[metadata prespecialization] Support for classes. #30089

nate-chandler · 2020-02-27T02:10:22Z

When generic metadata for a class is requested in the same module where the class is defined, rather than a call to the generic metadata accessor or to a variant of typeForMangledNode, a call to a new accessor--a canonical specialized generic metadata accessor--is emitted.
The new function is defined schematically as follows:

    MetadataResponse `canonical specialized metadata accessor for C<K>`(MetadataRequest request) {
      (void)`canonical specialized metadata accessor for superclass(C<K>)`(::Complete)
      (void)`canonical specialized metadata accessor for generic_argument_class(C<K>, 1)`(::Complete)
      ...
      (void)`canonical specialized metadata accessor for generic_argument_class(C<K>, count)`(::Complete)
      auto *metadata = objc_opt_self(`canonical specialized metadata for C<K>`);
      return {metadata, MetadataState::Complete};
    }

To enable these new canonical specialized generic metadata accessors, metadata for generic classes is prespecialized as needed. So are the metaclasses and the corresponding rodata.

Previously, the lazy objc naming hook was registered during process execution when the first generic class metadata was instantiated. Since that instantiation may occur "before process launch" (i.e. if the generic metadata is prespecialized), the lazy naming hook is now
installed at process launch.

nate-chandler · 2020-02-27T02:10:32Z

@swift-ci please benchmark

swift-ci · 2020-02-27T02:34:03Z

Build failed before running benchmark.

nate-chandler · 2020-02-28T01:37:41Z

@swift-ci please benchmark

swift-ci · 2020-02-28T02:47:46Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
FlattenListFlatMap	7092	7925	+11.7%	0.89x (?)

Improvement	OLD	NEW	DELTA	RATIO
LazilyFilteredArrayContains	35700	30200	-15.4%	1.18x
ArrayAppendLazyMap	6980	6070	-13.0%	1.15x
ObjectiveCBridgeStringHash	78	68	-12.8%	1.15x
ObjectiveCBridgeStringGetASCIIContents	328	305	-7.0%	1.08x

Code size: -O

Performance: -Osize

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStringHash	78	68	-12.8%	1.15x
String.data.LargeUnicode	109	96	-11.9%	1.14x (?)
ArrayAppendLazyMap	7900	6980	-11.6%	1.13x (?)
ObjectiveCBridgeStringGetASCIIContents	327	305	-6.7%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding_ascii	398	431	+8.3%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubFromNSDate	5930	5020	-15.3%	1.18x (?)
ObjectiveCBridgeStringHash	79	68	-13.9%	1.16x
ArrayOfPOD	1123	1042	-7.2%	1.08x (?)
ObjectiveCBridgeStringGetASCIIContents	328	305	-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

airspeedswift · 2020-02-29T22:32:32Z

Is it surprising we didn't see improvement at -Onone like we did with the struct metadata?

nate-chandler · 2020-03-04T21:17:05Z

@swift-ci please benchmark

nate-chandler · 2020-03-04T21:38:51Z

@airspeedswift This isn't quite done yet, but even when it is, we may need to directly refer to some specializations of generic classes to see similar performance changes.

swift-ci · 2020-03-04T22:28:29Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
EqualSubstringSubstring	40	44	+10.0%	0.91x (?)
EqualSubstringSubstringGenericEquatable	40	44	+10.0%	0.91x (?)
EqualStringSubstring	41	45	+9.8%	0.91x (?)
EqualSubstringString	41	45	+9.8%	0.91x (?)

Improvement	OLD	NEW	DELTA	RATIO
ArrayAppendLazyMap	7140	6100	-14.6%	1.17x
ObjectiveCBridgeStringHash	78	67	-14.1%	1.16x
LazilyFilteredArrayContains	34900	30200	-13.5%	1.16x
String.data.LargeUnicode	97	89	-8.2%	1.09x (?)
InsertCharacterEndIndexNonASCII	61	56	-8.2%	1.09x (?)
ObjectiveCBridgeStringGetASCIIContents	329	305	-7.3%	1.08x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
EqualStringSubstring	40	44	+10.0%	0.91x (?)
UTF8Decode_InitFromBytes_ascii	289	313	+8.3%	0.92x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStringHash	78	67	-14.1%	1.16x
ArrayAppendLazyMap	7900	6990	-11.5%	1.13x (?)
ObjectiveCBridgeStringGetASCIIContents	328	304	-7.3%	1.08x (?)
StringComparison_emoji	808	752	-6.9%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
UTF8Decode_InitFromData_ascii_as_ascii	675	738	+9.3%	0.91x (?)
UTF8Decode_InitFromBytes_ascii_as_ascii	493	535	+8.5%	0.92x (?)
LessSubstringSubstring	47	51	+8.5%	0.92x (?)
ArrayOfPOD	1035	1116	+7.8%	0.93x (?)

