[String] ASCII fast-path for UTF16View #20848

milseman · 2018-11-29T00:26:55Z

Add an isASCII fast-path for many UTF16View operations. These are
heavily utilized in random-access scenarios, allowing us to both be
more efficient and skip generating breadcrumbs for ASCII strings.

milseman · 2018-11-29T00:28:07Z

@swift-ci please test

milseman · 2018-11-29T00:28:15Z

@swift-ci please benchmark

swift-ci · 2018-11-29T01:12:25Z

Build comment file:

Performance: -O

TEST	OLD	NEW	DELTA	RATIO
Regression
Breadcrumbs.MutatedIdxToUTF16.Mixed	259	284	+9.7%	0.91x
Breadcrumbs.IdxToUTF16Range.longMixed	372	405	+8.9%	0.92x
Breadcrumbs.IdxToUTF16.longMixed	712	771	+8.3%	0.92x
Breadcrumbs.MutatedUTF16ToIdx.Mixed	261	282	+8.0%	0.93x
Improvement
Breadcrumbs.MutatedIdxToUTF16.ASCII	761	4	-99.5%	190.20x
Breadcrumbs.MutatedUTF16ToIdx.ASCII	757	4	-99.5%	189.20x
Breadcrumbs.IdxToUTF16Range.longASCII	290	32	-89.0%	9.06x
Breadcrumbs.IdxToUTF16.longASCII	594	137	-76.9%	4.34x
Breadcrumbs.UTF16ToIdxRange.longASCII	84	26	-69.0%	3.23x
Breadcrumbs.UTF16ToIdx.longASCII	311	137	-55.9%	2.27x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	47	24	-48.9%	1.96x

Performance: -Osize

TEST	OLD	NEW	DELTA	RATIO
Regression
Breadcrumbs.MutatedIdxToUTF16.Mixed	258	284	+10.1%	0.91x
Breadcrumbs.MutatedUTF16ToIdx.Mixed	259	285	+10.0%	0.91x
Breadcrumbs.IdxToUTF16Range.longMixed	372	404	+8.6%	0.92x
Breadcrumbs.IdxToUTF16.longMixed	710	767	+8.0%	0.93x
Improvement
Breadcrumbs.MutatedIdxToUTF16.ASCII	760	4	-99.5%	189.95x
Breadcrumbs.MutatedUTF16ToIdx.ASCII	755	4	-99.5%	188.70x
Breadcrumbs.IdxToUTF16Range.longASCII	291	34	-88.3%	8.56x
Breadcrumbs.IdxToUTF16.longASCII	579	138	-76.2%	4.20x
Breadcrumbs.UTF16ToIdxRange.longASCII	84	28	-66.7%	3.00x
Breadcrumbs.UTF16ToIdx.longASCII	310	138	-55.5%	2.25x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	47	24	-48.9%	1.96x
InsertCharacterEndIndex	163	149	-8.6%	1.09x

Performance: -Onone

TEST	OLD	NEW	DELTA	RATIO
Regression
Breadcrumbs.MutatedUTF16ToIdx.Mixed	269	298	+10.8%	0.90x
Breadcrumbs.MutatedIdxToUTF16.Mixed	265	292	+10.2%	0.91x
Improvement
Breadcrumbs.MutatedIdxToUTF16.ASCII	769	12	-98.4%	64.08x
Breadcrumbs.MutatedUTF16ToIdx.ASCII	768	12	-98.4%	63.99x
Breadcrumbs.IdxToUTF16Range.longASCII	429	163	-62.0%	2.63x
Breadcrumbs.IdxToUTF16.longASCII	1181	715	-39.5%	1.65x
Breadcrumbs.UTF16ToIdxRange.longASCII	216	154	-28.7%	1.40x
Breadcrumbs.UTF16ToIdx.longASCII	905	709	-21.7%	1.28x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	116	92	-20.7%	1.26x
ExclusivityIndependent	86	75	-12.8%	1.15x

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

This looks great. I probably mid-air collided part of it with the PR I just landed though; sorry!

gottesmm · 2018-11-29T02:44:58Z

Nice!

milseman · 2018-11-29T19:40:30Z

That's fine, I'll roll in the (less important but still a nice 30%) ASCII fast-path in _nativeCopyUTF16CodeUnits too.

