`DifferentiableFunctionInst` now stores result indices.
`SILAutoDiffIndices` now stores result indices instead of a source index.
`@differentiable` SIL function types may now have multiple differentiability
result indices and `@noDerivative` resutls.
`@differentiable` AST function types do not have `@noDerivative` results (yet),
so this functionality is not exposed to users.
Resolves TF-689 and TF-1256.
Infrastructural support for TF-983: supporting differentiation of `apply`
instructions with multiple active semantic results.
This will let us track class methods that must exist for pass ordering, interface, or ABI reasons, but which can
be given more efficient runtime representation because they have no overrides.
* a new [immutable] attribute on ref_element_addr and ref_tail_addr
* new instructions: begin_cow_mutation and end_cow_mutation
These new instructions are intended to be used for the stdlib's COW containers, e.g. Array.
They allow more aggressive optimizations, especially for Array.
This became necessary after recent function type changes that keep
substituted generic function types abstract even after substitution to
correctly handle automatic opaque result type substitution.
Instead of performing the opaque result type substitution as part of
substituting the generic args the underlying type will now be reified as
part of looking at the parameter/return types which happens as part of
the function convention apis.
rdar://62560867
Add `linear_function` and `linear_function_extract` instructions.
`linear_function` creates a `@differentiable(linear)` function-typed value from
an original function operand and a transpose function operand (optional).
`linear_function_extract` extracts either the original or transpose function
value from a `@differentiable(linear)` function.
Resolves TF-1142 and TF-1143.
Add `differentiable_function` and `differentiable_function_extract`
instructions.
`differentiable_function` creates a `@differentiable` function-typed
value from an original function operand and derivative function operands
(optional).
`differentiable_function_extract` extracts either the original or
derivative function value from a `@differentiable` function.
The differentiation transform canonicalizes `differentiable_function`
instructions, filling in derivative function operands if missing.
Resolves TF-1139 and TF-1140.
We saw this failure with a Clang module imported @_implementationOnly
with synthesized conformances by the ClangImporter. It caused
issues only in sil-opt as it reads all the witness tables.
rdar://problem/58924131
The `differentiability_witness_function` instruction looks up a
differentiability witness function (JVP, VJP, or transpose) for a referenced
function via SIL differentiability witnesses.
Add round-trip parsing/serialization and IRGen tests.
Notes:
- Differentiability witnesses for linear functions require more support.
`differentiability_witness_function [transpose]` instructions do not yet
have IRGen.
- Nothing currently generates `differentiability_witness_function` instructions.
The differentiation transform does, but it hasn't been upstreamed yet.
Resolves TF-1141.
SIL differentiability witnesses are a new top-level SIL construct mapping
an "original" SIL function and derivative configuration to derivative SIL
functions.
This patch adds `SILDifferentiabilityWitness` serialization/deserialization.
Resolves TF-1136.
Motivation: `GenericSignatureImpl::getCanonicalSignature` crashes for
`GenericSignature` with underlying `nullptr`. This led to verbose workarounds
when computing `CanGenericSignature` from `GenericSignature`.
Solution: `GenericSignature::getCanonicalSignature` is a wrapper around
`GenericSignatureImpl::getCanonicalSignature` that returns the canonical
signature, or `nullptr` if the underlying pointer is `nullptr`.
Rewrite all verbose workarounds using `GenericSignature::getCanonicalSignature`.
The cross-module-optimization can change the linkage of a function to public. Then the SILLinkage is "out of sync" with the linkage derived from the AST. We need to make sure to read the correct SILLinkage from the module file.
This is a first version of cross module optimization (CMO).
The basic idea for CMO is to use the existing library evolution compiler features, but in an automated way. A new SIL module pass "annotates" functions and types with @inlinable and @usableFromInline. This results in functions being serialized into the swiftmodule file and thus available for optimizations in client modules.
The annotation is done with a worklist-algorithm, starting from public functions and continuing with entities which are used from already selected functions. A heuristic performs a preselection on which functions to consider - currently just generic functions are selected.
The serializer then writes annotated functions (including function bodies) into the swiftmodule file of the compiled module. Client modules are able to de-serialize such functions from their imported modules and use them for optimiations, like generic specialization.
The optimization is gated by a new compiler option -cross-module-optimization (also available in the swift driver).
By default this option is off. Without turning the option on, this change is (almost) a NFC.
rdar://problem/22591518
SIL type lowering erases DynamicSelfType, so we generate
incorrect code when casting to DynamicSelfType. Fixing this
requires a fair amount of plumbing, but most of the
changes are mechanical.
Note that the textual SIL syntax for casts has changed
slightly; the target type is now a formal type without a '$',
not a SIL type.
Also, the unconditional_checked_cast_value and
checked_cast_value_br instructions now take the _source_
formal type as well, just like the *_addr forms they are
intended to replace.
* Fix Swift following bitstream reader API update
Upstream change in rL364464 broke downstream Swift.
(cherry picked from commit 50de105bf1)
Conflicts:
lib/Serialization/Deserialization.cpp
lib/Serialization/ModuleFile.cpp
tools/driver/modulewrap_main.cpp
Structurally prevent a number of common anti-patterns involving generic
signatures by separating the interface into GenericSignature and the
implementation into GenericSignatureBase. In particular, this allows
the comparison operators to be deleted which forces callers to
canonicalize the signature or ask to compare pointers explicitly.
The weak imported flag is now only set if the attribute is unconditionally
weak linked, which is the case when it or one of its parent contexts has a
@_weakLinked attribute.
