Skip to content

[stdlib] Optimize high-level Set operations #40012

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 14 commits into from
Nov 5, 2021
Merged

[stdlib] Optimize high-level Set operations #40012

merged 14 commits into from
Nov 5, 2021

Conversation

lorentey
Copy link
Member

@lorentey lorentey commented Nov 2, 2021 •
edited
Loading

This revives and replaces #21300, dramatically speeding up some high-level Set operations while also restoring the (undocumented) behavior that Set.intersection always returns items from self, not its argument.

As discussed on the forums, Swift 5.5 changed Set.intersection's behavior as an optimization (#36678). This method does not guarantee which input set it uses to construct its results, so arguably, code that relies on the original behavior is in the wrong. However, there is no reason to break such code: we can implement intersections even faster than #36678, while also preserving the original behavior. This PR does this and more. (rdar://84831592)

(I believe it would be worth documenting this behavior for both Set.intersection and Set.union. However, that is an API change that likely requires a Swift evolution proposal.)

This new incarnation of #21300 does not introduce a new type, which makes it possible to cleanly deploy these improvements to any shipping stdlib version. In exchange, the implementation turned slightly more complicated in places: the existing _UnsafeBitset type does not have a cached count, so we need to explicitly maintain it as a standalone variable. Temporary bitmaps are now allocated using the new temporary allocation facility, so they will be in most cases allocated on the stack.

The primary drawback of all these optimizations is a code size increase. I believe the performance increase will be worth this extra cost, but let's keep on eye on benchmark reports.

The original PR promised the following improvements: (these are very rough numbers, and they depend on the contents/size of the set)

Operation Speedup
Set.isSubset<S>(of:) 3x
Set.isStrictSubset<S>(of:) 3x
Set.isSuperset<S>(of:) 4x
Set.isStrictSuperset<S>(of:) 4x
Set.isDisjoint<S>(with:) 4x
Set.subtracting(_:) 1-4x
Set.filter(_:) 1-4x
Set.intersection<S>(_:) 4-6x
Set.intersection(_:) 1-4x

lorentey added 10 commits November 1, 2021 21:38
@lorentey
[stdlib] _UnsafeBitset.withTemporaryBitset: New internal function
e2415f7
@lorentey
[stdlib] _SetVariant.convertedToNative: New convenience function
6b1f6c0
@lorentey
[stdlib] Optimize Set.isSubset<S>(of: S)
5e642d2 
Use a temporary bitset to speed up the Sequence variant by roughly a factor of 3.
@lorentey
[stdlib] Optimize Set.isStrictSubset<S>(of: S)
9043c71 
Use a temporary bitset to speed up the Sequence variant by roughly a factor of 3.
@lorentey
[stdlib] Optimize Set.isSuperset<S>(of: S)
725ee55 
Call into the specialized overload if the argument happens to be a `Set`.
@lorentey
[stdlib] Optimize Set.isStrictSuperset(of:)
a2540c2 
- Use a temporary bitset to speed up the `Sequence` variant by roughly a factor of 4.
- Fix a logic error causing the `a == b` case for the set variant to be O(n) instead of O(1).
@lorentey
[stdlib] Add a fast path to Set.isDisjoint<S>(with: S)
8612f2f 
Have the generic variant call out to the specialized overload if the argument happens to be a `Set`.
@lorentey
[stdlib] Optimize Set.subtracting(_:)
80296bb 
Use a temporary bitset to avoid hashing elements more than once, and to prevent rehashings during the creation of the result set.

This leads to a speedup of about 0-4x, depending on the number of elements removed.
@lorentey
[stdlib] Optimize Set.filter(_:)
db80c78 
This works the same way as `Set.subtracting<S>(_:)`, and has similar performance benefits.
@lorentey
[stdlib] Optimize Set.intersection(_:)
2e3e88c 
Use a temporary bitset to speed up the `Sequence` variant by roughly a factor of ~4-6, and the set/set variant by a factor of ~1-4, depending on the ratio of overlapping elements.
@lorentey
Copy link
Member Author

lorentey commented Nov 2, 2021

@swift-ci test

@lorentey
Copy link
Member Author

lorentey commented Nov 2, 2021

@swift-ci benchmark

@lorentey lorentey mentioned this pull request Nov 2, 2021
Closed
@swift-ci
Copy link
Contributor

swift-ci commented Nov 2, 2021

Performance (x86_64): -O

Regression OLD NEW DELTA RATIO
Set.subtracting.Seq.Int.Empty 137 368 +168.6% 0.37x
Set.subtracting.Int.Empty 40 81 +102.5% 0.49x
Set.subtracting.Seq.Box.Empty 197 374 +89.8% 0.53x
Set.subtracting.Box.Empty 30 54 +80.0% 0.56x
Set.subtracting.Seq.Empty.Box 136 225 +65.4% 0.60x
Set.subtracting.Seq.Int25 420 575 +36.9% 0.73x
Set.subtracting.Empty.Box 15 19 +26.7% 0.79x
AnyHashableWithAClass 94000 119000 +26.6% 0.79x (?)
SetSubtractingInt25 98 122 +24.5% 0.80x
Set.subtracting.Seq.Empty.Int 134 164 +22.4% 0.82x
Set.subtracting.Empty.Int 27 32 +18.5% 0.84x (?)
CharacterLiteralsLarge 97 108 +11.3% 0.90x (?)
 
