Re-shim tagged NSString creation #26588

Catfish-Man · 2019-08-09T22:59:57Z

No description provided.

Catfish-Man · 2019-08-09T23:00:14Z

@swift-ci please smoke benchmark

Catfish-Man · 2019-08-09T23:06:51Z

This will almost certainly be a (hopefully smallish) perf regression. I have a plan for fixing it once other pieces are in place.

swift-ci · 2019-08-09T23:51:47Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	362	1605	+343.4%	0.23x
NSStringConversion.Rebridge	206	520	+152.4%	0.40x
ObjectiveCBridgeToNSArray	14300	27500	+92.3%	0.52x
NSStringConversion.Rebridge.Mutable	824	1476	+79.1%	0.56x
ObjectiveCBridgeStubToNSString	2370	4240	+78.9%	0.56x
SevenBoom	819	1452	+77.3%	0.56x
NSError	162	285	+75.9%	0.57x
ObjectiveCBridgeToNSSet	19250	32050	+66.5%	0.60x
ObjectiveCBridgeStubToArrayOfNSString2	4020	6240	+55.2%	0.64x
DictionaryBridgeToObjC_Access	870	1277	+46.8%	0.68x (?)
DictionaryBridge	846	1031	+21.9%	0.82x (?)
ObjectiveCBridgeStubURLAppendPathRef2	2430	2830	+16.5%	0.86x
ObjectiveCBridgeStubURLAppendPath2	2500	2850	+14.0%	0.88x (?)
PlistPerfEncode	511	570	+11.5%	0.90x
UTF8Decode_InitDecoding_ascii	257	286	+11.3%	0.90x (?)
FlattenListLoop	4338	4825	+11.2%	0.90x (?)
IterateData	1327	1453	+9.5%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubToNSDate2	1460	1260	-13.7%	1.16x (?)
RandomShuffleLCG2	768	704	-8.3%	1.09x
SuffixAnySequenceLazy	4252	3913	-8.0%	1.09x (?)
Set.isDisjoint.Seq.Box.Empty	127	117	-7.9%	1.09x
Array2D	7520	6928	-7.9%	1.09x (?)
Data.hash.Empty	77	71	-7.8%	1.08x (?)
RemoveWhereSwapInts	67	62	-7.5%	1.08x (?)
MapReduce	397	368	-7.3%	1.08x
MapReduceAnyCollection	398	369	-7.3%	1.08x
StringUTF16SubstringBuilder	3030	2820	-6.9%	1.07x (?)
Set.isSubset.Seq.Int.Empty	217	202	-6.9%	1.07x
NormalizedIterator_fastPrenormal	1060	990	-6.6%	1.07x
Set.isDisjoint.Seq.Int.Empty	91	85	-6.6%	1.07x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	378	1604	+324.3%	0.24x
NSStringConversion.Rebridge	206	518	+151.5%	0.40x
ObjectiveCBridgeToNSArray	14550	27250	+87.3%	0.53x
NSStringConversion.Rebridge.Mutable	792	1465	+85.0%	0.54x
ObjectiveCBridgeStubToNSString	2380	4230	+77.7%	0.56x
SevenBoom	865	1473	+70.3%	0.59x
ObjectiveCBridgeToNSSet	18900	31200	+65.1%	0.61x
ObjectiveCBridgeStubToArrayOfNSString2	4020	6180	+53.7%	0.65x (?)
NSError	234	354	+51.3%	0.66x
DictionaryBridgeToObjC_Access	903	1267	+40.3%	0.71x (?)
DictionaryBridge	864	1049	+21.4%	0.82x (?)
ObjectiveCBridgeStubURLAppendPathRef2	2530	2910	+15.0%	0.87x (?)
ObjectiveCBridgeStubURLAppendPath2	2610	3000	+14.9%	0.87x (?)
PlistPerfEncode	503	563	+11.9%	0.89x (?)
JSONPerfEncode	577	643	+11.4%	0.90x (?)
IterateData	1326	1451	+9.4%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
Set.isDisjoint.Int.Empty	168	152	-9.5%	1.11x (?)
FlattenListLoop	4428	4057	-8.4%	1.09x (?)
Set.isSubset.Int.Empty	97	89	-8.2%	1.09x (?)
RemoveWhereMoveInts	37	34	-8.1%	1.09x
Array2D	7520	6928	-7.9%	1.09x (?)
Data.hash.Empty	77	71	-7.8%	1.08x (?)
RemoveWhereSwapInts	65	60	-7.7%	1.08x
Set.isDisjoint.Seq.Box.Empty	170	157	-7.6%	1.08x (?)
NormalizedIterator_fastPrenormal	1070	990	-7.5%	1.08x (?)
Set.isSubset.Seq.Int.Empty	217	202	-6.9%	1.07x (?)
Set.isStrictSubset.Seq.Int.Empty	220	205	-6.8%	1.07x