To correctly handle weak linking based availability with serialized SIL
functions, we need to serialize the actual version tuple when the SIL function
was introduced. This is because the deployment target of the client app can
be older than the deployment target that the original module was built with.
Fixes <rdar://problem/52783668>.
Now that GenericSignatures store their single unique GenericEnvironment,
we can remove similar logic from deserialization to preserve identity
of GenericEnvironments.
...fulfilling the promised audit from 0747d9a339. No intended
functionality change /other/ than the order of already-unsorted lists.
This affected a number of SIL tests that relied on deserialization
order matching the original source order; I have no idea why the old
hash logic would make that the case. If we think that's a valuable
property, we should serialize a list of functions in addition to the
iterable table. (Maybe just in SIB mode?)
A generic environment is always serialized as a GenericSignature with
a lazily-recreated environment, though sometimes it has to include
extra info specifically for generic environments used by SIL. The code
that was doing this claimed a bit for disambiguating between the two,
shrinking the permitted size of a compiled module from 2^31 bits to
2^30. (The code isn't just needlessly complicated; GenericEnvironments
used to be serialized with more information.)
Rather than have two representations for GenericEnvironmentID, this
commit just drops it altogether in favor of referencing
GenericSignatures directly. This causes a negligible file size
shrinkage for swiftmodules in addition to eliminating the problematic
disambiguation bit.
For now, the Deserialization logic will continue to cache
GenericEnvironments that are used directly by Deserialization, but
really that should probably be done at the AST level. Then we can
simplify further to ModuleFile tracking a plain list of
GenericSignatures.
Rather than storing the set of input requirements in a
(SIL)SpecializeAttr, store the specialized generic signature. This
prevents clients from having to rebuild the same specialized generic
signature on every use.
This provides a singular instruction for convert an unmanaged value to a ref,
then strong_retain it. I expanded the definition of UNCHECKED_REF_STORAGE to
include these copy like instructions. This instruction is valid in all SIL.
The reason why I am adding this instruction is that currently when we emit an
access to an unowned (unsafe) ivar, we use an unmanaged_to_ref and a strong
retain. This can look to the optimizer like a strong retain that can potentially
be optimized. By combining the two together into a new instruction, we can avoid
this potential problem since the pattern matching will break.
`ModuleFile::error` was being used both for errors of initial parse
and configuration (non-fatal) and format errors during actual
deserialization (fatal, indicating a corrupted module). Split out the
latter to `ModuleFile::fatal()` (to go with the existing
`ModuleFile::fatal(llvm::Error)`) and be more consistent about
explicitly setting statuses for the former.
Since 'fatal()' is always fatal, this also allows deleting dummy
recovery code that would never be used in practice.
This flag is set by DefinitInitialization if the lifetime of the stored value is controlled dynamically.
If the flag is set, it's not (easily) possibly to statically calculate the lifetime of the stored value.
Ownership is truly a property not of a declaration, but of a function body. So
it makes sense to just match what we deserialize.
This also helps us to avoid mismatches if we lower ownership from a function,
delete it, and then relink it.
I also used this as an opportunity to clean up where we set that flag in
deserialization to only happen in one place for both declarations/definitions.
This indicates that the "self" argument to the current function is always dynamically of the exact
static base class type, allowing metadata accesses in IRGen to use the local self metadata to answer
metadata requests for the class type. Set this attribute on allocating entry points of designated
inits, which is one of the most common places where we emit redundant metadata accesses.
I found this issue by inspection. I do not think it is actually possible to hit
this bug today since:
1. To hit the bug one would need to link in a definition of a function when we
do not have a pre-existing function. If we create the declaration first, then we
set the flag correctly when we deserialize the body of the function.
2. We today always set a callback on functions when we link in the body and
always strip ownership/set that flag. So we /should/ always set it correctly.
That being said, it is better to match what was actually serialized as close as
possible.
To distinguish between classes which have the same name (but are in different contexts).
Fixes a miscompile if classes with the same name are used from a different module.
SR-10634
rdar://problem/50538534
The ownership kind is Any for trivial types, or Owned otherwise, but
whether a type is trivial or not will soon depend on the resilience
expansion.
This means that a SILModule now uniques two SILUndefs per type instead
of one, and serialization uses two distinct sentinel IDs for this
purpose as well.
For now, the resilience expansion is not actually used here, so this
change is NFC, other than changing the module format.
Using an anonymous union in KeyPathPatternComponent instead of the weird void * in SetterAndIdKind
Added TupleElement kind to KeyPathComponentKindEncoding
Written basic SIL keypath serialization tests
Deleted or edited some old Swift-level tuple key path tests
It does not take ownership of its non-trivial arguments, is a trivial
function type and therefore must not be destroyed. The compiler must
make sure to extend the lifetime of non-trivial arguments beyond the
last use of the closure.
%objc = copy_value %0 : $AnObject
%closure = partial_apply [stack] [callee_guaranteed] %16(%obj) : $@convention(thin) (@guaranteed AnObject) -> ()
%closure2 = mark_dependence %closure : $@noescape @callee_guaranteed () -> () on %obj : $AnObject
%user = function_ref @useClosure : $@convention(thin) (@noescape @callee_guaranteed () -> ()) -> ()
apply %user(%closure2) : $@convention(thin) (@noescape @callee_guaranteed () -> ()) -> ()
dealloc_stack %closure : $() ->()
destroy_value %obj : $AnObject // noescape closure does not take ownership
SR-904
rdar://35590578
