Suboptimal code generated for even a trivial ValueConverter

[hez2010]

Description

I'm seeing suboptimal code generated for ValueConverters even when what the converter does is trivial.

For example,

class Program
{
 private static readonly Foo foo = new();
 private static readonly CultureInfo culture = CultureInfo.CurrentCulture;

 public int Test()
 {
 var x = 3;
 var y = (int)foo.Convert(x, typeof(int), null, culture);
 return y;
 }
}

interface IValueConverter
{
 object Convert(object value, Type targetType, object? parameter, CultureInfo? culture);
}

sealed class Foo : IValueConverter
{
 public object Convert(object value, Type targetType, object? parameter, CultureInfo? culture)
 {
 return (int)value + 1;
 }
}

The codegen:

; Assembly listing for method Program:Test():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Windows
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data

G_M000_IG01: ;; offset=0000H
 push rdi
 push rsi
 sub rsp, 40
 
G_M000_IG02: ;; offset=0006H
 mov rsi, 0x7FF7BD400858
 mov rcx, rsi
 call CORINFO_HELP_NEWSFAST
 mov rdi, rax
 test byte ptr [(reloc 0x7ff7bd6f9344)], 1
 je SHORT G_M000_IG05
 
G_M000_IG03: ;; offset=0024H
 mov rcx, 0x23591401DA8
 mov rcx, gword ptr [rcx]
 mov dword ptr [rdi+08H], 3
 cmp byte ptr [rcx], cl
 mov rcx, rsi
 call CORINFO_HELP_NEWSFAST
 mov ecx, dword ptr [rdi+08H]
 inc ecx
 mov dword ptr [rax+08H], ecx
 mov eax, dword ptr [rax+08H]
 
G_M000_IG04: ;; offset=004DH
 add rsp, 40
 pop rsi
 pop rdi
 ret 
 
G_M000_IG05: ;; offset=0054H
 mov rcx, 0x7FF7BD6F9310
 mov edx, 4
 call CORINFO_HELP_GETSHARED_NONGCSTATIC_BASE
 jmp SHORT G_M000_IG03
 
; Total bytes of code 106

This is important because ValueConverters are heavily used in Desktop apps that use XAML (WPF, UWP, MAUI and etc.), and we want to minimize the overhead caused by ValueConverters.

Expected codegen:

mov eax, 4
ret

Configuration

.NET 8 preview 3

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Description

I'm seeing suboptimal code generated for ValueConverters even when the what the converter does is trivial.

For example,

class Program
{
 private static readonly Foo foo = new();
 private static readonly CultureInfo culture = CultureInfo.CurrentCulture;

 public int Test()
 {
 var x = 3;
 var y = (int)foo.Convert(x, typeof(int), null, culture);
 return y;
 }
}

interface IValueConverter
{
 object Convert(object value, Type targetType, object? parameter, CultureInfo? culture);
}

sealed class Foo : IValueConverter
{
 public object Convert(object value, Type targetType, object? parameter, CultureInfo? culture)
 {
 return (int)value + 1;
 }
}

The codegen:

; Assembly listing for method Program:Test():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Windows
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data

G_M000_IG01: ;; offset=0000H
 push rdi
 push rsi
 sub rsp, 40
 
G_M000_IG02: ;; offset=0006H
 mov rsi, 0x7FF7BD400858
 mov rcx, rsi
 call CORINFO_HELP_NEWSFAST
 mov rdi, rax
 test byte ptr [(reloc 0x7ff7bd6f9344)], 1
 je SHORT G_M000_IG05
 
G_M000_IG03: ;; offset=0024H
 mov rcx, 0x23591401DA8
 mov rcx, gword ptr [rcx]
 mov dword ptr [rdi+08H], 3
 cmp byte ptr [rcx], cl
 mov rcx, rsi
 call CORINFO_HELP_NEWSFAST
 mov ecx, dword ptr [rdi+08H]
 inc ecx
 mov dword ptr [rax+08H], ecx
 mov eax, dword ptr [rax+08H]
 
G_M000_IG04: ;; offset=004DH
 add rsp, 40
 pop rsi
 pop rdi
 ret 
 
G_M000_IG05: ;; offset=0054H
 mov rcx, 0x7FF7BD6F9310
 mov edx, 4
 call CORINFO_HELP_GETSHARED_NONGCSTATIC_BASE
 jmp SHORT G_M000_IG03
 
; Total bytes of code 106

This is important because ValueConverters are heavily used in Desktop apps that uses XAML (WPF, UWP, MAUI and etc.), and we want to minimize the overhead caused by ValueConverters.

Expected codegen:

mov eax, 4
ret

Configuration

.NET 8 preview 3

Author:	hez2010
Assignees:	-
Labels:	tenet-performance, area-CodeGen-coreclr
Milestone:	-

[EgorBo]

We call it "multi-use boxing" problem, #9118

[EgorBo]

cc @markples

[hez2010]

Note that the code in the original post is a trivial case. In real-world cases, there'll be some type tests in the converter to avoid casting exceptions.

For example,

if (value is int d) return d + 1;
else return default(int);

The return value is boxed.

[markples]

Thank you for the example.

• The order of the first allocation and type initialization check/call require the allocated object to be spilled. Of course, this is moot if the allocation is removed.
• The input allocation is unnecessary.
• The null check on foo (I think?) remains. I suppose readonly+nonnull in cctor could help here.
• The output allocation is unnecessary, and in the more complicated example there are two allocations, one on each path, making it a bit more complicated. (Well, I assume so.. either the source or JIT could merge them.)

I'll include this in #9118.