C#Vector.CopyTo几乎比非SIMD版本快？

概述更新：之前提到的跨度问题已在.net核心2.1版本(目前正在预览中)中修复.这些实际上使得Span Vector *比数组Vector更快* … 注意：在“Intel Xeon E5-1660 v4”上进行测试,CPU-Z告诉我有“MMX,SSE,SSE2,SSE3,SSSE3,SSE4.1,SSE4.2,EM64T,VT-x,AES, AVX,AVX2,FMA3,RSX“所以它应该没问题…… 在 更新：之前提到的跨度问题已在.net核心2.1版本(目前正在预览中)中修复.这些实际上使得Span Vector *比数组Vector更快* …

注意：在“Intel Xeon E5-1660 v4”上进行测试,cpu-Z告诉我有“MMX,SSE,SSE2,SSE3,SSSE3,SSE4.1,SSE4.2,EM64T,VT-x,AES,AVX,AVX2,FMA3,RSX“所以它应该没问题……

在回答Vector based question之后,我想我会尝试实现一些BLAS功能.我发现那些正在阅读/求和的产品如dot产品非常好,但是我回写一个阵列是坏的 – 比非SIMD更好,但几乎没有.

我做错了什么,或者是否需要在JIT中做更多的工作？

示例(假设x.Length = y.Length,not null等等等等等等)：

public static voID daxpy(double Alpha,double[] x,double[] y){    for (var i = 0; i < x.Length; ++i)        y[i] = y[i] + x[i] * Alpha;}

在矢量形式变为：

public static voID daxpy(double Alpha,double[] y){    var i = 0;    if (Vector.IsHarDWareAccelerated)    {        var length = x.Length + 1 - Vector<double>.Count;        for (; i < length; i += Vector<double>.Count)        {            var vAlpha = new Vector<double>(Alpha);            var vx = new Vector<double>(x,i);            var vy = new Vector<double>(y,i);            (vy + vx * vAlpha).copyTo(y,i);        }    }    for (; i < x.Length; ++i)        y[i] = y[i] + x[i] * Alpha;}

而且,在.NET Core 2.0中玩游戏,虽然我会尝试Span,无论是天真还是矢量形式：

public static voID daxpy(double Alpha,Span<double> x,Span<double> y){    for (var i = 0; i < x.Length; ++i)        y[i] += x[i] * Alpha;}

和矢量

public static voID daxpy(double Alpha,Span<double> y){    if (Vector.IsHarDWareAccelerated)    {        var vx = x.NonPortableCast<double,Vector<double>>();        var vy = y.NonPortableCast<double,Vector<double>>();        var vAlpha = new Vector<double>(Alpha);        for (var i = 0; i < vx.Length; ++i)            vy[i] += vx[i] * vAlpha;        x = x.Slice(Vector<double>.Count * vx.Length);        y = y.Slice(Vector<double>.Count * vy.Length);    }    for (var i = 0; i < x.Length; ++i)        y[i] += x[i] * Alpha;}

所以这些的相对时间是：

Naive       1.0Vector      0.8Span Naive  2.5 ==> Update: Span Naive  1.1Span Vector 0.9 ==> Update: Span Vector 0.6

我做错了什么？我几乎无法想到一个更简单的例子,所以我不这么认为？

解决方法 你可能想用2.1以上的测试;
在我的笔记本电脑上(SIMD与我的桌面相比较差),我得到：

daxpy_naive x10000: 144msdaxpy_arr_vector x10000: 77msdaxpy_span x10000: 173msdaxpy_vector x10000: 67msdaxpy_vector_no_slice x10000: 67ms

