SIL optimizer: a collection of optimizer changes to prepare for COW representation #32023

eeckstein · 2020-05-26T17:49:07Z

This PR contains a set of independent (and mostly small) optimizer changes which are needed - or beneficial - for the switch to the COW representation in Array types.

For details see the commit list.

I'd like to explicitly mention one larger addition: A new optimization pass which constant folds the uniqueness result of begin_cow_mutation instructions, if it can be proved that the buffer argument is uniquely referenced.
For example:

     %buffer = end_cow_mutation %mutable_buffer
     // ...
     // %buffer does not escape here
     // ...
     (%is_unique, %mutable_buffer2) = begin_cow_mutation %buffer
     cond_br %is_unique, ...

is replaced with

     %buffer = end_cow_mutation [keep_unique] %mutable_buffer
     // ...
     (%not_used, %mutable_buffer2) = begin_cow_mutation %buffer
     %true = integer_literal 1
     cond_br %true, ...

If someone is interested, https://github.com/apple/swift/pull/31730/commits contains all changes for supporting COW representation in Array types.

Constant folds the uniqueness result of begin_cow_mutation instructions, if it can be proved that the buffer argument is uniquely referenced. For example: %buffer = end_cow_mutation %mutable_buffer // ... // %buffer does not escape here // ... (%is_unique, %mutable_buffer2) = begin_cow_mutation %buffer cond_br %is_unique, ... is replaced with %buffer = end_cow_mutation [keep_unique] %mutable_buffer // ... (%not_used, %mutable_buffer2) = begin_cow_mutation %buffer %true = integer_literal 1 cond_br %true, ... Note that the keep_unique flag is set on the end_cow_mutation because the code now relies on that the buffer is really uniquely referenced. The optimization can also handle def-use chains between end_cow_mutation and begin_cow_mutation which involve phi-arguments. An additional peephole optimization is performed: if the begin_cow_mutation is the only use of the end_cow_mutation, the whole pair of instructions is eliminated.

… owned-to-guarantee transformation. If a function is annotated with @_semantics("optimize.sil.specialize.owned2guarantee.never") its arguments will not be converted from owned to guaranteed.

… are only destroyed. If the only use of an upcast, unchecked_ref_cast or end_cow_mutation is a destroy/release, just destroy the operand and remove the cast/end_cow_mutation.

(end_cow_mutation (upcast X)) -> (end_cow_mutation X) (end_cow_mutation (unchecked_ref_cast X)) -> (end_cow_mutation X)

…mantics Used to "finalize" an array literal. It's not used, yet. So this is NFC. Also handle the "array.finalize_intrinsic" function in various array specific optimizations.

…e optimizations. Mostly this is about "looking through" a begin_cow_mutation or end_cow_mutation.

This is in preparation for COW support. More optimizer tests require an optimized non-assert stdlib build.

…nction. Support a version of Array._checkSubscript which has no wasNativeTypeChecked parameter.

The COWOpts optimization relies more on LICM. This additional run of the pass ensures that there is no phase ordering issue between LICM and COWOpts

…ination

eeckstein · 2020-05-26T17:50:30Z

@swift-ci test

eeckstein · 2020-05-26T17:50:37Z

@swift-ci benchmark

swift-ci · 2020-05-26T18:16:50Z

Performance: -O

Code size: -O

Performance: -Osize

Improvement	OLD	NEW	DELTA	RATIO
SuffixArrayLazy	5	4	-20.0%	1.25x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubToNSDateRef	2500	2720	+8.8%	0.92x (?)

