[ownership] Move lowering past SILCombine and right before the inliner. #35802

gottesmm · 2021-02-06T01:06:25Z

Just a good place to test this functionality separately from the inliner.

NOTE: There is currently a draft since there is one commit I am going to get in via a different PR. This PR I am going to use initially for testing and maybe once that other commit lands, rebase on top and just move the passes as an independent change into the tree.

gottesmm · 2021-02-06T01:06:32Z

@swift-ci test

gottesmm · 2021-02-06T01:06:54Z

@swift-ci test source compatibility

gottesmm · 2021-02-06T01:07:00Z

@swift-ci benchmark

…unchecked_take_enum_data_addr + load -> load + unchecked_enum_data. We must make sure the new load is inserted where the old load was instead of inserting it at the unchecked_take_enum_data_addr, since the unchecked_take_enum_data_addr may be in a different block from old load /and/ old load may not post-dominate that point. This was just a thinko.

… run. I am doing this to separate the perf effect of moving ownership past the inliner with this earlier stuff.

gottesmm · 2021-02-06T02:04:01Z

@swift-ci test

gottesmm · 2021-02-06T02:04:08Z

@swift-ci benchmark

gottesmm · 2021-02-06T02:04:13Z

@swift-ci test source compatibility

gottesmm · 2021-02-06T04:37:50Z

@swift-ci smoke test compiler performance

gottesmm · 2021-02-06T05:05:22Z

Just to note, current source compat release failures:

18:12:47   FAIL: Dollar, 5.1, 4feadc, Dollar, generic/platform=macOS
18:12:47   FAIL: Dollar, 4.0, 433d4b, Dollar, generic/platform=macOS
18:12:47   FAIL: GRDB.swift, 5.0, 759c10, Swift Package
18:12:47   FAIL: GRDB.swift, 4.2, 759c10, Swift Package
18:12:47   FAIL: Kitura, 5.1, 8578c4, Swift Package
18:12:47   FAIL: Kitura, 4.0, f1c316, Swift Package
18:12:47   FAIL: ProcedureKit, 5.1, a7fa56, ProcedureKitCloud, platform=macOS
18:12:47   FAIL: ProcedureKit, 5.1, a7fa56, ProcedureKit, generic/platform=tvOS
18:12:47   FAIL: ProcedureKit, 5.1, a7fa56, ProcedureKit, generic/platform=iOS
18:12:47   FAIL: ProcedureKit, 5.1, a7fa56, ProcedureKit, platform=macOS
18:12:47   FAIL: ProcedureKit, 5.0, a7fa56, ProcedureKitCloud, platform=macOS
18:12:47   FAIL: ProcedureKit, 5.0, a7fa56, ProcedureKit, generic/platform=tvOS
18:12:47   FAIL: ProcedureKit, 5.0, a7fa56, ProcedureKit, generic/platform=iOS
18:12:47   FAIL: ProcedureKit, 5.0, a7fa56, ProcedureKit, platform=macOS
18:12:47   FAIL: ProcedureKit, 4.2, 94193d, ProcedureKitCloud, platform=macOS
18:12:47   FAIL: ProcedureKit, 4.2, 94193d, ProcedureKit, generic/platform=tvOS
18:12:47   FAIL: ProcedureKit, 4.2, 94193d, ProcedureKit, generic/platform=iOS
18:12:47   FAIL: ProcedureKit, 4.2, 94193d, ProcedureKit, platform=macOS
18:12:47   FAIL: swift-log, 5.2, 16cfcc, Swift Package
18:12:47   FAIL: swift-log, 5.0, 16cfcc, Swift Package
18:12:47   FAIL: async-http-client, 5.0, a72c5a, Swift Package
18:12:47   FAIL: swift-distributed-tracing, 5.2, 73f2b5, Swift Package
18:12:47   FAIL: swift-distributed-tracing, 5.0, 73f2b5, Swift Package
18:12:47   FAIL: swift-cluster-membership, 5.2, 0fd434, Swift Package
18:12:47   FAIL: SwiftyJSON, 5.1, bad5f2, SwiftyJSON watchOS, generic/platform=watchOS
18:12:47   FAIL: SwiftyJSON, 5.1, bad5f2, SwiftyJSON tvOS, generic/platform=tvOS
18:12:47   FAIL: SwiftyJSON, 5.1, bad5f2, SwiftyJSON iOS, generic/platform=iOS
18:12:47   FAIL: SwiftyJSON, 5.1, bad5f2, SwiftyJSON macOS, platform=macOS
18:12:47   FAIL: SwiftyJSON, 5.0, b93c5d, SwiftyJSON watchOS, generic/platform=watchOS
18:12:47   FAIL: SwiftyJSON, 5.0, b93c5d, SwiftyJSON tvOS, generic/platform=tvOS
18:12:47   FAIL: SwiftyJSON, 5.0, b93c5d, SwiftyJSON iOS, generic/platform=iOS
18:12:47   FAIL: SwiftyJSON, 5.0, b93c5d, SwiftyJSON macOS, platform=macOS
18:12:47   FAIL: SwiftyJSON, 4.2, b93c5d, SwiftyJSON watchOS, generic/platform=watchOS
18:12:47   FAIL: SwiftyJSON, 4.2, b93c5d, SwiftyJSON tvOS, generic/platform=tvOS
18:12:47   FAIL: SwiftyJSON, 4.2, b93c5d, SwiftyJSON iOS, generic/platform=iOS
18:12:47   FAIL: SwiftyJSON, 4.2, b93c5d, SwiftyJSON macOS, platform=macOS
18:12:47   FAIL: vapor_console, 5.1, fcd8c1, Swift Package
18:12:47 ========================================

