[SIL Optimization] Make ArraySemantics.cpp aware of "array.uninitialized_intrinsic" #29643

ravikandhadai · 2020-02-04T21:18:13Z

This semantics attribute is used by the top-level array initializer (in ArrayShared.swift),
which is the entry point used by the compiler to initialize array from array literals.
This initializer is "early inlined" so that other optimizations can work on its body.

This PR also updates DeadObjectElimination and ArrayCOWOpts optimization passes to work with this semantics attribute in addition to "array.uninitialized", which they already use. (But this will not make them more effective as they are also applied after early inlining.)

It also refactors mapInitializationStores function from ArrayElementValuePropagation.cpp to
ArraySemantic.cpp so that the array-initialization pattern matching functionality
implemented by the function can be reused by other optimizations.

The purpose of this PR is to reuse the functionality implemented in ArraySemantics such as getArrayValue, getStoragePointer etc. in a new "mandatory" pass that will unroll forEach loops over array literals, which itself is aimed at optimizing the new os log calls (see #29651).

ravikandhadai · 2020-02-04T21:22:08Z

@swift-ci Please test

ravikandhadai · 2020-02-04T21:24:08Z

@swift-ci Please benchmark

swift-ci · 2020-02-04T21:50:46Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
ClassArrayGetter2	50	60	+20.0%	0.83x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
SuffixCountableRangeLazy	4	6	+50.0%	0.67x (?)
PrefixWhileCountableRange	15	17	+13.3%	0.88x (?)

Code size: -Osize

Performance: -Onone

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: 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: 64 GB

ravikandhadai · 2020-02-04T22:33:18Z

@swift-ci Please benchmark

swift-ci · 2020-02-04T23:42:23Z

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubFromArrayOfNSString2	2920	3140	+7.5%	0.93x (?)

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Improvement	OLD	NEW	DELTA	RATIO
DictionaryBridgeToObjC_Access	1057	924	-12.6%	1.14x (?)

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: 8-Core Intel Xeon E5
  Processor Speed: 3 GHz
  Number of Processors: 1
  Total Number of Cores: 8
  L2 Cache (per Core): 256 KB
  L3 Cache: 25 MB
  Memory: 64 GB

eeckstein

lgtm!
I mentioned some cosmetic issues in my comments.

eeckstein · 2020-02-05T07:56:52Z

lib/SILOptimizer/Analysis/ArraySemantic.cpp

@@ -652,7 +671,20 @@ SILValue swift::ArraySemanticsCall::getArrayValue() const {
 }

 SILValue swift::ArraySemanticsCall::getArrayElementStoragePointer() const {
-  if (getKind() == ArrayCallKind::kArrayUninitialized) {


This function looks almost exactly the same as getArrayValue(), except the tuple element number (and the special handling of ArrayCallKind::kArrayInit in getArrayValue).
Can you extract the common code into a helper function? e.g. getArrayInitResult(int tupleElement)

Sure. That is a good idea.

eeckstein · 2020-02-05T08:09:54Z

lib/SILOptimizer/Analysis/ArraySemantic.cpp

+    } else if (SI)
+      return false;
+
+    // Store an index_addr projection.


typo: Store to an ...

…zed_intrinsic" semantics attribute that is used by the top-level array initializer (in ArrayShared.swift), which is the entry point used by the compiler to initialize array from array literals. This initializer is early-inlined so that other optimizations can work on its body. Fix DeadObjectElimination and ArrayCOWOpts optimization passes to work with this semantics attribute in addition to "array.uninitialized", which they already use. Refactor mapInitializationStores function from ArrayElementValuePropagation.cpp to ArraySemantic.cpp so that the array-initialization pattern matching functionality implemented by the function can be reused by other optimizations.

ravikandhadai · 2020-02-05T22:32:49Z

@swift-ci Please test and merge

ravikandhadai · 2020-02-05T22:33:19Z

@swift-ci Please test and merge

ravikandhadai · 2020-02-05T23:34:22Z

@swift-ci Please test macOS Platform

ravikandhadai · 2020-02-05T23:34:54Z

@swift-ci Please test macOS Platform

swift-ci · 2020-02-05T23:37:07Z

Build failed
Swift Test OS X Platform
Git Sha - 72243bfd7ff4b511e8966ada74eccfc94b176eac

ravikandhadai requested review from atrick and eeckstein February 4, 2020 21:18

ravikandhadai mentioned this pull request Feb 5, 2020

Add a new mandatory optimization pass for unrolling forEach calls over arrays created from array literals. #29651

Merged

eeckstein approved these changes Feb 5, 2020

View reviewed changes

ravikandhadai force-pushed the fix_dead_array_elim branch from 72243bf to a6bed21 Compare February 5, 2020 22:29

ravikandhadai merged commit bca5511 into swiftlang:master Feb 6, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[SIL Optimization] Make ArraySemantics.cpp aware of "array.uninitialized_intrinsic" #29643

[SIL Optimization] Make ArraySemantics.cpp aware of "array.uninitialized_intrinsic" #29643

Uh oh!

ravikandhadai commented Feb 4, 2020 •

edited

Loading

Uh oh!

ravikandhadai commented Feb 4, 2020

Uh oh!

ravikandhadai commented Feb 4, 2020

Uh oh!

swift-ci commented Feb 4, 2020

Uh oh!

ravikandhadai commented Feb 4, 2020

Uh oh!

swift-ci commented Feb 4, 2020

Uh oh!

eeckstein left a comment

Uh oh!

eeckstein Feb 5, 2020

Uh oh!

ravikandhadai Feb 5, 2020

Uh oh!

eeckstein Feb 5, 2020

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

swift-ci commented Feb 5, 2020

Uh oh!

Uh oh!

[SIL Optimization] Make ArraySemantics.cpp aware of "array.uninitialized_intrinsic" #29643

[SIL Optimization] Make ArraySemantics.cpp aware of "array.uninitialized_intrinsic" #29643

Uh oh!

Conversation

ravikandhadai commented Feb 4, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ravikandhadai commented Feb 4, 2020

Uh oh!

ravikandhadai commented Feb 4, 2020

Uh oh!

swift-ci commented Feb 4, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

ravikandhadai commented Feb 4, 2020

Uh oh!

swift-ci commented Feb 4, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

eeckstein left a comment

Choose a reason for hiding this comment

Uh oh!

eeckstein Feb 5, 2020

Choose a reason for hiding this comment

Uh oh!

ravikandhadai Feb 5, 2020

Choose a reason for hiding this comment

Uh oh!

eeckstein Feb 5, 2020

Choose a reason for hiding this comment

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

ravikandhadai commented Feb 5, 2020

Uh oh!

swift-ci commented Feb 5, 2020

Uh oh!

Uh oh!

ravikandhadai commented Feb 4, 2020 •

edited

Loading