To enable those clients which care only about _blocks_ whose
instructions or dead-defs make up the liveness boundary to avoid
iterating instruction lists, enhance boundary computation with a second
type (`PrunedLivenessBlockBoundary`) and migrate some methods from
liveness onto the boundary types to enable dispatching on the boundary
type to do only the amount of analysis that's necessary.
When checking whether an instruction is contained in a liveness
boundary, a pointer to a DeadEndBlocks instance must always be passed.
When the pointer is null, it is only checked that the instruction occurs
within the direct live region. When the pointer is non-null, it is
checked whether the instruction occurs within the region obtained by
extending the live region up to the availability boundary within
dead-end regions that are adjacent to the non-lifetime-ending portion of
the liveness boundary.
The areUsesWithinBoundary/areUsesOutsideBoundary methods take the
dead-ends because the region that they're checking for containment
within contains more than just (is a (non-strict) superset of) the
pruned live region. On the other hand, the region also contains _less_
than (is a (non-strict) subset of) the available region. Instructions
are in this extended region only if they're in a dead-end block which is
forwards reachable from the non-lifetime-ending liveness boundary.
And use it in lifetime extension/maximization.
The new member function differs from updateForUse in that it doesn't
overwrite the old value for lifetime ending associated with the
instruction (calling updateForUse with lifetimeEnding=false overwrites
the flag set by a previous call with lifetimeEnding=true because if an
instruction both consumes and doesn't consume a copy-extended value, the
value must be live after the instruction).
llvm::SmallSetVector changed semantics
(https://reviews.llvm.org/D152497) resulting in build failures in Swift.
The old semantics allowed usage of types that did not have an
`operator==` because `SmallDenseSet` uses `DenseSetInfo<T>::isEqual` to
determine equality. The new implementation switched to using
`std::find`, which internally uses `operator==`. This type is used
pretty frequently with `swift::Type`, which intentionally deletes
`operator==` as it is not the canonical type and therefore cannot be
compared in normal circumstances.
This patch adds a new type-alias to the Swift namespace that provides
the old semantic behavior for `SmallSetVector`. I've also gone through
and replaced usages of `llvm::SmallSetVector` with the
`Swift::SmallSetVector` in places where we're storing a type that
doesn't implement or explicitly deletes `operator==`. The changes to
`llvm::SmallSetVector` should improve compile-time performance, so I
left the `llvm::SmallSetVector` where possible.
This now supports liveness holes within blocks where previously we
expected a phi def.
PrunedLiveness is streamlined for SSA liveness where we expect the
defs and uses to be related in the SSA def-use graph.
MultiDefPrunedLiveness was added on top of SSAPrunedLivenes to handle
extended SSA liveness that occurs with phis, but are still connected
in the def-use graph. Recently MultiDefPrunedLiveness was repurposed
for liveness of addressible storage. This means that we can now have
uses before defs in the same block without a corresponding phi.
Fortunately, handling this case is a straightforward extension of
MultiDefPrunedLiveness that does not complicate or penalize the
streamlined SSA case.
PrunedLiveness only knows about the live blocks and uses.
The PrunedLiveRange subclass is now responsible for updating liveness
based on both the defs and uses. This is in preparation for handling
non-SSA liveness when uses occur before the first def.
This now supports liveness holes within blocks where previously we
expected a phi def.
PrunedLiveness is streamlined for SSA liveness where we expect the
defs and uses to be related in the SSA def-use graph.
MultiDefPrunedLiveness was added on top of SSAPrunedLivenes to handle
extended SSA liveness that occurs with phis, but are still connected
in the def-use graph. Recently MultiDefPrunedLiveness was repurposed
for liveness of addressible storage. This means that we can now have
uses before defs in the same block without a corresponding phi.
Fortunately, handling this case is a straightforward extension of
MultiDefPrunedLiveness that does not complicate or penalize the
streamlined SSA case.
PrunedLiveness only knows about the live blocks and uses.
The PrunedLiveRange subclass is now responsible for updating liveness
based on both the defs and uses. This is in preparation for handling
non-SSA liveness when uses occur before the first def.
Use BasicBlockBitfield to record per-block liveness state. This has
been the intention since BasicBlockBitfield was first introduced.
Remove the per-field bitfield from PrunedLiveBlocks. This
(re)specializes the data structure for scalar liveness and drastically
simplifies the implementation.
This utility is fundamental to all ownership utilities. It will be on
the critical path in many areas of the compiler, including at
-Onone. It needs to be minimal and as easy as possible for compiler
engineers to understand, investigate, and debug.
This is in preparation for fixing bugs related to multi-def liveness
as used by the move checker.
Specifically, previously if we emitted an error we just dumped all of the
consuming uses. Now instead for each consuming use that needs a copy, we perform
a search for a specific boundary use (consuming or non-consuming) that is
reachable from the former and emit a specialized error for it. Thus we emit for
the two consuming case the normal consumed twice error, and now for
non-consuming errors we emit the "use after consume" error.
For those who are unaware, CanonicalizeOSSALifetime::canonicalizeValueLifetime()
is really a high level driver routine for the functionality of
CanonicalizeOSSALifetime that computes liveness and then rewrites copies using
boundary information. This change introduces splits the implementation of
canonicalizeValueLifetime into two parts: a first part called computeLiveness
and a second part called rewriteLifetimes. Internally canonicalizeValueLifetime
still just calls these two methods.
