DeadCodeElimination stopped deleting dead copy_value and hoisting destroy_value
instructions.
LifetimeCompletion of valuesToComplete was necessary since DCE could delete blocks
that contain only destroy_value instructions or forwarding instructions with
destroy_value uses only.
Since this is not done in DeadCodeElimination, remove this support.
This new OSSA invariant simplifies many optimizations because they don't have to take care of the corner case of incomplete lifetimes in dead-end blocks.
The implementation basically consists of these changes:
* add the lifetime completion utility
* add a flag in SILFunction which tells optimization that they need to run the lifetime completion utility
* let all optimizations complete lifetimes if necessary
* enable the ownership verifier to check complete lifetimes
These two new invariants eliminate corner cases which caused bugs if optimization didn't handle them.
Also, it will significantly simplify lifetime completion.
The implementation basically consists of these changes:
* add a flag in SILFunction which tells optimization if they need to take care of infinite loops
* add a utility to break infinite loops
* let all optimizations remove unreachable blocks and break infinite loops if necessary
* add verification to check the new SIL invariants
The new `breakIfniniteLoops` utility breaks infinite loops in the control flow by inserting an "artificial" loop exit to a new dead-end block with an `unreachable`.
It inserts a `cond_br` with a `builtin "infinite_loop_true_condition"`:
```
bb0:
br bb1
bb1:
br bb1 // back-end branch
```
->
```
bb0:
br bb1
bb1:
%1 = builtin "infinite_loop_true_condition"() // always true, but the compiler doesn't know
cond_br %1, bb2, bb3
bb2: // new back-end block
br bb1
bb3: // new dead-end block
unreachable
```
When an owned value has no lifetime ending uses it means that it is in a dead-end region.
We must not remove and inserting compensating destroys for it because that would potentially destroy the value too early.
Initialization of an object might be cut off and removed after a dead-end loop or an `unreachable`.
In this case a class destructor would see uninitialized fields.
Fixes a mis-compile
https://github.com/swiftlang/swift/issues/85851
rdar://165876726
DeadCodeElimination can remove instructions including their destroys and then insert compensating destroys at a new place.
This is effectively destroy-hoisting which doesn't respect deinit-barriers.
Disable removing and re-creating `destroy_value` instructions. This is done by other optimizations.
Empty access scopes can be a result of e.g. redundant-load-elimination.
It's still important to keep those access scopes to detect access violations.
Even if the load is physically not done anymore, in case of a conflicting access a propagated load is still wrong and must be detected.
rdar://164571252
When DCE deletes instructions as dead, if the instruction ends one of
its operands lifetimes, it must insert a compensating lifetime end.
When the def block of the value and the parent block of the instruction
are different, it uses lifetime completion. Lifetime completion relies
on complete liveness, which doesn't and can't exist for values with
pointer escapes. The result is ending lifetimes too early.
Avoid this scenario by marking such instructions live.
In the fullness of time, it may be possible to track the deleted
instruction's "location" even in the face of deletions of adjacent
instructions and parent blocks and to insert the lifetime end at that
location.
rdar://149007151
DCE may enqueue a value for lifetime completion and later on erase the
instruction that defines that value. When erasing an instruction, erase
each of its results from the collection of values to complete.
rdar://141560546
DCE deletes ownership forwarding instructions when it doesn’t have useful users.
It inserts destroy_value/end_borrow for its operands to compensate their lifetimes.
DCE also deletes branches when its successor blocks does not have useful instructions.
It deletes blocks and creates a jump to the nearest post dominating block.
When DCE needs to delete a forwarding instruction in a dead block, it cannot just create
lifetime ends of its operands at its position. Use LifetimeCompletion utility in such cases.
rdar://140428721
DCE inserts end_borrow at phi operands when a guaranteed phi becomes dead.
This should be done for reborrows which end the lifetime of the incoming value.
The existing check was not accurate and ended up inserting end_borrow for forwarded values as well.
Fixes rdar://139283745
Although I don't plan to bring over new assertions wholesale
into the current qualification branch, it's entirely possible
that various minor changes in main will use the new assertions;
having this basic support in the release branch will simplify that.
(This is why I'm adding the includes as a separate pass from
rewriting the individual assertions)
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.
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
In cea0f00598, `InstructionDeleter` began
deleting `load [take]` instructions. Analogous to how it creates a
`destroy_value` when deleting an instruction which consumes a value, in
the case of deleting a `load [take]` the `InstructionDeleter` inserts a
compensating `destroy_addr`.
Previously, `DeadCodeElimination` did not observe the creation of any
instructions created by the `InstructionDeleter`. In the case of the
newly created `destroy_addr`, DCE didn't mark that the `destroy_addr`
was live and so deleted it. The result was a leak.
Here, this is fixed by passing an `InstModCallbacks`--with an
`onCreateNewInst` implementation--down into `erasePhiArgument` that
eventually invokes the `InstructionDeleter`. When the
`InstructionDeleter` creates a new instruction, DCE marks it live.
For very large control flow graphs the markControllingTerminatorsLive can stack overflow.
Fix this by doing the work iteratively instead of recursively.
rdar://106198943
