mahalex/fasterbase64
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add benchmarks

Browse Source
This commit is contained in:
Alexander Luzgarev 2021-12-05 21:34:21 +01:00
parent a802188106
commit f243320c53
1 changed files with 45 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
README.md
View File

@ -5,15 +5,16 @@ The algorithm is described in
[Faster Base64 Encoding and Decoding Using AVX2 Instructions](https://arxiv.org/abs/1704.00605)
by Wojciech Muła and Daniel Lemire.
Benchmarks show that this implementation is about 8-10 times faster
than the standard .NET implementation
Benchmarks (see below) show that this implementation is
about 8-10 times faster than the standard .NET implementation
in `System.Convert.TryToBase64Chars()` and `System.Convert.TryFromBase64Chars()`.
We provide re-implementations of these two methods
within a static class `FasterBase64.Convert`.
* `FasterBase64.Convert.TryToBase64Chars()` works
the same way as `System.Convert.TryToBase64Chars()`;
* `System.FasterBase64.TryFromBase64Chars()` works
* `FasterBase64.FasterBase64.TryFromBase64Chars()` works
the same way as`System.Convert.TryFromBase64Chars()`,
if the input does not contain whitespace. The standard .NET implementation
differs from the RFC4648 standard in that it allows whitespace in the data.
@ -23,15 +24,53 @@ within a static class `FasterBase64.Convert`.
We believe that (in all reasonable cases) this is still faster than
using the standard implementation.
Caveats:
## Benchmarks for encoding
AMD Ryzen 9 3950X 16-Core Processor, 3493 Mhz,
Windows 10.0.22509, .NET 6.0.0 (6.0.21.52210)
| Method | N | Mean | Error | StdDev |
|------- |------ |------------:|----------:|----------:|
| Old | 100 | 95.05 ns | 0.758 ns | 0.672 ns |
| New | 100 | 25.06 ns | 0.507 ns | 0.498 ns |
| Old | 1000 | 872.76 ns | 11.321 ns | 10.036 ns |
| New | 1000 | 92.05 ns | 0.633 ns | 0.592 ns |
| Old | 10000 | 8,723.65 ns | 79.673 ns | 74.526 ns |
| New | 10000 | 724.58 ns | 9.693 ns | 9.067 ns |
Here "Old" denotes usage of `System.Convert.TryToBase64Chars()`,
"New" denotes usage of `FasterBase64.Convert.TryToBase64Chars()`,
"N" is the size of input byte array.
## Benchmarks for decoding
AMD Ryzen 9 3950X 16-Core Processor, 3493 Mhz,
Windows 10.0.22509, .NET 6.0.0 (6.0.21.52210)
| Method | N | Mean | Error | StdDev |
|------- |------ |------------:|----------:|----------:|
| Old | 100 | 94.66 ns | 0.839 ns | 0.743 ns |
| New | 100 | 54.08 ns | 0.547 ns | 0.457 ns |
| Old | 1000 | 831.63 ns | 9.560 ns | 8.474 ns |
| New | 1000 | 131.31 ns | 1.085 ns | 0.906 ns |
| Old | 10000 | 8,254.94 ns | 64.406 ns | 57.094 ns |
| New | 10000 | 1,011.61 ns | 6.320 ns | 5.602 ns |
Here "Old" denotes usage of `System.Convert.TryFromBase64Chars()`,
"New" denotes usage of `FasterBase64.Convert.TryFromBase64Chars()`,
"N" is the size of byte array, that is then encoded in base64 using
`System.Convert.TryToBase64Chars()` to form input to the
benchmarked methods.
## Caveats
* Although we believe the implementation to be reasonably well tested,
there might still be bugs.
* The implementation uses AVX2 instructions, but does not check if AVX2
is available. AVX2 support can be easily checked by querying the property
`System.Runtime.Intrinsics.X86.Avx2.IsSupported`.
TODO:
## TODO
* NuGet package.
License:
## License
* Copyright (c) Alexander Luzgarev, 2021, under the GPL license.