All llvm::Functions created during IRGen will have target-cpu and target-features
attributes if they are non-null.
Update testing cases to expect the attribute in function definition.
Add testing case function-target-features.swift to verify target-cpu and
target-features.
rdar://20772331
Swift SVN r28186
This matches how dispatch_once works in C, dramatically cutting the cost of a global accessor by avoiding the runtime call in the hot path and giving the global a unique branch for the CPU to predict away. For now, only do this for Darwin; non-ObjC platforms don't necessarily expose their "done" value as ABI like ours do.
While we're here, change "once" to take a thin function pointer. We don't ever emit global initializers with context dependencies, and this simplifies the runtime glue between swift_once and dispatch_once/std::call_once a bit.
Swift SVN r28166
Calls to willThrow are marked as read-none so that the optimizer can remove
them. The willThrow builtin is still generated for all throw/rethrow sites,
but I plan to look at this next.
rdar://20356658
Swift SVN r27877
These aren't really orthogonal concerns--you'll never have a @thick @cc(objc_method), or an @objc_block @cc(witness_method)--and we have gross decision trees all over the codebase that try to hopscotch between the subset of combinations that make sense. Stop the madness by eliminating AbstractCC and folding its states into SILFunctionTypeRepresentation. This cleans up a ton of code across the compiler.
I couldn't quite eliminate AbstractCC's information from AST function types, since SIL type lowering transiently created AnyFunctionTypes with AbstractCCs set, even though these never occur at the source level. To accommodate type lowering, allow AnyFunctionType::ExtInfo to carry a SILFunctionTypeRepresentation, and arrange for the overlapping representations to share raw values.
In order to avoid disturbing test output, AST and SILFunctionTypes are still printed and parsed using the existing @thin/@thick/@objc_block and @cc() attributes, which is kind of gross, but lets me stage in the real source-breaking change separately.
Swift SVN r27095
The set of attributes that make sense at the AST level is increasingly divergent from those at the SIL level, so it doesn't really make sense for these to be the same. It'll also help prevent us from accidental unwanted propagation of attributes from the AST to SIL, which has caused bugs in the past. For staging purposes, start off with SILFunctionType's versions exactly the same as the FunctionType versions, which necessitates some ugly glue code but minimizes the potential disruption.
Swift SVN r27022
Previously some parts of the compiler referred to them as "fields",
and most referred to them as "elements". Use the more generic 'elements'
nomenclature because that's what we refer to other things in the compiler
(e.g. the elements of a bracestmt).
At the same time, make the API better by providing "getElement" consistently
and using it, instead of getElements()[i].
NFC.
Swift SVN r26894
As part of this, re-arrange the argument order so that
generic arguments come before the context, which comes
before the error result. Be more consistent about always
adding a context parameter on thick functions, even
when it's unused. Pull out the witness-method Self
argument so that it appears last after the error
argument.
Swift SVN r26667
The deallocating parameter convention is a new convention put on a
non-trivial parameter if the caller function guarantees to the callee
that the parameter has the deallocating bit set in its object header.
This means that retains and releases do not need to be emitted on these
parameters even though they are non-trivial. This helps to solve a bug
in +0 self and makes it trivial for the optimizer to perform
optimizations based on this property.
It is not emitted yet by SILGen and will only be put on the self
argument of Deallocator functions.
Swift SVN r26179
As per discussion on the mailing list we don't want different behavior for
programmer errors at Ounchecked than at O for errors that are not diagnosable.
Neither O nor Ounchecked will now emit LLVM's NoAlias attribute for inout
parameters.
rdar://20041458
Swift SVN r25784
We have to guarantee memory safety in the presence of the user violating the
inout assumption. Claiming NoAlias for parameters that might alias is not
memory safe because LLVM will optimize based on that assumption.
Unfortunately, this means that llvm can't optimize arrays as aggressively. For
example, the load of array->buffer won't get hoisted out of loops (this is the
Sim2DArray regression below).
-O numbers (before/after):
CaptureProp 0.888365
Chars 1.09143
ImageProc 0.917197
InsertionSort 0.895204
JSONHelperDeserialize 0.909717
NSDictionaryCastToSwift 0.923466
Sim2DArray 0.76296
SwiftStructuresBubbleSort 0.897483
Continue emitting noalias for inout when compiling Ounchecked.
rdar://20041458
Swift SVN r25770
Insert a block's context argument into the normal parameter set before lowering external arguments, instead of adding it later in a post-pass, so that we assign argument attributes to the correct indices. Fix signature expansion to put the block context parameter in the right place relative to struct returns. Fixes rdar://problem/20035272.
