Skip to content
/ Base64 Public

Base64 encoding / decoding with SIMD-support, also base64Url

License

Notifications You must be signed in to change notification settings

gfoidl/Base64

Repository files navigation

CI NuGet
Build Status NuGet

gfoidl.Base64

A .NET library for base64 encoding / decoding, as well as base64Url support. Encoding can be done to buffers of type byte (for UTF-8) or char. Decoding can read from buffers of type byte (for UTF-8) or char.

Encoding / decoding supports buffer-chains, for example for very large data or when the data arrives in chunks.

In .NET Core 3.01 onwards encoding / decoding is done with SIMD-support:

Framework scalar SSSE3 AVX2 arm64
.NET 7 ✔️ ✔️ ✔️ ✔️
.NET Core 3.0 ✔️ ✔️ ✔️
.NET Standard 2.0 / .NET 4.5 ✔️

If available AVX will "eat" up as much as possible, then SSE will "eat" up as much as possible, finally scalar code processes the rest (including padding).

Usage

Basically the entry to encoder / decoder is Base64.Default for base64, and Base64.Url for base64Url.

See demo for further examples.

Encoding

byte[] guid = Guid.NewGuid().ToByteArray();

string guidBase64     = Base64.Default.Encode(guid);
string guidBases64Url = Base64.Url.Encode(guid);

or Span<byte> based (for UTF-8 encoded output):

int guidBase64EncodedLength = Base64.Default.GetEncodedLength(guid.Length);
Span<byte> guidBase64UTF8   = stackalloc byte[guidBase64EncodedLength];
OperationStatus status      = Base64.Default.Encode(guid, guidBase64UTF8, out int consumed, out int written);

int guidBase64UrlEncodedLength = Base64.Url.GetEncodedLength(guid.Length);
Span<byte> guidBase64UrlUTF8   = stackalloc byte[guidBase64UrlEncodedLength];
status                         = Base64.Url.Encode(guid, guidBase64UrlUTF8, out consumed, out written);

Decoding

Guid guid = Guid.NewGuid();

string guidBase64    = Convert.ToBase64String(guid.ToByteArray());
string guidBase64Url = guidBase64.Replace('+', '-').Replace('/', '_').TrimEnd('=');

byte[] guidBase64Decoded    = Base64.Default.Decode(guidBase64);
byte[] guidBase64UrlDecoded = Base64.Url.Decode(guidBase64Url);

or Span<char> based:

int guidBase64DecodedLen    = Base64.Default.GetDecodedLength(guidBase64);
int guidBase64UrlDecodedLen = Base64.Url.GetDecodedLength(guidBase64Url);

Span<byte> guidBase64DecodedBuffer    = stackalloc byte[guidBase64DecodedLen];
Span<byte> guidBase64UrlDecodedBuffer = stackalloc byte[guidBase64UrlDecodedLen];

OperationStatus status = Base64.Default.Decode(guidBase64, guidBase64DecodedBuffer, out int consumed, out int written);
status                 = Base64.Url.Decode(guidBase64Url, guidBase64UrlDecodedBuffer, out consumed, out written);

Buffer chains

Buffer chains are handy when for encoding / decoding

  • very large data
  • data arrives is chunks, e.g. by reading from a (buffered) stream / pipeline
  • the size of data is initially unknown
  • ...
var rnd         = new Random();
Span<byte> data = new byte[1000];
rnd.NextBytes(data);

// exact length could be computed by Base64.Default.GetEncodedLength, here for demo exzessive size
Span<char> base64 = new char[5000];

OperationStatus status = Base64.Default.Encode(data.Slice(0, 400), base64, out int consumed, out int written, isFinalBlock: false);
status                 = Base64.Default.Encode(data.Slice(consumed), base64.Slice(written), out consumed, out int written1, isFinalBlock: true);

base64 = base64.Slice(0, written + written1);

Span<byte> decoded = new byte[5000];
status             = Base64.Default.Decode(base64.Slice(0, 100), decoded, out consumed, out written, isFinalBlock: false);
status             = Base64.Default.Decode(base64.Slice(consumed), decoded.Slice(written), out consumed, out written1, isFinalBlock: true);

decoded = decoded.Slice(0, written + written1);

