[stdlib] Set, Dictionary: Make Cocoa indices resilient for now

[lorentey]

This makes _CocoaSet.Index and _CocoaDictionary.Index fully resilient for now, allowing us to replace/fix its internals at any point in the future.

[lorentey]

@swift-ci please smoke benchmark

[lorentey]

@swift-ci please smoke test

[lorentey]

cc @airspeedswift What do you think? This leaves the door open for a Cocoa indexing overhaul.

[lorentey]

Hm, this seems a bit excessive.

[lorentey]

Gosh; Dictionary.Keys and .Values use IndexingIterator. That won't do at all.

[lorentey]

I think #19576 (for swapAt) and #19602 (for Histogram et al) will cut down these performance regressions.

[lorentey]

@swift-ci please smoke test

[lorentey]

@swift-ci please smoke benchmark

[swift-ci]

Build comment file:

Build failed before running benchmark.

---

[lorentey]

@swift-ci please smoke test

[lorentey]

@swift-ci please smoke benchmark

[lorentey]

Hm, I'll have to make the Cocoa Index storage representation @_fixed_layout to reduce these regressions.

I'll do that by converting it to a CoW type, which I was planning to do anyway.

[lorentey]

@swift-ci please smoke benchmark

[lorentey]

@swift-ci please smoke test

[lorentey]

Oops, I forgot to apply these changes to Dictionary; I'm afraid these next benchmarks will still disappoint

[lorentey]

@swift-ci please smoke benchmark

[lorentey]

@swift-ci please smoke test

[lorentey]

Sigh

@swift-ci please smoke test linux platform

[lorentey]

Okay, this should do the trick. (He said, with a naive look on his face.)

[swift-ci]

Build comment file:

Performance: -O

TEST	OLD	NEW	DELTA	RATIO
Improvement
CStringLongAscii	3552	3299	-7.1%	1.08x (?)

Code size: -O

TEST	OLD	NEW	DELTA	RATIO
Regression
DictOfArraysToArrayOfDicts.o	34702	35142	+1.3%	0.99x
Improvement
DictionarySwap.o	32475	31406	-3.3%	1.03x
RGBHistogram.o	28401	27517	-3.1%	1.03x

Performance: -Osize

TEST	OLD	NEW	DELTA	RATIO
Improvement
CStringLongAscii	3543	3288	-7.2%	1.08x
FloatingPointPrinting_Float_description_small	5673	5282	-6.9%	1.07x

Code size: -Osize

TEST	OLD	NEW	DELTA	RATIO
Improvement
DictionarySwap.o	28555	27331	-4.3%	1.04x
RGBHistogram.o	24881	24077	-3.2%	1.03x

Performance: -Onone

TEST	OLD	NEW	DELTA	RATIO
Regression
CharIteration_utf16_unicodeScalars_Backwards	248178	304188	+22.6%	0.82x
ArrayOfPOD	758	846	+11.6%	0.90x (?)
CharIteration_chinese_unicodeScalars_Backwards	225786	247267	+9.5%	0.91x

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

---

[lorentey]

Oh hi! I'm just sitting over here, improving benchmarks by making things non-inlinable.

(This probably just means we lack indexing benchmarks for bridged sets/dictionaries.)

[jrose-apple]

Is this correct? It writes to the hasher, after all.

[lorentey]

Oh, oops, good catch!