Swift SVN r25763
Upstream clang got better about preserving the type of lowered vector parameters, so float_vec3 lowers to a <3 x float> parameter rather than a generic <2 x double>. We still represent the opaque blob as an aligned i128, since we don't understand padding in imported structs, so we need to generalize coerceValue to deal with size mismatches in both directions. Fixes rdar://problem/20011424.
Swift SVN r25701
Clean up the messy code for putting byval arguments at the right argument index; we don't pretend to handle multiple uncurry levels in IRGen anymore, so we can just use the absolute index of lowered arguments for attribute indices. Add zeroext/signext attributes as needed for small integer arguments in calls. Fixes rdar://problem/19455987.
Swift SVN r25687
In partial apply forwarding thunks, we need to cast dependent result types of
polymorphic function calls to the specialized type.
For example for:
%0 = function_ref @closure :
$@thin <C where C : Base> (@owned Base) -> @owned C
%1 = partial_apply %0<Sub>() :
$@thin <C where C : Base> (@owned Base) -> @owned C
We need to cast the result type:
define %C3Sub* @_TPA_closure(%C4Base*) {
entry:
%1 = call %swift.type* @_TMaC3Sub()
%2 = tail call %C4Base* @closure(%C4Base* %0, %swift.type* %1)
--> %3 = bitcast %C4Base* %2 to %C3Sub*
ret %C3Sub* %3
}
rdar://19930705
Swift SVN r25521
When forming a closure over a dynamic function value, we were failing to set sret, byval, etc. attributes on the call of the underlying function in the partial application thunk. Fixes rdar://problem/19875360.
Swift SVN r25387
was incomplete; we do use it on i386 for arbitrary small
aggregates, and unfortunately this means we have to learn
how to expand quite a lot of types.
Tested by the i386-simulator build. I still think
we can completely remove this code relatively soon.
Swift SVN r24981
the C calling convention.
You might notice the absence of tests. Older versions of Clang
used Expand for ARM64 homogeneous aggregates. Trunk Clang no
longer does this, and indeed, on Apple platforms we no longer
use Expand to pass any kind of argument at all. The only target
in trunk that uses it is the Windows target's vectorcall
alternative CC, which I have no way to convince Swift to
generate a call against. Therefore this code is completely
dead in trunk, and thus untestable.
Removing Expand completely from Clang would be the ideal solution,
but that's politically tricky because this is a part of Clang
that's frequently edited by people supporting out-of-tree backends.
I've started the process, but until then, if we're going to have
dead code, it ought to at least be correct dead code.
This commit fixes rdar://19199427, or at least it will
when it's merged to branches that actually use Expand.
Swift SVN r24949
Per the previous commit we are no longer using this. Minor save in
simplicity and maybe a bit of compilation time as well.
In the long run IRGen probably shouldn't be pulling information from the
AST at all; the SILModule should be able to tell it what types it needs
to emit information for. But this is still an improvement for now.
No functionality change (that was the previous commit).
Swift SVN r24840
Instead, just fall through to the normal public/internal/private switch
added in the previous commit. Local declarations are always private.
Make sure we emit all local declarations by using the list in the SourceFile,
rather than walking the AST (which missed a few cases and was less efficient
anyway).
As an exception, declarations without accessibility at all still get private
linkage. These are things like local variables that don't get accessed by
symbol, even when using the debugger.
rdar://problem/19623016
Swift SVN r24839
OpaqueStorageTypeInfo uses iNNN types that don't always have the correct alloc size for an expected
size at the LLVM level. This needs to be fixed before my partial apply closure fixes can hold.
Swift SVN r24551
In order to deal with generic indirect value captures, we need to be able to bind their type metadata in the partial apply forwarder and heap object destructor to have access to the value operations for that type. NecessaryBindings gives us a way to do that and clean up our ad-hoc polymorphic argument forwarding we had before. N(intended)FC yet, aside from some harmless reordering of operations, since we also need to implement NonFixedOffsets for heap objects to be fully operational.