And the debug has had issues over the past bit just building. Just going to make sure I didn't make this worse.

swift-ci · 2021-02-06T06:03:16Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
FlattenListLoop	1682	2515	+49.5%	0.67x (?)
StringInterpolation	10800	12400	+14.8%	0.87x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced	4500	4960	+10.2%	0.91x (?)
StringBuilderLong	1350	1480	+9.6%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	8	5	-37.5%	1.60x
LessSubstringSubstring	44	39	-11.4%	1.13x
StringToDataMedium	3800	3400	-10.5%	1.12x (?)
EqualSubstringSubstring	42	39	-7.1%	1.08x (?)
EqualSubstringString	42	39	-7.1%	1.08x (?)
EqualStringSubstring	43	40	-7.0%	1.07x
EqualSubstringSubstringGenericEquatable	43	40	-7.0%	1.07x
LessSubstringSubstringGenericComparable	43	40	-7.0%	1.07x (?)
SortStringsUnicode	3100	2895	-6.6%	1.07x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
Diffing.Same	8	9	+12.5%	0.89x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced	4660	5120	+9.9%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	8	5	-37.5%	1.60x
String.data.LargeUnicode	126	106	-15.9%	1.19x (?)
FlattenListFlatMap	7092	6061	-14.5%	1.17x (?)
LessSubstringSubstringGenericComparable	44	39	-11.4%	1.13x
LessSubstringSubstring	43	39	-9.3%	1.10x (?)
InsertCharacterEndIndexNonASCII	65	59	-9.2%	1.10x (?)
EqualStringSubstring	44	40	-9.1%	1.10x (?)
EqualSubstringSubstringGenericEquatable	44	40	-9.1%	1.10x (?)
StringWalk	2720	2520	-7.4%	1.08x
StrComplexWalk	5150	4780	-7.2%	1.08x

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
FloatingPointPrinting_Float_description_uniform	19800	25100	+26.8%	0.79x (?)
FloatingPointPrinting_Float_description_small	22140	27216	+22.9%	0.81x (?)
StringBuilderLong	2350	2720	+15.7%	0.86x (?)
ArrayLiteral2	3286	3743	+13.9%	0.88x (?)
PrefixWhileCountableRangeLazy	52589	56776	+8.0%	0.93x (?)
ArrayValueProp	5196	5606	+7.9%	0.93x (?)
ArrayOfPOD	1061	1142	+7.6%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	99	90	-9.1%	1.10x (?)
DataReplaceSmallBuffer	6000	5500	-8.3%	1.09x (?)
ClassArrayGetter2	4830	4490	-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

gottesmm · 2021-02-06T06:10:59Z

@swift-ci benchmark

gottesmm · 2021-02-06T07:22:40Z

Release source compat failed due to a weird module map issue.

gottesmm · 2021-02-06T07:22:54Z

Not this PR for sure.

gottesmm · 2021-02-06T07:23:29Z

Debug one hit the same thing:

23:19:17 49%: Compile Triple.cpp (x86_64)
23:19:17 49%: Compile Swift source files (arm64)
23:19:17 fatal error: module map file '/Users/buildnode/jenkins/workspace-private/swift-PR-source-compat-suite-debug/build/compat_macos/swiftpm-macosx-x86_64/apple/Intermediates.noindex/GeneratedModuleMaps/macosx/CCryptoBoringSSL.modulemap' not found
23:19:17 
23:19:17 fatal error: module map file '/Users/buildnode/jenkins/workspace-private/swift-PR-source-compat-suite-debug/build/compat_macos/swiftpm-macosx-x86_64/apple/Intermediates.noindex/GeneratedModuleMaps/macosx/CCryptoBoringSSL.modulemap' not found
23:19:17 
23:19:17 fatal error: module map file '/Users/buildnode/jenkins/workspace-private/swift-PR-source-compat-suite-debug/build/compat_macos/swiftpm-macosx-x86_64/apple/Intermediates.noindex/GeneratedModuleMaps/macosx/CCryptoBoringSSL.modulemap' not found
23:19:17 
23:19:17 fatal error: module map file '/Users/buildnode/jenkins/workspace-private/swift-PR-source-compat-suite-debug/build/compat_macos/swiftpm-macosx-x86_64/apple/Intermediates.noindex/GeneratedModuleMaps/macosx/CCryptoBoringSSL.modulemap' not found
23:19:17 
23:19:18 49%: Compile Threading.cpp (x86_64)
23:19:18 49%: Compile TargetParser.cpp (x86_64)
23:19:18 1 error generated.
23:19:18 
23:19:18 2 errors generated.
23:19:18 
23:19:18 1 error generated.
23:19:18 
23:19:18 Build cancelled
23:19:18 env SDKROOT=/Applicat

