The copy operator has been implemented and doesn't use it. Remove
`Builtin.copy` and `_copy` as much as currently possible.
Source compatibility requires that `_copy` remain in the stdlib. It is
deprecated here and just uses the copy operator.
Handling old swiftinterfaces requires that `Builtin.copy` be defined.
Redefine it here as a passthrough--SILGen machinery will produce the
necessary copy_addr.
rdar://127502242
Changes in this CR add part of the, Swift based, Autodiff specific
closure specialization optimization pass. The pass does not modify any
code nor does it even exist in any of the optimization pipelines. The
rationale for pushing this partially complete optimization pass upstream
is to keep up with the breaking changes in the underlying Swift based
compiler infrastructure.
Compute, update and handle borrowed-from instruction in various utilities and passes.
Also, used borrowed-from to simplify `gatherBorrowIntroducers` and `gatherEnclosingValues`.
Replace those utilities by `Value.getBorrowIntroducers` and `Value.getEnclosingValues`, which return a lazily computed Sequence of borrowed/enclosing values.
Under OSSA, the instruction may still be structurally responsible for consuming
its operand even if the result is dead, so we can't remove it without breaking
invariants.
More generally, this should probably apply to any instruction which consumes
one or more of its operands, has no side effects, and doesn't produce any
nontrivial results that require further consumption to keep the value alive.
I went with this targeted fix, since it addresses a problem that shows up
in practice (rdar://125381446) and the more general change appears to
disturb the optimizer pipeline while building the standard library.
Don't treat StoreBorrow addresses as unknown bases. While they are never the base of a formal access, they are returned
as the AccessBase when querying the enclosing scope of an address.
Treat mark_dependence [nonescaping] as a dependent value even if the dependence base does not have a recognizable
scope (e.g. a multiply-defined alloc_stack). This happens because ClosureLifetimeFixup creates redundant mark_dependence
instructions for partial_apply captures. We constantly need to work around this broken representation of nonescaping closures.
Add PartialApplyInst.hasNoescapeCapture
Add PartialApplyInst.mayEscape
Refactor DiagnoseInvalidEscapingCaptures. This may change functionality because tuples containing a noescape closure are now correctly recognized. Although I'm not sure such tupes can ever be captured directly.
SILBoxTypes have their own generic signature and substitution
map. This means that every time we query isEscapable or mayEscape, we
need to extract the type of the box's field and perform type
substitution so that the AST query only sees types from the function's
generic environment.
Fixes rdar://124179106 (Assertion failed in SIL:
(!type->hasTypeParameter() && "caller forgot to mapTypeIntoContext!"))
A live range representing the ownership of addressible memory.
This live range represents the minimal guaranteed lifetime of the object being addressed. Uses of derived addresses
may be extended up to the ends of this scope without violating ownership.
.liveOut objects (@in_guaranteed, @out and globals) have no instruction range.
.local objects (alloc_stack, yield, @in, @inout) report the single live range of the full assignment that reaches
this address.
.owned values (boxes and references) simply report OSSA liveness.
.borrow values report each borrow scope's range. The effective live range is their intersection. A valid use must
lie within
Compute the live range for the borrow scopes of a guaranteed value. This returns a separate instruction range for
each of the value's borrow introducers. Unioning those ranges would be incorrect. We typically want their
intersection.
This avoids a lot of confusion because the callers expect this type. Fixing it just required some redundancy and
bridging in the EnclosigValues implementation.
For years, optimizer engineers have been hitting a common bug caused by passes
assuming all SILValues have a parent function only to be surprised by SILUndef.
Generally we see SILUndef not that often so we see this come up later in
testing. This patch eliminates that problem by making SILUndef uniqued at the
function level instead of the module level. This ensures that it makes sense for
SILUndef to have a parent function, eliminating this possibility since we can
define an API to get its parent function.
rdar://123484595
