SingleValueInstruction gets added to the worklist, causing cast assert at https://github.com/apple/swift/blame/master/lib/SILOptimizer/Utils/ConstantFolding.cpp#L1585.
One such example inst is the following (in the tensorflow branch), which produces a SILValue of type
MultipleValueInstructionResult, so ValueBase::getDefiningInstruction() still
returns a valid inst for it, even though that graph_op inst is not a SingleValueInstruction.
```
%94 = graph_op "Fill,i,i"(%73 : $TensorHandle<Int32>, %85 : $TensorHandle<Float>) {T: $Float, index_type: $Int32, __device: "/device:CPU:0"} : $TensorHandle<Float>
```
The same fix has been merged into the tensorflow branch: https://github.com/apple/swift/pull/18272
This name makes it clear that the function has not yet been deleted and also
contrasts with the past tense used in the API notifyAddedOrModifiedFunction to
show that said function has already added/modified the function.
The name notifyAddFunction is actively harmful since the pass manager uses this
entrypoint to notify analyses of added *OR* modified functions. It is up to the
caller analysis to distinguish in between these cases.
I am not vouching for the design, just trying to make names match the
current behavior.
Generally in the SIL/SILOptimizer libraries we have been putting kinds in the
swift namespace, not a nested scope in a type in swift (see ValueKind as an
example of this).
So far we immediately bailed once we detect a cycle in specializations. But it turned out that this prevented efficient code generation for some stdlib functions like compactMap.
With this change we allow specialization of cycles up to a depth of 1 (= still very limited to prevent code size explosion in some corner cases).
The effect of this optimization is tested with the existing benchmark FatCompactMap.
SR-7952, rdar://problem/41005326
This patch adds SIL-level debug info support for variables whose
static type is rewritten by an optimizer transformation. When a
function is (generic-)specialized or inlined, the static types of
inlined variables my change as they are remapped into the generic
environment of the inlined call site. With this patch all inlined
SILDebugScopes that point to functions with a generic signature are
recursively rewritten to point to clones of the original function with
new unique mangled names. The new mangled names consist of the old
mangled names plus the new substituions, similar (or exactly,
respectively) to how generic specialization is handled.
On libSwiftCore.dylib (x86_64), this yields a 17% increase in unique
source vars and a ~24% increase in variables with a debug location.
rdar://problem/28859432
rdar://problem/34526036
Now that access marker verification is strict and exhaustive, adjust some code
to handle the extra markers and extra checks produced by -enable-verify-exclusivity.
All this does is automate the creation of the ${DIRNAME}_SOURCES variables that we already create and allows for the author to avoid having to prefix with the directory name, i.e.:
set(FOOBAR_SOURCES
FooBar/Source.cpp
PARENT_SCOPE)
=>
silopt_register_sources(
Source.cpp)
Much easier and cleaner to read. I put the code that implements this in the
CMakeLists.txt file just for the SILOptimizer.
Fixes <rdar://40555427> [SR-7773]:
SILCombiner::propagateConcreteTypeOfInitExistential fails to full propagate type
substitutions.
Fixes <rdar://problem/40923849>
SILCombiner::propagateConcreteTypeOfInitExistential crashes on protocol
compositions.
This rewrite fixes several fundamental bugs in the SILCombiner optimization that
propagates concrete types. In particular, the pass needs to handle:
- Arguments of callee Self type in non-self position.
- Indirect and direct return values of Self type.
- Types that indirectly depend on Self within callee function signature.
- Protocol composition existentials.
- All of the above need to work for protocol extensions as well as witness methods.
- For protocol extensions, conformance lookup should be based on the existential's conformance list.
Additionally, the optimization should not depend on a SILFunction's DeclContext,
which is not serialized. (In fact, we should prevent SIL passes from using
DeclContext). Furthermore, the code needs to be expressed in a way that one can
reason about correctness and invariants.
The root cause of these bugs is that SIL passes are written based on untested
assumptions of Swift type system. A SIL pass needs to handle all verifiable SIL
input because passes need to be composable. Bail-out logic can be added to
simplify the design; however, _the bail-out logic itself cannot make any
assumptions about the language or type system_ that aren't clearly and
explicitly enforced in the SIL verifier. This is a common mistake and major
source of bugs.
I created as many unit tests as I reasonably could to prevent this code from
regressing. Creating enough unit tests to cover all corner cases that were
broken in the original code would be intractable. But the code has been
simplified such that many corner cases disappear.
This opens up some oportunity for generalizing the optimization and eliminating
special cases. However, I want this PR to be limited to fixing correctness
issues only. In the long term, it would be preferable to replace this
optimization entirely with a much more powerful general type propagation pass.
The "subclass scope" is meant to represent a connection to a vtable (and how
public something needs to be), for things that end up in class
vtables. Specializations and thunks are mostly internal implementation details
and do not end up there, so subclass scope is not applicable to them. This stops
the thunks and specializations being incorrectly public.
(Note, there are some thunks that _are_ public facing: if a function has its
signature optimized, the original entry point becomes a thunk, and this entry
point is what ends up in vtables etc., so needs to remain around, which means
keeping the same hacks for `private` members of an `open` class.)
Fixes rdar://problem/40738913.
Improve the check done in https://github.com/apple/swift/pull/14740
The class_method being devirtualized might be OK for most call sites, i.e. we are allowed to devirtualize it, however, the caller itself might have been marked with inlinable