Improvement OLD NEW DELTA RATIO
Set.isStrictSuperset.Seq.Box0 10780 38 -99.6% 283.68x
Set.isStrictSuperset.Seq.Int0 4809 27 -99.4% 178.10x
Set.filter.Int100.28k 1940 50 -97.4% 38.80x
Set.filter.Int100.16k 1021 28 -97.3% 36.46x
Set.filter.Int100.20k 1142 35 -96.9% 32.63x
Set.filter.Int100.24k 1289 42 -96.7% 30.69x
Set.isSubset.Seq.Box25 1231 101 -91.8% 12.19x
Set.isStrictSubset.Seq.Box25 1236 103 -91.7% 12.00x
Set.isStrictSuperset.Seq.Box25 1085 102 -90.6% 10.64x
Set.isSubset.Seq.Int25 555 74 -86.7% 7.50x
Set.intersection.Seq.Int100 433 59 -86.4% 7.34x
Set.isStrictSubset.Seq.Int25 553 76 -86.3% 7.28x
Set.isStrictSuperset.Seq.Int25 481 74 -84.6% 6.50x
SetIntersectionInt100 340 62 -81.8% 5.48x
Set.intersection.Seq.Box0 364 78 -78.6% 4.67x
Set.isSubset.Seq.Int50 629 151 -76.0% 4.17x
Set.isStrictSubset.Seq.Int50 635 153 -75.9% 4.15x
Set.intersection.Seq.Box25 476 118 -75.2% 4.03x
Set.isStrictSuperset.Seq.Int100 1088 306 -71.9% 3.56x
Set.isStrictSubset.Seq.Int100 1084 308 -71.6% 3.52x
Set.isStrictSubset.Box0 342 104 -69.6% 3.29x
Set.isStrictSuperset.Seq.Int50 486 150 -69.1% 3.24x
Set.intersection.Seq.Int50 305 96 -68.5% 3.18x
Set.intersection.Seq.Int25 237 77 -67.5% 3.08x
Set.isSubset.Seq.Box0 1081 389 -64.0% 2.78x
Set.intersection.Seq.Int0 157 57 -63.7% 2.75x
Set.isStrictSubset.Seq.Box0 1086 416 -61.7% 2.61x
Set.isSubset.Seq.Int100 783 310 -60.4% 2.53x
Set.subtracting.Seq.Int100 752 302 -59.8% 2.49x
Set.isStrictSubset.Int0 256 104 -59.4% 2.46x
Set.filter.Int50.28k 1085 449 -58.6% 2.42x
Set.filter.Int50.16k 583 242 -58.5% 2.41x
Set.filter.Int50.20k 670 301 -55.1% 2.23x
SetSubtractingInt100 153 72 -52.9% 2.12x
Set.filter.Int50.24k 765 362 -52.7% 2.11x
SetIntersectionInt50 205 99 -51.7% 2.07x
SetIntersectionBox25 240 127 -47.1% 1.89x
Set.subtracting.Seq.Box0 745 413 -44.6% 1.80x
SetIntersectionInt25 138 81 -41.3% 1.70x
Set.isSubset.Seq.Int0 482 283 -41.3% 1.70x
Set.isStrictSubset.Seq.Int0 480 291 -39.4% 1.65x
Set.isStrictSuperset.Seq.Empty.Int 278 174 -37.4% 1.60x
SetIntersectionBox0 132 87 -34.1% 1.52x
Set.subtracting.Seq.Box25 1212 969 -20.0% 1.25x
DictionaryBridgeToObjC_Access 1272 1052 -17.3% 1.21x (?)
SetSubtractingInt50 121 105 -13.2% 1.15x
Set.isStrictSubset.Int.Empty 80 70 -12.5% 1.14x
SetSubtractingBox0 148 131 -11.5% 1.13x (?)
FlattenListFlatMap 6751 6103 -9.6% 1.11x (?)
NSStringConversion.Rebridge.UTF8 652 592 -9.2% 1.10x (?)
SetSubtractingInt0 68 62 -8.8% 1.10x
Set.isSubset.Seq.Int.Empty 193 176 -8.8% 1.10x (?)
Set.isStrictSubset.Seq.Int.Empty 192 178 -7.3% 1.08x (?)
StringBuilder 331 308 -6.9% 1.07x (?)
StringBuilderSmallReservingCapacity 340 317 -6.8% 1.07x (?)

