[Wasm][AOT] Invoking Action<struct> is 7x slower than Action<ref>

[jeromelaban]

Description

Considering this code, using Mixed mode AOT without profile;

private static void TestRegisterColor()
{
    var sw = Stopwatch.StartNew();
    int value = 0;
    for (int i = 0; i < 10000000; i++)
    {
        RegisterColor(null, c => value++);
    }
    sw.Stop();
    Console.WriteLine($"Color: {sw.ElapsedMilliseconds}");
}

private static void TestRegisterOther()
{
    var sw = Stopwatch.StartNew();
    int value = 0;
    for (int i = 0; i < 10000000; i++)
    {
        RegisterOther(null, c => value++);
    }
    sw.Stop();
    Console.WriteLine($"Other: {sw.ElapsedMilliseconds}");
}

private static void RegisterColor(object o, Action<Color> action)
{
    action(new Color());
}

private static void RegisterOther(object o, Action<IDisposable> action)
{
    action(null);
}

and

public struct Color
{
  public byte A { get; set; }
  public byte B { get; set; }
  public byte G { get; set; }
  public byte R { get; set; }
}

Results are:

Color: 1710
Other: 259

Repro: https://github.com/jeromelaban/Wasm.Samples/tree/master/Bug50757/Bug50757

Configuration

b7a1648

Regression?

No

Other information

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[kg]

I doubt it will significantly remove the gap, but maybe return an existing Color instance instead of a new one? This test case has to create a new instance and zero it (not THAT expensive, but definitely Work) then pass it as an argument value, vs the ref one just passing a nullptr.

[jeromelaban]

There are workarounds for that case specifically for sure (e.g. not use a generic delegate is a very good start), yet there are many cases where this is not possible to change (in the BCL or existing nuget packages).

I'd find interesting for the sharedvt generics exclusion when running mixed-mode AOT to be configurable, as the performance impact is far from neutral.

[kg]

I just meant from the perspective of "if this is fixed the two timing values should be ~equal", I absolutely agree

[ghost]

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

Issue Details

---

Description

Considering this code, using Mixed mode AOT without profile;

private static void TestRegisterColor()
{
    var sw = Stopwatch.StartNew();
    int value = 0;
    for (int i = 0; i < 10000000; i++)
    {
        RegisterColor(null, c => value++);
    }
    sw.Stop();
    Console.WriteLine($"Color: {sw.ElapsedMilliseconds}");
}

private static void TestRegisterOther()
{
    var sw = Stopwatch.StartNew();
    int value = 0;
    for (int i = 0; i < 10000000; i++)
    {
        RegisterOther(null, c => value++);
    }
    sw.Stop();
    Console.WriteLine($"Other: {sw.ElapsedMilliseconds}");
}

private static void RegisterColor(object o, Action<Color> action)
{
    action(new Color());
}

private static void RegisterOther(object o, Action<IDisposable> action)
{
    action(null);
}

and

public struct Color
{
  public byte A { get; set; }
  public byte B { get; set; }
  public byte G { get; set; }
  public byte R { get; set; }
}

Results are:

Color: 1710
Other: 259

Repro: https://github.com/jeromelaban/Wasm.Samples/tree/master/Bug50757/Bug50757

Configuration

b7a1648

Regression?

No

Other information

Author:	jeromelaban
Assignees:	-
Labels:	area-VM-meta-mono, untriaged
Milestone:	-

[CoffeeFlux]

FYI @jeffschwMSFT most of these wasm AOT-specific issues should just get the AOT label.

[jeromelaban]

As per @vargaz's help, adding -O=gsharedvt to aot-cross allows for gsharedvt code to be AOTed, and gives ~x2 performance improvement in Action<struct> invocation performance.

[SamMonoRT]

@jeromelaban @vargaz - based on above comment from @jeromelaban - this seems completed ?

[SamMonoRT]

Closing this, @jeromelaban please reopen if this is still an issue.