swift-ci · 2021-02-06T08:22:59Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitFromData_ascii_as_ascii	628	742	+18.2%	0.85x (?)
StringToDataLargeUnicode	3450	3900	+13.0%	0.88x (?)
StringInterpolation	10900	12300	+12.8%	0.89x (?)
UTF8Decode_InitFromBytes_ascii_as_ascii	445	497	+11.7%	0.90x (?)
ObjectiveCBridgeToNSDictionary	15000	16200	+8.0%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	8	5	-37.5%	1.60x
FlattenListFlatMap	4547	3889	-14.5%	1.17x (?)
LessSubstringSubstring	44	39	-11.4%	1.13x (?)
DictionaryKeysContainsCocoa	35	32	-8.6%	1.09x (?)
ObjectiveCBridgeStubFromArrayOfNSString2	2630	2440	-7.2%	1.08x (?)
EqualSubstringString	42	39	-7.1%	1.08x (?)
EqualStringSubstring	43	40	-7.0%	1.07x (?)

Code size: -O

Performance: -Osize

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	8	5	-37.5%	1.60x
EqualStringSubstring	44	39	-11.4%	1.13x (?)
LessSubstringSubstringGenericComparable	44	39	-11.4%	1.13x
LessSubstringSubstring	43	39	-9.3%	1.10x
InsertCharacterEndIndexNonASCII	65	59	-9.2%	1.10x (?)
EqualSubstringSubstringGenericEquatable	44	40	-9.1%	1.10x (?)
StringWalk	2720	2520	-7.4%	1.08x (?)
EqualSubstringSubstring	42	39	-7.1%	1.08x (?)
EqualSubstringString	42	39	-7.1%	1.08x (?)
StrComplexWalk	5150	4790	-7.0%	1.08x (?)
InsertCharacterEndIndex	177	165	-6.8%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
FloatingPointPrinting_Float_description_uniform	19800	25000	+26.3%	0.79x (?)
FloatingPointPrinting_Float_description_small	22140	27216	+22.9%	0.81x
UTF8Decode_InitFromBytes_ascii_as_ascii	468	523	+11.8%	0.89x (?)
StringToDataLargeUnicode	8650	9450	+9.2%	0.92x (?)
ArrayOfPOD	1060	1142	+7.7%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	99	90	-9.1%	1.10x (?)
ClassArrayGetter2	4830	4490	-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

swift-ci · 2021-02-06T08:35:49Z

Compilation-performance test failed

LucianoPAlmeida · 2021-02-06T18:16:16Z

Release source compat failed due to a weird module map issue.

@gottesmm There is a discussion about that on #35503 :)

tbkka · 2021-02-06T21:04:14Z

This one I happen to know something about: FloatingPointPrinting_Float_description_uniform. That test just converts a bunch of float values to string in a tight loop. Interesting point: those strings should all be small, so it should not be creating refcounted objects at all.

gottesmm · 2021-02-06T21:19:52Z

I don't think the perf changes are real. Going to run again.

gottesmm · 2021-02-06T21:19:59Z

@swift-ci benchmark

gottesmm · 2021-02-06T21:20:11Z

Not real in the sense that the regressions are changing each time suggesting noise.

swift-ci · 2021-02-06T22:26:11Z

Performance: -O

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	7	5	-28.6%	1.40x
UTF8Decode_InitFromData_ascii_as_ascii	748	659	-11.9%	1.14x (?)
String.data.LargeUnicode	111	101	-9.0%	1.10x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
FlattenListFlatMap	3888	6093	+56.7%	0.64x (?)
UTF8Decode_InitFromData_ascii_as_ascii	612	662	+8.2%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	7	5	-28.6%	1.40x
InsertCharacterEndIndexNonASCII	65	59	-9.2%	1.10x (?)
InsertCharacterEndIndex	181	165	-8.8%	1.10x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
FloatingPointPrinting_Float_description_uniform	19700	25100	+27.4%	0.78x (?)
FloatingPointPrinting_Float_description_small	22032	27108	+23.0%	0.81x
StringBuilderLong	2280	2610	+14.5%	0.87x (?)
ErrorHandling	4590	5240	+14.2%	0.88x (?)
StringBuilderSmallReservingCapacity	3012	3264	+8.4%	0.92x (?)
DropWhileAnySeqCntRange	69710	75477	+8.3%	0.92x (?)
StringUTF16Builder	3080	3330	+8.1%	0.92x (?)
StringBuilder	3044	3279	+7.7%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
StringFromLongWholeSubstringGeneric	99	90	-9.1%	1.10x (?)
ClassArrayGetter2	4840	4490	-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

