Benchmarks for #116241 (EgorBo)

[EgorBot]

Processing dotnet/runtime#116241 (comment) command:

Command

-amd -arm

using System.Runtime.CompilerServices;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;

BenchmarkSwitcher.FromAssembly(typeof(Benchmarks).Assembly).Run(args);

public class Benchmarks
{
    int class1 = 0;
    int class2 = 0;

    BaseClass b1 = new DerivedClass1();
    BaseClass b2 = new DerivedClass2();

    [Benchmark]
    public void Bench()
    {
        for (int i = 0; i < 10000; i++)
            Problem(i);
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    public void Problem(int i)
    {
        BaseClass b = i % 3 == 0 ? b1 : b2;
        if (b is DerivedClass1)
            class1++;
        else if (IsDerivedClass2(b))
            class2++;
    }

    public bool IsDerivedClass2(BaseClass b) => b is DerivedClass2;

    public class BaseClass { }
    public class DerivedClass1 : BaseClass { }
    public class DerivedClass2 : BaseClass { }
}

(EgorBot will reply in this issue)

[EgorBot]

Benchmark results on linux_azure_genoa

BenchmarkDotNet v0.14.0, Ubuntu 24.04.2 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 16 logical and 8 physical cores
  Job-QABILF : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-TWRPVJ : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

Method	Toolchain	Mean	Error	Ratio
Bench	Main	30.91 μs	0.007 μs	1.00
Bench	PR	38.21 μs	0.028 μs	1.24

BDN_Artifacts.zip

cc @EgorBo (agent_logs.txt). EgorBot manual: link.

[EgorBot]

Benchmark results on linux_azure_cobalt100

BenchmarkDotNet v0.14.0, Ubuntu 24.04.2 LTS (Noble Numbat)
azure_cobalt100
  Job-DTXCCP : .NET 10.0.0 (42.42.42.42424), Arm64 RyuJIT AdvSIMD
  Job-EYDNGE : .NET 10.0.0 (42.42.42.42424), Arm64 RyuJIT AdvSIMD

Method	Toolchain	Mean	Error	Ratio
Bench	Main	36.93 μs	0.014 μs	1.00
Bench	PR	45.16 μs	0.042 μs	1.22

BDN_Artifacts.zip

cc @EgorBo (agent_logs.txt). EgorBot manual: link.

[EgorBo]

@EgorBot -amd --envvars DOTNET_JitDisasm:Problem

using System.Runtime.CompilerServices;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;

BenchmarkSwitcher.FromAssembly(typeof(Benchmarks).Assembly).Run(args);

public class Benchmarks
{
    int class1 = 0;
    int class2 = 0;

    BaseClass b1 = new DerivedClass1();
    BaseClass b2 = new DerivedClass2();

    [Benchmark]
    public void Bench()
    {
        for (int i = 0; i < 10000; i++)
            Problem(i);
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    public void Problem(int i)
    {
        BaseClass b = i % 3 == 0 ? b1 : b2;
        if (b is DerivedClass1)
            class1++;
        else if (IsDerivedClass2(b))
            class2++;
    }

    public bool IsDerivedClass2(BaseClass b) => b is DerivedClass2;

    public class BaseClass { }
    public class DerivedClass1 : BaseClass { }
    public class DerivedClass2 : BaseClass { }
}

[EgorBot]

Benchmark results on linux_azure_genoa

BenchmarkDotNet v0.14.0, Ubuntu 24.04.2 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 16 logical and 8 physical cores
  Job-OOZDCU : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-DKTLLQ : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
EnvironmentVariables=DOTNET_JitDisasm=Problem

Method	Toolchain	Mean	Error	Ratio
Bench	Main	30.90 μs	0.007 μs	1.00
Bench	PR	38.17 μs	0.004 μs	1.24

BDN_Artifacts.zip

cc @EgorBo (agent_logs.txt). EgorBot manual: link.

[EgorBo]

; Assembly listing for method Benchmarks:Problem(int):this (Tier1)
; Emitting BLENDED_CODE for generic X64 + VEX + EVEX on Unix
; Tier1 code
; optimized code
; optimized using Synthesized PGO
; rbp based frame
; partially interruptible
; with Synthesized PGO: fgCalledCount is 4119
; 1 inlinees with PGO data; 0 single block inlinees; 0 inlinees without PGO data

G_M000_IG01:                ;; offset=0x0000
       push     rbp
       push     r15
       push     rbx
       lea      rbp, [rsp+0x10]
       mov      rbx, rdi
 
G_M000_IG02:                ;; offset=0x000C
       mov      edx, 0x55555556
; ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ (mov: 1) 32B boundary ...............................
       mov      eax, edx
       imul     edx:eax, esi
       mov      edi, edx
       shr      edi, 31
       add      edi, edx
       lea      edi, [rdi+2*rdi]
       sub      esi, edi
       je       SHORT G_M000_IG07
 
G_M000_IG03:                ;; offset=0x0023
       mov      r15, gword ptr [rbx+0x10]
 
G_M000_IG04:                ;; offset=0x0027
       mov      rsi, r15
       mov      rdi, 0x769F22D06560
; ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ (mov: 4) 32B boundary ...............................
       mov      rax, 0x769F229C6868
       call     [rax]CORINFO_HELP_ISINSTANCEOFCLASS
       test     rax, rax
       jne      SHORT G_M000_IG08
 
G_M000_IG05:                ;; offset=0x0045
       mov      rsi, r15
       mov      rdi, 0x769F22D065F8
; ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ (mov: 2) 32B boundary ...............................
       mov      rax, 0x769F229C6868
       call     [rax]CORINFO_HELP_ISINSTANCEOFCLASS
       test     rax, rax
       je       SHORT G_M000_IG06
       inc      dword ptr [rbx+0x1C]
 
G_M000_IG06:                ;; offset=0x0066
       pop      rbx
       pop      r15
       pop      rbp
       ret      
 
G_M000_IG07:                ;; offset=0x006B
       mov      r15, gword ptr [rbx+0x08]
       jmp      SHORT G_M000_IG04
; ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ (jmp: 1 ; jcc erratum) 32B boundary ...............................
 
G_M000_IG08:                ;; offset=0x0071
       inc      dword ptr [rbx+0x18]
       jmp      SHORT G_M000_IG06
 
; Total bytes of code 118

looks like the casts no longer expand