[IRGen][interop] do not add 'nocapture' to not bitwise takable types #68481

hyp · 2023-09-13T01:19:49Z

The use of 'nocapture' for parameters and return values is incorrect for C++ types, as they can actually capture a pointer into its own value (e.g. std::string in libstdc++)

rdar://115062687

hyp · 2023-09-13T01:20:37Z

@swift-ci please test

hyp · 2023-09-13T01:20:42Z

@swift-ci please benchmark

hyp · 2023-09-13T15:08:41Z

Performance (x86_64): -O

Regression	OLD	NEW	DELTA	RATIO
StringWithCString2	0.0	0.002	+200.0%	0.33x
UTF8Decode_InitDecoding	134.917	165.0	+22.3%	0.82x (?)
UTF8Decode_InitFromCustom_contiguous	137.071	164.923	+20.3%	0.83x (?)
UTF8Decode_InitFromCustom_noncontiguous	174.818	202.889	+16.1%	0.86x (?)
Array.removeAll.keepingCapacity.Int	0.097	0.11	+13.3%	0.88x (?)

Improvement	OLD	NEW	DELTA	RATIO
SequenceAlgosRange	2178.75	1742.727	-20.0%	1.25x (?)
CxxStringConversion.cxx.to.swift	146.5	128.0	-12.6%	1.14x (?)
StringComparison_ascii	208.091	187.0	-10.1%	1.11x (?)
Calculator	157.75	144.063	-8.7%	1.10x (?)
Chars2	3523.81	3231.25	-8.3%	1.09x (?)

Code size: -O

Regression	OLD	NEW	DELTA	RATIO
BinaryFloatingPointConversionFromBinaryInteger.o	28860	29292	+1.5%	0.99x
COWArrayGuaranteedParameterOverhead.o	1314	1330	+1.2%	0.99x
InsertCharacter.o	4643	4691	+1.0%	0.99x

Performance (x86_64): -Osize

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	133.636	161.444	+20.8%	0.83x (?)
UTF8Decode_InitFromCustom_contiguous	133.545	160.929	+20.5%	0.83x (?)
UTF8Decode_InitFromCustom_noncontiguous	258.75	285.833	+10.5%	0.91x (?)
Set.subtracting.Empty.Box	19.594	21.333	+8.9%	0.92x (?)
CxxStringConversion.cxx.to.swift	142.4	154.8	+8.7%	0.92x (?)
Set.filter.Int50.28k	317.286	342.25	+7.9%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
SequenceAlgosRange	2614.444	1742.727	-33.3%	1.50x
FlattenListLoop	1622.0	1381.0	-14.9%	1.17x (?)
Set.isSuperset.Seq.Empty.Int	50.08	44.647	-10.8%	1.12x (?)
Calculator	157.833	141.077	-10.6%	1.12x (?)
UTF8Decode_InitDecoding_ascii	183.8	169.2	-7.9%	1.09x (?)

Code size: -Osize

Improvement	OLD	NEW	DELTA	RATIO
BinaryFloatingPointConversionFromBinaryInteger.o	28406	28094	-1.1%	1.01x

Performance (x86_64): -Onone

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	141.333	168.0	+18.9%	0.84x (?)
UTF8Decode_InitFromCustom_contiguous	145.25	169.4	+16.6%	0.86x (?)
CharacterRecognizer.ascii	127.786	139.0	+8.8%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
CxxStringConversion.cxx.to.swift	161.0	129.4	-19.6%	1.24x (?)

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: 32 GB

hyp · 2023-09-13T15:09:03Z

@swift-ci please benchmark

hyp · 2023-09-14T16:32:40Z

Performance (x86_64): -O

Regression	OLD	NEW	DELTA	RATIO
ArrayAppendToGeneric	300.0	621.667	+107.2%	0.48x (?)
ArrayAppendGenericStructs	1380.0	1786.667	+29.5%	0.77x (?)
UTF8Decode_InitFromCustom_contiguous	133.824	161.333	+20.6%	0.83x (?)
UTF8Decode_InitDecoding	137.7	160.9	+16.8%	0.86x (?)
UTF8Decode_InitFromCustom_noncontiguous	173.1	201.111	+16.2%	0.86x (?)

Improvement	OLD	NEW	DELTA	RATIO
SequenceAlgosRange	2178.75	1742.727	-20.0%	1.25x (?)
Array.removeAll.keepingCapacity.Object	6.707	5.823	-13.2%	1.15x (?)
StringComparison_ascii	208.636	187.0	-10.4%	1.12x (?)
Chars2	3590.476	3235.714	-9.9%	1.11x (?)
Calculator	158.333	144.462	-8.8%	1.10x (?)
CxxStringConversion.cxx.to.swift	143.0	130.5	-8.7%	1.10x (?)
Data.hash.Medium	27.423	25.067	-8.6%	1.09x (?)
OpenClose	56.625	52.286	-7.7%	1.08x (?)