Code size: -O

Regression OLD NEW DELTA RATIO
SetTests.o 131325 138749 +5.7% 0.95x
 
Improvement OLD NEW DELTA RATIO
DictTest3.o 15981 15085 -5.6% 1.06x

Performance (x86_64): -Osize

Regression OLD NEW DELTA RATIO
Set.subtracting.Seq.Int.Empty 138 367 +165.9% 0.38x
Set.subtracting.Int.Empty 39 81 +107.7% 0.48x
Set.subtracting.Seq.Box.Empty 195 397 +103.6% 0.49x
Set.subtracting.Box.Empty 30 54 +80.0% 0.56x
Set.subtracting.Seq.Empty.Box 137 232 +69.3% 0.59x
Set.subtracting.Empty.Int 27 45 +66.7% 0.60x
Set.subtracting.Seq.Empty.Int 134 222 +65.7% 0.60x
FlattenListFlatMap 4393 6858 +56.1% 0.64x (?)
Set.subtracting.Seq.Int25 435 573 +31.7% 0.76x
Set.subtracting.Empty.Box 15 19 +26.7% 0.79x
AnyHashableWithAClass 94000 119000 +26.6% 0.79x
SetSubtractingInt25 99 122 +23.2% 0.81x
Array2D 7520 8112 +7.9% 0.93x (?)
 
Improvement OLD NEW DELTA RATIO
Set.isStrictSuperset.Seq.Box0 12906 45 -99.7% 286.79x
Set.isStrictSuperset.Seq.Int0 4699 29 -99.4% 162.03x
Set.filter.Int100.28k 1915 52 -97.3% 36.83x
Set.filter.Int100.16k 1005 29 -97.1% 34.65x
Set.filter.Int100.20k 1125 36 -96.8% 31.25x
Set.filter.Int100.24k 1265 43 -96.6% 29.42x
Set.isSubset.Seq.Box25 1467 166 -88.7% 8.84x
Set.isStrictSubset.Seq.Box25 1469 168 -88.6% 8.74x
Set.isStrictSuperset.Seq.Box25 1307 162 -87.6% 8.07x
Set.isSubset.Seq.Int25 551 74 -86.6% 7.45x
Set.intersection.Seq.Int100 430 59 -86.3% 7.29x
Set.isStrictSubset.Seq.Int25 552 77 -86.1% 7.17x
Set.isStrictSuperset.Seq.Int25 475 74 -84.4% 6.42x
SetIntersectionInt100 342 63 -81.6% 5.43x
Set.isSubset.Seq.Int50 630 152 -75.9% 4.14x
Set.isStrictSubset.Seq.Int50 629 155 -75.4% 4.06x
Set.isStrictSuperset.Seq.Int100 1090 305 -72.0% 3.57x
Set.isStrictSubset.Box0 352 101 -71.3% 3.49x
Set.isStrictSubset.Seq.Int100 1089 317 -70.9% 3.44x
Set.isStrictSuperset.Seq.Int50 473 149 -68.5% 3.17x
Set.intersection.Seq.Int50 303 96 -68.3% 3.16x
Set.intersection.Seq.Box25 542 175 -67.7% 3.10x
Set.intersection.Seq.Int25 234 77 -67.1% 3.04x
Set.intersection.Seq.Box0 407 136 -66.6% 2.99x
Set.intersection.Seq.Int0 156 57 -63.5% 2.74x
Set.subtracting.Seq.Int100 771 304 -60.6% 2.54x
Set.isStrictSubset.Int0 259 104 -59.8% 2.49x
Set.isSubset.Seq.Int100 786 316 -59.8% 2.49x
Set.filter.Int50.28k 1094 440 -59.8% 2.49x
Set.filter.Int50.16k 588 239 -59.4% 2.46x
Set.filter.Int50.20k 675 296 -56.1% 2.28x
SetSubtractingInt100 157 70 -55.4% 2.24x
Set.filter.Int50.24k 771 356 -53.8% 2.17x
SetIntersectionInt50 207 101 -51.2% 2.05x
Set.isStrictSubset.Seq.Box0 1300 682 -47.5% 1.91x
Set.isSubset.Seq.Box0 1298 692 -46.7% 1.88x
SetIntersectionBox25 277 163 -41.2% 1.70x
Set.isSubset.Seq.Int0 478 284 -40.6% 1.68x
SetIntersectionInt25 138 82 -40.6% 1.68x
Set.isStrictSubset.Seq.Int0 479 303 -36.7% 1.58x
Set.isStrictSuperset.Seq.Empty.Int 279 202 -27.6% 1.38x
Set.subtracting.Seq.Box0 895 683 -23.7% 1.31x
SetSubtractingBox0 178 137 -23.0% 1.30x
SetIntersectionBox0 147 117 -20.4% 1.26x
Set.isStrictSubset.Int.Empty 84 71 -15.5% 1.18x
SetSubtractingInt50 124 105 -15.3% 1.18x
SetSubtractingBox25 291 251 -13.7% 1.16x
Set.subtracting.Seq.Box25 1410 1239 -12.1% 1.14x (?)
Set.subtracting.Seq.Int50 544 486 -10.7% 1.12x (?)
DictionaryBridgeToObjC_Access 1193 1079 -9.6% 1.11x (?)
RemoveWhereFilterString 322 293 -9.0% 1.10x
SetSubtractingInt0 69 63 -8.7% 1.10x (?)
ObjectiveCBridgeStubFromNSString 1766 1623 -8.1% 1.09x (?)
Set.isStrictSubset.Empty.Int 169 157 -7.1% 1.08x (?)

