继续进行F#性能测试。有关更多背景信息,请参见此处:
f# NativePtr.stackalloc in Struct Constructor
F# NativePtr.stackalloc Unexpected Stack Overflow
现在我已经让堆栈数组在F#中工作了。然而,出于某种原因,等效的C#速度大约快50倍。我已经包含了以下ILSpy反编译版本,它似乎只有1行真正不同(在stackAlloc内部)。这里发生了什么?难道未经检查的算术确实导致了如此大的差异吗?不确定我该如何测试这个?
https://msdn.microsoft.com/en-us/library/a569z7k8.aspx
F#代码:
f# NativePtr.stackalloc in Struct Constructor
F# NativePtr.stackalloc Unexpected Stack Overflow
现在我已经让堆栈数组在F#中工作了。然而,出于某种原因,等效的C#速度大约快50倍。我已经包含了以下ILSpy反编译版本,它似乎只有1行真正不同(在stackAlloc内部)。这里发生了什么?难道未经检查的算术确实导致了如此大的差异吗?不确定我该如何测试这个?
https://msdn.microsoft.com/en-us/library/a569z7k8.aspx
F#代码:
#nowarn "9"
open Microsoft.FSharp.NativeInterop
open System
open System.Diagnostics
open System.Runtime.CompilerServices
[<MethodImpl(MethodImplOptions.NoInlining)>]
let stackAlloc x =
let mutable ints:nativeptr<byte> = NativePtr.stackalloc x
()
[<EntryPoint>]
let main argv =
printfn "%A" argv
let size = 8192
let reps = 10000
stackAlloc size // JIT
let clock = Stopwatch()
clock.Start()
for i = 1 to reps do
stackAlloc size
clock.Stop()
let elapsed = clock.Elapsed.TotalMilliseconds
let description = "F# NativePtr.stackalloc"
Console.WriteLine("{0} ({1} bytes, {2} reps): {3:#,##0.####}ms", description, size, reps, elapsed)
Console.ReadKey() |> ignore
0
C# 代码
using System;
using System.Diagnostics;
namespace CSharpLanguageFeatures
{
class CSharpStackArray
{
static void Main(string[] args)
{
int size = 8192;
int reps = 10000;
stackAlloc(size); // JIT
Stopwatch clock = new Stopwatch();
clock.Start();
for (int i = 0; i < reps; i++)
{
stackAlloc(size);
}
clock.Stop();
string elapsed = clock.Elapsed.TotalMilliseconds.ToString("#,##0.####");
string description = "C# stackalloc";
Console.WriteLine("{0} ({1} bytes, {2} reps): {3:#,##0.####}ms", description, size, reps, elapsed);
Console.ReadKey();
}
public unsafe static void stackAlloc(int arraySize)
{
byte* pArr = stackalloc byte[arraySize];
}
}
}
using Microsoft.FSharp.Core;
using System;
using System.Diagnostics;
using System.IO;
using System.Runtime.CompilerServices;
[CompilationMapping(SourceConstructFlags.Module)]
public static class FSharpStackArray
{
[MethodImpl(MethodImplOptions.NoInlining)]
public unsafe static void stackAlloc(int x)
{
IntPtr ints = stackalloc byte[x * sizeof(byte)];
}
[EntryPoint]
public static int main(string[] argv)
{
PrintfFormat<FSharpFunc<string[], Unit>, TextWriter, Unit, Unit> format = new PrintfFormat<FSharpFunc<string[], Unit>, TextWriter, Unit, Unit, string[]>("%A");
PrintfModule.PrintFormatLineToTextWriter<FSharpFunc<string[], Unit>>(Console.Out, format).Invoke(argv);
FSharpStackArray.stackAlloc(8192);
Stopwatch clock = new Stopwatch();
clock.Start();
for (int i = 1; i < 10001; i++)
{
FSharpStackArray.stackAlloc(8192);
