Memoize the ObjCBridgeable conformance for Swift.String #30329

tbkka · 2020-03-10T16:11:59Z

This seems to be the single most common bridge cast,
and repeated lookup seems to be a performance issue
for the common operation of bridging a string-valued
NSDictionary into Swift. Spending a couple of static
words of memory to memoize it could be a nice perf win.

Resolves rdar://55237013

This seems to be the single most common bridge cast, and repeated lookup seems to be a performance issue for the common operation of bridging a string-valued NSDictionary into Swift. Spending a couple of static words of memory to memoize it should be a nice perf win.

tbkka · 2020-03-10T16:12:14Z

@swift-ci Please benchmark

swift-ci · 2020-03-10T16:40:30Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
FlattenListFlatMap	3695	5598	+51.5%	0.66x (?)
FlattenListLoop	2773	3183	+14.8%	0.87x (?)
PrefixArrayLazy	13	14	+7.7%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9400	7200	-23.4%	1.31x
ObjectiveCBridgeToNSSet	11450	9000	-21.4%	1.27x
ObjectiveCBridgeFromNSStringForced	1780	1535	-13.8%	1.16x (?)
ObjectiveCBridgeStubFromArrayOfNSString2	2120	1860	-12.3%	1.14x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToStringForced	29400	26600	-9.5%	1.11x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced	3300	3000	-9.1%	1.10x (?)
ObjectiveCBridgeFromNSSetAnyObjectToString	46000	42000	-8.7%	1.10x (?)
DictionaryBridgeToObjC_Access	558	510	-8.6%	1.09x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToString	28900	26600	-8.0%	1.09x (?)
ObjectiveCBridgeStubToArrayOfNSString2	2800	2580	-7.9%	1.09x (?)
ObjectiveCBridgeStubFromNSDateRef	2540	2370	-6.7%	1.07x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
StringInterpolationManySmallSegments	8800	10700	+21.6%	0.82x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9250	7200	-22.2%	1.28x
ObjectiveCBridgeToNSSet	11450	9050	-21.0%	1.27x
ObjectiveCBridgeFromNSStringForced	1765	1540	-12.7%	1.15x (?)
ObjectiveCBridgeStubFromArrayOfNSString2	2100	1870	-11.0%	1.12x (?)
DictionaryBridgeToObjC_Access	558	511	-8.4%	1.09x (?)

Code size: -Osize

Performance: -Onone

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9800	7600	-22.4%	1.29x
ObjectiveCBridgeToNSSet	11850	9500	-19.8%	1.25x
ObjectiveCBridgeStubFromArrayOfNSString2	2080	1860	-10.6%	1.12x (?)
ObjectiveCBridgeStubToArrayOfNSString2	2840	2560	-9.9%	1.11x (?)
ObjectiveCBridgeFromNSString	2165	1955	-9.7%	1.11x (?)
ObjectiveCBridgeFromNSStringForced	1840	1665	-9.5%	1.11x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToStringForced	30200	27600	-8.6%	1.09x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToString	30000	27500	-8.3%	1.09x (?)
DictionaryBridgeToObjC_Access	650	601	-7.5%	1.08x (?)
TypeFlood	148	137	-7.4%	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: 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

tbkka · 2020-03-10T16:48:21Z

@Catfish-Man -- Any thoughts about FlattenListFlatMap and FlattenListLoop here? I don't offhand see any reason those benchmarks would touch the Obj-C bridging path, which means they should be unaffected by this change. Am I missing something? Apart from that, the benchmarks here look nice.

Catfish-Man · 2020-03-10T17:15:17Z

FlattenListFlatMap has been odder than one might expect; I've been looking into it for the array changes I've been working on, and it doesn't respond to changes in a way I can make sense of.