milseman · 2018-11-29T19:41:51Z

@swift-ci please benchmark

milseman · 2018-11-29T19:41:57Z

@swift-ci please test

swift-ci · 2018-11-29T19:48:12Z

Build failed
Swift Test Linux Platform
Git Sha - 4e63bd906e89ef167f0d1482706f00bc283ed689

swift-ci · 2018-11-29T20:24:36Z

Build failed
Swift Test OS X Platform
Git Sha - 4e63bd906e89ef167f0d1482706f00bc283ed689

swift-ci · 2018-11-29T21:05:07Z

Build comment file:

Performance: -O

TEST	OLD	NEW	DELTA	RATIO
Regression
Breadcrumbs.MutatedIdxToUTF16.Mixed	256	284	+10.9%	0.90x
Breadcrumbs.MutatedUTF16ToIdx.Mixed	259	286	+10.4%	0.91x
Improvement
Breadcrumbs.MutatedUTF16ToIdx.ASCII	755	4	-99.5%	188.70x
Breadcrumbs.UTF16ToIdxRange.longASCII	83	27	-67.5%	3.07x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	50	19	-62.0%	2.63x
Breadcrumbs.UTF16ToIdx.longASCII	311	137	-55.9%	2.27x

Performance: -Osize

TEST	OLD	NEW	DELTA	RATIO
Regression
Breadcrumbs.MutatedIdxToUTF16.Mixed	258	285	+10.5%	0.91x
Breadcrumbs.MutatedUTF16ToIdx.Mixed	262	287	+9.5%	0.91x
Improvement
Breadcrumbs.MutatedUTF16ToIdx.ASCII	757	4	-99.5%	189.20x
Breadcrumbs.UTF16ToIdxRange.longASCII	84	28	-66.7%	3.00x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	50	19	-62.0%	2.63x
Breadcrumbs.UTF16ToIdx.longASCII	323	138	-57.3%	2.34x
ObjectiveCBridgeFromNSArrayAnyObjectToString	55870	47802	-14.4%	1.17x
EqualStringSubstring	13	12	-7.7%	1.08x (?)

Performance: -Onone

TEST	OLD	NEW	DELTA	RATIO
Improvement
Breadcrumbs.MutatedUTF16ToIdx.ASCII	794	23	-97.1%	34.52x

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

--------------

Add an isASCII fast-path for many UTF16View operations. These are heavily utilized in random-access scenarios, allowing us to both be more efficient and skip generating breadcrumbs for ASCII strings.

milseman · 2018-11-29T22:12:21Z

@swift-ci please test

milseman · 2018-11-29T22:12:27Z

@swift-ci please benchmark

swift-ci · 2018-11-29T22:14:14Z

Build failed
Swift Test Linux Platform
Git Sha - 2a46550fdf5e75603c0a38ed1a7f57931ade116e

swift-ci · 2018-11-29T22:14:14Z

Build failed
Swift Test OS X Platform
Git Sha - 2a46550fdf5e75603c0a38ed1a7f57931ade116e

swift-ci · 2018-11-29T23:26:30Z

Build comment file:

Performance: -O

TEST	OLD	NEW	DELTA	RATIO
Regression
Breadcrumbs.MutatedIdxToUTF16.Mixed	234	259	+10.7%	0.90x
Breadcrumbs.MutatedUTF16ToIdx.Mixed	236	260	+10.2%	0.91x
Improvement
Breadcrumbs.MutatedUTF16ToIdx.ASCII	685	4	-99.4%	171.21x
Breadcrumbs.UTF16ToIdxRange.longASCII	82	24	-70.7%	3.42x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	48	16	-66.7%	3.00x
Breadcrumbs.UTF16ToIdx.longASCII	274	123	-55.1%	2.23x

Performance: -Osize

TEST	OLD	NEW	DELTA	RATIO
Regression
StringComparison_abnormal	7972	8929	+12.0%	0.89x
Breadcrumbs.MutatedIdxToUTF16.Mixed	232	259	+11.6%	0.90x
Improvement
Breadcrumbs.MutatedUTF16ToIdx.ASCII	685	4	-99.4%	171.21x
Breadcrumbs.UTF16ToIdxRange.longASCII	82	25	-69.5%	3.28x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	49	16	-67.3%	3.06x
Breadcrumbs.UTF16ToIdx.longASCII	277	124	-55.2%	2.23x
StringEqualPointerComparison	564	512	-9.2%	1.10x

