[stdlib] Make use of protocol requirements to convert from concrete floating-point types

[xwu]

This PR is about picking the low-hanging fruit in terms of recovering performance during generic floating-point conversion. It is so low-hanging, in fact, it's actually partly a reversion to the original placeholder implementation of generic floating-point conversion.

In brief, we never removed the requirements for init(_: Float), init(_: Double), and init(_: Float80) from the protocol, and now that we have ABI stability, these requirements will be present forever. So, we can call them from the generic implementation.

There is one caveat to be highlighted: As in other situations (see #30417), the compiler regards the generic initializer itself as a suitable default implementation of init(_: Float) and friends, and therefore it will not warn if a concrete type does not provide its own implementations of the latter. Currently, this means that the concrete type will unwittingly have a very slow implementation; with this change, the concrete type will have an implementation that crashes at runtime due to infinite recursion. I think it's arguable as to which one is the worse outcome; this PR doesn't cause or otherwise change the underlying issue.

[xwu]

cc @troughton

I figure it's best to start small and chip away at this issue gradually. This should be an easy win.

[stephentyrone]

@swift-ci benchmark

[stephentyrone]

This should handle Float16 as well (implemented as self = Self(Float(value_)) in that case).

[xwu]

@stephentyrone The required checks for the availability of Float16 are...not pretty. See the commit "🤮" for details.

[xwu]

There's a benchmark PR not yet merged, by the way.

[xwu]

@swift-ci test

[xwu]

@swift-ci test

[xwu]

@swift-ci test Linux platform

[xwu]

Hmm

@swift-ci please benchmark

[swift-ci]

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
EqualSubstringSubstring	22	29	+31.8%	0.76x
LessSubstringSubstring	22	29	+31.8%	0.76x (?)
EqualStringSubstring	22	29	+31.8%	0.76x
EqualSubstringSubstringGenericEquatable	22	29	+31.8%	0.76x
EqualSubstringString	22	29	+31.8%	0.76x
LessSubstringSubstringGenericComparable	22	29	+31.8%	0.76x
StringComparison_longSharedPrefix	341	379	+11.1%	0.90x (?)
Set.isSuperset.Seq.Empty.Int	49	54	+10.2%	0.91x (?)
Set.isDisjoint.Empty.Int	84	92	+9.5%	0.91x (?)
CSVParsing.Scalar	275	298	+8.4%	0.92x (?)
Set.isDisjoint.Empty.Box	87	94	+8.0%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
ConvertFloatingPoint.GenericDoubleToDouble	2129	57	-97.3%	37.35x
AngryPhonebook.ASCII2	144	110	-23.6%	1.31x
UTF8Decode_InitDecoding	167	142	-15.0%	1.18x
UTF8Decode_InitFromCustom_contiguous	163	145	-11.0%	1.12x (?)
UTF8Decode_InitFromCustom_noncontiguous	318	294	-7.5%	1.08x (?)
StringHashing_fastPrenormal	670	620	-7.5%	1.08x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
EqualSubstringSubstring	22	29	+31.8%	0.76x
EqualSubstringSubstringGenericEquatable	22	29	+31.8%	0.76x
EqualSubstringString	22	29	+31.8%	0.76x
LessSubstringSubstring	23	29	+26.1%	0.79x
EqualStringSubstring	23	29	+26.1%	0.79x
LessSubstringSubstringGenericComparable	23	29	+26.1%	0.79x (?)
StringComparison_longSharedPrefix	340	380	+11.8%	0.89x (?)
Data.hash.Medium	30	33	+10.0%	0.91x (?)
Set.isDisjoint.Seq.Empty.Int	84	92	+9.5%	0.91x (?)
Set.isSuperset.Seq.Int0	118	128	+8.5%	0.92x (?)
Set.isDisjoint.Seq.Int.Empty	50	54	+8.0%	0.93x (?)
Set.isDisjoint.Empty.Box	88	95	+8.0%	0.93x (?)
Set.isSuperset.Seq.Int.Empty	88	95	+8.0%	0.93x (?)
Set.isStrictSubset.Int.Empty	51	55	+7.8%	0.93x (?)
DropFirstAnySequenceLazy	1643	1768	+7.6%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
ConvertFloatingPoint.GenericDoubleToDouble	2132	61	-97.1%	34.95x
AngryPhonebook.ASCII2	144	110	-23.6%	1.31x
UTF8Decode_InitDecoding	167	144	-13.8%	1.16x
UTF8Decode_InitFromCustom_contiguous	165	145	-12.1%	1.14x (?)
IterateData	922	859	-6.8%	1.07x (?)

Code size: -Osize

Regression	OLD	NEW	DELTA	RATIO
FloatingPointConversion.o	1681	1706	+1.5%	0.99x

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
LessSubstringSubstring	26	34	+30.8%	0.76x
EqualSubstringSubstringGenericEquatable	26	33	+26.9%	0.79x
EqualSubstringSubstring	27	34	+25.9%	0.79x (?)
EqualStringSubstring	27	34	+25.9%	0.79x (?)
LessSubstringSubstringGenericComparable	27	33	+22.2%	0.82x (?)
EqualSubstringString	28	34	+21.4%	0.82x
Set.subtracting.Seq.Empty.Box	1019	1160	+13.8%	0.88x (?)
Set.subtracting.Empty.Box	196	221	+12.8%	0.89x (?)
Set.isDisjoint.Box.Empty	1055	1187	+12.5%	0.89x (?)
Set.isDisjoint.Seq.Empty.Box	816	915	+12.1%	0.89x (?)
Set.subtracting.Seq.Box.Empty	1245	1394	+12.0%	0.89x (?)
Set.subtracting.Box.Empty	221	247	+11.8%	0.89x (?)
Set.isDisjoint.Seq.Box.Empty	1010	1128	+11.7%	0.90x (?)
SetIsSubsetBox0	1112	1233	+10.9%	0.90x (?)
Set.isStrictSubset.Empty.Int	511	566	+10.8%	0.90x (?)
Set.isStrictSubset.Box0	1129	1247	+10.5%	0.91x (?)
Set.isDisjoint.Empty.Box	971	1072	+10.4%	0.91x (?)
DictionaryLiteral	8060	8780	+8.9%	0.92x (?)
Combos	1826	1983	+8.6%	0.92x (?)
NSDictionaryCastToSwift	3180	3450	+8.5%	0.92x (?)
Set.isStrictSubset.Int.Empty	294	318	+8.2%	0.92x (?)
StackPromo	69700	75100	+7.7%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
ConvertFloatingPoint.GenericDoubleToDouble	2059	1132	-45.0%	1.82x
AngryPhonebook.ASCII2	145	111	-23.4%	1.31x
UTF8Decode_InitDecoding	178	148	-16.9%	1.20x
UTF8Decode_InitFromCustom_contiguous	180	151	-16.1%	1.19x (?)

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 mini
  Model Identifier: Macmini8,1
  Processor Name: 6-Core Intel Core i7
  Processor Speed: 3.2 GHz
  Number of Processors: 1
  Total Number of Cores: 6
  L2 Cache (per Core): 256 KB
  L3 Cache: 12 MB
  Memory: 64 GB