"used" (e.g., we will refer to the vtable in the generated code) and
when they are defined (i.e., because we've seen the key function
definition). Previously, we were effectively tracking "potential
definitions" rather than uses, so we were a bit too eager about emitting
vtables for classes without key functions.
The new scheme:
- For every use of a vtable, Sema calls MarkVTableUsed() to indicate
the use. For example, this occurs when calling a virtual member
function of the class, defining a constructor of that class type,
dynamic_cast'ing from that type to a derived class, casting
to/through a virtual base class, etc.
- For every definition of a vtable, Sema calls MarkVTableUsed() to
indicate the definition. This happens at the end of the translation
unit for classes whose key function has been defined (so we can
delay computation of the key function; see PR6564), and will also
occur with explicit template instantiation definitions.
- For every vtable defined/used, we mark all of the virtual member
functions of that vtable as defined/used, unless we know that the key
function is in another translation unit. This instantiates virtual
member functions when needed.
- At the end of the translation unit, Sema tells CodeGen (via the
ASTConsumer) which vtables must be defined (CodeGen will define
them) and which may be used (for which CodeGen will define the
vtables lazily).
From a language perspective, both the old and the new schemes are
permissible: we're allowed to instantiate virtual member functions
whenever we want per the standard. However, all other C++ compilers
were more lazy than we were, and our eagerness was both a performance
issue (we instantiated too much) and a portability problem (we broke
Boost test cases, which now pass).
Notes:
(1) There's a ton of churn in the tests, because the order in which
vtables get emitted to IR has changed. I've tried to isolate some of
the larger tests from these issues.
(2) Some diagnostics related to
implicitly-instantiated/implicitly-defined virtual member functions
have moved to the point of first use/definition. It's better this
way.
(3) I could use a review of the places where we MarkVTableUsed, to
see if I missed any place where the language effectively requires a
vtable.
Fixes PR7114 and PR6564.
llvm-svn: 103718
shadowing it in the GlobalDeclMap. Eliminates the string-uniquing
requirement for mangled names, which should help C++ codegen times a little.
Forces us to do string lookups instead of pointer lookups, which might hurt
codegen times a little across the board. We'll see how it plays out.
Removing the string-uniquing requirement implicitly fixes any bugs like
PR6635 which arose from the fact that we had multiple uniquing tables for
different kinds of identifiers.
llvm-svn: 99012
iterations of this patch gave explicit template instantiation
link-once ODR linkage, which permitted the back end to eliminate
unused symbols. Weak ODR linkage still requires the symbols to be
generated.
llvm-svn: 98441
Clang's support for weakref is now better than llvm-gcc's :-)
We don't introduce a new symbol and we correctly mark undefined references weak only if there is no
definition or regular undefined references in the same file.
llvm-svn: 97733
1) emit base destructors as aliases to their unique base class destructors
under some careful conditions. This is enabled for the same targets that can
support complete-to-base aliases, i.e. not darwin.
2) Emit non-variadic complete constructors for classes with no virtual bases
as calls to the base constructor. This is enabled on all targets and in
theory can trigger in situations that the alias optimization can't (mostly
involving virtual bases, mostly not yet supported).
These are bundled together because I didn't think it worthwhile to split them,
not because they really need to be.
llvm-svn: 96842
and destructors when the two entities are semantically identical, i.e. when
the class has no virtual base classes. We only do this for linkage types
for which aliases are supported, i.e. internal and external, i.e. not linkonce.
llvm-svn: 96451
1. Add helper class for sema checks for target attributes
2. Add helper class for codegen of target attributes
As a proof-of-concept - implement msp430's 'interrupt' attribute.
llvm-svn: 93118
linkage of vtables. Before this, we were emitting RTTI names for
template instantiations with strong external linkage rather than with
weak ODR linkage.
llvm-svn: 92857
the constructor. This doesn't handle cases requiring the VTT at the moment,
and generates unnecessary stores, but I think it's essentially correct.
llvm-svn: 91731
operators, and destructors. Avoids generating declarations/definitions of
trivial constructors/destructors, and makes sure the trivial copy assignment
operator is generated when necessary.
llvm-svn: 89943
Refine the VTT entries for virtual bases to refer to the complete
object's vtable instead of constructor vtables.