}
clock.Stop();
double elapsed = clock.Elapsed.TotalMilliseconds;
Console.WriteLine("{0} ({1} bytes, {2} reps): {3:#,##0.####}ms", "F# NativePtr.stackalloc", 8192, 10000, elapsed);
ConsoleKeyInfo consoleKeyInfo = Console.ReadKey();
return 0;
}
}
C# 版本反编译
using System;
using System.Diagnostics;
namespace CSharpLanguageFeatures
{
internal class CSharpStackArray
{
private static void Main(string[] args)
{
int size = 8192;
int reps = 10000;
CSharpStackArray.stackAlloc(size);
Stopwatch clock = new Stopwatch();
clock.Start();
for (int i = 0; i < reps; i++)
{
CSharpStackArray.stackAlloc(size);
}
clock.Stop();
string elapsed = clock.Elapsed.TotalMilliseconds.ToString("#,##0.####");
string description = "C# stackalloc";
Console.WriteLine("{0} ({1} bytes, {2} reps): {3:#,##0.####}ms", new object[]
{
description,
size,
reps,
elapsed
});
Console.ReadKey();
}
public unsafe static void stackAlloc(int arraySize)
{
IntPtr arg_06_0 = stackalloc byte[checked(unchecked((UIntPtr)arraySize) * 1)];
}
}
}
F#版本的IL字节分配
.method public static
void stackAlloc (
int32 x
) cil managed noinlining
{
// Method begins at RVA 0x2050
// Code size 13 (0xd)
.maxstack 4
.locals init (
[0] native int ints
)
IL_0000: nop
IL_0001: ldarg.0
IL_0002: sizeof [mscorlib]System.Byte
IL_0008: mul
IL_0009: localloc
IL_000b: stloc.0
IL_000c: ret
} // end of method FSharpStackArray::stackAlloc
C# 版本 IL - 字节分配
.method public hidebysig static
void stackAlloc (
int32 arraySize
) cil managed
{
// Method begins at RVA 0x2094
// Code size 8 (0x8)
.maxstack 8
IL_0000: ldarg.0
IL_0001: conv.u
IL_0002: ldc.i4.1
IL_0003: mul.ovf.un
IL_0004: localloc
IL_0006: pop
IL_0007: ret
} // end of method CSharpStackArray::stackAlloc
更新的F# IL - IntPtr分配
(该内容是关于IT技术方面的).method public static
void stackAlloc (
int32 x
) cil managed noinlining
{
// Method begins at RVA 0x2050
// Code size 13 (0xd)
.maxstack 4
.locals init (
[0] native int ints
)
IL_0000: nop
IL_0001: ldarg.0
IL_0002: sizeof [mscorlib]System.IntPtr
IL_0008: mul
IL_0009: localloc
IL_000b: stloc.0
IL_000c: ret
} // end of method FSharpStackArray::stackAlloc
更新的C# IL - IntPtr分配
.method public hidebysig static
void stackAlloc (
int32 arraySize
) cil managed
{
// Method begins at RVA 0x2415
// Code size 13 (0xd)
.maxstack 8
IL_0000: ldarg.0
IL_0001: conv.u
IL_0002: sizeof [mscorlib]System.IntPtr
IL_0008: mul.ovf.un
IL_0009: localloc
IL_000b: pop
IL_000c: ret
} // end of method CSharpStackArray::stackAlloc
stackAlloc
方法的实际 IL 吗? - Fyodor Soikinintptr
本身。似乎 C# 编译器通过使用sizeof(byte)
总是等于1
的知识来优化这个计算。 - Fyodor SoikinF#
被标记为noinline
。将其添加到C#
版本中,但没有太大的区别。在这种情况下,确实需要看到执行的确切x64代码。不幸的是,JIT:er会根据是否连接了调试器来更改JIT:ed代码,这使得这一点更加困难。可以在循环内部生成崩溃并获取内存转储,然后使用windebug检查该代码。 - Just another metaprogrammer