This gives big code size wins for unused types and also for types, which are never used in a generic context.
Also it reduces the amount of symbols in the symbol table.
The size wins heavily depend on the project. I have seen binary size reductions from 0 to 20% on real world projects.
rdar://problem/30119960
This is NFC in intent, but I had to restructure the code to emit more
of the lists "inline", which means I inevitably altered some IRGen
emission patterns in ways that are visible to tests:
- GenClass emits property/ivar/whatever descriptors in a somewhat
different order.
- An ext method type list is now emitted as just an array, not a struct
containing only that array.
- Protocol descriptors are no longer emitted as packed structs.
I was sorely tempted to stop using packed structs for all the metadata
emission, but didn't really want to update that many tests in one go.
The list of directly inherited protocols of a ProtocolDecl is already
encoded in the requirement signature, as conformance constraints where
the subject is Self. Gather the list from there rather than separately
computing/storing the list of "inherited protocols".
Separate formal lowered types from SIL types.
The SIL type of an argument will depend on the SIL module's conventions.
The module conventions are determined by the SIL stage and LangOpts.
Almost NFC, but specialized manglings are broken incidentally as a result of
fixes to the way passes handle book-keeping of aruments. The mangler is fixed in
the subsequent commit.
Otherwise, NFC is intended, but quite possible do to rewriting the logic in many
places.
The typedef `swift::Module` was a temporary solution that allowed
`swift::Module` to be renamed to `swift::ModuleDecl` without requiring
every single callsite to be modified.
Modify all the callsites, and get rid of the typedef.
Changes:
* Terminate all namespaces with the correct closing comment.
* Make sure argument names in comments match the corresponding parameter name.
* Remove redundant get() calls on smart pointers.
* Prefer using "override" or "final" instead of "virtual". Remove "virtual" where appropriate.
When building exemplar archetypes we would build different exemplars for the
archetype "S: A<T>" of the example below because we would just use the pointer
to the class type ": A<T>" instead of profiling (decending) into the type.
This caused us to end up with different exemplar types for different uses of M
and ultimately several llvm type instances for the same generic type M<S,T>.
class A<T> {}
class M<T, S: A<T>> {...}
Also reverts "IRGen: Cast the type of class pointer to the storage type"
from commit 2d73b3cef2 but keeps its test case.
Proper fix for rdar//28684642
Allow it only to have one context parameter, whose ownership convention matches the convention of the resulting thick function, effectively limiting it to binding a closure invocation function to its context.
It's the same thing as for alloc_ref: the optional [tail_elems ...] attribute specify the tail elements to allocate.
For details see docs/SIL.rst
This feature is needed so that we can allocate a MangedBuffer with alloc_ref_dynamic.
The ManagedBuffer.create() function uses the dynamic self type to create the buffer instance.
Resilient classes are not fully implemented yet, and can cause
crashes at runtime; add a flag disabling them until the code is
done, to unblock standard library testing with resilience
enabled.
Remove special-casing that makes NSObject fragile since
this messes up layout. The optimization probably has little
practical benefit anyway.
Fixes <https://bugs.swift.org/browse/SR-2586>.
The new instructions are: ref_tail_addr, tail_addr and a new attribute [ tail_elems ] for alloc_ref.
For details see docs/SIL.rst
As these new instructions are not generated so far, this is a NFC.
The new instructions are: ref_tail_addr, tail_addr and a new attribute [ tail_elems ] for alloc_ref.
For details see docs/SIL.rst
As these new instructions are not generated so far, this is a NFC.
I apologize in advance to @jrose-apple, who is not a fan
of this fix ;-)
In unoptimized builds, the convenience initializers on
DispatchQueue allocate and immediately deallocate an
instance of OS_dispatch_queue prior to calling the
C function that returns the "real" instance.
This is because we don't have a way to write user-defined
factory initializers yet; convenience initializers still
have an 'initializing' entry point that takes an existing
instance, which we have no choice but to throw away.
Unfortunately, when we perform the fake allocation, we
look up class metadata by calling the wrong Swift runtime
function, causing a crash when we send +allocWithZone:.
Fix this so that the metadata is accessed via a lookup
from the Objective-C runtime, instead of making a totally
fake 'foreign metadata' object -- it looks like there was
code for this already, it just wasn't used in all cases.
While getting metadata for a runtime-only class should be
rare, this feels like a real bug fix, to me.
Second, we would ultimately free the fake object by sending
-release, however OS_dispatch_queue has an override of
-dealloc which doesn't like to be called with a completely
uninitialized instance.
Here, I'm going to drop all pretense of sanity. The patch
just changes IRGen to lower the dealloc_partial_ref instruction
as a call to the object_dispose() Objective-C runtime function
when the class in question is a runtime-only class. This
frees the object without running -dealloc, which *happens*
to work for OS_dispatch_queue.
Fixes <rdar://problem/27226313>.
