IRGen: don't create ObjC methods with the unnamed_addr attribute

[eeckstein]

In some places, Foundation is comparing ObjC method pointers.
Therefore LLVM's function merging pass must not create aliases for identica; functions, but instead create thunks.
This can be ensured by not creating ObjC methods with the unnamed_addr attribute.

rdar://problem/58483698

[eeckstein]

@swift-ci test

[eeckstein]

@swift-ci benchmark

[swift-ci]

Performance: -O

Regression	OLD	NEW	DELTA	RATIO
EqualSubstringSubstring	21	28	+33.3%	0.75x (?)
EqualSubstringSubstringGenericEquatable	21	28	+33.3%	0.75x (?)
EqualSubstringString	21	28	+33.3%	0.75x (?)
LessSubstringSubstringGenericComparable	21	28	+33.3%	0.75x (?)
LessSubstringSubstring	21	27	+28.6%	0.78x
EqualStringSubstring	22	28	+27.3%	0.79x (?)
NSStringConversion.Long	523	608	+16.3%	0.86x (?)
StringComparison_longSharedPrefix	320	356	+11.2%	0.90x (?)
NSStringConversion.Medium	252	272	+7.9%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	124	106	-14.5%	1.17x (?)

Code size: -O

Performance: -Osize

Regression	OLD	NEW	DELTA	RATIO
LessSubstringSubstring	21	27	+28.6%	0.78x
EqualSubstringString	21	27	+28.6%	0.78x
LessSubstringSubstringGenericComparable	21	27	+28.6%	0.78x
EqualSubstringSubstring	22	28	+27.3%	0.79x (?)
EqualStringSubstring	22	28	+27.3%	0.79x (?)
EqualSubstringSubstringGenericEquatable	22	28	+27.3%	0.79x (?)
SuffixCountableRange	4	5	+25.0%	0.80x (?)
NSStringConversion.Long	493	570	+15.6%	0.86x
StringComparison_longSharedPrefix	317	351	+10.7%	0.90x (?)
IterateData	869	954	+9.8%	0.91x (?)
NSStringConversion.Medium	236	255	+8.1%	0.93x (?)
Improvement	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	126	104	-17.5%	1.21x
ObjectiveCBridgeStringHash	43	40	-7.0%	1.07x (?)

Code size: -Osize

Performance: -Onone

Regression	OLD	NEW	DELTA	RATIO
EqualSubstringSubstringGenericEquatable	25	31	+24.0%	0.81x
LessSubstringSubstringGenericComparable	25	31	+24.0%	0.81x
EqualSubstringSubstring	26	32	+23.1%	0.81x
LessSubstringSubstring	26	32	+23.1%	0.81x (?)
EqualStringSubstring	26	32	+23.1%	0.81x (?)
EqualSubstringString	26	32	+23.1%	0.81x
ArrayOfGenericPOD2	714	836	+17.1%	0.85x (?)
Improvement	OLD	NEW	DELTA	RATIO
UTF8Decode_InitDecoding	143	124	-13.3%	1.15x (?)
ObjectiveCBridgeStringHash	43	40	-7.0%	1.07x (?)
DataSubscriptMedium	58	54	-6.9%	1.07x (?)

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

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 4fc55830bcd4d4a8f86151d1176f90b9bcbe81e1

[eeckstein]

@swift-ci test macOS

[eeckstein]

@swift-ci test macOS

[aschwaighofer]

Maybe in a follow-up can you add a comment reminding us in the future to not do this again?

[compnerd]

Could you add:

CHECK-SAME-NOT: unnamed_addr

here and the other locations where it was removed?

[eeckstein]

Good point. I'll add a regexp at the end of the line so that unnamed_addr is not matched.

[jckarter]

This seems like an unfortunate blanket policy. Do we know what Foundation is using the imp pointers for? Is it for an isolated enough set of cases that we could special case them instead of eliminating this optimization for all ObjC entry points?

[eeckstein]

@jckarter It's used in Foundation to check if a method has an override. I'm not sure if there are more cases. Anyway, I think we doesn't harm to be on the safe side here. I don't think this attribute has a significant performance/code-size effect (for objc methods).

[eeckstein]

@swift-ci smoke test