Support for @noescape SILFunctionTypes.
These are the underlying SIL changes necessary to implement the new
closure capture ABI.
Note: This includes a change to function name mangling that
primarily affects reabstraction thunks.
The new ABI will allow stack allocation of non-escaping closures as a
simple optimization.
The new ABI, and the stack allocation optimization, also require
closure context to be @guaranteed. That will be implemented as the
next step.
Many SIL passes pattern match partial_apply sequences. These all
needed to be fixed to handle the convert_function that SILGen now
emits. The conversion is now needed whenever a function declaration,
which has an escaping type, is passed into a @NoEscape argument.
In addition to supporting new SIL patterns, some optimizations like
inlining and SIL combine are now stronger which could perturb some
benchmark results.
These underlying SIL changes should be merged now to avoid conflicting
with other work. Minor benchmark discrepancies can be investigated as part of
the stack-allocation work.
* Add a noescape attribute to SILFunctionType.
And set this attribute correctly when lowering formal function types to SILFunctionTypes based on @escaping.
This will allow stack allocation of closures, and unblock a related ABI change.
* Flip the polarity on @noescape on SILFunctionType and clarify that
we don't default it.
* Emit withoutActuallyEscaping using a convert_function instruction.
It might be better to use a specialized instruction here, but I'll leave that up to Andy.
Andy: And I'll leave that to Arnold who is implementing SIL support for guaranteed ownership of thick function types.
* Fix SILGen and SIL Parsing.
* Fix the LoadableByAddress pass.
* Fix ClosureSpecializer.
* Fix performance inliner constant propagation.
* Fix the PartialApplyCombiner.
* Adjust SILFunctionType for thunks.
* Add mangling for @noescape/@escaping.
* Fix test cases for @noescape attribute, mangling, convert_function, etc.
* Fix exclusivity test cases.
* Fix AccessEnforcement.
* Fix SILCombine of convert_function -> apply.
* Fix ObjC bridging thunks.
* Various MandatoryInlining fixes.
* Fix SILCombine optimizeApplyOfConvertFunction.
* Fix more test cases after merging (again).
* Fix ClosureSpecializer. Hande convert_function cloning.
Be conservative when combining convert_function. Most of our code doesn't know
how to deal with function type mismatches yet.
* Fix MandatoryInlining.
Be conservative with function conversion. The inliner does not yet know how to
cast arguments or convert between throwing forms.
* Fix PartialApplyCombiner.
This is a loop that attempts to find the index of something with the longest
prefix, but wasn't actually saving the new longest length when finding something
with a longer length, meaning the loop always chose the last index. The update was
accidentally lost in f40e9ef0ae.
With the option -sil-print-debuginfo the printing of debug locations and scopes can be enabled.
I made the default for the option “false”, because in 99% of the time I don’t need the debug info in the printed SIL and I prefer better readability.
This commit does a few things:
1. It uses SwitchEnumBuilder so we are not re-inventing any wheels.
2. Instead of hacking around not putting in a destroy for .None on the fail
pass, just *do the right thing* and recognize that we have a binary case enum
and in such a case, just emit code for the other case rather than use a default
case (meaning no cleanup on .none).
rdar://31145255
Instead of appending a character for each substitution, we now prefix the substitution with the repeat count, e.g.
AbbbbB -> A5B
The same is done for known-type substitutions, e.g.
SiSiSi -> S3i
This significantly shrinks mangled names which contain large lists of the same type, like
func foo(_ x: (Int, Int, Int, Int, Int, Int, Int, Int, Int, Int, Int, Int))
rdar://problem/30707433
Use a syntax that declares the layout's generic parameters and fields,
followed by the generic arguments to apply to the layout:
{ var Int, let String } // A concrete box layout with a mutable Int
// and immutable String field
<T, U> { var T, let U } <Int, String> // A generic box layout,
// applied to Int and String
// arguments
Keep in mind that these are approximations that will not impact correctness
since in all cases I ensured that the SIL will be the same after the
OwnershipModelEliminator has run. The cases that I was unsure of I commented
with SEMANTIC ARC TODO. Once we have the verifier any confusion that may have
occurred here will be dealt with.
rdar://28685236
This ensures that ownership is properly propagated forward through the use-def
graph.
This was the work that was stymied by issues relating to SILBuilder performing
local ARC dataflow. I ripped out that local dataflow in 6f4e2ab and added a
cheap ARC guaranteed dataflow pass that performs the same optimization.
Also in the process of doing this work, I found that there were many SILGen
tests that were either pattern matching in the wrong functions or had wrong
CHECK lines (for instance CHECK_NEXT). I fixed all of these issues and also
expanded many of the tests so that they verify ownership. The only work I left
for a future PR is that there are certain places in tests where we are using the
projection from an original value, instead of a copy. I marked those with a
message SEMANTIC ARC TODO so that they are easy to find.
rdar://28685236
This flips the switch to have @noescape be the default semantics for
function types in argument positions, for everything except property
setters. Property setters are naturally escaping, so they keep their
escaping-by-default behavior.
Adds contentual printing, and updates the test cases.
There is some further (non-source-breaking) work to be done for
SE-0103:
- We need the withoutActuallyEscaping function
- Improve diagnostics and QoI to at least @noescape's standards
- Deprecate / drop @noescape, right now we allow it
- Update internal code completion printing to be contextual
- Add more tests to explore tricky corner cases
- Small regressions in fixits in attr/attr_availability.swift
Similarly to how we've always handled parameter types, we
now recursively expand tuples in result types and separately
determine a result convention for each result.