One minor revision: this lifts the proposed restriction against
overriding a non-open method with an open one. On reflection,
that was inconsistent with the existing rule permitting non-public
methods to be overridden with public ones. The restriction on
subclassing a non-open class with an open class remains, and is
in fact consistent with the existing access rule.
Instead of hooking into nominal type and extension emission
and walking all conformances of those declarations, let's
just directly hook into the logic for emitting conformances.
This fixes an issue where we would apparently emit duplicate
conformances, as well as unnecessary conformances that are
defined elsewhere.
The ObjC runtime on OS X 10.10 and older and iOS 9 and older can't
handle them, so for these targets, emit nil for all class property
lists.
It's a little unfortunate that this is target-dependent, but there's
not much we can do about it.
rdar://problem/25605427
Rather than collection nominal type and extension decls and emit
reflection metadata records in one go, we can emit them as they
are encountered and instead collection builtin types referenced
by those at the end.
specialization to be separately lowered in IRGen, use the mangling
of the specialized type as the name of the llvm::StructType instead
of the base, unspecialized type.
This tends to produce fewer collisions between IR type names.
LLVM does unique the names on its own, so that's not strictly
necessary, but it's still a good idea because it makes the test
output more reliable and somewhat easier to read (modulo the
impact of bigger type names). Collisions will still occur if
the type is specialized at an archetype, since in this case we
will fall back on the unspecialized type.
properties of classes with generic layouts.
Previously we were falling back on accessing them via the field
offset vector even when we knew everything about the type.
As a minor benefit, this allows RemoteAST to also determine offsets
for members of classes with generic layout.
Half of the test changes are IR type-name uniquing; I'm going to
explore mangling these with the full type where possible.
If a subclass grounds all the type parameters from its base class, we don't have to worry about any erasure edge cases. We should be able to support this, giving existing code that subclasses the nongeneric form of the class a migration path. Spot-fix some places in IRGen where we assume we can't emit static references to ObjC generic classes or metaclasses.
Give empty fields a notional byte offset so we can still use the offset of the first field as an approximation of the instanceStart of a class.
Fixes SR-1055.
initialization in-place on demand. Initialize parent metadata
references correctly on struct and enum metadata.
Also includes several minor improvements related to relative
pointers that I was using before deciding to simply switch the
parent reference to an absolute reference to get better access
patterns.
Includes a fix since the earlier commit to make enum metadata
writable if they have an unfilled payload size. This didn't show
up on Darwin because "constant" is currently unenforced there in
global data containing relocations.
This patch requires an associated LLDB change which is being
submitted in parallel.
initialization in-place on demand. Initialize parent metadata
references correctly on struct and enum metadata.
Also includes several minor improvements related to relative
pointers that I was using before deciding to simply switch the
parent reference to an absolute reference to get better access
patterns.
This reorganization allows adding attributes that refer to types.
I need this for a @_specialize attribute with a type list.
PrintOptions.h and other headers depend on these enums. But Attr.h
defines a lot of classes that almost never need to be included.
Similarly to how we've always handled parameter types, we
now recursively expand tuples in result types and separately
determine a result convention for each result.
The most important code-generation change here is that
indirect results are now returned separately from each
other and from any direct results. It is generally far
better, when receiving an indirect result, to receive it
as an independent result; the caller is much more likely
to be able to directly receive the result in the address
they want to initialize, rather than having to receive it
in temporary memory and then copy parts of it into the
target.
The most important conceptual change here that clients and
producers of SIL must be aware of is the new distinction
between a SILFunctionType's *parameters* and its *argument
list*. The former is just the formal parameters, derived
purely from the parameter types of the original function;
indirect results are no longer in this list. The latter
includes the indirect result arguments; as always, all
the indirect results strictly precede the parameters.
Apply instructions and entry block arguments follow the
argument list, not the parameter list.
A relatively minor change is that there can now be multiple
direct results, each with its own result convention.
This is a minor change because I've chosen to leave
return instructions as taking a single operand and
apply instructions as producing a single result; when
the type describes multiple results, they are implicitly
bound up in a tuple. It might make sense to split these
up and allow e.g. return instructions to take a list
of operands; however, it's not clear what to do on the
caller side, and this would be a major change that can
be separated out from this already over-large patch.
Unsurprisingly, the most invasive changes here are in
SILGen; this requires substantial reworking of both call
emission and reabstraction. It also proved important
to switch several SILGen operations over to work with
RValue instead of ManagedValue, since otherwise they
would be forced to spuriously "implode" buffers.
- Implement emission of type references for nominal type field
reflection, using a small custom encoder resulting in packed
structs, not strings. This will let us embed 7-bit encoded
32-bit relative offsets directly in the structure (not yet
hooked in).
- Use the AST Mangler for encoding type references
Archetypes and internal references were complicating this before, so we
can take the opportunity to reuse this machinery and avoid unique code
and new ABI.
Next up: Tests for reading the reflection sections and converting the
demangle tree into a tree of type references.
Todo: For concrete types, serialize the types for associated types of
their conformances to bootstrap the typeref substitution process.
rdar://problem/15617914