I wouldn't worry about it too much but taking a quick sample/instruments profile locally might be worthwhile just in case.

stdlib/public/runtime/Casting.cpp

tbkka · 2020-03-10T20:18:38Z

@swift-ci Please benchmark

swift-ci · 2020-03-10T20:49:03Z

Performance: -O

Improvement	OLD	NEW	DELTA	RATIO
FlattenListFlatMap	6249	4124	-34.0%	1.52x (?)
ObjectiveCBridgeToNSArray	9500	7250	-23.7%	1.31x (?)
ObjectiveCBridgeToNSSet	11350	9100	-19.8%	1.25x (?)
FlattenListLoop	3365	2829	-15.9%	1.19x (?)
UTF8Decode_InitDecoding	167	141	-15.6%	1.18x (?)
ObjectiveCBridgeStubFromArrayOfNSString2	2130	1820	-14.6%	1.17x (?)
ObjectiveCBridgeFromNSStringForced	1800	1580	-12.2%	1.14x (?)
NormalizedIterator_fastPrenormal	660	580	-12.1%	1.14x (?)
Chars2	3500	3100	-11.4%	1.13x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToStringForced	30600	27200	-11.1%	1.12x (?)
Calculator	155	138	-11.0%	1.12x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToString	29800	26800	-10.1%	1.11x (?)
ObjectiveCBridgeStubToArrayOfNSString2	2820	2540	-9.9%	1.11x (?)
IterateData	878	794	-9.6%	1.11x (?)
StringHashing_fastPrenormal	590	540	-8.5%	1.09x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced	3320	3080	-7.2%	1.08x (?)
OpenClose	61	57	-6.6%	1.07x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
FlattenListLoop	2653	2924	+10.2%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9300	7000	-24.7%	1.33x
ObjectiveCBridgeToNSSet	11550	9050	-21.6%	1.28x (?)
ObjectiveCBridgeStubFromArrayOfNSString2	2130	1800	-15.5%	1.18x (?)
UTF8Decode_InitDecoding	164	139	-15.2%	1.18x (?)
Calculator	154	138	-10.4%	1.12x (?)
StringHashing_fastPrenormal	590	530	-10.2%	1.11x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToStringForced	29800	26800	-10.1%	1.11x (?)
Chars2	3400	3100	-8.8%	1.10x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubFromNSDate	3020	3270	+8.3%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9800	7350	-25.0%	1.33x
ObjectiveCBridgeToNSSet	11800	9450	-19.9%	1.25x
UTF8Decode_InitDecoding	185	156	-15.7%	1.19x
ObjectiveCBridgeStubFromArrayOfNSString2	2110	1850	-12.3%	1.14x (?)
ObjectiveCBridgeFromNSStringForced	1835	1615	-12.0%	1.14x (?)
ObjectiveCBridgeFromNSString	2135	1895	-11.2%	1.13x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToStringForced	30600	27400	-10.5%	1.12x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToString	30200	27200	-9.9%	1.11x (?)
ObjectiveCBridgeStubToArrayOfNSString2	2820	2580	-8.5%	1.09x (?)

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: 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

Depending on benchmark results, this may get backed out.

tbkka · 2020-03-10T21:08:36Z

@swift-ci Please benchmark

swift-ci · 2020-03-10T21:31:49Z

Performance: -O

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9350	7100	-24.1%	1.32x
ObjectiveCBridgeToNSSet	11350	8750	-22.9%	1.30x
UTF8Decode_InitDecoding	164	138	-15.9%	1.19x
ObjectiveCBridgeStubFromArrayOfNSString2	2130	1840	-13.6%	1.16x (?)
ObjectiveCBridgeFromNSStringForced	1770	1540	-13.0%	1.15x (?)
NormalizedIterator_fastPrenormal	640	570	-10.9%	1.12x
ObjectiveCBridgeFromNSArrayAnyObjectToStringForced	29400	26600	-9.5%	1.11x (?)
ObjectiveCBridgeStubToArrayOfNSString2	2800	2540	-9.3%	1.10x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToString	29100	26400	-9.3%	1.10x (?)
Chars2	3400	3100	-8.8%	1.10x (?)
Calculator	152	139	-8.6%	1.09x (?)
DictionaryBridgeToObjC_Access	552	506	-8.3%	1.09x (?)
OpenClose	61	56	-8.2%	1.09x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced	3240	3020	-6.8%	1.07x (?)

