[stdlib] Add fast string interpolation for metatypes #32113

airspeedswift · 2020-05-31T05:42:30Z

"\(T.self)" is pretty common, and sometimes performance sensitive, but currently takes the slowest of paths. This gives them a custom string interpolation that cuts straight to what a long process of conformance checks and mirror reflection used to do.

This makes _typeName ABI (it was already public but underscored) which I think is OK.

airspeedswift · 2020-05-31T05:42:49Z

@swift-ci please test

swift-ci · 2020-05-31T06:50:24Z

Build failed
Swift Test OS X Platform
Git Sha - ecdca44b9bef843f979f609b61435ebfcf33944e

swift-ci · 2020-05-31T06:51:59Z

Build failed
Swift Test Linux Platform
Git Sha - ecdca44b9bef843f979f609b61435ebfcf33944e

airspeedswift · 2020-06-01T05:05:58Z

@swift-ci please test

airspeedswift · 2020-06-01T05:06:03Z

@swift-ci please benchmark

airspeedswift · 2020-06-01T05:06:13Z

@swift-ci please test source compatibility

swift-ci · 2020-06-01T05:08:34Z

Build failed
Swift Test OS X Platform
Git Sha - ecdca44b9bef843f979f609b61435ebfcf33944e

swift-ci · 2020-06-01T05:08:46Z

Build failed
Swift Test Linux Platform
Git Sha - ecdca44b9bef843f979f609b61435ebfcf33944e

swift-ci · 2020-06-01T05:32:11Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	141	167	+18.4%	0.84x
ObjectiveCBridgeStubFromNSDate	3970	4650	+17.1%	0.85x (?)
UTF8Decode_InitFromCustom_contiguous	143	167	+16.8%	0.86x
Breadcrumbs.MutatedIdxToUTF16.Mixed	186	216	+16.1%	0.86x
Breadcrumbs.MutatedUTF16ToIdx.Mixed	183	212	+15.8%	0.86x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringToDataSmall	600	550	-8.3%	1.09x (?)
NormalizedIterator_fastPrenormal	670	620	-7.5%	1.08x (?)
NormalizedIterator_latin1	228	212	-7.0%	1.08x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	144	169	+17.4%	0.85x
UTF8Decode_InitFromCustom_contiguous	144	168	+16.7%	0.86x
Breadcrumbs.MutatedUTF16ToIdx.Mixed	182	212	+16.5%	0.86x (?)
Breadcrumbs.MutatedIdxToUTF16.Mixed	186	216	+16.1%	0.86x (?)
Dictionary4OfObjects	278	317	+14.0%	0.88x
ObjectiveCBridgeStubNSDateRefAccess	479	534	+11.5%	0.90x (?)
ObjectiveCBridgeStubToNSStringRef	68	74	+8.8%	0.92x (?)
ObjectiveCBridgeStringHash	70	76	+8.6%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
MapReduce	252	219	-13.1%	1.15x (?)
PrefixWhileSequence	225	196	-12.9%	1.15x
MapReduceAnyCollection	250	218	-12.8%	1.15x
RemoveWhereSwapInts	40	36	-10.0%	1.11x (?)
PrefixWhileAnySeqCntRange	202	182	-9.9%	1.11x (?)
RandomShuffleLCG2	496	448	-9.7%	1.11x (?)
PrefixWhileAnySeqCRangeIter	203	185	-8.9%	1.10x (?)
RangeAssignment	248	227	-8.5%	1.09x (?)
ArrayPlusEqualFiveElementCollection	4921	4514	-8.3%	1.09x (?)
FlattenListLoop	3068	2818	-8.1%	1.09x (?)
RemoveWhereFilterInts	25	23	-8.0%	1.09x (?)
NormalizedIterator_fastPrenormal	690	640	-7.2%	1.08x (?)
Array2D	4624	4304	-6.9%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitFromCustom_contiguous	150	176	+17.3%	0.85x
UTF8Decode_InitDecoding	150	173	+15.3%	0.87x (?)
Breadcrumbs.MutatedUTF16ToIdx.Mixed	198	228	+15.2%	0.87x
Breadcrumbs.MutatedIdxToUTF16.Mixed	202	231	+14.4%	0.87x (?)
ObjectiveCBridgeStringIsEqualAllSwift	56	62	+10.7%	0.90x (?)
ObjectiveCBridgeStubNSDateRefAccess	3372	3705	+9.9%	0.91x (?)
ObjectiveCBridgeStringIsEqual2	187	202	+8.0%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
ArrayOfPOD	722	643	-10.9%	1.12x (?)
StringWalk	2920	2640	-9.6%	1.11x (?)
DataSubscriptMedium	69	63	-8.7%	1.10x (?)
ArrayOfGenericPOD2	574	531	-7.5%	1.08x (?)