Code size: -O

Regression	OLD	NEW	DELTA	RATIO
BinaryFloatingPointConversionFromBinaryInteger.o	28860	29292	+1.5%	0.99x
COWArrayGuaranteedParameterOverhead.o	1314	1330	+1.2%	0.99x
InsertCharacter.o	4643	4691	+1.0%	0.99x

Performance (x86_64): -Osize

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	136.167	165.091	+21.2%	0.82x (?)
UTF8Decode_InitFromCustom_contiguous	138.923	165.909	+19.4%	0.84x (?)
UTF8Decode_InitFromCustom_noncontiguous	258.714	285.286	+10.3%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
SequenceAlgosRange	2615.0	1742.727	-33.4%	1.50x
UTF8Decode_InitFromCustom_contiguous_ascii	201.5	173.0	-14.1%	1.16x (?)
Calculator	155.933	141.462	-9.3%	1.10x (?)
Set.isSuperset.Seq.Empty.Int	49.0	44.64	-8.9%	1.10x (?)
Set.isDisjoint.Seq.Int.Empty	49.0	44.758	-8.7%	1.09x (?)
CxxStringConversion.cxx.to.swift	181.5	169.0	-6.9%	1.07x (?)
DataAppendDataMediumToMedium	3189.231	2975.0	-6.7%	1.07x (?)

Code size: -Osize

Improvement	OLD	NEW	DELTA	RATIO
BinaryFloatingPointConversionFromBinaryInteger.o	28406	28094	-1.1%	1.01x

Performance (x86_64): -Onone

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	140.667	168.5	+19.8%	0.83x (?)
UTF8Decode_InitFromCustom_contiguous	143.5	171.545	+19.5%	0.84x (?)
CharacterRecognizer.ascii	127.786	137.692	+7.8%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
CxxStringConversion.cxx.to.swift	137.5	125.0	-9.1%	1.10x (?)

Code size: -swiftlibs

Regression	OLD	NEW	DELTA	RATIO
libswiftSwiftPrivateLibcExtras.dylib	16384	32768	+100.0%	0.50x

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: 32 GB

hyp · 2023-09-14T18:02:50Z

@swift-ci please test windows platform

hyp · 2023-09-14T18:31:48Z

@swift-ci please test

hyp · 2023-09-15T22:22:02Z

@swift-ci please test

hyp · 2023-09-18T18:03:36Z

@swift-ci please test

hyp · 2023-09-25T15:50:19Z

@swift-ci please test

hyp · 2023-09-25T23:03:49Z

@swift-ci please test

hyp · 2023-09-25T23:04:11Z

should be ready now.

hyp · 2023-09-25T23:04:21Z

@swift-ci please benchmark

hyp · 2023-09-25T23:04:28Z

@swift-ci please test source compatibility

The use of 'nocapture' for parameters and return values is incorrect for C++ types, as they can actually capture a pointer into its own value (e.g. std::string in libstdc++) rdar://115062687

hyp · 2023-09-26T00:45:21Z

@swift-ci please test

hyp · 2023-09-26T00:45:31Z

@swift-ci please test source compatibility

hyp · 2023-09-26T00:45:43Z

@swift-ci please benchmark

The use of 'nocapture' for parameters and return values is incorrect for C++ types, as they can actually capture a pointer into its own value (e.g. std::string in libstdc++) rdar://115062687 Cherrypick commit 4858cb6 Cherrypick PR swiftlang#68481

hyp added c++ interop Feature: Interoperability with C++ IRGen LLVM IR generation labels Sep 13, 2023

hyp force-pushed the eng/no-nocapture branch from 2ce2367 to ce195bf Compare September 15, 2023 22:21

hyp marked this pull request as ready for review September 15, 2023 23:29

hyp requested review from zoecarver, egorzhdan, ktoso and adrian-prantl as code owners September 15, 2023 23:29

[IRGen][interop] do not add 'nocapture' to not bitwise takable types

4858cb6

The use of 'nocapture' for parameters and return values is incorrect for C++ types, as they can actually capture a pointer into its own value (e.g. std::string in libstdc++) rdar://115062687

hyp force-pushed the eng/no-nocapture branch from 4ad4a4c to 4858cb6 Compare September 26, 2023 00:43

hyp merged commit 6ecea1a into swiftlang:main Sep 26, 2023

finagolfin mentioned this pull request Oct 14, 2023

[android][test] Fix five tests that are failing on the community Android CI #69189

Merged

hjyamauchi mentioned this pull request Nov 15, 2024

Cherrypick fixes for issue 74866 to release/5.10 #77277

Closed

[IRGen][interop] do not add 'nocapture' to not bitwise takable types #68481

[IRGen][interop] do not add 'nocapture' to not bitwise takable types #68481

Uh oh!

Conversation

hyp commented Sep 13, 2023

Uh oh!

hyp commented Sep 13, 2023

Uh oh!

hyp commented Sep 13, 2023

Uh oh!

hyp commented Sep 13, 2023

Performance (x86_64): -O

Code size: -O

Performance (x86_64): -Osize

Code size: -Osize

Performance (x86_64): -Onone

Code size: -swiftlibs

Uh oh!

hyp commented Sep 13, 2023

Uh oh!

hyp commented Sep 14, 2023

Performance (x86_64): -O

Code size: -O

Performance (x86_64): -Osize

Code size: -Osize

Performance (x86_64): -Onone

Code size: -swiftlibs

Uh oh!

hyp commented Sep 14, 2023

Uh oh!

hyp commented Sep 14, 2023

Uh oh!

hyp commented Sep 15, 2023

Uh oh!

hyp commented Sep 18, 2023

Uh oh!

hyp commented Sep 25, 2023

Uh oh!

hyp commented Sep 25, 2023

Uh oh!

hyp commented Sep 25, 2023

Uh oh!

hyp commented Sep 25, 2023

Uh oh!

hyp commented Sep 25, 2023

Uh oh!

hyp commented Sep 26, 2023

Uh oh!

hyp commented Sep 26, 2023

Uh oh!

hyp commented Sep 26, 2023

Uh oh!

Uh oh!