[stdlib] Fix bug in small string uninitialized init

[milseman]

Fix a bug and enforce the invariant that all bits between the last code unit
and the descriminator in a small string should be unset.

[milseman]

@atrick is this a proper use of initializeMemory? It's basically a convenient way to mask off our unused bytes and make sure they're set to 0.

[xwu]

@milseman Now that I'm looking at this again, I think this needs to rebind the memory; otherwise, after initializing memory as UInt8, portions of the memory are bound simultaneously as RawBitPattern.self and UInt8.self 😕

let count = self.count
let len = try withUnsafeMutableBytes(of: &self._storage) {
  (rawBufPtr: UnsafeMutableRawBufferPointer) -> Int in
  let len = try f(rawBufPtr)
  rawBufPtr.bindMemory(to: UInt8.self, capacity: count)
  defer {
    _ = rawBufPtr.bindMemory(to: RawBitPattern.self, capacity: 1)
  }
  UnsafeMutableRawBufferPointer(
    rebasing: rawBufPtr[len...]
  ).initializeMemory(as: UInt8.self, repeating: 0)
  return len
}

[atrick]

I don't see any API for initializing raw bytes without using a typed pointer. You can do the silly rebinding thing above or just write your own raw initialize that doesn't care about the in-memory type

var nextPtr = self
    for _ in 0..<count {
      Builtin.initialize(repeatedValue, nextPtr._rawValue)
      nextPtr += MemoryLayout<T>.stride
    }

[milseman]

@swift-ci please smoke test

[milseman]

@swift-ci please benchmark

[swift-ci]

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
AngryPhonebook	350	497	+42.0%	0.70x
AngryPhonebook.ASCII2.Small	159	219	+37.7%	0.73x
StringFromLongWholeSubstring	4	5	+25.0%	0.80x
NSStringConversion	450	546	+21.3%	0.82x (?)
NSStringConversion.Mutable	765	830	+8.5%	0.92x (?)
Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	6	5	-16.7%	1.20x
FlattenListLoop	2508	2198	-12.4%	1.14x (?)
LessSubstringSubstring	42	39	-7.1%	1.08x (?)
EqualStringSubstring	42	39	-7.1%	1.08x (?)
EqualSubstringString	42	39	-7.1%	1.08x (?)
LessSubstringSubstringGenericComparable	42	39	-7.1%	1.08x
SortStringsUnicode	3110	2905	-6.6%	1.07x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
AngryPhonebook.ASCII2.Small	159	223	+40.3%	0.71x
AngryPhonebook	350	490	+40.0%	0.71x
NSStringConversion	450	564	+25.3%	0.80x
StringFromLongWholeSubstring	4	5	+25.0%	0.80x
NSStringConversion.Rebridge	228	263	+15.4%	0.87x (?)
Array2D	6944	7520	+8.3%	0.92x (?)
String.replaceSubrange.String.Small	61	66	+8.2%	0.92x (?)
RandomShuffleLCG2	416	448	+7.7%	0.93x
Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	6	5	-16.7%	1.20x
EqualSubstringSubstringGenericEquatable	43	39	-9.3%	1.10x
FindString.Loop1.Substring	511	471	-7.8%	1.08x
LessSubstringSubstring	42	39	-7.1%	1.08x (?)
EqualSubstringSubstring	42	39	-7.1%	1.08x (?)
EqualStringSubstring	42	39	-7.1%	1.08x (?)
EqualSubstringString	42	39	-7.1%	1.08x (?)
LessSubstringSubstringGenericComparable	42	39	-7.1%	1.08x (?)
DropLastAnyCollectionLazy	29698	27720	-6.7%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
AngryPhonebook	546	686	+25.6%	0.80x
AngryPhonebook.ASCII2.Small	255	310	+21.6%	0.82x
NSStringConversion	1246	1372	+10.1%	0.91x (?)
Data.hash.Medium	54	59	+9.3%	0.92x (?)
SevenBoom	1431	1539	+7.5%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
SortStringsUnicode	4815	4490	-6.7%	1.07x (?)

Code size: -swiftlibs

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

[milseman]

@swift-ci please smoke test macOS platform

[milseman]

@swift-ci please smoke test macOS platform

[milseman]

@swift-ci please smoke test macOS platform

[milseman]

@swift-ci please smoke test macOS platform.

[milseman]

Waiting on swiftlang/llvm-project#2770

[milseman]

@swift-ci please smoke test

[milseman]

Seriously...

@swift-ci please smoke test

[milseman]

@swift-ci please smoke test linux platform

[milseman]

@swift-ci please smoke test windows platform

[milseman]

@swift-ci please smoke test windows platform