SimpleSIMD

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

★ 94 Open GitHub ↗

Easy to use SIMD accelerated Array and Span methods

Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# SimpleSIMD

[![NuGet version (SimpleSIMD)](https://img.shields.io/nuget/v/SimpleSIMD.svg?style=flat-square)](https://www.nuget.org/packages/SimpleSIMD/)

### What is SIMD?
Single Instruction, Multiple Data (SIMD) units refer to hardware components that perform the same operation on multiple data operands concurrently.
The concurrency is performed on a single thread, while utilizing the full size of the processor register to perform several operations at one.  
This approach could be combined with standard multithreading for massive performence boosts in numeric computations.

## Goals And Purpose
* Single API to unify SIMD for **All supported types**
* Gain performence boost for mathematical computations using a simple API
* Simplifies SIMD usage, and to make it easy to integrate it into an already existing solutions
* Helps generalize several methemathical functions for supported types
* Performs less allocations compared to standard LINQ implementations

## Available Functions

<details>
  <summary><h4>Comparison</h4></summary>
  
  * Equal
  * Greater
  * GreaterOrEqual
  * Less
  * LessOrEqual
</details>
<details>
  <summary><h4>Elementwise</h4></summary>
  
  * Negate
  * Abs
  * Add
  * Divide
  * Multiply
  * Subtract
  * And
  * AndNot
  * Or
  * Xor
  * Not
  * Select
  * Ternary (Conditional Select)
  * Concat
  * Sqrt
</details>
<details>
  <summary><h4>Reduction</h4></summary>
  
  * Aggregate
  * Sum
  * Average
  * Max
  * Min
  * Dot
</details>
<details>
  <summary><h4>General</h4></summary>
  
  * All
  * Any
  * Contains
  * IndexOf
  * Fill
  * Foreach
</details>

### Auto-Generated Functions
For any of the ``Elementwise`` functions, an auto-generated overload is created, which doesn't accept ```Span<T> result```, 
and instead returns ```T[]``` as the result.  
  
For any of the functions with the Value Delagate pattern, an auto-generated overload is created, which accepts regular delegates.
Note that using this overload results in performence losses. Check [Value Delegates](#value-delegates) section for more info.

## Performance Benefits

A simple benchmark to demonstrate performance gains of using SIMD.  
Benchmarked method was a ``Sum`` over an ``int[]``.

| Method | Length |          Mean |       Error |      StdDev |        Median | Ratio |
|------- |------- |--------------:|------------:|------------:|--------------:|------:|
|   SIMD |     10 |      3.556 ns |   0.0655 ns |   0.0581 ns |      3.537 ns |  0.66 |
|  Naive |     10 |      5.357 ns |   0.0568 ns |   0.0531 ns |      5.348 ns |  1.00 |
|   SIMD |    100 |      9.079 ns |   0.1948 ns |   0.1822 ns |      9.032 ns |  0.20 |
|  Naive |    100 |     46.178 ns |   0.5255 ns |   0.4658 ns |     46.203 ns |  1.00 |
|   SIMD |   1000 |     66.018 ns |   0.6931 ns |   0.6483 ns |     65.802 ns |  0.17 |
|  Naive |   1000 |    388.244 ns |   3.0852 ns |   2.8859 ns |    389.093 ns |  1.00 |
|   SIMD |   3000 |    185.507 ns |   1.3070 ns |   1.1587 ns |    185.375 ns |  0.16 |
|  Naive |   3000 |  1,139.552 ns |  11.9608 ns |  11.1881 ns |  1,139.374 ns |  1.00 |
|   SIMD |   6000 |    365.993 ns |   3.2114 ns |   3.0039 ns |    365.075 ns |  0.16 |
|  Naive |   6000 |  2,274.374 ns |  14.2898 ns |  12.6675 ns |  2,271.185 ns |  1.00 |
|   SIMD |  10000 |    585.275 ns |   5.2631 ns |   4.1091 ns |    586.638 ns |  0.15 |
|  Naive |  10000 |  3,938.198 ns |  46.8599 ns |  43.8328 ns |  3,926.622 ns |  1.00 |
|   SIMD |  30000 |  1,791.966 ns |  30.4379 ns |  48.2777 ns |  1,778.255 ns |  0.15 |
|  Naive |  30000 | 11,848.767 ns | 184.5488 ns | 163.5977 ns | 11,773.515 ns |  1.00 |
|   SIMD |  60000 |  3,612.872 ns |  71.7281 ns | 113.7683 ns |  3,580.606 ns |  0.15 |
|  Naive |  60000 | 23,606.125 ns | 249.0765 ns | 232.9863 ns | 23,542.178 ns |  1.00 |
|   SIMD | 100000 |  7,325.734 ns | 156.6350 ns | 451.9279 ns |  7,138.866 ns |  0.19 |
|  Naive | 100000 | 40,283.073 ns | 464.1261 ns | 434.1439 ns | 40,328.790 ns |  1.00 |

<details>
  <summary>Benchmark Details</summary>
  
  ```
  BenchmarkDotNet=v0.13.2, OS=Windows 11 (10.0.22621.819)
  Intel Core i7-10510U CPU 1.80GHz, 1 CPU, 8 logical and 4 physical cores
  .NET SDK=7.0.100
    [Host]     : .NET 7.0.0 (7.0.22.51805), X64 RyuJIT AVX2
    DefaultJob : .NET 7.0.0 (7.0.22.51805), X64 RyuJIT AVX2
```
</details>

## Value Delegates
This library uses the value delegate pattern. This pattern is used as a replacement for regular delegates.  
Calling functions using this patten may feel unusual since it requires creation of structs to pass as arguments instead of delegates, but it is very beneficial performance-wise. 
The performance difference makes using this pattern worthwhile in performance critical places.  
Since the focus of this library is **pure performance**, we use this pattern wherever possible.