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

airspeedswift · 2020-06-03T19:07:18Z

@swift-ci please benchmark

swift-ci · 2020-06-03T20:08:07Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
AngryPhonebook.Strasse	182	219	+20.3%	0.83x
AngryPhonebook.Armenian	206	241	+17.0%	0.85x
AngryPhonebook.Cyrillic	221	258	+16.7%	0.86x
CharIndexing_punctuatedJapanese_unicodeScalars	1040	1200	+15.4%	0.87x
CharIndexing_punctuated_unicodeScalars	1120	1240	+10.7%	0.90x
RemoveWhereSwapInts	53	58	+9.4%	0.91x
String.replaceSubrange.Substring.Small	66	72	+9.1%	0.92x
FlattenListLoop	3579	3889	+8.7%	0.92x (?)
ArrayPlusEqualFiveElementCollection	6956	7511	+8.0%	0.93x (?)
String.replaceSubrange.String.Small	51	55	+7.8%	0.93x (?)
MapReduce	334	360	+7.8%	0.93x (?)
CharIteration_punctuated_unicodeScalars_Backwards	1040	1120	+7.7%	0.93x
MapReduceAnyCollection	332	357	+7.5%	0.93x

Improvement	OLD	NEW	DELTA	RATIO
TypeName	13567	1900	-86.0%	7.14x
DictionaryBridgeToObjC_Access	838	667	-20.4%	1.26x (?)
EqualSubstringSubstring	37	34	-8.1%	1.09x (?)
LessSubstringSubstring	37	34	-8.1%	1.09x
EqualSubstringSubstringGenericEquatable	37	34	-8.1%	1.09x
EqualSubstringString	37	34	-8.1%	1.09x
LessSubstringSubstringGenericComparable	37	34	-8.1%	1.09x
EqualStringSubstring	38	35	-7.9%	1.09x (?)
ObjectiveCBridgeToNSSet	14100	13000	-7.8%	1.08x (?)
ObjectiveCBridgeFromNSSetAnyObjectToString	65500	60500	-7.6%	1.08x (?)
StringComparison_slowerPrenormal	1450	1340	-7.6%	1.08x (?)
Set.subtracting.Empty.Box	14	13	-7.1%	1.08x

Code size: -O

Regression	OLD	NEW	DELTA	RATIO
Substring.o	15731	15907	+1.1%	0.99x

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
AngryPhonebook.Strasse	181	219	+21.0%	0.83x
DropWhileAnySequenceLazy	1906	2231	+17.1%	0.85x (?)
AngryPhonebook.Armenian	206	241	+17.0%	0.85x
AngryPhonebook.Cyrillic	221	258	+16.7%	0.86x
RemoveWhereSwapInts	53	58	+9.4%	0.91x
FlattenListLoop	3561	3888	+9.2%	0.92x (?)
ArrayPlusEqualFiveElementCollection	6956	7511	+8.0%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
TypeName	13655	1963	-85.6%	6.96x
Breadcrumbs.UTF16ToIdxRange.longASCII	29	25	-13.8%	1.16x
SubstringComparable	15	13	-13.3%	1.15x
PrefixAnySeqCRangeIterLazy	158	142	-10.1%	1.11x (?)
PrefixAnySeqCntRangeLazy	158	142	-10.1%	1.11x (?)
EqualSubstringSubstring	37	34	-8.1%	1.09x
EqualStringSubstring	37	34	-8.1%	1.09x (?)
EqualSubstringSubstringGenericEquatable	37	34	-8.1%	1.09x
EqualSubstringString	37	34	-8.1%	1.09x (?)
LessSubstringSubstringGenericComparable	38	35	-7.9%	1.09x (?)
StringComparison_nonBMPSlowestPrenormal	1510	1400	-7.3%	1.08x
FindString.Rec3.String	214	199	-7.0%	1.08x (?)