gottesmm · 2021-02-06T23:40:44Z

If you look at the 3 groups of results for -O, -Osize every time the regressions are changing suggesting they are noise. Consider for -O:

FlattenListLoop | 1682 | 2515 | +49.5% | 0.67x (?)
StringInterpolation | 10800 | 12400 | +14.8% | 0.87x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced | 4500 | 4960 | +10.2% | 0.91x (?)
StringBuilderLong | 1350 | 1480 | +9.6% | 0.91x (?)

UTF8Decode_InitFromData_ascii_as_ascii | 628 | 742 | +18.2% | 0.85x (?)
StringToDataLargeUnicode | 3450 | 3900 | +13.0% | 0.88x (?)
StringInterpolation | 10900 | 12300 | +12.8% | 0.89x (?)
UTF8Decode_InitFromBytes_ascii_as_ascii | 445 | 497 | +11.7% | 0.90x (?)
ObjectiveCBridgeToNSDictionary | 15000 | 16200 | +8.0% | 0.93x (?)

StringFromLongWholeSubstringGeneric | 7 | 5 | -28.6% | 1.40x
UTF8Decode_InitFromData_ascii_as_ascii | 748 | 659 | -11.9% | 1.14x (?)
String.data.LargeUnicode | 111 | 101 | -9.0% | 1.10x (?)

And for -Osize:

Diffing.Same | 8 | 9 | +12.5% | 0.89x (?)
ObjectiveCBridgeFromNSArrayAnyObjectForced | 4660 | 5120 | +9.9% | 0.91x (?)

StringFromLongWholeSubstringGeneric | 8 | 5 | -37.5% | 1.60x
EqualStringSubstring | 44 | 39 | -11.4% | 1.13x (?)
LessSubstringSubstringGenericComparable | 44 | 39 | -11.4% | 1.13x
LessSubstringSubstring | 43 | 39 | -9.3% | 1.10x
InsertCharacterEndIndexNonASCII | 65 | 59 | -9.2% | 1.10x (?)
EqualSubstringSubstringGenericEquatable | 44 | 40 | -9.1% | 1.10x (?)
StringWalk | 2720 | 2520 | -7.4% | 1.08x (?)
EqualSubstringSubstring | 42 | 39 | -7.1% | 1.08x (?)
EqualSubstringString | 42 | 39 | -7.1% | 1.08x (?)
StrComplexWalk | 5150 | 4790 | -7.0% | 1.08x (?)
InsertCharacterEndIndex | 177 | 165 | -6.8% | 1.07x (?)

FlattenListFlatMap | 3888 | 6093 | +56.7% | 0.64x (?)
UTF8Decode_InitFromData_ascii_as_ascii | 612 | 662 | +8.2% | 0.92x (?)

gottesmm · 2021-02-06T23:41:05Z

Merging!

gottesmm added 2 commits February 5, 2021 17:55

[ownership] Move ownership lowering to right before the first inliner…

5ab8022

… run. I am doing this to separate the perf effect of moving ownership past the inliner with this earlier stuff.

gottesmm force-pushed the move-ownership-past-sil-combine-before-inliner branch from da69a5f to 5ab8022 Compare February 6, 2021 02:02

gottesmm changed the title ~~[DRAFT][ownership] Move lowering past SILCombine and right before the inliner.~~ [ownership] Move lowering past SILCombine and right before the inliner. Feb 6, 2021

gottesmm marked this pull request as ready for review February 6, 2021 06:12

gottesmm merged commit fae2564 into swiftlang:main Feb 6, 2021

gottesmm deleted the move-ownership-past-sil-combine-before-inliner branch February 6, 2021 23:41

[ownership] Move lowering past SILCombine and right before the inliner. #35802

[ownership] Move lowering past SILCombine and right before the inliner. #35802

Uh oh!

Conversation

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

swift-ci commented Feb 6, 2021

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

swift-ci commented Feb 6, 2021

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

swift-ci commented Feb 6, 2021

Uh oh!

LucianoPAlmeida commented Feb 6, 2021

Uh oh!

tbkka commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

swift-ci commented Feb 6, 2021

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

gottesmm commented Feb 6, 2021

Uh oh!

Uh oh!