Add a step that recognizes if two subtrees of a union node are
topologically equivalent and merges them. This corresponds to removing
duplicate fields in unions.
This deduplication was prevously done while emitting layouts.
A check is inserted into DLAMakeLayouts to assert that, after
constructing unions, no union has only one child, which could be the
case if we didn't deduplicate union fields in the graph.
This commit changes the `TupleTree::deserialize` method to return an
`llvm::ErrorOr<TupleTree>`, enabling users to choose their policies on
error reporting.
callOnPathSteps was using linear scan search in KeyedObjectContainer.
This commit fixes it for using binary search, dramatically improving
performance in workloads that do lots of lookups.
This commit introduces the type system of the model along with several
various other improvements to the model and its users.
* Introduce the type system.
* Introduce possibility to tag certain fields in the model as to be
optional during YAML serialization.
* All the `Name` fields have been replaced in favor of `CustomName` plus
a `name` method that will use `CustomName` if available, or an
automatically generated name otherwise.
* Make TupleTreeReferences behavior more robust: now you either need to
have a valid pointer to `Root` and a `Path` or be default constructed
(`nullptr` for `Root` and an empty `Path`). Any other configuration is
invalid.
* The type system introduces `RawFunctionType`: this superseds the
previous way in which we were specifying arguments and return
values. Users of such information have been updated accordingly.
This commit drops the KeyTraits in favor of a std::any-like solution.
Basically, we type erase any key the user wants to employ, just exposing
a virtual version of the destructor, a comparison operator and a clone
primitive.
Before this commit, the execution logic of RecursiveCoroutine used an
underlying std::stack allocated on the heap to hold all the
coroutine_handles. It also manually managed passing return values from
callees to callers.
This commit drops this unnecessary auxiliary stack.
When a RecursiveCoroutine co_awaits another one, the handle of the
awaiter is injected into the awaitee, so that when the awaitee is done
it can directly execute the remaining part of the awaiter as a
continuation.
This commit takes out of EnforceABI the part taking care of creating
wrappers for calls to helpers and promoting CSV to local variables.
This decoupling, enables to run -promote-csvs multiple times, for
instance after inlining.
This commit greatly improves the performance by ensuring that, when
computing the set of nodes we want to consider for AVI, we do not
traverse the dispatcher.
Doing so, means including *a lot* of irrelevant nodes and wasting a lot
of computation, since the CFG usually is not influenced by stuff
happening before an indirect jump.
In at least a situation the speedup is in the order of 20x, however this
depends on the size of the binary, since traversing the dispatcher means
including all the binary in the computations (as opposed to just the set
of blocks involved in the dataflow to compute a certain expression).
This commit ensures that FunctionIsolation and EnforceABI do only
thing. This means that they no longer modify `root`.
Instead, we have a new pass, `invoke-isolated-functions` that needs to
be run after them and replaces the entry point of the functions with
invokes to the isolated functions, possibly with the appropriate
arguments.
When we compare StackAnalysis test results, the order in which things
appear in the JSON is relevant. However, the output was
non-deterministic due to a `std::map` using a pointer as key.
This commit improves the situation by sorting the elements by name
before dumping them in JSON.
Pietro Fezzardi