Performance: -Onone

TEST	OLD	NEW	DELTA	RATIO
Improvement
Breadcrumbs.MutatedUTF16ToIdx.ASCII	720	22	-96.9%	32.73x
Breadcrumbs.CopyUTF16CodeUnits.ASCII	205	169	-17.6%	1.21x
ArrayOfPOD	764	692	-9.4%	1.10x

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: Quad-Core Intel Xeon E5
  Processor Speed: 3.7 GHz
  Number of Processors: 1
  Total Number of Cores: 4
  L2 Cache (per Core): 256 KB
  L3 Cache: 10 MB
  Memory: 16 GB

--------------

Add an isASCII fast-path for many UTF16View operations. These are heavily utilized in random-access scenarios, allowing us to both be more efficient and skip generating breadcrumbs for ASCII strings.

* Assorted bridging changes: • Convert _AbstractStringStorage to a protocol, and the free functions used to deduplicate implementations to extensions on that protocol. • Move 'start' into the abstract type and use that to simplify some code • Move the ASCII fast path for length into UTF16View. • Add a weirder but faster way to check which (if any) of our NSString subclasses a given object is, and adopt it (cherry picked from commit 8b57921) * [String] ASCII fast-path for UTF16View (#20848) Add an isASCII fast-path for many UTF16View operations. These are heavily utilized in random-access scenarios, allowing us to both be more efficient and skip generating breadcrumbs for ASCII strings.

Add an isASCII fast-path for many UTF16View operations. These are heavily utilized in random-access scenarios, allowing us to both be more efficient and skip generating breadcrumbs for ASCII strings.

milseman requested a review from Catfish-Man November 29, 2018 00:30

Catfish-Man approved these changes Nov 29, 2018

View reviewed changes

BasThomas mentioned this pull request Nov 29, 2018

[123] Issue #123 - November 29, 2018 SwiftWeekly/swiftweekly.github.io#424

Closed

milseman force-pushed the ascii_hotness branch from 4e63bd9 to 2a46550 Compare November 29, 2018 19:41

[String] ASCII fast-path for UTF16View

2fb2114

Add an isASCII fast-path for many UTF16View operations. These are heavily utilized in random-access scenarios, allowing us to both be more efficient and skip generating breadcrumbs for ASCII strings.

milseman force-pushed the ascii_hotness branch from 2a46550 to 2fb2114 Compare November 29, 2018 22:11

milseman merged commit c3c6fdc into swiftlang:master Nov 30, 2018

milseman deleted the ascii_hotness branch November 30, 2018 02:19

milseman mentioned this pull request Nov 30, 2018

[5.0] [String] ASCII fast-path for UTF16View #20900

Merged

milseman mentioned this pull request Feb 15, 2019

[benchmark] Add some benchmarks for String breadcrumbs #20769

Merged

[String] ASCII fast-path for UTF16View #20848

[String] ASCII fast-path for UTF16View #20848

Uh oh!

Conversation

milseman commented Nov 29, 2018

Uh oh!

milseman commented Nov 29, 2018

Uh oh!

milseman commented Nov 29, 2018

Uh oh!

swift-ci commented Nov 29, 2018

Build comment file:

Performance: -O

Performance: -Osize

Performance: -Onone

Uh oh!

Catfish-Man left a comment

Choose a reason for hiding this comment

Uh oh!

gottesmm commented Nov 29, 2018

Uh oh!

milseman commented Nov 29, 2018

Uh oh!

milseman commented Nov 29, 2018

Uh oh!

milseman commented Nov 29, 2018

Uh oh!

swift-ci commented Nov 29, 2018

Uh oh!

swift-ci commented Nov 29, 2018

Uh oh!

swift-ci commented Nov 29, 2018

Build comment file:

Performance: -O

Performance: -Osize

Performance: -Onone

Uh oh!

milseman commented Nov 29, 2018

Uh oh!

milseman commented Nov 29, 2018

Uh oh!

swift-ci commented Nov 29, 2018

Uh oh!

swift-ci commented Nov 29, 2018

Uh oh!

swift-ci commented Nov 29, 2018

Build comment file:

Performance: -O

Performance: -Osize

Performance: -Onone

Uh oh!

Uh oh!