With PE/COFF, one cannot reference a data symbol directly across the
binary module boundary. Instead, the reference must be indirected
through the Import Address Table (IAT) to allow for position
independence.
When generating a reference to a AsyncFunctionPointer ({i8*, i32}), we
tag the pointer as being indirected by tagging bit 1 (with the
assumption that native alignment will ensure 4/8 byte alignment, freeing
the bottom 2 bits at least for bit-packing). We tweak the
v-table/witness table emission such that all references to the
AsyncFunctionPointer are replaced with the linker synthetic import
symbol with the bit packing:
~~~
.quad __imp_$s1L1CC1yyYaKFTu+1
~~~
rather than
~~~
.quad $s1L1CC1yyYaKFTu
~~~
Upon access of the async function pointer reference, we open-code the
check for the following:
~~~
pointer = (pointer & 1) ? *(void **)(pointer & ~1) : pointer;
~~~
Thanks to @DougGregor for the discussion and the suggestion for the
pointer tagging. Thanks to @aschwaighofer for pointers to the code that
I had missed. Also, thanks to @SeanROlszewski for the original code
sample that led to the reduced test case.
Fixes: SR-15399
These include _move and @_noImplicitCopy. I still need to wire up the parsing of
those behind this feature.
The reason that I am adding this now is that I am going to now need to make some
changes behind a feature flag and I have not yet needed to add one. The specific
reason I needed to add one here is to ensure that I properly guard inside _move
the call to Builtin.move so as to prevent a "cond_fail" incident.
P.S.: This work depends on experimental lexical lifetimes being enabled as well,
so I did that at the same time in this PR.
Introduce a few changes to the logic for watchOS concurrency
back-deployment with respect to the async frame pointer.
* Only apply the change to watchOS device targets, not simulator targets
* Only introduce the override when no specific
`-swift-async-frame-pointer=<value>` option is provided on the
command line
* Only override the default when deploying to watchOS < 8, and
* Use "never" for the default rather than "always".
This represents a different but safer trade-off than before. Setting
the async bit in the frame pointer can cause older APIs (such as
backtrace APIs in the OS) to crash when they encounter such frame
pointers. So, with this change we never set the bit when back-deploying
for watchOS device, to avoid said crashes.
The trade-off here is that a back-deployed watchOS app will never have
the async frame pointer bit set, so async backtraces will be
unavailable even when running on watchOS 8 or newer.
The asm definition of `swift_async_extendedFramePointerFlags` prevents
the use of bitcode with the back-deployment libraries, so remove the
definition and use of this symbol from watchOS binaries entirely.
Instead, always force the async frame bit to be set. This trades off
backtraces on older OS's for debuggability of newer ones. If it causes
problems, it can be disabled via the option
`-swift-async-frame-pointer=never`.
Fixes rdar://84687579.
Previously, the flag was a LangOptioins. That didn't make much sense because
this isn't really a user-facing behavior. More importantly, as a member
of that type type it couldn't be accessed when setting up pass
pipelines. Here, the flag is moved to SILOptions.
Intro the frontend flag `-check-api-availability-only` that limits
availability checking to the API and SPI. This mode doesn't check the
availability of non-inlinable function bodies and of non-ABI-public decl
signatures.
This mode goal is to check all that is printed in the swiftinterface
file. It should be used in place of the wider
`-disable-availability-checking` when generating an interface for
platforms with no binaries.
rdar://81679692
Introduce a compiler flag that warnings about any public types defined in
a module that are neither explicitly `Sendable` nor explicitly
non-`Sendable` (the latter of which has no spelling currently), which
is intended to help with auditing a module for Sendable conformances.
* Fix unnecessary one-time recompile of stdlib with -enable-ossa-flag
This includes a bit in the module format to represent if the module was
compiled with -enable-ossa-modules flag. When compiling a client module
with -enable-ossa-modules flag, all dependent modules are checked for this bit,
if not on, recompilation is triggered with -enable-ossa-modules.
* Updated tests
This enables optimizing / dead-stripping of witness methods across modules at
LTO time.
- Under -internalize-at-link, restrict visibility of wtables to linkage unit.
- Emit thunks for cross-module wcalls when WME is enabled.
- Use thunks for wcalls across modules when WME is enabled.
- Adjust TBDGen to account for witness method thunks when WME is enabled.
- Add an IR test to check that thunks are used when doing cross-module calls.
- Add an end-to-end test case for cross-module WME.