Code size: -Osize

Regression OLD NEW DELTA RATIO
SetTests.o 110066 120179 +9.2% 0.92x

Performance (x86_64): -Onone

Regression OLD NEW DELTA RATIO
Set.isDisjoint.Seq.Empty.Int 435 733 +68.5% 0.59x
Set.isDisjoint.Seq.Empty.Box 456 747 +63.8% 0.61x
Set.isSuperset.Seq.Empty.Int 624 894 +43.3% 0.70x (?)
ArrayAppendAscii 15334 21624 +41.0% 0.71x (?)
ArrayAppendUTF16 15402 21624 +40.4% 0.71x
Set.isDisjoint.Seq.Int.Empty 698 973 +39.4% 0.72x (?)
ArrayAppendLatin1 15640 21692 +38.7% 0.72x (?)
Set.isDisjoint.Seq.Box.Empty 722 1001 +38.6% 0.72x
Set.isSubset.Seq.Empty.Int 311 418 +34.4% 0.74x
Set.subtracting.Box.Empty 149 189 +26.8% 0.79x
Set.subtracting.Int.Empty 160 200 +25.0% 0.80x
AnyHashableWithAClass 129500 157000 +21.2% 0.82x (?)
Set.isDisjoint.Seq.Int100 1846 2166 +17.3% 0.85x (?)
Set.isSuperset.Seq.Int0 1744 2041 +17.0% 0.85x (?)
Set.isSuperset.Seq.Int.Empty 1686 1969 +16.8% 0.86x (?)
SetSubtractingBox25 855 993 +16.1% 0.86x (?)
Set.isSuperset.Seq.Box0 2007 2304 +14.8% 0.87x (?)
Set.subtracting.Seq.Box.Empty 894 980 +9.6% 0.91x (?)
Set.subtracting.Seq.Box25 8726 9540 +9.3% 0.91x (?)
Set.subtracting.Seq.Int.Empty 872 953 +9.3% 0.92x (?)
StringInterpolation 11500 12400 +7.8% 0.93x (?)
 
Improvement OLD NEW DELTA RATIO
Set.isStrictSuperset.Seq.Box0 84279 255 -99.7% 330.50x
Set.isStrictSuperset.Seq.Int0 64464 227 -99.6% 283.98x
Set.filter.Int100.28k 5348 326 -93.9% 16.40x
Set.filter.Int100.16k 2886 185 -93.6% 15.60x
Set.filter.Int100.20k 3390 230 -93.2% 14.74x
Set.filter.Int100.24k 3969 275 -93.1% 14.43x
SetIntersectionInt100 876 117 -86.6% 7.49x
Set.isStrictSubset.Seq.Int25 6707 1320 -80.3% 5.08x
Set.isSubset.Seq.Int25 6648 1319 -80.2% 5.04x
Set.isStrictSuperset.Seq.Int25 6467 1314 -79.7% 4.92x
Set.isSubset.Seq.Box25 9282 1916 -79.4% 4.84x
Set.isStrictSubset.Seq.Box25 9257 1936 -79.1% 4.78x
Set.isStrictSuperset.Seq.Box25 8547 1927 -77.5% 4.44x
SetSubtractingInt100 491 167 -66.0% 2.94x
Set.filter.Int50.28k 3108 1059 -65.9% 2.93x
SetIntersectionInt50 529 181 -65.8% 2.92x
Set.filter.Int50.16k 1701 609 -64.2% 2.79x
Set.isStrictSubset.Box0 858 318 -62.9% 2.70x
Set.isStrictSubset.Seq.Int50 6875 2599 -62.2% 2.65x
Set.isSubset.Seq.Int50 6888 2619 -62.0% 2.63x
Set.filter.Int50.20k 2012 796 -60.4% 2.53x
Set.isStrictSuperset.Seq.Int50 6521 2608 -60.0% 2.50x
Set.filter.Int50.24k 2366 1019 -56.9% 2.32x
SetIntersectionInt25 345 149 -56.8% 2.32x
Set.isStrictSubset.Int0 603 287 -52.4% 2.10x
Set.intersection.Seq.Int100 2163 1039 -52.0% 2.08x
SetSubtractingInt0 307 156 -49.2% 1.97x
SetSubtractingInt50 433 239 -44.8% 1.81x
Set.intersection.Seq.Box25 2772 1607 -42.0% 1.72x
SetIntersectionBox25 1049 638 -39.2% 1.64x
Set.intersection.Seq.Int50 1832 1129 -38.4% 1.62x
Set.intersection.Seq.Box0 2227 1454 -34.7% 1.53x
Set.isStrictSubset.Seq.Int100 7957 5198 -34.7% 1.53x
Set.isStrictSuperset.Seq.Int100 7958 5233 -34.2% 1.52x
Set.intersection.Seq.Int25 1651 1095 -33.7% 1.51x
Set.isStrictSubset.Int.Empty 421 291 -30.9% 1.45x
SetSubtractingInt25 391 281 -28.1% 1.39x
Set.isSubset.Seq.Int100 7267 5255 -27.7% 1.38x
Set.intersection.Seq.Int0 1438 1053 -26.8% 1.37x
Set.subtracting.Seq.Int100 7037 5195 -26.2% 1.35x
SetSubtractingBox0 682 522 -23.5% 1.31x
Set.subtracting.Seq.Empty.Int 610 477 -21.8% 1.28x
Set.subtracting.Seq.Empty.Box 634 497 -21.6% 1.28x
Set.isStrictSubset.Seq.Int0 6575 5258 -20.0% 1.25x
Set.isSubset.Seq.Int0 6465 5220 -19.3% 1.24x
Set.isStrictSuperset.Seq.Empty.Int 1374 1122 -18.3% 1.22x (?)
SetIntersect 1400 1150 -17.9% 1.22x
SetIntersectionInt0 140 115 -17.9% 1.22x
Set.isStrictSubset.Seq.Box0 8519 7254 -14.8% 1.17x
Set.isSubset.Seq.Box0 8500 7253 -14.7% 1.17x
Set.subtracting.Seq.Int50 6525 5641 -13.5% 1.16x
Set.subtracting.Seq.Int0 6060 5263 -13.2% 1.15x (?)
Set.subtracting.Seq.Box0 8103 7235 -10.7% 1.12x (?)
Set.isStrictSubset.Seq.Int.Empty 1267 1133 -10.6% 1.12x (?)
Set.subtracting.Seq.Int25 6343 5885 -7.2% 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

