peephole optimization for lea command #67187
Description
Description
I am not sure whether this is a bug or known issue or just that my environment is out of time. Anyway, let the code talks:
public struct Foo
{
public int a;
public int b;
}
public class StructContainer
{
Foo foo;
public void Update()
{
this.foo.a += 1;
this.foo.b += 1;
}
}
public class NormalContainer
{
int a;
int b;
public void Update()
{
this.a += 1;
this.b += 1;
}
}
[Benchmark]
public StructContainer StructContainerUpdate()
{
var sc = new StructContainer();
sc.Update();
return sc;
}
[Benchmark]
public NormalContainer NormalContainerUpdate()
{
var nc = new NormalContainer();
nc.Update();
return nc;
}Generated code:
## .NET 6.0.3 (6.0.322.12309), X64 RyuJIT
; Benchmarks.StructInline.StructContainerUpdate()
sub rsp,28
mov rcx,offset MT_Benchmarks.StructInline+StructContainer
call CORINFO_HELP_NEWSFAST
lea rdx,[rax+8] ; we do not really need this line
inc dword ptr [rdx]
lea rdx,[rax+0C]
inc dword ptr [rdx]
add rsp,28
ret
; Total bytes of code 36
; Benchmarks.StructInline.NormalContainerUpdate()
sub rsp,28
mov rcx,offset MT_Benchmarks.StructInline+NormalContainer
call CORINFO_HELP_NEWSFAST
inc dword ptr [rax+8]
inc dword ptr [rax+0C]
add rsp,28
ret
; Total bytes of code 30I know little about CLR, so I may have seriously underestimated the complexity of it. But it feels like the lea command is bit of not clever, right?
The memory layout will not change. So something like this.struct1.struct2.struct3.struct4.property could always be calculated by [rax+offset](if the offset is not too large), isn't it?
We does have this, so this problem is pretty limited, should be another peephole optimization I think.
Configuration
BenchmarkDotNet=v0.13.1, OS=Windows 10.0.22000
Intel Core i9-10900X CPU 3.70GHz, 1 CPU, 20 logical and 10 physical cores
.NET SDK=6.0.201
Regression?
Data
Analysis
category:cq
theme:optimization