Previously we were:
1. Doing a linear scan, performing certain optimizations, and setting up a
worklist for future processing.
2. Draining the worklist of changed instructions until we reached a fix point.
The key thing here is that for (1) to be safe, we would need to not perform any
optimizations on block arguments since there was an ssumption that we would only
perform SSA-like optimizations.
I am going to be adding such optimizations soon, so it makes sense to just
convert the initial traversal to non-destructively seed the worklist and
eliminate that initial optimization pass.
This should be NFC.
This change could impact Swift programs that previously appeared
well-behaved, but weren't fully tested in debug mode. Now, when running
in release mode, they may trap with the message "error: overlapping
accesses...".
Recent optimizations have brought performance where I think it needs
to be for adoption. More optimizations are planned, and some
benchmarks should be further improved, but at this point we're ready
to begin receiving bug reports. That will help prioritize the
remaining work for Swift 5.
Of the 656 public microbenchmarks in the Swift repository, there are
still several regressions larger than 10%:
TEST OLD NEW DELTA RATIO
ClassArrayGetter2 139 1307 +840.3% **0.11x**
HashTest 631 1233 +95.4% **0.51x**
NopDeinit 21269 32389 +52.3% **0.66x**
Hanoi 1478 2166 +46.5% **0.68x**
Calculator 127 158 +24.4% **0.80x**
Dictionary3OfObjects 391 455 +16.4% **0.86x**
CSVParsingAltIndices2 526 604 +14.8% **0.87x**
Prims 549 626 +14.0% **0.88x**
CSVParsingAlt2 1252 1411 +12.7% **0.89x**
Dictionary4OfObjects 206 232 +12.6% **0.89x**
ArrayInClass 46 51 +10.9% **0.90x**
The common pattern in these benchmarks is to define an array of data
as a class property and to repeatedly access that array through the
class reference. Each of those class property accesses now incurs a
runtime call. Naturally, introducing a runtime call in a loop that
otherwise does almost no work incurs substantial overhead. This is
similar to the issue caused by automatic reference counting. In some
cases, more sophistacated optimization will be able to determine the
same object is repeatedly accessed. Furthermore, the overhead of the
runtime call itself can be improved. But regardless of how well we
optimize, there will always a class of microbenchmarks in which the
runtime check has a noticeable impact.
As a general guideline, avoid performing class property access within
the most performance critical loops, particularly on different objects
in each loop iteration. If that isn't possible, it may help if the
visibility of those class properties is private or internal.
While trying to emit implicit load expression make sure that it's
done in a way where force/paren expression is always top level,
that leads to better fix-its and, in case of forcing, more compact
SIL by intermediate optional container.
Resolves: [SR-8150](https://bugs.swift.org/browse/SR-8150) / rdar://problem/41725207
Otherwise, when we mangle a signature of bridged/not_bridged arguments
we can see the state from the previous match of another instruction.
SR-7426
rdar://39414272
Stop creating ImplicitlyUnwrappedOptional<T> so that we can remove it
from the type system.
Enable the code that generates disjunctions for Optional<T> and
rewrites expressions based on the original declared type being 'T!'.
Most of the changes supporting this were previously merged to master,
but some things were difficult to merge to master without actually
removing IUOs from the type system:
- Dynamic member lookup and dynamic subscripting
- Changes to ensure the bridging peephole still works
Past commits have attempted to retain as much fidelity with how we
were printing things as possible. There are some cases where we still
are not printing things the same way:
- In diagnostics we will print '?' rather than '!'
- Some SourceKit and Code Completion output where we print a Type
rather than Decl.
Things like module printing via swift-ide-test attempt to print '!'
any place that we now have Optional types that were declared as IUOs.
There are some diagnostics regressions related to the fact that we can
no longer "look through" IUOs. For the same reason some output and
functionality changes in Code Completion. I have an idea of how we can
restore these, and have opened a bug to investigate doing so.
There are some small source compatibility breaks that result from
this change:
- Results of dynamic lookup that are themselves declared IUO can in
rare circumstances be inferred differently. This shows up in
test/ClangImporter/objc_parse.swift, where we have
var optStr = obj.nsstringProperty
Rather than inferring optStr to be 'String!?', we now infer this to
be 'String??', which is in line with the expectations of SE-0054.
The fact that we were only inferring the outermost IUO to be an
Optional in Swift 4 was a result of the incomplete implementation of
SE-0054 as opposed to a particular design. This should rarely cause
problems since in the common-case of actually using the property rather
than just assigning it to a value with inferred type, we will behave
the same way.
- Overloading functions with inout parameters strictly by a difference
in optionality (i.e. Optional<T> vs. ImplicitlyUnwrappedOptional<T>)
will result in an error rather than the diagnostic that was added
in Swift 4.1.
- Any place where '!' was being used where it wasn't supposed to be
allowed by SE-0054 will now treat the '!' as if it were '?'.
Swift 4.1 generates warnings for these saying that putting '!'
in that location is deprecated. These locations include for example
typealiases or any place where '!' is nested in another type like
`Int!?` or `[Int!]`.
This commit effectively means ImplicitlyUnwrappedOptional<T> is no
longer part of the type system, although I haven't actually removed
all of the code dealing with it yet.
ImplicitlyUnwrappedOptional<T> is is dead, long live implicitly
unwrapped Optional<T>!
Resolves rdar://problem/33272674.
This replaces the '[volatile]' flag. Now, class_method and
super_method are only used for vtable dispatch.
The witness_method instruction is still overloaded for use
with both ObjC protocol requirements and Swift protocol
requirements; the next step is to make it only mean the
latter, also using objc_method for ObjC protocol calls.