@swift-ci
Copy link
Contributor

swift-ci commented Nov 2, 2021

Build failed
Swift Test OS X Platform
Git Sha - 2e3e88c

@lorentey
Copy link
Member Author

lorentey commented Nov 2, 2021 •
edited
Loading

Huh, a mere 5-10% overall code size increase on SetTests.o seems like a welcome surprise!

I have some assertion failures I'll need to investigate. (That isn't cause for alarm, this PR is a draft for a reason. 🙈)

@lorentey
Copy link
Member Author

lorentey commented Nov 3, 2021

@swift-ci benchmark

@lorentey
Copy link
Member Author

lorentey commented Nov 3, 2021

@swift-ci test

@lorentey lorentey marked this pull request as ready for review November 3, 2021 03:19
@swift-ci
Copy link
Contributor

swift-ci commented Nov 3, 2021

Performance (x86_64): -O

Regression OLD NEW DELTA RATIO
Set.subtracting.Empty.Box 15 36 +140.0% 0.42x
Set.subtracting.Seq.Int.Empty 137 242 +76.6% 0.57x
Set.subtracting.Seq.Empty.Int 135 233 +72.6% 0.58x
Set.subtracting.Seq.Empty.Box 136 233 +71.3% 0.58x
Set.subtracting.Empty.Int 27 46 +70.4% 0.59x
Set.subtracting.Seq.Int25 421 582 +38.2% 0.72x (?)
Set.subtracting.Seq.Box.Empty 197 272 +38.1% 0.72x
SetSubtractingInt25 97 123 +26.8% 0.79x
UTF8Decode_InitDecoding_ascii 253 288 +13.8% 0.88x (?)
NSStringConversion.Rebridge.UTF8 603 673 +11.6% 0.90x (?)
CharacterLiteralsLarge 97 108 +11.3% 0.90x (?)
LessSubstringSubstring 39 43 +10.3% 0.91x (?)
EqualSubstringSubstring 39 42 +7.7% 0.93x (?)
EqualStringSubstring 39 42 +7.7% 0.93x (?)
EqualSubstringString 39 42 +7.7% 0.93x (?)
ConvertFloatingPoint.MockFloat64Exactly2 26 28 +7.7% 0.93x (?)
 
