Teach stripCasts (and getUnderlyingObject) about begin_access.

[atrick]

This was blocking EscapeAnalysis and many other analyses from handling
access markers.

[atrick]

@eeckstein I don't know why stripCasts does not check EndCOWMutation

[eeckstein]

It's a longer story: when I added end_cow_mutation, I came to the conclusion that those global stripWhatEver utility functions are not such a good idea.
Let's take stripCasts as example: every optimization has (or could have) a different view of what a cast instruction is. For EscapeAnalysis a begin_access is the same as a cast. But that's not necessarily true for other optimizations, which might need to treat begin_access separately.

I think it's better to let each optimization/analysis contain there own strip... functions, even if this means some code duplication.
Then it's much easier to modify them and add instructions as needed without worrying about other optimizations.

That's why I added end_cow_mutation only at those places where it was really needed and not in the global strip functions.
... and unfortunately, never came back to remove the global strip functions at all.

[atrick]

The strip utilities are not good abstractions for a high-level IR like SIL. I also did not want to modify them when adding the access markers. But not handling access markers has been blocking progress recently, and migrating passes away from the strip functions will take more time.

We now have an AccessUseDefVisitor that more properly abstracts over address and pointer operations, allowing passes to override some of the behavior without copying a giant switch statement everywhere. I'll demonstrate this in some upcoming PRs... And for every use-def abstraction, we need a matching def-use abstraction.

More importantly, an abstraction is useful if we can use it impose some order on SIL, defining which patterns are valid SIL and enforcing that in the verifier. In some situations, we can't simply bail-out on unrecognized patterns, so we need an abstraction that's guaranteed to work everywhere.

[eeckstein]

Don't get me wrong: I think this PR is fine. I just wanted to answer the question why end_cow_mutation is not handled in stripCasts.

[atrick]

@swift-ci test

[atrick]

@swift-ci benchmark

[atrick]

@swift-ci test source compatibility

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 3766962

[swift-ci]

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
CharacterLiteralsLarge	63	75	+19.0%	0.84x
ObjectiveCBridgeStubFromNSString	505	579	+14.7%	0.87x (?)
PrefixArrayLazy	13	14	+7.7%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
Set.isSuperset.Seq.Empty.Int	50	46	-8.0%	1.09x (?)

Code size: -O

Improvement	OLD	NEW	DELTA	RATIO
ObserverClosure.o	2446	2398	-2.0%	1.02x
ObserverPartiallyAppliedMethod.o	2509	2461	-1.9%	1.02x
Prims.o	12763	12635	-1.0%	1.01x
PrimsSplit.o	12815	12687	-1.0%	1.01x

Performance: -Osize

Improvement	OLD	NEW	DELTA	RATIO
MapReduceLazyCollectionShort	50	30	-40.0%	1.67x
FlattenListFlatMap	4410	2832	-35.8%	1.56x (?)
MapReduceString	40	35	-12.5%	1.14x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
ObjectiveCBridgeStubToNSDateRef	2540	2780	+9.4%	0.91x (?)
ObjectiveCBridgeStubToNSDate2	390	420	+7.7%	0.93x (?)

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

[atrick]

@swift-ci test macos platform