In analogy to the ownership verifier, the MemoryLifetimeVerifier checks the lifetime of memory locations.
It's based on MemoryLocations and MemoryDataflow, which are general utilities for analysing memory locations and calculating the lifetime of locations.
Memory locations are limited to addresses which are guaranteed to be not aliased, like @in/inout parameters or alloc_stack.
This flag is set by DefinitInitialization if the lifetime of the stored value is controlled dynamically.
If the flag is set, it's not (easily) possibly to statically calculate the lifetime of the stored value.
Ownership is truly a property not of a declaration, but of a function body. So
it makes sense to just match what we deserialize.
This also helps us to avoid mismatches if we lower ownership from a function,
delete it, and then relink it.
I also used this as an opportunity to clean up where we set that flag in
deserialization to only happen in one place for both declarations/definitions.
This indicates that the "self" argument to the current function is always dynamically of the exact
static base class type, allowing metadata accesses in IRGen to use the local self metadata to answer
metadata requests for the class type. Set this attribute on allocating entry points of designated
inits, which is one of the most common places where we emit redundant metadata accesses.
Now we just always mark functions as having or not having ownership. So we know
when parsing/deserializing what state we are in already. There are also
assertions in SILBuilder around most of the instructions that can exist in only
a single mode, meaning even the error was sort of redundant.
NOTE:
1. To test this I changed UnaryOp_match to use this under the hood.
2. These types of m_##ID##Inst matchers now will only accept compound types and
I added a static assert to verify that this mistake doesn't happen. We
previously had matchers that would take an int or the like to match tuple
extract patterns. I converted those to use TupleExtractOperation that also
properly handles destructures.
This improves on the previous situation:
- The request ensures that the backing storage for lazy properties
and property wrappers gets synthesized first; previously it was
only somewhat guaranteed by callers.
- Instead of returning a range this just returns an ArrayRef,
which simplifies clients.
- Indexing into the ArrayRef is O(1), which addresses some FIXMEs
in the SIL optimizer.
This patch comes out of my reading some generic code using .none in transparent
functions to conditionally compile out code at -Onone. Sadly, before this the
dead code in question wouldn't be compiled out unless the protocol was
constrained to be a class protocol.
I added a test that validates that this conditional compilation property can be
relied on in -Onone code in both cases.
This can be really useful when using conditional break points in
TypeLowering.
NOTE: Certain SILTypes do not have manglings and this will assert! We should
change the mangling infrastructure to just return Optional<None> or false
instead.
Specifically:
1. I removed an extra defensive copy that we put in place some time ago that
isn't really warranted. We know that we have an @owned value, so can safely just
pass the value as a @guaranteed parameter. This also eliminates an ownership
error that would occur due to my not having updated this code for ownership in
tree.
2. I also ensured that if we are performing a loadable address bridging cast ->
value bridging cast that we store the loadable value back into memory after we
perform the cast. Otherwise, it appears to leak to the ownership verifier.
I also centralized the non-ownership tests for this into one place
(const_fold_objc_bridge.sil => constant_propagation_objc.sil).
Specifically, when we optimize conversions such as:
Optional<@escaping () -> ()>
->
Optional<@noescape () -> ()>
->
Optional<@noescape @convention(block) () -> ()>
previously we were lifetime extending over the @noescape lifetime barrier by
making a copy and then putting a mark_dependence from the copy onto the original
value. This was just a quick way to tell the ownership verifier that the copy
was tied to the other value and thus should not be eliminated. The correctness
of the actual lifetime extension comes from the optimizer being conservative
around rr insts.
This commit instead changes our optimization to borrow the copied optional
value, extract the payload, and use that instead.
Assign separate SILProfiler instances to stored property initializers
and constructors.
Starting with rdar://39460313, coverage reporting for these constructs
was bundled up into a single SILProfiler uniqued by the NominalTypeDecl.
There are two problems with doing this.
First, the shared SILProfiler is given a fake name that can't be
demangled. That breaks Xcode's reports. Second, the relationship
between SILProfiler and SILFunction is supposed to be 1:1. Having a
shared SILProfiler muddies things a bit and requires extra bookkeeping.
rdar://47467864
Add `llvm_unreachable` to mark covered switches which MSVC does not
analyze correctly and believes that there exists a path through the
function without a return value.
If an override B.f() is more visible than a base method A.f(), it is
possible that an override C.f() of B.f() cannot see the original method
A.f().
In this case, we would encounter linker errors if we referenced the
method descriptor or method dispatch thunk for A.f().
Make this work by treating B.f() as the least derived method in this
case, and ensuring that the vtable thunk for B.f() dispatches through
the vtable again.
Fixes <rdar://problem/48330571>, <https://bugs.swift.org/browse/SR-10648>.
With the advent of dynamic_function_ref the actual callee of such a ref
my vary. Optimizations should not assume to know the content of a
function referenced by dynamic_function_ref. Introduce
getReferencedFunctionOrNull which will return null for such function
refs. And getInitialReferencedFunction to return the referenced
function.
Use as appropriate.
rdar://50959798