Code size: -Osize

Regression	OLD	NEW	DELTA	RATIO
RomanNumbers.o	6138	6779	+10.4%	0.91x
CSVParsing.o	50669	52030	+2.7%	0.97x
StringMatch.o	4786	4898	+2.3%	0.98x
FindStringNaive.o	8770	8966	+2.2%	0.98x
Substring.o	15694	15950	+1.6%	0.98x
StringBuilder.o	7434	7531	+1.3%	0.99x

Improvement	OLD	NEW	DELTA	RATIO
Mirror.o	13192	12370	-6.2%	1.07x

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
AngryPhonebook.Strasse	181	218	+20.4%	0.83x
AngryPhonebook.Armenian	206	241	+17.0%	0.85x
AngryPhonebook.Cyrillic	222	258	+16.2%	0.86x
StringBuilderWithLongSubstring	2690	2980	+10.8%	0.90x (?)
FindString.Rec3.Substring	729	799	+9.6%	0.91x (?)
StringUTF16SubstringBuilder	16010	17350	+8.4%	0.92x (?)
ObjectiveCBridgeStringHash	111	120	+8.1%	0.93x
RangeReplaceableCollectionPlusDefault	5672	6120	+7.9%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
TypeName	14527	2377	-83.6%	6.11x
DictionaryBridgeToObjC_Access	989	774	-21.7%	1.28x (?)
StringRemoveDupes	832	681	-18.1%	1.22x
String.data.LargeUnicode	120	105	-12.5%	1.14x (?)
String.data.Medium	115	101	-12.2%	1.14x (?)
ParseFloat.Double.Exp	43	38	-11.6%	1.13x (?)
ObjectiveCBridgeFromNSSetAnyObjectToStringForced	86000	77500	-9.9%	1.11x (?)
FrequenciesUsingReduceInto	4010	3730	-7.0%	1.08x (?)
EqualSubstringSubstringGenericEquatable	43	40	-7.0%	1.07x (?)
LessSubstringSubstringGenericComparable	43	40	-7.0%	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: 8-Core Intel Xeon E5
  Processor Speed: 3 GHz
  Number of Processors: 1
  Total Number of Cores: 8
  L2 Cache (per Core): 256 KB
  L3 Cache: 25 MB
  Memory: 64 GB

beccadax

I'm okay with this change in principle, but I have some questions about the specifics.

stdlib/public/core/StringInterpolation.swift

test/IDE/complete_at_top_level.swift

airspeedswift · 2020-06-26T00:16:17Z

test/IDE/complete_at_top_level.swift

@@ -471,7 +471,7 @@ var stringInterp = "\(#^STRING_INTERP_3^#)"
 _ = "" + "\(#^STRING_INTERP_4^#)" + ""
 // STRING_INTERP: Begin completions
 // STRING_INTERP-DAG: Decl[InstanceMethod]/CurrNominal/IsSystem: ['(']{#(value): T#}[')'][#Void#];
-// STRING_INTERP-DAG: Decl[Struct]/CurrModule: FooStruct[#FooStruct#];
+// STRING_INTERP-DAG: Decl[Struct]/CurrModule/TypeRelation[Convertible]: FooStruct[#FooStruct#]; name=FooStruct


@rintaro I tracked down the real difference and I think this is it. Does that make sense, given we've overloaded interpolation to take a metatype?