#### Usage:

``` csharp
using System;
using System.Numerics;
using SimpleSimd;

namespace MyProgram
{
    class Program
    {
        static void Main()
        {
            // Creating the data
            // Can be int[], Span<int>, ReadOnlySpan<int>
            int[] Data = GetData();
            
            // We need to create 2 structs which will serve as a replacement for delegates
            SimdOps.Sum(Data, new VecSelector(), new Selector());
        }
    }             
    
    // A struct which is used as Vector<int> selector
    // Inheritence from IFunc<Vector<T>, Vector<T>> is according to Sum() signature
    struct VecSelector : IFunc<Vector<int>, Vector<int>>
    {
        public Vector<int> Invoke(Vector<int> param) => DoSomething(param);
    }

// A struct which is used as int selector
    // Inheritence from IFunc<T, T> is according to Sum() signature
    struct Selector : IFunc<int, int>
    {
        public int Invoke(int param) => DoSomething(param);
    }   
}
```

#### benchmark:

Both of the benchmarked methods have the exactly same code, both of them are accelerated using SIMD,  
the only difference is the argument types.

``` csharp
// Delegate, baseline
public static T Sum<T>(ReadOnlySpan<T> span, Func<Vector<T>, Vector<T>> vSelector, Func<T, T> selector) 
            where T : struct, INumber<T>;

// ValueDelegate
public static T Sum<T, F1, F2>(ReadOnlySpan<T> span, F1 vSelector, F2 selector)
            where T  : struct, INumber<T>
            where F1 : struct, IFunc<Vector<T>, Vector<T>>
            where F2 : struct, IFunc<T, T>;
```

|        Method | Length |          Mean |       Error |      StdDev |        Median | Ratio |
|-------------- |------- |--------------:|------------:|------------:|--------------:|------:|
|      Delegate |     10 |      9.477 ns |   0.0910 ns |   0.0851 ns |      9.467 ns |  1.00 |
| ValueDelegate |     10 |      3.969 ns |   0.1078 ns |   0.1107 ns |      3.961 ns |  0.42 |
|      Delegate |    100 |     37.747 ns |   0.6666 ns |   0.6236 ns |     37.698 ns |  1.00 |
| ValueDelegate |    100 |      9.295 ns |   0.1697 ns |   0.1587 ns |      9.276 ns |  0.25 |
|      Delegate |   1000 |    264.978 ns |   5.2711 ns |   4.9306 ns |    263.820 ns |  1.00 |
| ValueDelegate |   1000 |     66.474 ns |   1.0799 ns |   1.0101 ns |     66.471 ns |  0.25 |
|      Delegate |   3000 |    773.737 ns |  11.6963 ns |  10.9407 ns |    773.347 ns |  1.00 |
| ValueDelegate |   3000 |    186.632 ns |   3.7407 ns |   4.1578 ns |    185.751 ns |  0.24 |
|      Delegate |   6000 |  1,554.745 ns |  26.9752 ns |  25.2326 ns |  1,559.120 ns |  1.00 |
| ValueDelegate |   6000 |    369.259 ns |   6.3982 ns |   5.6719 ns |    368.428 ns |  0.24 |
|      Delegate |  10000 |  2,612.493 ns |  51.2703 ns |  47.9583 ns |  2,615.721 ns |  1.00 |
| ValueDelegate |  10000 |    624.057 ns |  12.4864 ns |  16.2358 ns |    622.558 ns |  0.24 |
|      Delegate |  30000 |  8,718.167 ns | 173.5442 ns | 170.4436 ns |  8,719.592 ns |  1.00 |
| ValueDelegate |  30000 |  1,860.125 ns |  35.8075 ns |  47.8020 ns |  1,865.076 ns |  0.22 |
|      Delegate |  60000 | 17,259.904 ns | 330.4238 ns | 429.6443 ns | 17,109.451 ns |  1.00 |
| ValueDelegate |  60000 |  3,715.645 ns |  72.8741 ns | 121.7563 ns |  3,689.114 ns |  0.22 |
|      Delegate | 100000 | 27,357.138 ns | 534.2404 ns | 548.6255 ns | 27,176.126 ns |  1.00 |
| ValueDelegate | 100000 |  7,485.716 ns | 150.0830 ns | 440.1676 ns |  7,313.833 ns |  0.27 |

## Limitations
* Methods are not lazily evaluated as IEnumerable
* Old hardware might not support SIMD
* Supported collection types:
  * ```T[]```
  * ```Span<T>```
  * ```ReadOnlySpan<T>```
* Supports only **Primitive Numeric Types** as array elements. Supported types are:
  * ```byte, sbyte```
  * ```short, ushort```
  * ```int, uint```
  * ```long, ulong```
  * ```nint, nuint```
  * ```float```
  * ```double```

## Contributing
**All ideas and suggestions are welcome.**
Feel free to open an issue if you have an idea or a suggestion that might improve this project.
If you encounter a bug or have a feature request, please open a relevent issue.

## License
This project is licensed under MIT license. For more info see the [License File](LICENSE)

SimpleSIMD

Download

Today's picks

Was this content helpful?

Recent Feedback

SimpleSIMD

Download

Today's picks

Was this content helpful?

Recent Feedback

Download README