c# Weird Method Inlining Behavior

Question

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.

The code:

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;
    }
}

Odd behavior that I noticed:

• Inside GenericOps.Min(), GenericOps.Less() IS NOT inlined, opposed to IntOps.Min() where IntOps.Less() IS inlined.
• Inside C.M2(), GenericOps<T>.Min IS NOT inlined, opposed to C.M2Loop where GenericOps<T>.Min IS inlined.

Decompilation was done using SharpLab on X64 platform, Core CLR v4.700.20.41105 on amd64

Decompiled 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

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?

Answer 1

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.

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.