Improvement	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStringHash	78	67	-14.1%	1.16x
String.data.Empty	38	35	-7.9%	1.09x (?)
String.data.Small	41	38	-7.3%	1.08x (?)
ObjectiveCBridgeStringGetASCIIContents	330	306	-7.3%	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

nate-chandler · 2020-03-13T16:43:25Z

@swift-ci please test

swift-ci · 2020-03-13T16:45:32Z

Build failed
Swift Test Linux Platform
Git Sha - ad7b89e311905b6eb9f9888f11222430b1b1e2e3

swift-ci · 2020-03-13T16:45:52Z

Build failed
Swift Test OS X Platform
Git Sha - ad7b89e311905b6eb9f9888f11222430b1b1e2e3

nate-chandler · 2020-03-14T00:47:36Z

@swift-ci please test

swift-ci · 2020-03-14T01:27:34Z

Build failed
Swift Test Linux Platform
Git Sha - 1f8bdd46247df2bfcbcc2b40a680a860d8ddf947

swift-ci · 2020-03-14T01:55:41Z

Build failed
Swift Test OS X Platform
Git Sha - 1f8bdd46247df2bfcbcc2b40a680a860d8ddf947

nate-chandler · 2020-03-17T21:41:55Z

@swift-ci please test

swift-ci · 2020-03-17T21:44:24Z

Build failed
Swift Test OS X Platform
Git Sha - 1f8bdd46247df2bfcbcc2b40a680a860d8ddf947

nate-chandler · 2020-05-27T21:57:35Z

@swift-ci please test

swift-ci · 2020-05-27T22:00:27Z

Build failed
Swift Test Linux Platform
Git Sha - 71bb637e2fa144b8a2ba60de611ed61a0e1230b0

swift-ci · 2020-05-27T22:00:43Z

Build failed
Swift Test OS X Platform
Git Sha - 71bb637e2fa144b8a2ba60de611ed61a0e1230b0

nate-chandler · 2020-05-27T22:03:01Z

@swift-ci please test

swift-ci · 2020-05-27T22:03:24Z

Build failed
Swift Test Linux Platform
Git Sha - 3cbed7b1834a59421d26b2d0cb6b618b90230a73

swift-ci · 2020-05-27T22:06:48Z

Build failed
Swift Test OS X Platform
Git Sha - 3cbed7b1834a59421d26b2d0cb6b618b90230a73

nate-chandler · 2020-05-28T16:27:50Z

@swift-ci please clean test linux platform

swift-ci · 2020-05-28T17:43:09Z

Build failed
Swift Test Linux Platform
Git Sha - 668b088450eed2a44d6b3d4b61d4e65433851e8a

nate-chandler · 2020-05-28T21:56:44Z

@swift-ci please test

swift-ci · 2020-05-28T21:57:06Z

Build failed
Swift Test Linux Platform
Git Sha - 668b088450eed2a44d6b3d4b61d4e65433851e8a

swift-ci · 2020-05-28T21:59:02Z

Build failed
Swift Test OS X Platform
Git Sha - 668b088450eed2a44d6b3d4b61d4e65433851e8a

nate-chandler · 2020-05-28T22:33:21Z

@swift-ci please test macos platform

When generic metadata for a class is requested in the same module where the class is defined, rather than a call to the generic metadata accessor or to a variant of typeForMangledNode, a call to a new accessor--a canonical specialized generic metadata accessor--is emitted. The new function is defined schematically as follows: MetadataResponse `canonical specialized metadata accessor for C<K>`(MetadataRequest request) { (void)`canonical specialized metadata accessor for superclass(C<K>)`(::Complete) (void)`canonical specialized metadata accessor for generic_argument_class(C<K>, 1)`(::Complete) ... (void)`canonical specialized metadata accessor for generic_argument_class(C<K>, count)`(::Complete) auto *metadata = objc_opt_self(`canonical specialized metadata for C<K>`); return {metadata, MetadataState::Complete}; } where generic_argument_class(C<K>, N) denotes the Nth generic argument which is both (1) itself a specialized generic type and is also (2) a class. These calls to the specialized metadata accessors for these related types ensure that all generic class types are registered with the Objective-C runtime. To enable these new canonical specialized generic metadata accessors, metadata for generic classes is prespecialized as needed. So are the metaclasses and the corresponding rodata. Previously, the lazy objc naming hook was registered during process execution when the first generic class metadata was instantiated. Since that instantiation may occur "before process launch" (i.e. if the generic metadata is prespecialized), the lazy naming hook is now installed at process launch.

