Main	Byte[250]	97.97 ns	0.077 ns	0.46
GetUtf8Bytes	PR	Byte[250]	109.18 ns	0.098 ns	0.52
GetUtf8Bytes	Main	Byte[500]	150.22 ns	0.184 ns	0.71
GetUtf8Bytes	PR	Byte[500]	113.65 ns	0.109 ns	0.54
ValidateUtf8	Main	Byte[1000]	331.35 ns	0.245 ns	1.00
ValidateUtf8	PR	Byte[1000]	95.34 ns	0.021 ns	0.29
ValidateUtf8	Main	Byte[100]	36.10 ns	0.032 ns	0.11
ValidateUtf8	PR	Byte[100]	22.56 ns	0.009 ns	0.07
ValidateUtf8	Main	Byte[250]	77.43 ns	0.056 ns	0.23
ValidateUtf8	PR	Byte[250]	60.51 ns	0.034 ns	0.18
ValidateUtf8	Main	Byte[500]	167.75 ns	0.581 ns	0.51
ValidateUtf8	PR	Byte[500]	73.30 ns	0.022 ns	0.22

BDN_Artifacts.zip

[lemire]

Feel free to adapt as needed! :-)

[gfoidl]

Is this better for the registers?

[gfoidl]

@EgorBo another try 😉

[huoyaoyuan]

Question regarding the PR title: it seems using AVX2 (256), not AVX-512

[lemire]

@huoyaoyuan I think that's a good point. This does seem to be AVX2 (which is not a bad idea).