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	451	1699	+276.7%	0.27x
NSStringConversion.Rebridge	232	547	+135.8%	0.42x
ObjectiveCBridgeToNSArray	14900	27500	+84.6%	0.54x
NSStringConversion.Rebridge.Mutable	873	1515	+73.5%	0.58x
ObjectiveCBridgeStubToNSString	2480	4300	+73.4%	0.58x
ObjectiveCBridgeToNSSet	19450	32600	+67.6%	0.60x
SevenBoom	987	1615	+63.6%	0.61x
ObjectiveCBridgeStubToArrayOfNSString2	4060	6060	+49.3%	0.67x (?)
DictionaryBridgeToObjC_Access	1039	1404	+35.1%	0.74x (?)
NSError	596	718	+20.5%	0.83x
DictionaryBridge	977	1116	+14.2%	0.88x (?)
JSONPerfEncode	610	681	+11.6%	0.90x (?)
ObjectiveCBridgeStubURLAppendPathRef2	2680	2980	+11.2%	0.90x (?)
PlistPerfEncode	536	592	+10.4%	0.91x (?)
ObjectiveCBridgeStubURLAppendPath2	2710	2960	+9.2%	0.92x (?)
DictionaryOfAnyHashableStrings_insert	7294	7924	+8.6%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringInterpolation	13700	12300	-10.2%	1.11x (?)
FloatingPointPrinting_Float_interpolated	58600	53400	-8.9%	1.10x (?)
EqualStringSubstring	51	47	-7.8%	1.09x (?)
Data.hash.Empty	154	142	-7.8%	1.08x (?)
ArrayOfGenericPOD2	1432	1321	-7.8%	1.08x (?)
ArrayOfPOD	1140	1062	-6.8%	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

Catfish-Man · 2019-08-09T23:52:05Z

Nooooooo :( :(

Catfish-Man · 2019-08-09T23:56:52Z

@swift-ci please smoke benchmark

swift-ci · 2019-08-10T00:21:32Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	362	1595	+340.6%	0.23x
NSStringConversion.Rebridge	206	519	+151.9%	0.40x
ObjectiveCBridgeToNSArray	14450	27500	+90.3%	0.53x
NSStringConversion.Rebridge.Mutable	810	1467	+81.1%	0.55x
NSError	161	286	+77.6%	0.56x
SevenBoom	815	1443	+77.1%	0.56x
ObjectiveCBridgeStubToNSString	2380	4200	+76.5%	0.57x
ObjectiveCBridgeToNSSet	18800	32250	+71.5%	0.58x
ObjectiveCBridgeStubToArrayOfNSString2	4020	6120	+52.2%	0.66x
DictionaryBridgeToObjC_Access	872	1264	+45.0%	0.69x (?)
DictionaryBridge	843	1037	+23.0%	0.81x (?)
ObjectiveCBridgeStubURLAppendPath2	2400	2910	+21.2%	0.82x (?)
ObjectiveCBridgeStubURLAppendPathRef2	2390	2820	+18.0%	0.85x (?)
JSONPerfEncode	582	649	+11.5%	0.90x (?)
PlistPerfEncode	514	568	+10.5%	0.90x (?)
IterateData	1326	1451	+9.4%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
RandomShuffleLCG2	768	704	-8.3%	1.09x
SuffixAnySequenceLazy	4251	3910	-8.0%	1.09x (?)
Set.isDisjoint.Seq.Box.Empty	127	117	-7.9%	1.09x (?)
Array2D	7520	6928	-7.9%	1.09x (?)
ObjectiveCBridgeStubFromNSDateRef	4090	3770	-7.8%	1.08x (?)
Data.hash.Empty	77	71	-7.8%	1.08x (?)
RemoveWhereSwapInts	67	62	-7.5%	1.08x (?)
MapReduce	397	368	-7.3%	1.08x
MapReduceAnyCollection	398	369	-7.3%	1.08x (?)
FlattenListLoop	5189	4821	-7.1%	1.08x (?)
Set.isSubset.Seq.Int.Empty	217	202	-6.9%	1.07x (?)
Set.isDisjoint.Seq.Int.Empty	91	85	-6.6%	1.07x (?)
Set.isSuperset.Seq.Empty.Int	91	85	-6.6%	1.07x