Yeah, it makes sense 👍

airspeedswift · 2020-06-26T00:16:33Z

@swift-ci please test

airspeedswift · 2020-06-26T00:16:38Z

@swift-ci please benchmark

airspeedswift · 2020-06-26T02:44:07Z

@swift-ci please benchmark

swift-ci · 2020-06-26T03:16:38Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
EqualSubstringSubstring	21	29	+38.1%	0.72x
LessSubstringSubstring	21	29	+38.1%	0.72x
EqualSubstringSubstringGenericEquatable	21	29	+38.1%	0.72x
EqualSubstringString	21	29	+38.1%	0.72x
LessSubstringSubstringGenericComparable	21	29	+38.1%	0.72x
EqualStringSubstring	21	28	+33.3%	0.75x
AngryPhonebook.Strasse	120	146	+21.7%	0.82x
CharIndexing_punctuatedJapanese_unicodeScalars	760	920	+21.1%	0.83x
AngryPhonebook.Armenian	141	162	+14.9%	0.87x (?)
AngryPhonebook.Cyrillic	152	174	+14.5%	0.87x (?)
StringComparison_longSharedPrefix	337	373	+10.7%	0.90x
FlattenListFlatMap	2372	2615	+10.2%	0.91x (?)
CharIndexing_punctuated_unicodeScalars	840	920	+9.5%	0.91x (?)
DataAppendSequence	5900	6400	+8.5%	0.92x (?)
IterateData	810	876	+8.1%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
TypeName	9261	1468	-84.1%	6.31x

Code size: -O

Regression	OLD	NEW	DELTA	RATIO
Substring.o	17155	17347	+1.1%	0.99x

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
FlattenListLoop	928	1516	+63.4%	0.61x (?)
LessSubstringSubstring	21	29	+38.1%	0.72x
EqualStringSubstring	21	29	+38.1%	0.72x
LessSubstringSubstringGenericComparable	21	29	+38.1%	0.72x
EqualSubstringSubstring	21	28	+33.3%	0.75x
EqualSubstringSubstringGenericEquatable	21	28	+33.3%	0.75x
EqualSubstringString	21	28	+33.3%	0.75x
Dictionary4	161	203	+26.1%	0.79x
AngryPhonebook.Strasse	120	145	+20.8%	0.83x (?)
AngryPhonebook.Armenian	141	162	+14.9%	0.87x (?)
Dictionary4OfObjects	230	264	+14.8%	0.87x (?)
AngryPhonebook.Cyrillic	152	174	+14.5%	0.87x (?)
StringComparison_longSharedPrefix	337	373	+10.7%	0.90x (?)
IterateData	803	869	+8.2%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
TypeName	9220	1539	-83.3%	5.99x
CharacterLiteralsLarge	71	63	-11.3%	1.13x (?)
FindString.Rec3.String	180	161	-10.6%	1.12x (?)
FindString.Rec3.Substring	212	194	-8.5%	1.09x (?)
CharacterLiteralsSmall	233	215	-7.7%	1.08x (?)

Code size: -Osize

Regression	OLD	NEW	DELTA	RATIO
RomanNumbers.o	6106	6747	+10.5%	0.90x
CSVParsing.o	50493	51710	+2.4%	0.98x
StringMatch.o	4786	4898	+2.3%	0.98x
FindStringNaive.o	8628	8824	+2.3%	0.98x
Substring.o	17315	17571	+1.5%	0.99x
StringBuilder.o	7259	7356	+1.3%	0.99x

Improvement	OLD	NEW	DELTA	RATIO
Mirror.o	13248	12442	-6.1%	1.06x

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
EqualSubstringSubstringGenericEquatable	25	32	+28.0%	0.78x (?)
LessSubstringSubstringGenericComparable	25	32	+28.0%	0.78x
EqualSubstringSubstring	26	33	+26.9%	0.79x
LessSubstringSubstring	26	33	+26.9%	0.79x
EqualStringSubstring	26	33	+26.9%	0.79x (?)
EqualSubstringString	26	33	+26.9%	0.79x (?)
AngryPhonebook.Strasse	120	145	+20.8%	0.83x
AngryPhonebook.Armenian	141	162	+14.9%	0.87x
AngryPhonebook.Cyrillic	152	174	+14.5%	0.87x (?)
String.replaceSubrange.String.Small	38	42	+10.5%	0.90x (?)
StringWalk	3040	3320	+9.2%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
TypeName	9811	1826	-81.4%	5.37x