The reason why I am doing this is that it lets the move only object checker use
the raw liveness information computed before the rewriting mucks with the
analysis information. This information is used by the checker to compute the raw
liveness boundary of a value and use that information to determine the list of
consuming uses not on the boundary, consuming uses on the boundary, and
non-consuming uses on the boundary. This is then used by later parts of the
checker to emit our errors.
Some additional benefits of doing this are:
1. I was able to eliminate callbacks in the rewriting stage of
CanonicalOSSALifetimes which previously gave the checker this information.
2. Previously the move checker did not have access to the non-consuming boundary
uses causing us to always fail appropriately, but sadly not emit a note showing
the non-consuming use. I am going to wire this up in a subsequent commit.
The other change to the implementation of the move checker that this caused is
that I needed to add an extra diagnostic check for instructions that consume the
value twice or consume the value and use the value. The reason why this must be
done is that liveness does not distinguish in between different operands on the
same instruction meaning such an error would be lost.
In preparation for adding OwnershipLiveness.
Rename Simple LiveRangeSummary to LiveRangeSummary.
Add initializeDefNode helpers to avoid confusion about the argument
type.
Add defBegin/defEnd iterators in MultiDefPrunedLiveness.
In a sense, liveness is wrong for dead guaranteed phis that aren't
enclosed in a separate borrow scope:
left: // live-within
br merge(%v)
right: // live-within
br merge(%v)
merge(%deadPhi) // dead
But this is the way that liveness layering works. PrunedLiveness
simply doesn't know which phis will "end" liveness. This will be the
job of OwnershipLiveness, which will have safe APIs for situations
where this weirdness matters.
I did this by doing the following:
1. I renamed computeUseBlockLiveness to computeScalarUseBlockLiveness and
changed it to take a specific bit that it is testing for.
2. I changed the logic that already existed in this code path that worked scalar
by scalar to use scalar logic rather than call the broken multi-bit at a time
code path.
3. We took advantage of resultingFoundLiveness now only returning the requested
bits instead of all bits. This ensures that we do not run
computeScalarUseBlockLiveness for those unneeded dead bits resulting in liveness
being inappropriately propagated into predecessors.
Previously, we would pan the array with isDead. The thinking is that this would
make it easier to bitmask directly against. In practice, this usage was a minor
use case and doing this lead to a bunch of logic mistakes in forthcoming code.
We can remove a ton of complexity by assuming that SSA uses never
occur before a def in the same block. This will hold true as long as
useless phis are only removed after all unreachable code is first
removed. Then we don't have the self-loop problem.
The SILVerifier already checks this invariant and I added an in-depth
comment in SimplifyCFG.
Computing simple liveness is distinct from computing transitive
liveness. But for safety and consistency, always handle the first
level of liveness transitively. This way, computeSimple can be used on
guaranteed values that need OSSA lifetime fixup.
Simple liveness just means that *inner* borrow and address scopes are
assumed to be complete.
This utility is still only used conservatively because OSSA lifetime
completion is not yet enabled. But, in the future, computeSimple can
be used in the common case.
The API for computing simple liveness now returns a
SimpleLiveRangeSummary. Callers need to decide how to handle reborrows
and pointer escapes. If either condition exists then the resulting
liveness does not necessarily encapsulate the definition's ownership.
Fixes some number of latent bugs w.r.t. liveness clients.
First restore the basic PrunedLiveness abstraction to its original
intention. Move code outside of the basic abstraction that polutes the
abstraction and is fundamentally wrong from the perspective of the
liveness abstraction.
Most clients need to reason about live ranges, including the def
points, not just liveness based on use points. Add a PrunedLiveRange
layer of types that understand where the live range is
defined. Knowing where the live range is defined (the kill set) helps
reliably check that arbitrary points are within the boundary. This
way, the client doesn't need to be manage this on its own. We can also
support holes in the live range for non-SSA liveness. This makes it
safe and correct for the way liveness is now being used. This layer
safety handles:
- multiple defs
- instructions that are both uses and defs
- dead values
- unreachable code
- self-loops
So it's no longer the client's responsibility to check these things!
Add SSAPrunedLiveness and MultiDefPrunedLiveness to safely handle each
situation.
Split code that I can't figure out into
DiagnosticPrunedLiveness. Hopefully it will be deleted soon.
Some notes:
1. I added support for both loadable/address only types.
2. These tests are based off of porting the move only object tests for inout,
vars, mutating self, etc.
3. I did not include already written tests for address only types in this
specific merge since I need to change us to borrow move only var like types.
Without that, we get a lot of spurious error msgs and the burden of writing that
is not worth it. So instead in a forthcoming commit where I fix that issue in
SILGen, I will commit the corresponding address only tests for this work.
4. I did not include support for trivial types in this. I am going to do
object/address for that at the same time.
This is the same algorithm as pruned liveness but assumes that one is tracking
liveness from an address and uses the same scheme as DI to track liveness as a
bit vector.
The reason why I am doing this is that in order to be able to use
PrunedLivenessBlocks with multiple elements, we can no longer rely on dead being
represented by a block not having any state, since a dead bit could have a
neighboring live bit.
That being said, the only place that this is used now is the current
PrunedLiveness implementation which only stores a single bit... but that still
at least exercises the code and lets us know that it works.
This is useful for making a form of PrunedLiveness that is sensitive to address
fields. I wired up the current PrunedLiveness to just use PrunedLiveBlocks with
num bits set to 1.