Improvement OLD NEW DELTA RATIO
Set.isStrictSuperset.Seq.Box0 10790 38 -99.6% 283.94x
Set.isStrictSuperset.Seq.Int0 4804 26 -99.5% 184.76x
Set.filter.Int100.28k 1915 50 -97.4% 38.30x
Set.filter.Int100.16k 1007 28 -97.2% 35.96x
Set.filter.Int100.20k 1126 35 -96.9% 32.17x
Set.filter.Int100.24k 1267 42 -96.7% 30.17x
Set.isSubset.Seq.Box25 1230 101 -91.8% 12.18x
Set.isStrictSubset.Seq.Box25 1236 103 -91.7% 12.00x
Set.isStrictSuperset.Seq.Box25 1087 102 -90.6% 10.66x
Set.isSubset.Seq.Int25 555 74 -86.7% 7.50x
Set.intersection.Seq.Int100 434 59 -86.4% 7.36x
Set.isStrictSubset.Seq.Int25 553 76 -86.3% 7.28x
Set.isStrictSuperset.Seq.Int25 481 73 -84.8% 6.59x
SetIntersectionInt100 342 62 -81.9% 5.52x
Set.intersection.Seq.Box0 364 79 -78.3% 4.61x
Set.isSubset.Seq.Int50 630 151 -76.0% 4.17x
Set.isStrictSubset.Seq.Int50 636 154 -75.8% 4.13x
Set.intersection.Seq.Box25 476 118 -75.2% 4.03x
Set.isStrictSuperset.Seq.Int100 1086 306 -71.8% 3.55x
Set.isStrictSubset.Seq.Int100 1083 307 -71.7% 3.53x
Set.isStrictSubset.Box0 340 104 -69.4% 3.27x
Set.isStrictSuperset.Seq.Int50 487 149 -69.4% 3.27x
Set.intersection.Seq.Int50 305 96 -68.5% 3.18x
Set.intersection.Seq.Int25 237 77 -67.5% 3.08x
Set.intersection.Seq.Int0 157 56 -64.3% 2.80x
Set.isSubset.Seq.Box0 1081 391 -63.8% 2.76x
Set.isStrictSubset.Seq.Box0 1085 415 -61.8% 2.61x
Set.isSubset.Seq.Int100 782 310 -60.4% 2.52x
Set.isStrictSubset.Int0 257 104 -59.5% 2.47x
Set.subtracting.Seq.Int100 753 307 -59.2% 2.45x
Set.filter.Int50.28k 1086 450 -58.6% 2.41x
Set.filter.Int50.16k 583 244 -58.1% 2.39x
SetSubtractingInt100 153 67 -56.2% 2.28x
Set.filter.Int50.20k 670 302 -54.9% 2.22x
Set.filter.Int50.24k 765 363 -52.5% 2.11x
SetIntersectionInt50 204 100 -51.0% 2.04x
SetIntersectionBox25 240 125 -47.9% 1.92x
Set.isSubset.Seq.Int0 481 280 -41.8% 1.72x
SetIntersectionInt25 138 81 -41.3% 1.70x
Set.subtracting.Seq.Box0 744 444 -40.3% 1.68x
Set.isStrictSubset.Seq.Int0 480 292 -39.2% 1.64x
Set.isStrictSuperset.Seq.Empty.Int 278 174 -37.4% 1.60x
SetIntersectionBox0 132 84 -36.4% 1.57x
Set.subtracting.Seq.Box25 1214 956 -21.3% 1.27x
SetSubtractingInt50 121 104 -14.0% 1.16x
Set.isStrictSubset.Int.Empty 80 70 -12.5% 1.14x
SetSubtractingBox0 148 130 -12.2% 1.14x (?)
SetSubtractingInt0 68 61 -10.3% 1.11x (?)
Set.isSubset.Seq.Int.Empty 193 174 -9.8% 1.11x (?)
Breadcrumbs.MutatedUTF16ToIdx.Mixed 307 279 -9.1% 1.10x (?)
Set.isStrictSubset.Seq.Int.Empty 192 178 -7.3% 1.08x (?)
StringBuilder 331 308 -6.9% 1.07x (?)

Code size: -O

Regression OLD NEW DELTA RATIO
SetTests.o 131325 142173 +8.3% 0.92x
 
Improvement OLD NEW DELTA RATIO
DictTest3.o 15981 15085 -5.6% 1.06x

Performance (x86_64): -Osize

Regression OLD NEW DELTA RATIO
Set.subtracting.Empty.Box 15 36 +140.0% 0.42x
Set.subtracting.Seq.Empty.Box 137 237 +73.0% 0.58x
Set.subtracting.Seq.Int.Empty 138 238 +72.5% 0.58x
Set.subtracting.Seq.Empty.Int 134 226 +68.7% 0.59x
Set.subtracting.Empty.Int 27 45 +66.7% 0.60x
Set.subtracting.Seq.Box.Empty 195 274 +40.5% 0.71x
Set.subtracting.Seq.Int25 438 580 +32.4% 0.76x
AnyHashableWithAClass 93500 119000 +27.3% 0.79x (?)
SetSubtractingInt25 99 122 +23.2% 0.81x
LessSubstringSubstring 40 44 +10.0% 0.91x (?)
Set.isSubset.Seq.Int.Empty 195 212 +8.7% 0.92x
Array2D 7520 8112 +7.9% 0.93x (?)
 