ReadOnlySequence / IBufferWriter

Encoding / decoding with ReadOnlySequence<byte> and IBufferWriter<byte> can be used together with System.IO.Pipelines.

var pipeOptions = PipeOptions.Default;
var pipe        = new Pipe(pipeOptions);

var rnd  = new Random(42);
var data = new byte[4097];
rnd.NextBytes(data);

pipe.Writer.Write(data);
await pipe.Writer.CompleteAsync();

ReadResult readResult = await pipe.Reader.ReadAsync();

var resultPipe = new Pipe();
Base64.Default.Encode(readResult.Buffer, resultPipe.Writer, out long consumed, out long written);
await resultPipe.Writer.FlushAsync();
await resultPipe.Writer.CompleteAsync();

pipe.Reader.AdvanceTo(readResult.Buffer.End);

(Functional) Comparison to classes in .NET

General

.NET provides the classes System.Convert and System.Buffers.Text.Base64 for base64 operations.

base64Url isn't supported, so hacky solutions like

string base64    = Convert.ToBase64String(data);
string base64Url = base64.Replace('+', '-').Replace('/', '_').TrimEnd('=');

are needed. This isn't ideal, as there are avoidable allocations and several iterations over the encoded string (see here and here for benchmark results).

gfoidl.Base64 supports encoding / decoding to / from base64Url in a direct way. Encoding byte[] -> byte[] for UTF-8 is supported, as well as byte[] -> char[]. Decoding byte[] -> byte[] for UTF-8 is supported, as well as char[] -> byte[].

Further SIMD isn't utilized in the .NET classes. SIMD is also supported in the .NET classes. (Note: I've opened an issue to add SIMD-support to these classes).

Convert.ToBase64XYZ / Convert.FromBase64XYZ

These methods only support byte[] -> char[] as types for encoding, and char[] -> byte[] as types for decoding, where char[] can also be string or (ReadOnly)Span<char>.

To support UTF-8 another method call like

byte[] utf8Encoded = Encoding.ASCII.GetBytes(base64String);

is needed.

An potential advantage of this class is that it allows the insertion of line-breaks (cf. Base64FormattingOptions.InsertLineBreaks).

System.Buffers.Text.Base64

This class only supports byte[] -> byte[] for encoding / decoding. So in order to get a string Encoding has to be used.

An potential advantage of this class is the support for in-place encoding / decoding (cf. Base64.EncodeToUtf8InPlace, Base64.DecodeFromUtf8InPlace )

Benchmarks

For old (.NET Core 3.1) benchmarks see results. Benchmarks are run regularly on CI, see there for current run-results (in the artifacts).

Performance gain depends, among a lot of other things, on the workload size, so here no table with superior results will be shown.

Direct encoding to a string is for small inputs slower than Convert.ToBase64String (has less overhead, and can write to string-buffer in a direct way). But the larger the workload, the better this library works.

Direct decoding from a string is generally (a lot) faster than Convert.ConvertFromBase64CharArray, also depending on workload size, but in the benchmark the speedup is from 1.5 to 12x.

Note: please measure / profile in your real usecase, as this are just micro-benchmarks.

Acknowledgements

The scalar version of the base64 encoding / decoding is based on System.Buffers.Text.Base64.

The scalar version of the base64Url encoding / decoding is based on dotnet/extensions#334 and dotnet/extensions#338.

Vectorized versions (SSE, AVX) for base64 encoding / decoding is based on https://github.com/aklomp/base64 (see also Acknowledgements in that repository).

Vectorized versions (SSE, AVX) for base64Url encoding / decoding is based on https://github.com/aklomp/base64 (see Acknowledgements in that repository). For decoding (SSE, AVX) code is based on Vector lookup (pshufb) by Wojciech Mula.

Development channel

To get packages from the development channel use a nuget.config similar to this one:

<?xml version="1.0" encoding="utf-8"?>
<configuration>
    <packageSources>
        <add key="gfoidl-public" value="https://pkgs.dev.azure.com/gh-gfoidl/github-Projects/_packaging/gfoidl-public/nuget/v3/index.json" />
    </packageSources>
</configuration>

Footnotes

  1. For targets pre .NET 7 use the v1.x versions.