nate-chandler · 2020-05-29T20:21:02Z

@swift-ci please test

swift-ci · 2020-05-29T20:21:24Z

Build failed
Swift Test Linux Platform
Git Sha - a8e949c787042dafc2bdbbd8817d9ebce27f49f2

swift-ci · 2020-05-29T20:23:53Z

Build failed
Swift Test OS X Platform
Git Sha - a8e949c787042dafc2bdbbd8817d9ebce27f49f2

nate-chandler · 2020-05-29T21:08:28Z

@swift-ci please test macos platform

nate-chandler · 2020-05-29T21:55:34Z

@swift-ci please test linux platform

compnerd · 2020-05-30T00:53:26Z

This breaks the Windows bots because the test case file names are far too long. The file system limits files to 260 characters. Could you please rename the files?

nate-chandler · 2020-05-30T00:55:38Z

@compnerd as mentioned there, let's just fix the filename, rather than reverting and recommitting.

compnerd · 2020-05-30T00:59:06Z

@nate-chandler responded there as well, but renaming definitely is better.

nate-chandler · 2020-05-30T01:08:27Z

@compnerd Renaming PR at #32098 .

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from e469b54 to ad7b89e Compare February 28, 2020 01:36

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from ad7b89e to 821659b Compare March 4, 2020 21:16

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch 4 times, most recently from 232cc58 to 35a6172 Compare March 11, 2020 18:23

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch 4 times, most recently from cbd1d12 to d3cfca1 Compare March 13, 2020 16:43

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch 3 times, most recently from bd8d35a to 1f8bdd4 Compare March 14, 2020 00:47

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from 1f8bdd4 to da6a524 Compare March 17, 2020 20:57

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from 71bb637 to 3cbed7b Compare May 22, 2020 01:45

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from 3cbed7b to da20de1 Compare May 27, 2020 21:57

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from da20de1 to 668b088 Compare May 27, 2020 22:02

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from 668b088 to a8e949c Compare May 28, 2020 21:56

nate-chandler force-pushed the generic-metadata-prespecialization-components/classes branch from a8e949c to bdef1ce Compare May 29, 2020 20:20

nate-chandler merged commit 5204af3 into swiftlang:master May 30, 2020

nate-chandler deleted the generic-metadata-prespecialization-components/classes branch May 30, 2020 00:32

compnerd mentioned this pull request May 30, 2020

Revert "[metadata prespecialization] Support for classes." #32097

Closed

[metadata prespecialization] Support for classes. #30089

[metadata prespecialization] Support for classes. #30089

Uh oh!

Conversation

nate-chandler commented Feb 27, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nate-chandler commented Feb 27, 2020

Uh oh!

swift-ci commented Feb 27, 2020

Uh oh!

nate-chandler commented Feb 28, 2020

Uh oh!

swift-ci commented Feb 28, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

airspeedswift commented Feb 29, 2020

Uh oh!

nate-chandler commented Mar 4, 2020

Uh oh!

nate-chandler commented Mar 4, 2020

Uh oh!

swift-ci commented Mar 4, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

nate-chandler commented Mar 13, 2020

Uh oh!

swift-ci commented Mar 13, 2020

Uh oh!

swift-ci commented Mar 13, 2020

Uh oh!

nate-chandler commented Mar 14, 2020

Uh oh!

swift-ci commented Mar 14, 2020

Uh oh!

swift-ci commented Mar 14, 2020

Uh oh!

nate-chandler commented Mar 17, 2020

Uh oh!

swift-ci commented Mar 17, 2020

Uh oh!

nate-chandler commented May 27, 2020

Uh oh!

swift-ci commented May 27, 2020

Uh oh!

swift-ci commented May 27, 2020

Uh oh!

nate-chandler commented May 27, 2020

Uh oh!

swift-ci commented May 27, 2020

Uh oh!

swift-ci commented May 27, 2020

Uh oh!

nate-chandler commented May 28, 2020

Uh oh!

swift-ci commented May 28, 2020

Uh oh!

nate-chandler commented May 28, 2020

Uh oh!

swift-ci commented May 28, 2020

Uh oh!

swift-ci commented May 28, 2020

Uh oh!

nate-chandler commented May 28, 2020

Uh oh!

nate-chandler commented May 29, 2020

Uh oh!

swift-ci commented May 29, 2020

Uh oh!

swift-ci commented May 29, 2020

nate-chandler commented Feb 27, 2020 •

edited

Loading