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	378	1603	+324.1%	0.24x
NSStringConversion.Rebridge	206	518	+151.5%	0.40x
ObjectiveCBridgeToNSArray	14650	27150	+85.3%	0.54x
NSStringConversion.Rebridge.Mutable	792	1465	+85.0%	0.54x
ObjectiveCBridgeStubToNSString	2380	4220	+77.3%	0.56x
SevenBoom	866	1479	+70.8%	0.59x
ObjectiveCBridgeToNSSet	18800	32050	+70.5%	0.59x
NSError	233	356	+52.8%	0.65x
ObjectiveCBridgeStubToArrayOfNSString2	4020	6140	+52.7%	0.65x
DictionaryBridgeToObjC_Access	901	1199	+33.1%	0.75x (?)
DictionaryBridge	869	1051	+20.9%	0.83x (?)
ObjectiveCBridgeStubURLAppendPath2	2530	2930	+15.8%	0.86x (?)
ObjectiveCBridgeStubURLAppendPathRef2	2470	2790	+13.0%	0.89x (?)
JSONPerfEncode	580	648	+11.7%	0.90x (?)
PlistPerfEncode	510	567	+11.2%	0.90x (?)
IterateData	1326	1458	+10.0%	0.91x

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubFromNSDateRef	4190	3760	-10.3%	1.11x (?)
Set.isDisjoint.Int.Empty	168	152	-9.5%	1.11x (?)
Set.isSubset.Int.Empty	98	90	-8.2%	1.09x (?)
RemoveWhereMoveInts	37	34	-8.1%	1.09x (?)
Array2D	7520	6928	-7.9%	1.09x (?)
Data.hash.Empty	77	71	-7.8%	1.08x
RemoveWhereSwapInts	65	60	-7.7%	1.08x
NormalizedIterator_fastPrenormal	1070	990	-7.5%	1.08x
FlattenListLoop	5277	4907	-7.0%	1.08x (?)
Set.isSubset.Seq.Int.Empty	217	202	-6.9%	1.07x (?)
Set.isStrictSubset.Seq.Int.Empty	220	205	-6.8%	1.07x (?)
String.replaceSubrange.String.Small	90	84	-6.7%	1.07x (?)
Set.isDisjoint.Seq.Box.Empty	168	157	-6.5%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	451	1695	+275.8%	0.27x
NSStringConversion.Rebridge	232	547	+135.8%	0.42x
ObjectiveCBridgeToNSArray	14900	27450	+84.2%	0.54x
NSStringConversion.Rebridge.Mutable	873	1518	+73.9%	0.58x
ObjectiveCBridgeStubToNSString	2480	4290	+73.0%	0.58x
ObjectiveCBridgeToNSSet	19800	32450	+63.9%	0.61x
SevenBoom	990	1616	+63.2%	0.61x
ObjectiveCBridgeStubToArrayOfNSString2	4020	6160	+53.2%	0.65x (?)
DictionaryBridgeToObjC_Access	1036	1416	+36.7%	0.73x (?)
NSError	598	727	+21.6%	0.82x
ObjectiveCBridgeStubURLAppendPath2	2630	3140	+19.4%	0.84x (?)
DictionaryBridge	966	1148	+18.8%	0.84x (?)
ObjectiveCBridgeStubURLAppendPathRef2	2610	3030	+16.1%	0.86x (?)
PlistPerfEncode	532	594	+11.7%	0.90x (?)
JSONPerfEncode	613	674	+10.0%	0.91x (?)
DictionaryOfAnyHashableStrings_insert	7350	7952	+8.2%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
FloatingPointPrinting_Float_interpolated	58800	53600	-8.8%	1.10x (?)
ArrayOfGenericPOD2	1434	1319	-8.0%	1.09x (?)
EqualStringSubstring	51	47	-7.8%	1.09x (?)
Data.hash.Empty	154	142	-7.8%	1.08x (?)
ArrayOfPOD	1144	1061	-7.3%	1.08x (?)
CSVParsingAlt2	2673	2486	-7.0%	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 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

Catfish-Man · 2019-08-10T04:01:54Z

ok I profiled this and apparently the protocol ends up getting boxed or something wacky, so the work being done is totally unrelated to the objc overhead. That's… somewhat promising I guess. Means I don't have to throw out the whole approach.

Catfish-Man · 2019-08-10T05:18:44Z

@swift-ci please smoke test

Catfish-Man · 2019-08-10T05:20:18Z

@swift-ci please smoke test

Catfish-Man · 2019-08-10T05:20:32Z

@swift-ci please smoke benchmark

swift-ci · 2019-08-10T06:12:05Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	362	402	+11.0%	0.90x (?)