Code size: -O

Performance: -Osize

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9300	6950	-25.3%	1.34x
ObjectiveCBridgeToNSSet	11200	8800	-21.4%	1.27x
UTF8Decode_InitDecoding	164	136	-17.1%	1.21x
ObjectiveCBridgeStubFromArrayOfNSString2	2090	1800	-13.9%	1.16x (?)
ObjectiveCBridgeFromNSStringForced	1770	1545	-12.7%	1.15x (?)
Calculator	154	136	-11.7%	1.13x
ObjectiveCBridgeFromNSArrayAnyObjectToStringForced	29600	26600	-10.1%	1.11x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToString	29100	26200	-10.0%	1.11x (?)
DictionaryBridgeToObjC_Access	577	523	-9.4%	1.10x (?)
Chars2	3400	3100	-8.8%	1.10x (?)
ObjectiveCBridgeStubToArrayOfNSString2	2760	2520	-8.7%	1.10x (?)
ObjectiveCBridgeFromNSDictionaryAnyObjectForced	4300	3950	-8.1%	1.09x (?)
NormalizedIterator_fastPrenormal	670	620	-7.5%	1.08x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced	3260	3020	-7.4%	1.08x (?)
ObjectiveCBridgeFromNSSetAnyObjectToString	43500	40500	-6.9%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubNSDateRefAccess	3323	3601	+8.4%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeToNSArray	9600	7250	-24.5%	1.32x
ObjectiveCBridgeToNSSet	11750	9400	-20.0%	1.25x
UTF8Decode_InitDecoding	183	157	-14.2%	1.17x
ObjectiveCBridgeStubFromArrayOfNSString2	2130	1850	-13.1%	1.15x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced	5020	4400	-12.4%	1.14x (?)
ObjectiveCBridgeFromNSArrayAnyObjectToString	30300	26900	-11.2%	1.13x (?)
DictionaryBridgeToObjC_Access	645	599	-7.1%	1.08x (?)
StringToDataEmpty	750	700	-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 mini
  Model Identifier: Macmini8,1
  Processor Name: 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

tbkka · 2020-03-10T21:36:44Z

@Catfish-Man -- Thoughts? Given the lack of interesting difference between these benchmarks, I'm inclined to back out the last change (caching Dictionary and Array conformances) and just keep the String conformance cache.

stdlib/public/runtime/Casting.cpp

This reverts commit 2b93747.

tbkka · 2020-03-10T23:23:06Z

@swift-ci Please test

swift-ci · 2020-03-10T23:25:29Z

Build failed
Swift Test OS X Platform
Git Sha - b13fb96

swift-ci · 2020-03-10T23:25:41Z

Build failed
Swift Test Linux Platform
Git Sha - b13fb96

tbkka requested a review from Catfish-Man March 10, 2020 16:11

Catfish-Man approved these changes Mar 10, 2020

View reviewed changes

BasThomas mentioned this pull request Mar 10, 2020

[155] Issue #155 - March 12, 2020 SwiftWeekly/swiftweekly.github.io#502

Closed

gottesmm reviewed Mar 10, 2020

View reviewed changes

stdlib/public/runtime/Casting.cpp Outdated Show resolved Hide resolved

gottesmm reviewed Mar 10, 2020

View reviewed changes

stdlib/public/runtime/Casting.cpp Outdated Show resolved Hide resolved

gottesmm reviewed Mar 10, 2020

View reviewed changes

stdlib/public/runtime/Casting.cpp Outdated Show resolved Hide resolved

Simplify this code along lines suggested by @gottesmm

5c536e3

Try memoizing Dictionary and Array bridge witnesses as well.

2b93747

Depending on benchmark results, this may get backed out.

gottesmm reviewed Mar 10, 2020

View reviewed changes

stdlib/public/runtime/Casting.cpp Show resolved Hide resolved

Revert "Try memoizing Dictionary and Array bridge witnesses as well."

5890c00

This reverts commit 2b93747.

tbkka merged commit 46bba7a into swiftlang:master Mar 11, 2020

tbkka deleted the tbkka-casting-stringBridgePerf branch October 16, 2020 00:33

Memoize the ObjCBridgeable conformance for Swift.String #30329

Memoize the ObjCBridgeable conformance for Swift.String #30329

Uh oh!

Conversation

tbkka commented Mar 10, 2020

Uh oh!

tbkka commented Mar 10, 2020

Uh oh!

swift-ci commented Mar 10, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

tbkka commented Mar 10, 2020

Uh oh!

Catfish-Man commented Mar 10, 2020

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tbkka commented Mar 10, 2020

Uh oh!

swift-ci commented Mar 10, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

tbkka commented Mar 10, 2020

Uh oh!

swift-ci commented Mar 10, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

tbkka commented Mar 10, 2020

Uh oh!

Uh oh!

tbkka commented Mar 10, 2020

Uh oh!

swift-ci commented Mar 10, 2020

Uh oh!

swift-ci commented Mar 10, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants