[SILGen] force immeidate LValue assignment cleanups.

[atrick]

In SILGenFunction::emitAssignToLValue, use an ArgumentScope instead of
a FormalEvaluationScope. This ensures that the lifetime of objects
needed to materialize each LValue component do not overlap.

For example in this tuple assignment:

public func testTupleAssign(x: inout [Int]) {
(x[0], x[1]) = (0, 1)
}

Array.subscript.nativeOwningMutableAddressor keeps a reference to the
array across the assignment, unnecessarily forcing a copy of the
array.

Fixes SR-8621: SILGen creates destroys for tuple assignment at the
wrong places.

[atrick]

@swift-ci test.

[atrick]

@swift-ci test source compatibility.

[rjmccall]

LGTM.

[eeckstein]

@swift-ci please smoke benchmark staging

[swift-ci]

Build comment file:

Performance: -O

TEST	OLD	NEW	DELTA	SPEEDUP
Improvement
Prims	5543	894	-83.9%	6.20x
PrimsSplit	5498	907	-83.5%	6.06x

Performance: -Osize

TEST	OLD	NEW	DELTA	SPEEDUP
Improvement
PrimsSplit	5687	978	-82.8%	5.81x
Prims	5566	977	-82.4%	5.70x

How to read the data

The tables contain differences in performance which are larger than 8% and differences in code size which are larger than 1%.

If you see any unexpected regressions, you should consider fixing the regressions before you merge the PR.

Noise: Sometimes the performance results (not code size!) contain false alarms. Unexpected regressions which are marked with '(?)' are probably noise. If you see regressions which you cannot explain you can try to run the benchmarks again. If regressions still show up, please consult with the performance team (@eeckstein).

Hardware Overview

  Model Name: Mac Pro
  Model Identifier: MacPro6,1
  Processor Name: 12-Core Intel Xeon E5
  Processor Speed: 2.7 GHz
  Number of Processors: 1
  Total Number of Cores: 12
  L2 Cache (per Core): 256 KB
  L3 Cache: 30 MB
  Memory: 64 GB

---