Add Base64url encoding/decoding

[buyaa-n]

• Base64Url encoding doesn't append padding, therefore encoded byte count differs from Base64 encoding:

  Source byte count	Base64 encoded byte count	Base64Url encoded byte count
  1	4	2
  2	4	3
  3	4	4
  4	8	6
  5	8	7
  6	8	8

• Base64Url decoding ignore whitespace and padding, therefore decodable byte count differs from Base64 decoding, the exact decoding result also depend on isFinalBlock and if padding involved

  Source byte count	Base64 decoded byte count	Base64Url decoded byte count
  1	max 0	max 0
  2	max 0	max 1
  3	max 0	max 2
  4	max 3	max 3
  5	max 3	max 3
  6	max 3	max 4
  7	max 3	max 5
  8	max 6	max 6

Approved API shape:

namespace System.Buffers.Text;

public static class Base64Url
{
    public static int GetMaxDecodedLength(int base64Length);
    public static int GetEncodedLength(int bytesLength);

    public static OperationStatus EncodeToUtf8(ReadOnlySpan<byte> source, Span<byte> destination, out int bytesConsumed, out int bytesWritten, bool isFinalBlock = true);
    public static int EncodeToUtf8(ReadOnlySpan<byte> source, Span<byte> destination);
    public static bool TryEncodeToUtf8(ReadOnlySpan<byte> source, Span<byte> destination, out int charsWritten);
    public static byte[] EncodeToUtf8(ReadOnlySpan<byte> source);

    public static OperationStatus EncodeToChars(ReadOnlySpan<byte> source, Span<char> destination, out int bytesConsumed, out int charsWritten, bool isFinalBlock = true);
    public static int EncodeToChars(ReadOnlySpan<byte> source, Span<char> destination);
    public static bool TryEncodeToChars(ReadOnlySpan<byte> source, Span<char> destination, out int charsWritten);
    public static char[] EncodeToChars(ReadOnlySpan<byte> source);
    public static string EncodeToString(ReadOnlySpan<byte> source);

    public static bool TryEncodeToUtf8InPlace(Span<byte> buffer, int dataLength, out int bytesWritten);

    public static OperationStatus DecodeFromUtf8(ReadOnlySpan<byte> source, Span<byte> destination, out int bytesConsumed, out int bytesWritten, bool isFinalBlock = true);
    public static int DecodeFromUtf8(ReadOnlySpan<byte> source, Span<byte> destination);
    public static bool TryDecodeFromUtf8(ReadOnlySpan<byte> source, Span<byte> destination, out int bytesWritten);
    public static byte[] DecodeFromUtf8(ReadOnlySpan<byte> source);
    public static OperationStatus DecodeFromChars(ReadOnlySpan<char> source, Span<byte> destination, out int charsConsumed, out int bytesWritten, bool isFinalBlock = true);
    public static int DecodeFromChars(ReadOnlySpan<char> source, Span<byte> destination);
    public static bool TryDecodeFromChars(ReadOnlySpan<char> source, Span<byte> destination, out int bytesWritten);
    public static byte[] DecodeFromChars(ReadOnlySpan<char> source);

    public static int DecodeFromUtf8InPlace(Span<byte> buffer);

    public static bool IsValid(ReadOnlySpan<char> base64UrlText);
    public static bool IsValid(ReadOnlySpan<char> base64UrlText, out int decodedLength);
    public static bool IsValid(ReadOnlySpan<byte> utf8Base64UrlText);
    public static bool IsValid(ReadOnlySpan<byte> utf8Base64UrlText, out int decodedLength);
}

Draft PR until below list is completed, for now mainly for checking perf, run tests on different CI legs, and get early feedback.

• Handle '%' in decoding as a valid padding character for Base64Url
• Doc needs improvement
• A few APIs missing unit tests
• Fix failures on ARM CI legs

Fixes #1658

[ghost]

Note regarding the new-api-needs-documentation label:

This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change.

[gfoidl]

Just skimmed over it...
I saw your comment, but for checking perf some of my comments may help (a bit).

[buyaa-n]

/benchmark

[buyaa-n]

Just skimmed over it... I saw your comment, but for checking perf some of my comments may help (a bit).

Thank you, sure perf related feedbacks appreciated, I updated the description, any feedbacks are welcome.

[pr-benchmarks]

Crank Pull Request Bot

/benchmark <benchmark[,...]> <profile[,...]> <component,[...]> <arguments>

Benchmarks:

• micro: .NET Performance micro benchmarks (set filter by adding --variable filter=...; by default filter=*LinqBenchmarks*)
• plaintext: TechEmpower Plaintext Scenario - ASP.NET Platform implementation
• json: TechEmpower JSON Scenario - ASP.NET Platform implementation
• fortunes: TechEmpower Fortunes Scenario - ASP.NET Platform implementation
• fortunes_ef: TechEmpower Fortunes Scenario with EF Core - ASP.NET Platform implementation
• httpclient: HttpClient Benchmark (change HTTP version by adding e.g. --variable httpVersion=3.0; change response size by adding e.g. --variable responseSize=256; default: HTTP/1.1 GET 8K)

Profiles:

• aspnet-citrine-lin: Intel/Linux 28 Cores
• aspnet-perf-lin: Intel/Linux 12 Cores
• aspnet-citrine-amd: Amd/Linux 48 Cores

Components:

• runtime
• libs

Arguments: any additional arguments to pass through to crank, e.g. --variable name=value

[stephentoub]

What guarantees this is in bounds?

[buyaa-n]

This is only called with a buffer that has size of 4

runtime/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Base64Decoder.cs

Lines 498 to 499 in e617b8f

	const int BlockSize = 4;
	Span<byte> buffer = stackalloc byte[BlockSize];

runtime/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Base64Decoder.cs

Line 545 in e617b8f

int paddingCount = GetPaddingCount<TBase64Decoder>(ref buffer[^1]);

[stephentoub]

Why did this change from 64 to 32?

[buyaa-n]

It was 32 originally, but unintentionally changed with this PR

The method comment also mentions it require 32 byte

runtime/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Base64Encoder.cs

Lines 316 to 322 in e617b8f

	private static unsafe void Avx2Encode<TBase64Encoder, T>(ref byte* srcBytes, ref T* destBytes, byte* srcEnd, int sourceLength, int destLength, byte* srcStart, T* destStart)
	where TBase64Encoder : IBase64Encoder<T>
	where T : unmanaged
	{
	// If we have AVX2 support, pick off 24 bytes at a time for as long as we can.
	// But because we read 32 bytes at a time, ensure we have enough room to do a
	// full 32-byte read without segfaulting.

[stephentoub]

Thanks!

[MihaZupan]

Would anything bad happen if this were left at a const 4?
This not being const would penalize the non-url path as well a bit.

[buyaa-n]

For Base64Url when the source is not multiple of 4 we need to adjust the buffer size accordingly, the BlockSize value used to fill the whitespace with padding (row 653) so that it could decoded correctly (in case remaining bytes were decodable or valid)

[MihaZupan]

@MihuBot fuzz Base64

[lewing]

also showing regressions in dotnet/perf-autofiling-issues#36512

and improvements dotnet/perf-autofiling-issues#36643