The most important code-generation change here is that
indirect results are now returned separately from each
other and from any direct results. It is generally far
better, when receiving an indirect result, to receive it
as an independent result; the caller is much more likely
to be able to directly receive the result in the address
they want to initialize, rather than having to receive it
in temporary memory and then copy parts of it into the
target.
The most important conceptual change here that clients and
producers of SIL must be aware of is the new distinction
between a SILFunctionType's *parameters* and its *argument
list*. The former is just the formal parameters, derived
purely from the parameter types of the original function;
indirect results are no longer in this list. The latter
includes the indirect result arguments; as always, all
the indirect results strictly precede the parameters.
Apply instructions and entry block arguments follow the
argument list, not the parameter list.
A relatively minor change is that there can now be multiple
direct results, each with its own result convention.
This is a minor change because I've chosen to leave
return instructions as taking a single operand and
apply instructions as producing a single result; when
the type describes multiple results, they are implicitly
bound up in a tuple. It might make sense to split these
up and allow e.g. return instructions to take a list
of operands; however, it's not clear what to do on the
caller side, and this would be a major change that can
be separated out from this already over-large patch.
Unsurprisingly, the most invasive changes here are in
SILGen; this requires substantial reworking of both call
emission and reabstraction. It also proved important
to switch several SILGen operations over to work with
RValue instead of ManagedValue, since otherwise they
would be forced to spuriously "implode" buffers.
And use project_box to get to the address value.
SILGen now generates a project_box for each alloc_box.
And IRGen re-uses the address value from the alloc_box if the operand of project_box is an alloc_box.
This lets the generated code be the same as before.
Other than that most changes of this (quite large) commit are straightforward.
Having a separate address and container value returned from alloc_stack is not really needed in SIL.
Even if they differ we have both addresses available during IRGen, because a dealloc_stack is always dominated by the corresponding alloc_stack in the same function.
Although this commit quite large, most changes are trivial. The largest non-trivial change is in IRGenSIL.
This commit is a NFC regarding the generated code. Even the generated SIL is the same (except removed #0, #1 and @local_storage).
Sema models enum case constructors as ApplyExprs. Formerly SILGen
would emit a case constructor function for each enum case,
constructing the enum value in the constructor body. ApplyExprs
of case constructors were lowered like any other call.
This is nice and straightforward but has several downsides:
1) Case constructor functions are very repetitive and trivial,
in particular for no-payload cases. They were declared
@_transparent and so were inlined at call sites, but for
public enums they still had to be emitted into the final
object file.
2) If the enum is generic, the substituted type may be loadable
even if the unsubstituted type is not, but since the case
constructor is polymorphic we had to allocate stack buffers
anyway, to pass the payload and result at the right abstration
level. This meant that for example Optional.Some(foo)
generated less-efficient SIL than the equivalent implicit
conversion.
3) We were missing out on peephole optimization opportunities when
the payload of an indirect case or address-only enum could be
emitted directly into the destination buffer, avoiding a copy.
One example would be when an enum payload is the result of
calling a function that returns an address-only value indirectly.
It appears we had unnecessary copies and takes even with -O.
Again, optional implicit conversions special-cased this.
This patch implements a new approach where a fully-formed call to
a element constructor is handled via a special code path where
the 'enum' or 'init_enum_data_addr' / 'inject_enum_addr'
instructions are emitted directly. These always work on the
substituted type, avoiding stack allocations unless needed.
An additional optimization is that the ArgumentSource abstraction
is used to delay evaluation of the payload argument until the
indirect box or address-only payload was set up.
If a element constructor is partially applied, we still emit a
reference to the constant as before.
It may seem like case constructor functions are at least useful
for resilience, but case constructors are transparent, so making
them resilient would require a new "transparent but only in this
module, and don't serialize the SIL body" declaration.
@inline(always) is almost what we need here, but this affect
mandatory inlining, only the optimizer, so it would be a
regression for non-resilient enums, or usages of resilient enums
in the current module.
A better approach is to construct resilient enums with a new
destructiveInjectEnumTag value witness function, which is
coming soon, and the general improvement from that approach
is what prompted this patch.
And include some supplementary mangling changes:
- Give the first generic param (depth=0, index=0) a single character mangling. Even after removing the self type from method declaration types, 'Self' still shows up very frequently in protocol requirement signatures.
- Fix the mangling of generic parameter counts to elide the count when there's only one parameter at the starting depth of the mangling.
Together these carve another 154KB out of a debug standard library. There's some awkwardness in demangled strings that I'll clean up in subsequent commits; since decl types now only mangle the number of generic params at their own depth, it's context-dependent what depths those represent, which we get wrong now. Currying markers are also wrong, but since free function currying is going away, we can mangle the partial application thunks in different ways.
Swift SVN r32896
And fix some bugs in the switch implementation I ran into on the way:
- Make getManagedSubobject(CopyOnSuccess) really produce a CopyOnSuccess ConsumableManagedValue;
- Avoid invalidating address-only enums when they can't be unconditionally taken by copying the enum before projecting it. Ideally there'd be a copy_enum_data_addr instruction to do this more efficiently.
Swift SVN r29817