Improvement	OLD	NEW	DELTA	RATIO
String.data.Medium	95	87	-8.4%	1.09x (?)
NormalizedIterator_fastPrenormal	1060	990	-6.6%	1.07x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	378	419	+10.8%	0.90x

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSString	451	491	+8.9%	0.92x (?)
ObjectiveCBridgeStubFromNSDateRef	4310	4670	+8.4%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
EqualStringSubstring	51	47	-7.8%	1.09x (?)
ArrayOfPOD	1144	1064	-7.0%	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 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

Catfish-Man · 2019-08-10T06:18:38Z

BOOM

Catfish-Man · 2019-08-10T09:41:09Z

stdlib/public/Darwin/Foundation/String.swift

This selector has to start with "new" to get swiftc to not autorelease the return value

Are the trailing underscores intended to convey privateness without screwing up the new prefix?

Catfish-Man · 2019-08-10T09:42:05Z

stdlib/public/Darwin/Foundation/String.swift

Casting directly to AnyObject here causes swiftc to emit a bridging call to bridge the optional out to objc; using an objc type explicitly avoids that

I'd suggest translating these PR comments to source code comments to lessen the pain of future generations. This all seems like valuable info.

Catfish-Man · 2019-08-10T09:42:43Z

stdlib/public/Darwin/Foundation/String.swift

Unclear what exactly I'm claiming is not being released here, but it does cause something to not be retain-released

Catfish-Man · 2019-08-10T09:44:13Z

stdlib/public/core/StringBridge.swift

Using '__StringStorage.self as AnyClass' instead of my original '_StringSelectorHolder.self as AnyObject' avoids a bunch of machinery that appeared to be boxing something.

Unfortunately it still emits a metadata accessor call that costs about 2/3 as much as objc_msgSend, but I haven't found a good way to avoid that yet.

Catfish-Man · 2019-08-10T09:44:45Z

stdlib/public/core/StringBridge.swift

When we switch to _CFStringCreateTaggedPointerString this will start returning nil more

mikeash · 2019-08-12T17:38:46Z

stdlib/public/Darwin/Foundation/String.swift

I'd suggest translating these PR comments to source code comments to lessen the pain of future generations. This all seems like valuable info.

mikeash · 2019-08-12T17:39:38Z

stdlib/public/Darwin/Foundation/String.swift

Are the trailing underscores intended to convey privateness without screwing up the new prefix?

Catfish-Man · 2019-08-12T22:16:40Z

@swift-ci please smoke test and merge

Catfish-Man · 2019-08-12T22:20:26Z

@swift-ci please smoke test and merge

Catfish-Man · 2019-08-12T23:33:02Z

Ugh, it claims it failed but I can't find the failure. Gonna re-run it

Catfish-Man · 2019-08-12T23:33:08Z

@swift-ci please smoke test

Catfish-Man marked this pull request as ready for review August 10, 2019 06:47

Catfish-Man self-assigned this Aug 10, 2019

Catfish-Man requested a review from milseman August 10, 2019 06:48

Catfish-Man commented Aug 10, 2019

View reviewed changes

mikeash approved these changes Aug 12, 2019

View reviewed changes

Redo CFStringCreateWithBytes shim

f559a94

Catfish-Man force-pushed the shadow-backflip branch from 0c5d8f4 to f559a94 Compare August 12, 2019 22:15

Catfish-Man merged commit bd3643a into swiftlang:master Aug 13, 2019

Catfish-Man deleted the shadow-backflip branch August 13, 2019 03:00

Re-shim tagged NSString creation #26588

Re-shim tagged NSString creation #26588

Uh oh!

Conversation

Catfish-Man commented Aug 9, 2019

Uh oh!

Catfish-Man commented Aug 9, 2019

Uh oh!

Catfish-Man commented Aug 9, 2019

Uh oh!

swift-ci commented Aug 9, 2019

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

Catfish-Man commented Aug 9, 2019

Uh oh!

Catfish-Man commented Aug 9, 2019

Uh oh!

swift-ci commented Aug 10, 2019

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

Catfish-Man commented Aug 10, 2019

Uh oh!

Catfish-Man commented Aug 10, 2019

Uh oh!

Catfish-Man commented Aug 10, 2019

Uh oh!

Catfish-Man commented Aug 10, 2019

Uh oh!

swift-ci commented Aug 10, 2019

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

Catfish-Man commented Aug 10, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Catfish-Man commented Aug 12, 2019

Uh oh!

Catfish-Man commented Aug 12, 2019

Uh oh!

Catfish-Man commented Aug 12, 2019

Uh oh!

Catfish-Man commented Aug 12, 2019

Uh oh!

Reviewers

Assignees

Labels

Projects