c#奇怪的方法内联行为 [英] c# Weird Method Inlining Behavior
问题描述
反编译我编写的代码时，我发现一个奇怪的内联行为.
我注意到的是，一个方法不会内联 UNLESS ，它处于循环状态.
奇怪的是，以非通用方式编写的此方法的等效项总是内联 .
I noticed an odd inlining behavior when decompiling a code that I wrote.
What I noticed is that a method will not get inlined UNLESS it is in a loop.
The weird thing is that the equivalent of this method, written in a non-generic way is getting inlined always.
代码:
using System;
using System.Runtime.CompilerServices;
using SharpLab.Runtime;
[JitGeneric(typeof(int))]
public static class GenericOps<T> where T : unmanaged
{
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static bool Less(T left, T right)
{
if (typeof(T) == typeof(byte)) return (byte)(object)left < (byte)(object)right;
if (typeof(T) == typeof(sbyte)) return (sbyte)(object)left < (sbyte)(object)right;
if (typeof(T) == typeof(ushort)) return (ushort)(object)left < (ushort)(object)right;
if (typeof(T) == typeof(short)) return (short)(object)left < (short)(object)right;
if (typeof(T) == typeof(uint)) return (uint)(object)left < (uint)(object)right;
if (typeof(T) == typeof(int)) return (int)(object)left < (int)(object)right;
if (typeof(T) == typeof(ulong)) return (ulong)(object)left < (ulong)(object)right;
if (typeof(T) == typeof(long)) return (long)(object)left < (long)(object)right;
if (typeof(T) == typeof(float)) return (float)(object)left < (float)(object)right;
if (typeof(T) == typeof(double)) return (double)(object)left < (double)(object)right;
throw new NotSupportedException(typeof(T).Name);
}
public static T Min(T left, T right)
{
return Less(left, right) ? left : right;
}
}
public static class IntOps
{
public static bool Less(int left, int right)
{
return left < right;
}
public static int Min(int left, int right)
{
return Less(left, right) ? left : right;
}
}
[JitGeneric(typeof(int))]
public static class C<T> where T : unmanaged
{
public static T M2(T a, T b)
{
return GenericOps<T>.Min(a, b);
}
public static T M2Loop(Span<T> a, Span<T> b)
{
T num = default;
for(int i = 0; i < a.Length; i++)
{
num = GenericOps<T>.Min(a[i], b[i]);
}
return num;
}
}
我注意到的奇怪行为:
- 在
GenericOps.Min()内，GenericOps.Less()未内联，与IntOps.Min()内嵌IntOps.Less()IS 的相反. - 在
C.M2()内部，GenericOps<T>.Min未内联，与C.M2Loop内嵌GenericOps<T>.MinIS 的相反.
- Inside
GenericOps.Min(),GenericOps.Less()IS NOT inlined, opposed toIntOps.Min()whereIntOps.Less()IS inlined. - Inside
C.M2(),GenericOps<T>.MinIS NOT inlined, opposed toC.M2LoopwhereGenericOps<T>.MinIS inlined.
反编译是使用SharpLab在X64平台Core CLR v4.700.20.41105 on amd64
Decompilation was done using SharpLab on X64 platform, Core CLR v4.700.20.41105 on amd64
反编译的JIT:
Microsoft.CodeAnalysis.EmbeddedAttribute..ctor()
L0000: ret
System.Runtime.CompilerServices.IsUnmanagedAttribute..ctor()
L0000: ret
GenericOps`1[[System.Int32, System.Private.CoreLib]].Less(Int32, Int32)
L0000: cmp ecx, edx
L0002: setl al
L0005: movzx eax, al
L0008: ret
GenericOps`1[[System.Int32, System.Private.CoreLib]].Min(Int32, Int32)
L0000: push rdi
L0001: push rsi
L0002: sub rsp, 0x28
L0006: mov esi, ecx
L0008: mov edi, edx
L000a: mov ecx, esi
L000c: mov edx, edi
L000e: call GenericOps`1[[System.Int32, System.Private.CoreLib]].Less(Int32, Int32)
L0013: test eax, eax
L0015: jne short L0020
L0017: mov eax, edi
L0019: add rsp, 0x28
L001d: pop rsi
L001e: pop rdi
L001f: ret
L0020: mov eax, esi
L0022: add rsp, 0x28
L0026: pop rsi
L0027: pop rdi
L0028: ret
IntOps.Less(Int32, Int32)
L0000: cmp ecx, edx
L0002: setl al
L0005: movzx eax, al
L0008: ret
IntOps.Min(Int32, Int32)
L0000: cmp ecx, edx
L0002: jl short L0007
L0004: mov eax, edx
L0006: ret
L0007: mov eax, ecx
L0009: ret
C`1[[System.Int32, System.Private.CoreLib]].M2(Int32, Int32)
L0000: push rdi
L0001: push rsi
L0002: sub rsp, 0x28
L0006: mov esi, ecx
L0008: mov edi, edx
L000a: mov ecx, esi
L000c: mov edx, edi
L000e: call GenericOps`1[[System.Int32, System.Private.CoreLib]].Less(Int32, Int32)
L0013: test eax, eax
L0015: jne short L001b
L0017: mov eax, edi
L0019: jmp short L001d
L001b: mov eax, esi
L001d: add rsp, 0x28
L0021: pop rsi
L0022: pop rdi
L0023: ret
C`1[[System.Int32, System.Private.CoreLib]].M2Loop(System.Span`1<Int32>, System.Span`1<Int32>)
L0000: push rsi
L0001: sub rsp, 0x20
L0005: mov r8, [rcx]
L0008: mov ecx, [rcx+8]
L000b: xor eax, eax
L000d: xor r9d, r9d
L0010: test ecx, ecx
L0012: jle short L0046
L0014: mov r10d, [rdx+8]
L0018: mov rdx, [rdx]
L001b: movsxd rax, r9d
L001e: mov r11d, [r8+rax*4]
L0022: cmp r9d, r10d
L0025: jae short L004c
L0027: mov eax, [rdx+rax*4]
L002a: cmp r11d, eax
L002d: setl sil
L0031: movzx esi, sil
L0035: test esi, esi
L0037: jne short L003b
L0039: jmp short L003e
L003b: mov eax, r11d
L003e: inc r9d
L0041: cmp r9d, ecx
L0044: jl short L001b
L0046: add rsp, 0x20
L004a: pop rsi
L004b: ret
L004c: call 0x00007ffbf05bfc60
L0051: int3
有人能对此怪异行为做出解释吗?
并有一种方法可以确保GenericOps.Less()和GenericOps.Min()始终内联吗?
Can anyone offer an explanation for this weird behavoir?
And is there a way to guarantee that GenericOps.Less() and GenericOps.Min() will always get inlined?
推荐答案
您的第一点很容易解释. throw语句可防止内联.请参考.net源代码，以了解它们如何通过使用ThrowHelper类将抛出异常移出主要方法主体.将throw语句移至其他方法以增加内联的机会.没有什么可以保证的.
Your first point is easy to explain. throw statements prevent inlining. Refer to .net source code to see how they move throwing exceptions out of main method body by using a ThrowHelper class. Move the throw statement to some other method to increase your chances of inlining. Nothing ever guarantees it.
关于另一点，我猜想jit决定了调用开销将足够大，以至于循环体中的内联将抵消增加的代码大小.
As for the other point, I am guessing the jit determines the call overhead will be large enough that inlining within the loop body will more than offset the increased code size.
这篇关于c#奇怪的方法内联行为的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
