Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/Cysharp/SimdLinq
Drop-in replacement of LINQ aggregation operations extremely faster with SIMD.
https://github.com/Cysharp/SimdLinq
Last synced: about 1 month ago
JSON representation
Drop-in replacement of LINQ aggregation operations extremely faster with SIMD.
- Host: GitHub
- URL: https://github.com/Cysharp/SimdLinq
- Owner: Cysharp
- License: mit
- Created: 2023-01-29T17:44:20.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2024-09-18T08:19:46.000Z (3 months ago)
- Last Synced: 2024-11-05T13:13:03.407Z (about 1 month ago)
- Language: C#
- Homepage:
- Size: 85.9 KB
- Stars: 775
- Watchers: 14
- Forks: 11
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE.md
Awesome Lists containing this project
- awesome-reference-tools - SimdLinq
README
# SimdLinq
[](https://github.com/Cysharp/SimdLinq/actions) [](https://github.com/Cysharp/SimdLinq/releases)
SimdLinq is a drop-in replacement of LINQ aggregation operations(`Sum`, `Average`, `Min`, `Max`) extremely faster with SIMD.
[.NET 7 LINQ supports SIMD](https://devblogs.microsoft.com/dotnet/performance_improvements_in_net_7/#linq) but it is very limited, due to compatibility and safety issues, it is only enabled for `int[]` `Average`, `Min`, `Max` and `long[]` `Min`, `Max`.
SimdLinq accelerates many methods (`Sum`, `LongSum`, `Average`, `Min`, `Max`, `MinMax`, `Contains`, `SequenceEqual`) and types (`byte`, `sbyte`, `short`, `ushort`, ` int`, `uint`, `long`, `ulong`, `float`, `double`). It can also be used with `List`, `Span`, `ReadOnlySpan`, `Memory`, `ReadOnlyMemory` in addition to `T[]`.
Using the overload resolution priority, all target methods are automatically SIMD by simply referencing the library and setting global using.
Quick Start
---
This library is distributed via NuGet.
> PM> Install-Package [SimdLinq](https://www.nuget.org/packages/SimdLinq)
Note that this library requires **.NET 7** or above because this library uses static abstract members and the new cross platform **.NET 7** SIMD Api.
```csharp
using SimdLinq; // enable SimdLinq extension methods
var array = Enumerable.Range(1, 100000).ToArray();
var sum = array.Sum(); // uses SimdLinqExtensions.Sum
```
To enable SimdLinq per file, add the `using SimdLinq;` using directive. To enable SimdLinq across the project, use global usings in the csproj file.
```xml
```
`SimdLinqExtensions` has methods for concrete data types, like `int[]` or `Span`, with the same names as LINQ's methods. If a method is eligible for SIMD optimization (such as the Sum of int[]), the `SimdLinqExtensions` method will be used for improved performance.
Unlike base LINQ, SimdLinq supports `Span`, allowing you to call methods such as `Sum`, `Min`, etc. on `Span` or `ReadOnlySpan` collections.
```csharp
(double Min, double Max) GetMinMax(Span span)
{
return span.MinMax();
}
```
`MinMax` is an original SimdLinq extension, that returns a tuple of `Min` and `Max`.
Compatibility
---
One of the reasons why LINQ's SIMD support in .NET 7 is incomplete, is compatibility. SimdLinq prioritises performance over some safety features and full compatibility with LINQ. Please note the following differences.
### Sum/Average
LINQ Sum is `checked` but SimdLinq is `unchecked`(SIMD operation is not supported overflow). To reduce the risk of overflow, `Sum` and `Average` only support types that are 32-bit or higher(`int`, `long`, `uint`, `ulong`, `double`, `float`).
SimdLinq provides `LongSum` for `int` and `uint`, that returns `long`/`ulong` so avoid overflow.
### float/double
Unlike LINQ, SimdLinq does not check for NaN in float/double Min/Max calculations. Additionally, the order of calculation for Sum is not sequential, leading to slight differences in floating-point operations compared to regular LINQ. For instance, LINQ returns 1.5710588F, while SimdLinq returns 1.5710589F. If compatibility is not a concern, this difference is not significant, but be aware of potential small tolerance differences.
Supported collections
---
`T[]`, `List`, `Span`, `Memory`, `ReadOnlyMemory`, `Span`, `ReadOnlySpan`.
Supported methods
---
* `Sum` for `int`, `uint`, `long`, `ulong`, `float`, `double`
* `LongSum` for `int`, `uint`
* `Average` for `int`, `uint`, `long`, `ulong`, `float`, `double`
* `Min` for `byte`, `sbyte`, `short`, `ushort`, `int`, `uint`, `long`, `ulong`, `float`, `double`
* `Max` for `byte`, `sbyte`, `short`, `ushort`, `int`, `uint`, `long`, `ulong`, `float`, `double`
* `MinMax` for `byte`, `sbyte`, `short`, `ushort`, `int`, `uint`, `long`, `ulong`, `float`, `double`
* `Contains` for `byte`, `sbyte`, `short`, `ushort`, `int`, `uint`, `long`, `ulong`, `float`, `double`
* `SequenceEqual` for `byte`, `sbyte`, `short`, `ushort`, `int`, `uint`, `long`, `ulong`, `float`, `double`
SoA
---
SIMD with LINQ requires primitive array(or Span). [Cysharp/StructureOfArraysGenerator](https://github.com/Cysharp/StructureOfArraysGenerator) makes easy to create SoA array to use SIMD easily.
License
---
This library is licensed under the MIT License.