could JIT can inline closure Action?

[egmkang]

void Foreach(Action<int> fn)
{
    foreach(var item in list)
    {
        fn(item);
    }
}

void test()
{
    var sum = 0;
    Foreach((item) => sum += item);
}

in the test function, call Foreach function will new a small object closure.

can CoreCLR JIT Engine inline this simple closure call?

[omariom]

Theoretically, if JIT inlines call to Foreach at a particular callsite, it could find that the delegate points to a closure, get rid of the closure allocation, inline the call and "inline" the closure's fields as local variables.

[EgorBo]

Theoretically, if JIT inlines call to Foreach

Methods with loops never inline, they simply don't pass this check:

runtime/src/coreclr/src/jit/inlinepolicy.cpp

Lines 492 to 495 in a605729

	else if (!m_IsForceInline && (basicBlockCount > MAX_BASIC_BLOCKS))
	{
	SetNever(InlineObservation::CALLEE_TOO_MANY_BASIC_BLOCKS);
	}

(well, if [AgressiveInlining] is not set of course)

[sakno]

Related to #6498

Methods with loops never inline, they simply don't pass this check:

Author asked about inlining of delegate invocation inside of this method.

[AndyAyersMS]

Methods with loops never inline

Are you sure? Simple loops can have 5 or fewer BBs.

can CoreCLR JIT Engine inline this simple closure call?

Following up on what @sakno said, there are some notes on what it would take to inline a delegate invoke starting at this comment, and a number of issues to sort through before we're anywhere close to being able to do this.

[jkotas]

Duplicate of #6498 and #4584