使用代码：

using System;using System.Diagnostics;using system.numerics;class Program{    static voID Main(string[] args)    {        double Alpha = 0.5;        double[] x = new double[16 * 1024],y = new double[x.Length];        var rand = new Random(12345);        for (int i = 0; i < x.Length; i++)            x[i] = rand.NextDouble();        RunAll(Alpha,x,y,1,false);        RunAll(Alpha,10000,true);    }    private static voID RunAll(double Alpha,double[] y,int loop,bool log)    {        GC.Collect(GC.MaxGeneration);        GC.WaitForPendingFinalizers();        var watch = Stopwatch.StartNew();        for(int i = 0; i < loop; i++)        {            daxpy_naive(Alpha,y);        }        watch.Stop();        if (log) Console.Writeline($"{nameof(daxpy_naive)} x{loop}: {watch.ElapsedMilliseconds}ms");        watch = Stopwatch.StartNew();        for (int i = 0; i < loop; i++)        {            daxpy_arr_vector(Alpha,y);        }        watch.Stop();        if (log) Console.Writeline($"{nameof(daxpy_arr_vector)} x{loop}: {watch.ElapsedMilliseconds}ms");        watch = Stopwatch.StartNew();        for (int i = 0; i < loop; i++)        {            daxpy_span(Alpha,y);        }        watch.Stop();        if (log) Console.Writeline($"{nameof(daxpy_span)} x{loop}: {watch.ElapsedMilliseconds}ms");        watch = Stopwatch.StartNew();        for (int i = 0; i < loop; i++)        {            daxpy_vector(Alpha,y);        }        watch.Stop();        if (log) Console.Writeline($"{nameof(daxpy_vector)} x{loop}: {watch.ElapsedMilliseconds}ms");        watch = Stopwatch.StartNew();        for (int i = 0; i < loop; i++)        {            daxpy_vector_no_slice(Alpha,y);        }        watch.Stop();        if (log) Console.Writeline($"{nameof(daxpy_vector_no_slice)} x{loop}: {watch.ElapsedMilliseconds}ms");    }    public static voID daxpy_naive(double Alpha,double[] y)    {        for (var i = 0; i < x.Length; ++i)            y[i] = y[i] + x[i] * Alpha;    }    public static voID daxpy_arr_vector(double Alpha,double[] y)    {        var i = 0;        if (Vector.IsHarDWareAccelerated)        {            var length = x.Length + 1 - Vector<double>.Count;            for (; i < length; i += Vector<double>.Count)            {                var vAlpha = new Vector<double>(Alpha);                var vx = new Vector<double>(x,i);                var vy = new Vector<double>(y,i);                (vy + vx * vAlpha).copyTo(y,i);            }        }        for (; i < x.Length; ++i)            y[i] = y[i] + x[i] * Alpha;    }    public static voID daxpy_span(double Alpha,Span<double> y)    {        for (var i = 0; i < x.Length; ++i)            y[i] += x[i] * Alpha;    }    public static voID daxpy_vector(double Alpha,Span<double> y)    {        if (Vector.IsHarDWareAccelerated)        {            var vx = x.NonPortableCast<double,Vector<double>>();            var vy = y.NonPortableCast<double,Vector<double>>();            var vAlpha = new Vector<double>(Alpha);            for (var i = 0; i < vx.Length; ++i)                vy[i] += vx[i] * vAlpha;            x = x.Slice(Vector<double>.Count * vx.Length);            y = y.Slice(Vector<double>.Count * vy.Length);        }        for (var i = 0; i < x.Length; ++i)            y[i] += x[i] * Alpha;    }    public static voID daxpy_vector_no_slice(double Alpha,Span<double> y)    {        int i = 0;        if (Vector.IsHarDWareAccelerated)        {            var vx = x.NonPortableCast<double,Vector<double>>();            var vAlpha = new Vector<double>(Alpha);            for (i = 0; i < vx.Length; ++i)                vy[i] += vx[i] * vAlpha;            i = Vector<double>.Count * vx.Length;        }        for (; i < x.Length; ++i)            y[i] += x[i] * Alpha;    }}

这是使用dotnet build -c Release和dotnet run -c Release,dotnet –version报告“2.2.0-prevIEw1-008000”(不久之前的“每日”).

在我的桌面上,我希望差异会更好.

总结

以上是内存溢出为你收集整理的C#Vector.CopyTo几乎比非SIMD版本快？全部内容，希望文章能够帮你解决C#Vector.CopyTo几乎比非SIMD版本快？所遇到的程序开发问题。

如果觉得内存溢出网站内容还不错，欢迎将内存溢出网站推荐给程序员好友。