Improvement OLD NEW DELTA RATIO
Set.isStrictSuperset.Seq.Box0 12899 45 -99.7% 286.64x
Set.isStrictSuperset.Seq.Int0 4697 30 -99.4% 156.56x
Set.filter.Int100.28k 1912 51 -97.3% 37.49x
Set.filter.Int100.16k 1004 29 -97.1% 34.62x
Set.filter.Int100.20k 1124 35 -96.9% 32.11x
Set.filter.Int100.24k 1263 42 -96.7% 30.07x
Set.isSubset.Seq.Box25 1469 167 -88.6% 8.80x
Set.isStrictSubset.Seq.Box25 1469 168 -88.6% 8.74x
Set.isStrictSuperset.Seq.Box25 1307 162 -87.6% 8.07x
Set.isSubset.Seq.Int25 552 74 -86.6% 7.46x
Set.intersection.Seq.Int100 430 59 -86.3% 7.29x
Set.isStrictSubset.Seq.Int25 553 76 -86.3% 7.28x
Set.isStrictSuperset.Seq.Int25 476 74 -84.5% 6.43x
SetIntersectionInt100 342 64 -81.3% 5.34x
Set.isSubset.Seq.Int50 629 152 -75.8% 4.14x
Set.isStrictSubset.Seq.Int50 629 154 -75.5% 4.08x
Set.isStrictSuperset.Seq.Int100 1094 307 -71.9% 3.56x
Set.isStrictSubset.Box0 352 101 -71.3% 3.49x
Set.isStrictSubset.Seq.Int100 1090 313 -71.3% 3.48x
Set.intersection.Seq.Int50 303 96 -68.3% 3.16x
Set.isStrictSuperset.Seq.Int50 472 150 -68.2% 3.15x
Set.intersection.Seq.Box25 542 174 -67.9% 3.11x
Set.intersection.Seq.Int25 234 77 -67.1% 3.04x
Set.intersection.Seq.Box0 407 136 -66.6% 2.99x
Set.intersection.Seq.Int0 156 57 -63.5% 2.74x
Set.isStrictSubset.Int0 260 101 -61.2% 2.57x
Set.isSubset.Seq.Int100 785 311 -60.4% 2.52x
Set.subtracting.Seq.Int100 769 308 -59.9% 2.50x
Set.filter.Int50.28k 1094 447 -59.1% 2.45x
Set.filter.Int50.16k 588 243 -58.7% 2.42x
SetSubtractingInt100 158 68 -57.0% 2.32x
Set.filter.Int50.20k 675 302 -55.3% 2.24x
Set.filter.Int50.24k 771 361 -53.2% 2.14x
SetIntersectionInt50 208 101 -51.4% 2.06x
Set.isStrictSubset.Seq.Box0 1299 679 -47.7% 1.91x
Set.isSubset.Seq.Box0 1298 688 -47.0% 1.89x
SetIntersectionInt25 139 82 -41.0% 1.70x
SetIntersectionBox25 277 164 -40.8% 1.69x
Set.isSubset.Seq.Int0 478 286 -40.2% 1.67x
Set.isStrictSubset.Seq.Int0 477 306 -35.8% 1.56x
FlattenListLoop 2540 1659 -34.7% 1.53x (?)
Set.isStrictSuperset.Seq.Empty.Int 278 207 -25.5% 1.34x
Set.subtracting.Seq.Box0 894 672 -24.8% 1.33x
SetSubtractingBox0 178 136 -23.6% 1.31x
SetIntersectionBox0 146 119 -18.5% 1.23x
SetSubtractingInt50 124 104 -16.1% 1.19x
Set.isStrictSubset.Int.Empty 84 71 -15.5% 1.18x
SetSubtractingBox25 291 251 -13.7% 1.16x
SetSubtractingInt0 69 61 -11.6% 1.13x
Set.subtracting.Seq.Box25 1410 1249 -11.4% 1.13x (?)
Set.subtracting.Seq.Int50 547 492 -10.1% 1.11x
StringBuilder 330 304 -7.9% 1.09x (?)
String.data.LargeUnicode 120 112 -6.7% 1.07x (?)

Code size: -Osize

Regression OLD NEW DELTA RATIO
SetTests.o 110066 124097 +12.7% 0.89x

Performance (x86_64): -Onone

Regression OLD NEW DELTA RATIO
Set.isDisjoint.Seq.Empty.Int 436 730 +67.4% 0.60x
Set.isDisjoint.Seq.Empty.Box 455 753 +65.5% 0.60x
Set.isSuperset.Seq.Empty.Int 626 907 +44.9% 0.69x
Set.isDisjoint.Seq.Int.Empty 700 997 +42.4% 0.70x (?)
ArrayAppendAscii 15334 21794 +42.1% 0.70x (?)
ArrayAppendUTF16 15436 21624 +40.1% 0.71x (?)
Set.isDisjoint.Seq.Box.Empty 724 1012 +39.8% 0.72x
ArrayAppendLatin1 15640 21658 +38.5% 0.72x (?)
Set.isSubset.Seq.Empty.Int 311 420 +35.0% 0.74x
SetSubtractingBox25 855 1127 +31.8% 0.76x
Set.subtracting.Seq.Box25 8687 10268 +18.2% 0.85x (?)
Set.isDisjoint.Seq.Int100 1864 2186 +17.3% 0.85x (?)
Set.subtracting.Empty.Box 116 135 +16.4% 0.86x (?)
Set.subtracting.Empty.Int 129 147 +14.0% 0.88x
DataCreateMediumArray 6360 7220 +13.5% 0.88x (?)
RandomDoubleOpaqueDef 51100 56000 +9.6% 0.91x (?)
RandomDoubleDef 50700 55300 +9.1% 0.92x (?)
 