A new LLVM IR affordance that allows expressing conditions under which globals
can be removed/dropped (even when marked with @llvm.used) is being discussed at:
- <https://reviews.llvm.org/D104496>
- <https://lists.llvm.org/pipermail/llvm-dev/2021-September/152656.html>
This is a preliminary implementation that marks runtime lookup records (namely
protocol records, type descriptors records and protocol conformance records)
with the !llvm.used.conditional descriptors. That allows link-time / LTO-time
removal of these records (by GlobalDCE) based on whether they're actually used
within the linkage unit. Effectively, this allows libraries that have a limited
and known set of clients, to be optimized against the client at LTO time, and
significantly reduce the code size of that library.
Parts of the implementation:
- New -conditional-runtime-records frontend flag to enable using !llvm.used.conditional
- IRGen code that emits these records can now emit these either as a single contiguous
array (asContiguousArray = true, the old way), which is used for JIT mode, or
as indivial globals (asContiguousArray = false), which is necessary for the
!llvm.used.conditional stripping to work.
- When records are emitted as individual globals, they have new names of
"\01l_protocol_" + mangled name of the protocol descriptor, and similarly for
other records.
- Fixed existing tests to account for individual records instead of a single array
- Added an IR level test, and an end-to-end execution test to demonstrate that
the !llvm.used.conditional-based stripping actually works.
"add inits to toplevel" and "call pattern heuristics" are only used in
code completion. Move them from LangOptions to CodeCompletionContext so
that they don't affect compiler arguments.
Added ForceStructTypeLayouts. When enabled, IRGen will lower structs using the
aligned group of TypeLayout rather than using TypeInfos. This potentially leads
to a size increase as TypeInfos currently produce better code than the
TypeLayout route.
- Under -internalize-at-link, stop unconditionally marking all globals as used.
- Under -internalize-at-link, restrict visibility of vtables to linkage unit.
- Emit virtual method thunks for cross-module vcalls when VFE is enabled.
- Use thunks for vcalls across modules when VFE is enabled.
- Adjust TBDGen to account for virtual method thunks when VFE is enabled.
- Add an end-to-end test case for cross-module VFE.
- Witness method calls are done via @llvm.type.checked.load instrinsic call with a type identifier
- Type id of a witness method is the requirement's mangled name
- Witness tables get !type markers that list offsets and type ids of all methods in the wtable
- Added -enable-llvm-wme to enable Witness Method Elimination
- Added IR test and execution test
control swift extended frame information emission
On linux we default to disable the extended frame info (since the system
libraries don't support it).
On darwin the default is to automatically choose based on the deployment target.
The Concurrency library explicitly forces extended frame information and the
back deployment library explicitly disables it.
Changed the frontend flag to -enable-experimental-lexical-lifetimes from
-enable-experimental-defined-lifetimes.
Changed the attribute on begin_borrow from [defined] to [lexical].
- Virtual calls are done via a @llvm.type.checked.load instrinsic call with a type identifier
- Type identifier of a vfunc is the base method's mangling
- Type descriptors and class metadata get !type markers that list offsets and type identifiers of all vfuncs
- The -enable-llvm-vfe frontend flag enables VFE
- Two added tests verify the behavior on IR and by executing a program
Serialize the canonical name of the SDK used when building a swiftmodule
file and use it to ensure that the swiftmodule file is loaded only with
the same SDK. The SDK name must be passed down from the frontend.
This will report unsupported configurations like:
- Installing roots between incompatible SDKs without deleting the
swiftmodule files.
- Having multiple targets in the same project using different SDKs.
- Loading a swiftmodule created with a newer SDK (and stdlib) with an
older SDK.
All of these lead to hard to investigate deserialization failures and
this change should detect them early, before reaching a deserialization
failure.
rdar://78048939
Swift 5.5 didn't support back deployment of concurrency features, so a
Swift 5.5 compiler processing the _Concurrency .swiftinterface will
produce errors for each `async` function and actor with pre-macOS
12/iOS 15 availability. Emit `-disable-availability-checking` into the
generated `.swiftinterface` files to allow Swift 5.5 to continue to
build them.
Finishes rdar://82602353.
Remove the option that explicitly enables concurrency back-deployment,
and instead always enable its support in the compiler. Remove the use
of the extraneous CMake option as well.
Add a frontend-only flag `-enable-experimental-back-deploy-concurrency`
to be used to stage in the back deployment of concurrency. At present,
all it does is lower the availability minimums for use of concurrency
features.