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

swift-ci · 2020-06-26T04:01:27Z

Build failed
Swift Test Linux Platform
Git Sha - 7b5a4b9

airspeedswift · 2020-06-26T13:09:31Z

@swift-ci please smoke test Linux platform

* Add fast string interpolation for metatypes (#32113) * Whitespace fixes Co-authored-by: Ben Cohen <[email protected]>

airspeedswift marked this pull request as draft May 31, 2020 05:43

airspeedswift force-pushed the typename-fastpath branch 2 times, most recently from 5bcdc82 to 215806f Compare May 31, 2020 21:41

airspeedswift marked this pull request as ready for review June 3, 2020 19:07

airspeedswift requested a review from beccadax June 3, 2020 19:08

beccadax requested changes Jun 3, 2020

View reviewed changes

stdlib/public/core/StringInterpolation.swift Show resolved Hide resolved

test/IDE/complete_at_top_level.swift Outdated Show resolved Hide resolved

airspeedswift force-pushed the typename-fastpath branch from 215806f to aa52ed5 Compare June 25, 2020 23:52

Add fast string interpolation for metatypes

7b5a4b9

airspeedswift force-pushed the typename-fastpath branch from aa52ed5 to 7b5a4b9 Compare June 26, 2020 00:14

airspeedswift commented Jun 26, 2020

View reviewed changes

airspeedswift requested review from rintaro and beccadax June 26, 2020 00:17

airspeedswift merged commit 87df996 into swiftlang:master Jun 26, 2020

kylemacomber pushed a commit to kylemacomber/swift that referenced this pull request Jul 17, 2020

Add fast string interpolation for metatypes (swiftlang#32113)

ebf46f8

kylemacomber mentioned this pull request Jul 17, 2020

[5.3] Add fast string interpolation for metatypes #32958

Merged

airspeedswift added a commit that referenced this pull request Jul 17, 2020

[5.3] Add fast string interpolation for metatypes (#32958)

cb76d0a

* Add fast string interpolation for metatypes (#32113) * Whitespace fixes Co-authored-by: Ben Cohen <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[stdlib] Add fast string interpolation for metatypes #32113

[stdlib] Add fast string interpolation for metatypes #32113

airspeedswift commented May 31, 2020 •

edited

Loading

airspeedswift commented May 31, 2020

swift-ci commented May 31, 2020

swift-ci commented May 31, 2020

airspeedswift commented Jun 1, 2020

airspeedswift commented Jun 1, 2020

airspeedswift commented Jun 1, 2020

swift-ci commented Jun 1, 2020

swift-ci commented Jun 1, 2020

swift-ci commented Jun 1, 2020

airspeedswift commented Jun 3, 2020

swift-ci commented Jun 3, 2020

beccadax left a comment

airspeedswift Jun 26, 2020

rintaro Jun 26, 2020

airspeedswift commented Jun 26, 2020

airspeedswift commented Jun 26, 2020

airspeedswift commented Jun 26, 2020

swift-ci commented Jun 26, 2020

swift-ci commented Jun 26, 2020

airspeedswift commented Jun 26, 2020

[stdlib] Add fast string interpolation for metatypes #32113

[stdlib] Add fast string interpolation for metatypes #32113

Conversation

airspeedswift commented May 31, 2020 • edited Loading

airspeedswift commented May 31, 2020

swift-ci commented May 31, 2020

swift-ci commented May 31, 2020

airspeedswift commented Jun 1, 2020

airspeedswift commented Jun 1, 2020

airspeedswift commented Jun 1, 2020

swift-ci commented Jun 1, 2020

swift-ci commented Jun 1, 2020

swift-ci commented Jun 1, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

airspeedswift commented Jun 3, 2020

swift-ci commented Jun 3, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

beccadax left a comment

Choose a reason for hiding this comment

airspeedswift Jun 26, 2020

Choose a reason for hiding this comment

rintaro Jun 26, 2020

Choose a reason for hiding this comment

airspeedswift commented Jun 26, 2020

airspeedswift commented Jun 26, 2020

airspeedswift commented Jun 26, 2020

swift-ci commented Jun 26, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

swift-ci commented Jun 26, 2020

airspeedswift commented Jun 26, 2020

airspeedswift commented May 31, 2020 •

edited

Loading