All SILArgument types are "block arguments". There are three kinds:
1. Function arguments
2. Phis
3. Terminator results
In every situation where the source of the block argument matters, we
need to distinguish between these three. Accidentally failing to
handle one of the cases is an perpetual source of compiler
bugs. Attempting to handle both phis and terminator results uniformly
is *always* a bug, especially once OSSA has phi flags. Even when all
cases are handled correctly, the code that deals with data flow across
blocks is incomprehensible without giving each case a type. This
continues to be a massive waste of time literally every time I review
code that involves cross-block control flow.
Unfortunately, we don't have these C++ types yet (nothing big is
blocking that, it just wasn't done). That's manageable because we can
use wrapper types on the Swift side for now. Wrapper types don't
create any more complexity than protocols, but they do sacrifice some
usability in switch cases.
There is no reason for a BlockArgument type. First, a function
argument is a block argument just as much as any other. BlockArgument
provides no useful information beyond Argument. And it is nearly
always a mistake to care about whether a value is a function argument
and not care whether it is a phi or terminator result.
This instructions marks the point where all let-fields of a class are initialized.
This is important to ensure the correctness of ``ref_element_addr [immutable]`` for let-fields,
because in the initializer of a class, its let-fields are not immutable, yet.
Codegen is the same, but `begin_dealloc_ref` consumes the operand and produces a new SSA value.
This cleanly splits the liferange to the region before and within the destructor of a class.
We inline a function (e.g. a struct initializer) into a global init function if the result is part of the initialized global.
Now, also handle functions with indirect return values. Such function can result from not-reabstracted generic specializations.
Handle cases where the result is stored into a temporary alloc_stack or directly stored to (a part) of the global variable.
A begin_apply can yield multiple addresses. We need to store the result of the apply in order to distinguish between two AccessBases with different results from the same begin_apply.
And simplify it.
This struct is not really needed by clients. It's just needed internally in 'Value.accessPath` (and similar properties) to compute the access path.
A projected value consists of the original value and a projection path.
For example, if the `value` is of type `struct S { var x: Int }` and `path` is `s0`, then the projected value represents field `x` of the original value.
Also, use ProjectedValue instead of AccessStoragePath.
It doesn't make sense to let getAccessPathWithScope return an `EnclosingScope` as the second tuple element, because in case it's a `base`, it duplicates the `AccessBase` (which is returned in the first tuple element).
Instead just return an optional `BeginAccessInst` which is not nil if such an "scope" is found.
Now that `AccessBase` is an enum, it makes sense to add an `unidentified` case. This avoids dealing with optional AccessBases in several place.
Clients don't need to make both, an optional check and a switch, but can check for unidentified access bases just in a single switch statement.
While I was using the new AccessUtils for a new optimization pass I discovered some areas for improvements. Also I found some bugs.
Changes:
* AccessBase: remove the unhealthy redundancy between `kind` and `baseAddress` types. Now AccessBase is single enum with the relevant base objects/addresses as payloads.
* AccessBase: for `global`, store the `GlobalValue` and not a `global_address` instruction, which is more accurate (because there can be multiple `global_addr`s for a single global variable)
* AccessBase: drop the support for function argument "pointers". The `pointer` is now always a `pointer_to_address` instruction. This also simplifies `PointerIdentification`: either it finds a matching `address_to_pointer` or it bails.
* AccessBase: improve `func isDistinct(from:)`. There are more possibilities to prove that two access bases do not alias.
* AccessBase: replace `var isUniquelyIdentified` with `var hasKnownStorageKind` which is more useful for aliasing checking.
* AccessPath: fix `func isDistinct(from:)`. `SmallProjectionPath.matches` is the wrong way to check if two expression paths may overlap. Instead use the new `SmallProjectionPath.mayOverlap`.
* AccessStoragePathWalker: rename `getAccessStorage` -> `visitAccessStorageRoots` and let it return false if it's not a class/reference AccessBase.
* add tests for `AccessPath.isDistinct(from:)`