Improvement OLD NEW DELTA RATIO
Set.isStrictSuperset.Seq.Box0 84431 254 -99.7% 332.40x
Set.isStrictSuperset.Seq.Int0 65051 228 -99.6% 285.31x
Set.filter.Int100.28k 5356 326 -93.9% 16.43x
Set.filter.Int100.16k 2890 185 -93.6% 15.62x
Set.filter.Int100.20k 3392 231 -93.2% 14.68x
Set.filter.Int100.24k 3970 276 -93.0% 14.38x
SetIntersectionInt100 875 119 -86.4% 7.35x
Set.isSubset.Seq.Int25 6733 1311 -80.5% 5.14x
Set.isStrictSubset.Seq.Int25 6716 1324 -80.3% 5.07x
Set.isStrictSuperset.Seq.Int25 6508 1310 -79.9% 4.97x
Set.isSubset.Seq.Box25 9267 1916 -79.3% 4.84x
Set.isStrictSubset.Seq.Box25 9271 1947 -79.0% 4.76x
Set.isStrictSuperset.Seq.Box25 8615 1918 -77.7% 4.49x
Set.filter.Int50.28k 3109 1053 -66.1% 2.95x
SetIntersectionInt50 528 184 -65.2% 2.87x
Set.filter.Int50.16k 1700 602 -64.6% 2.82x
Set.isStrictSubset.Box0 859 307 -64.3% 2.80x
Set.isSubset.Seq.Int50 6908 2599 -62.4% 2.66x
Set.isStrictSubset.Seq.Int50 6849 2610 -61.9% 2.62x
Set.filter.Int50.20k 2013 790 -60.8% 2.55x
Set.isStrictSuperset.Seq.Int50 6525 2591 -60.3% 2.52x
Set.filter.Int50.24k 2365 1020 -56.9% 2.32x
SetIntersectionInt25 344 150 -56.4% 2.29x
Set.isStrictSubset.Int0 603 287 -52.4% 2.10x
Set.intersection.Seq.Int100 2176 1039 -52.3% 2.09x
SetSubtractingInt0 307 157 -48.9% 1.96x
SetSubtractingInt100 491 279 -43.2% 1.76x
Set.intersection.Seq.Box25 2787 1610 -42.2% 1.73x
SetIntersectionBox25 1050 637 -39.3% 1.65x
Set.intersection.Seq.Int50 1838 1123 -38.9% 1.64x
Set.intersection.Seq.Box0 2237 1443 -35.5% 1.55x
Set.isStrictSubset.Seq.Int100 7976 5187 -35.0% 1.54x
Set.isStrictSuperset.Seq.Int100 7955 5230 -34.3% 1.52x
Set.intersection.Seq.Int25 1656 1093 -34.0% 1.52x
Set.isStrictSubset.Int.Empty 422 291 -31.0% 1.45x
Set.isSubset.Seq.Int100 7297 5226 -28.4% 1.40x (?)
Set.intersection.Seq.Int0 1447 1052 -27.3% 1.38x
SetSubtractingBox0 683 518 -24.2% 1.32x
Set.isSubset.Seq.Int0 6569 5266 -19.8% 1.25x
Set.isStrictSubset.Seq.Int0 6509 5290 -18.7% 1.23x
SetIntersectionInt0 140 115 -17.9% 1.22x
Set.subtracting.Seq.Int100 7060 5843 -17.2% 1.21x (?)
SetIntersect 1400 1160 -17.1% 1.21x
Set.isStrictSuperset.Seq.Empty.Int 1374 1152 -16.2% 1.19x (?)
Set.isSubset.Seq.Box0 8527 7203 -15.5% 1.18x
Set.isStrictSubset.Seq.Box0 8516 7249 -14.9% 1.17x
SetSubtractingInt50 433 374 -13.6% 1.16x
Set.subtracting.Seq.Int0 6085 5267 -13.4% 1.16x (?)
Breadcrumbs.MutatedUTF16ToIdx.Mixed 316 287 -9.2% 1.10x (?)
Set.subtracting.Seq.Box0 7948 7231 -9.0% 1.10x (?)
Set.isStrictSubset.Seq.Int.Empty 1266 1152 -9.0% 1.10x (?)
Breadcrumbs.MutatedIdxToUTF16.Mixed 330 304 -7.9% 1.09x (?)
Data.init.Sequence.64kB.Count0.RE.I 31942 29574 -7.4% 1.08x (?)
Data.append.Sequence.64kB.Count0.RE.I 31806 29524 -7.2% 1.08x (?)
Data.append.Sequence.809B.Count0.RE 39439 36642 -7.1% 1.08x (?)
Data.append.Sequence.809B.Count0.RE.I 39356 36714 -6.7% 1.07x (?)
Data.init.Sequence.64kB.Count0.RE 31730 29603 -6.7% 1.07x (?)
Data.init.Sequence.809B.Count0.RE 39648 37033 -6.6% 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

@swift-ci
Copy link
Contributor

swift-ci commented Nov 3, 2021

Build failed
Swift Test Linux Platform
Git Sha - af06df9

@lorentey
Copy link
Member Author

lorentey commented Nov 3, 2021

@swift-ci test

@lorentey lorentey merged commit 6d33683 into swiftlang:main Nov 5, 2021
@lorentey lorentey deleted the set-on-fire2 branch November 5, 2021 20:48
@lorentey lorentey mentioned this pull request Dec 9, 2021
Draft
7 tasks
@lorentey lorentey mentioned this pull request Jan 13, 2025
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@timvermeulen timvermeulen Awaiting requested review from timvermeulen

@glessard glessard Awaiting requested review from glessard

@natecook1000 natecook1000 Awaiting requested review from natecook1000

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@lorentey @swift-ci