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: 64 GB

swift-ci · 2020-05-26T19:11:07Z

Build failed
Swift Test Linux Platform
Git Sha - 3f42ad7

swift-ci · 2020-05-26T19:59:05Z

Build failed
Swift Test OS X Platform
Git Sha - 3f42ad7

eeckstein · 2020-05-27T07:15:59Z

@swift-ci test

swift-ci · 2020-05-27T08:29:46Z

Build failed
Swift Test Linux Platform
Git Sha - 3f42ad7

eeckstein · 2020-05-27T13:58:03Z

@swift-ci test linux

eeckstein added 10 commits May 26, 2020 18:01

FunctionSignatureTransforms: add a semantics attribute to prevent the…

e3ac98c

… owned-to-guarantee transformation. If a function is annotated with @_semantics("optimize.sil.specialize.owned2guarantee.never") its arguments will not be converted from owned to guaranteed.

SILCombine: remove dead casts and end_cow_mutation instructions which…

4ca6b31

… are only destroyed. If the only use of an upcast, unchecked_ref_cast or end_cow_mutation is a destroy/release, just destroy the operand and remove the cast/end_cow_mutation.

SILCombine: Remove a cast if it's only used by an end_cow_mutation.

01465d9

(end_cow_mutation (upcast X)) -> (end_cow_mutation X) (end_cow_mutation (unchecked_ref_cast X)) -> (end_cow_mutation X)

SIL: new "array.end_mutation" and "array.finalize_intrinsic" array se…

2403e56

…mantics Used to "finalize" an array literal. It's not used, yet. So this is NFC. Also handle the "array.finalize_intrinsic" function in various array specific optimizations.

SIL optimizer: Support begin_cow_mutation and end_cow_mutation in som…

33c8e16

…e optimizations. Mostly this is about "looking through" a begin_cow_mutation or end_cow_mutation.

tests: make some test more robust for optimizer changes.

16bd756

This is in preparation for COW support. More optimizer tests require an optimized non-assert stdlib build.

ArraySemantics: support "array.check_subscript" as a two-parameter fu…

99f4d85

…nction. Support a version of Array._checkSubscript which has no wasNativeTypeChecked parameter.

SIL optimizer: add an additional LICM pass to the pipeline.

216eec2

The COWOpts optimization relies more on LICM. This additional run of the pass ensures that there is no phase ordering issue between LICM and COWOpts

DeadObjectElimination: don't let fix_lifetime prevent dead array elim…

3f42ad7

…ination

eeckstein requested a review from atrick May 26, 2020 17:49

eeckstein merged commit 5013a02 into swiftlang:master May 28, 2020

eeckstein deleted the cow-optimizer-changes branch May 28, 2020 07:40

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

SIL optimizer: a collection of optimizer changes to prepare for COW representation #32023

SIL optimizer: a collection of optimizer changes to prepare for COW representation #32023

Uh oh!

eeckstein commented May 26, 2020 •

edited

Loading

Uh oh!

eeckstein commented May 26, 2020

Uh oh!

eeckstein commented May 26, 2020

Uh oh!

swift-ci commented May 26, 2020

Uh oh!

swift-ci commented May 26, 2020

Uh oh!

swift-ci commented May 26, 2020

Uh oh!

eeckstein commented May 27, 2020

Uh oh!

swift-ci commented May 27, 2020

Uh oh!

eeckstein commented May 27, 2020

Uh oh!

Uh oh!

SIL optimizer: a collection of optimizer changes to prepare for COW representation #32023

SIL optimizer: a collection of optimizer changes to prepare for COW representation #32023

Uh oh!

Conversation

eeckstein commented May 26, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

eeckstein commented May 26, 2020

Uh oh!

eeckstein commented May 26, 2020

Uh oh!

swift-ci commented May 26, 2020

Performance: -O

Code size: -O

Performance: -Osize

Code size: -Osize

Performance: -Onone

Code size: -swiftlibs

Uh oh!

swift-ci commented May 26, 2020

Uh oh!

swift-ci commented May 26, 2020

Uh oh!

eeckstein commented May 27, 2020

Uh oh!

swift-ci commented May 27, 2020

Uh oh!

eeckstein commented May 27, 2020

Uh oh!

Uh oh!

eeckstein commented May 26, 2020 •

edited

Loading