Refine the AddressPoint calculations for VTT entries for virtual bases.
llvm-svn: 87021
The decl wasn't being passed down, which meant that function attributes were not
being set correctly. This is particularly important for ARM, since it wants to
override the calling convention. Instead we would emit the builtin with the
wrong calling convention, and instcombine would come along and merrily shred all
the calls to it. :)
llvm-svn: 81756
- Track implicit instantiations vs. the not-yet-supported explicit
specializations
- Give implicit instantiations of function templates (and member
functions of class templates) linkonce_odr linkage.
- Improve name mangling for function template specializations,
including the template arguments of the instantiation and the return
type of the function.
Note that our name-mangling is improved, but not correct: we still
don't mangle substitutions, although the manglings we produce can be
demangled.
llvm-svn: 74466
to allow us to support generation of deferred ctors/dtors.
It looks like codegen isn't emitting a call to the dtor in
member-functions.cpp:test2, but when it does, its body should
get emitted.
llvm-svn: 71594
tentative definitions off to the ASTConsumer at the end of the
translation unit.
Eliminate CodeGen's internal tracking of tentative definitions, and
instead hook into ASTConsumer::CompleteTentativeDefinition. Also,
tweak the definition-deferal logic for C++, where there are no
tentative definitions.
Fixes <rdar://problem/6808352>, and will make it much easier for
precompiled headers to cope with tentative definitions in the future.
llvm-svn: 69681
- PR3980.
- <rdar://problem/6762287> [irgen] crash when generating tentative
definition of incomplete structure
- This also avoids creating common definitions for things which are
later overwritten.
- XFAIL'ed external-defs.c, it isn't completing types properly yet.
llvm-svn: 69231
- Pull out SetCommonAttributes, which handles the things common to
aliases, methods, functions, and variables.
- Pull out SetLLVMFunctionAttributesForDefinition, which handles the
LLVM attributes which we only want to apply to a definition (like
noinline and alwaysinline).
- Kill SetGVDeclarationAttributes (inlined into SetFunctionAttributes
and specialized).
- Kill SetFunctionAttributesForDefinition (inlined into sole caller).
- Inline SetGVDefinitionAttributes into SetMethodAttributes and
specialize.
- Rename SetGVDefinitionAttributes to SetFunctionDefinitionAttributes.
This is supposed to be a no functionality change commit, but I may
have made a mistake.
llvm-svn: 69036
functionality, fixing a crash on the attached testcase. Eliminate the
BuiltinFunctions cache, as it can contain dangling pointers. This fixes
a bunch of valgrind errors on test/CodeGen/builtins.c
llvm-svn: 67484
CodeGenModule::CreateRuntimeVariable.
- No real functionality change; although we now assert on silly
things like:
--
int objc_exception_throw;
void f0() { @throw(@"A"); }
--
instead of accepting it.
llvm-svn: 66292
BlockModule. No functionality change. This should help people that
don't want to know anything about blocks not be confused by the
overloaded use of the term block or nor want to see all the blocks
goop.
llvm-svn: 66042
we ensure that things added to the module can be found even when they
are not in GlobalDeclMap. The later is for increased flexibility,
should someone want to do something tricky like extern "Ada" in the
same module.
llvm-svn: 65657
IRgen no longer relies on isConstantInitializer, instead we just try
to emit the constant. If that fails then in C we emit an error
unsupported (this occurs when Sema accepted something that it doesn't
know how to fold, and IRgen doesn't know how to emit) and in C++ we
emit a guarded initializer.
This ends up handling a few more cases, because IRgen was actually
able to emit some of the constants Sema accepts but can't Evaluate().
For example, PR3398.
llvm-svn: 64780
Douglas Gregor