Swift SVN r24526
Until now, we treated all value parameters as +1, and all indirect parameters as +0 by reference, which is totally bogus. Now that we'll be partially applying guaranteed parameters, for instance to implement SomeType.foo curried method references, we need to get this right.
This breaks curries that capture dependent-layout parameters by value, but they were already broken due to us treating them as 'inout' captures. I'll fix this by adding NecessaryBindings and NonFixedOffsets to HeapLayout next.
Swift SVN r24475
Teach IRGen and the runtime about the extra inhabitants
of function pointers, and take advantage of that in
thin and thick function types.
Also add runtime entrypoints for thin function type
metadata.
Swift SVN r24346
IRGen uses a typedef, SpareBitVector, for its principal
purpose of tracking spare bits. Other uses should not
use this typedef, and I've tried to follow that, but I
did this rewrite mostly with sed and may have missed
some fixups.
This should be almost completely NFC. There may be
some subtle changes in spare bits for witness tables
and other off-beat pointer types. I also fixed a bug
where IRGen thought that thin functions were two
pointers wide, but this wouldn't have affected anything
because we never store thin functions anyway, since
they're not a valid AST type.
This commit repplies r24305 with two fixes:
- It fixes the computation of spare bits for unusual
integer types to use the already-agreed-upon type
size instead of recomputing it. This fixes the
i386 stdlib build. Joe and I agreed that we should
also change the size to use the LLVM alloc size
instead of the next power of 2, but this patch
does not do that yet.
- It changes the spare bits in function types back
to the empty set. I'll be changing this in a
follow-up, but it needs to be tied to runtime
changes. This fixes the regression test failures.
Swift SVN r24324
IRGen uses a typedef, SpareBitVector, for its principal
purpose of tracking spare bits. Other uses should not
use this typedef, and I've tried to follow that, but I
did this rewrite mostly with sed and may have missed
some fixups.
This should be almost completely NFC. There may be
some subtle changes in spare bits for witness tables
and other off-beat pointer types. I also fixed a bug
where IRGen thought that thin functions were two
pointers wide, but this wouldn't have affected anything
because we never store thin functions anyway, since
they're not a valid AST type.
Swift SVN r24305
This happens with GLKVector3 on x86_64, whose ABI type is { <2 x float>, float }, which LLVM considers to have 16 bytes size due to alignment, but has a storage type of { float, float, float }. Relax the assertion that the coerced type is exactly the same size as the storage type.
Swift SVN r23888
isn't used yet, but will be for modeling the self argument passed to an
address-only witness implementation. NFC since all this code is dead :-)
Swift SVN r23857
If an imported C struct has no __nonnull pointer fields, then we can give a default initializer that zeroes all of its fields. This becomes a requirement when working with partially-imported types like NSDecimal. NSDecimal has bitfields Swift can't see yet, so it's impossible to DI, but the Foundation functions that work with NSDecimal all emit their result by out parameter, and without access to its fields it is impossible to initialize an NSDecimal for use with one of these functions. Implement the initializer using a builtin that gets lowered by IRGen; this is also made necessary by the fact that Swift has only a partial view of the struct, so we can't form a complete zero initializer until we have the definitive type layout from Clang.
Swift SVN r23727
It returns the argument and specifies that the value is not negative.
It has only an effect if the argument is a load or call.
The effect of this builtin is that for the load/call argument a positive range metadata is created in llvm ir.
I also added a public function _assumeNonNegative(x: Int) -> Int in the stdlib.
To be on the save side, I prefixed it with an underscore. But maybe it makes sense to make it available for all users.
Swift SVN r23582
CmpXChg builtins now return (T, Bool) to match the LLVM return value.
Turn the tests back on and check extractvalue / inttoptr instructions.
<rdar://problem/17309776> Update modeling of cmpxchg builtin to handle weak-ness and separate success bit
Swift SVN r23104
Modeling builtins as first-class function values doesn't really make sense because there's no real function value to emit, and modeling them this way complicates passes that work with builtins because they have to invent function types for builtin invocations. It's much more straightforward to have a single instruction that references the builtin by ID, along with the type information for the necessary values, type parameters, and results, so add a new "builtin" instruction that directly represents a builtin invocation. NFC yet.
Swift SVN r22690
