SharpImmintrin 1.0.0

SharpImmintrin

A simple .NET library which acts as a porting assistant that makes it much easier to port SIMD C code into C#.

Sounds cool, but how?

Suppose that we have this C code and we want to port it into C#:

#define SIZE 4
#include <emmintrin.h>  // SSE2 intrinsics

double a[SIZE] = {1.0, 2.0, 3.0, 4.0};
double b[SIZE] = {5.0, 6.0, 7.0, 8.0};
double result[SIZE];

for (int i = 0; i < SIZE; i += 2) {
    __m128d va = _mm_loadu_pd(&a[i]);
    __m128d vb = _mm_loadu_pd(&b[i]);

    __m128d vr = _mm_add_pd(va, vb);

    _mm_storeu_pd(&result[i], vr);
}

Right off the bat, we would port arrays and the outer for loop:

const int SIZE = 4;

// Collection Expressions

Span<double> a = [1.0, 2.0, 3.0, 4.0];
Span<double> b = [5.0, 6.0, 7.0, 8.0];
Span<double> result = stackalloc double[4];

for (int i = 0; i < SIZE; i += 2)
{
}

There's now a problem: methods like _mm_loadu_pd come from C SSE2 intrinsics, and to someone who doesn't know how to use them, it can be difficult to port into C#.

Let's go ahead and install SharpImmintrin, as well as SharpImmintrin.GlobalUsings, from NuGet.

Now let's add these usings:

using static SharpImmintrin.CIntrinsicX86Sse2;

And let's paste that code:

__m128d va = _mm_loadu_pd(&a[i]);
__m128d vb = _mm_loadu_pd(&b[i]);

__m128d vr = _mm_add_pd(va, vb);

_mm_storeu_pd(&result[i], vr);

Suddenly, methods like _mm_loadu_pd are now defined.

<img src="Resources/CodeSSE2.png">

Now, when we hover over them, we can see the exact equivalent in the .NET BCL that this method maps into, in the XML documentation.

<img src="Resources/EquivalentSSE2.png">

So, we can just replace that method with Sse2.LoadVector128 accordingly.

Following the XML documentation, we now get this:

using System.Runtime.Intrinsics.X86;

const int SIZE = 4;

// Collection Expressions

Span<double> a = [1.0, 2.0, 3.0, 4.0];
Span<double> b = [5.0, 6.0, 7.0, 8.0];
Span<double> result = stackalloc double[4];

for (int i = 0; i < SIZE; i += 2)
{
    __m128d va = Sse2.LoadVector128(&a[i]);
    __m128d vb = Sse2.LoadVector128(&b[i]);

    __m128d vr = Sse2.Add(va, vb);

    Sse2.Store(&result[i], vr);
}

From then on, we can fix the code slightly, wrapping it in an unsafe block and making sure we can access pointers to arrays.

using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;

const int SIZE = 4;

// Collection Expressions

Span<double> a = [1.0, 2.0, 3.0, 4.0];
Span<double> b = [5.0, 6.0, 7.0, 8.0];
Span<double> result = stackalloc double[4];

unsafe
{
    fixed (double* aP = a)
    fixed (double* bP = b)
    fixed (double* resultP = result)
    {
        for (int i = 0; i < SIZE; i += 2)
        {
            Vector128<double> va = Sse2.LoadVector128(&aP[i]);
            Vector128<double> vb = Sse2.LoadVector128(&bP[i]);

            Vector128<double> vr = Sse2.Add(va, vb);

            Sse2.Store(&resultP[i], vr);
        }
    }
}

Which in turn gives us a C# representation of the C code.

And now when we're done with porting C code into C#, we can go ahead and uninstall the SharpImmintrin package. Simple as that.

Supported architectures

• X86
  • AES
  • AVX
  • AVX2
  • AVX512BMM
  • AVX512BW
  • AVX512CD
  • AVX512F
  • AVX512VBMI
  • AVX512VBMI2
  • AVXVNNI
  • BMI1
  • BMI2
  • FMA
  • GFNI
  • LZCNT
  • PCLMULQDQ
  • POPCNT
  • SSE
  • SSE2
  • SSE3
  • SSE4.1
  • SSE4.2
  • SSSE3
• ARM
  • NEON (a.k.a. AdvSimd)

License

Licensed under MIT, see LICENSE.txt

Building

git clone https://github.com/winscripter/SharpImmintrin.git
cd SharpImmintrin
dotnet build

To start tests: dotnet test