Sanjiv Gupta [Tue, 3 Feb 2009 18:07:49 +0000 (18:07 +0000)]
Targets that don't have stack use global address space for parameters.
Specify external linkage for such globals so that llvm optimizer do
not assume there values initialized as zero.
Chris Lattner [Tue, 3 Feb 2009 07:30:45 +0000 (07:30 +0000)]
switch SourceManager from using an std::map and std::list of
ContentCache objects to using a densemap and list, and allocating
the ContentCache objects from a bump pointer. This does not speed
up or slow down things substantially, but gives us control over
their alignment.
Daniel Dunbar [Tue, 3 Feb 2009 06:17:37 +0000 (06:17 +0000)]
Add ABIArgInfo::Direct kind, which passes arguments using whatever the
native IRgen type is. This is like Default, but without any extra
semantics (like automatic tweaking of structures or void).
Chris Lattner [Tue, 3 Feb 2009 01:28:04 +0000 (01:28 +0000)]
Content Cache only needs to be 4-byte aligned. Since it is stored in
an std::set, we can't make a strong guarantee about what its alignment
will be. Since I don't need the 3rd bit anyway yet, just change the
assertion.
Douglas Gregor [Tue, 3 Feb 2009 00:34:39 +0000 (00:34 +0000)]
Simplify the way in which we inject the names of tag definitions and
elaborated-type-specifier declarations into outer scopes while
retaining their proper lexical scope. This way is simpler and more
consistent with the way DeclContexts work, and also fixes
Douglas Gregor [Mon, 2 Feb 2009 23:39:07 +0000 (23:39 +0000)]
Add a macro-based enumeration of all of the Decl nodes (like we do
with Stmt/Expr nodes), and convert some of the more mundane
switch-on-all-decl-kinds uses over to use this new file.
Douglas Gregor [Mon, 2 Feb 2009 22:11:10 +0000 (22:11 +0000)]
Steve set me straight on this one. GCC was right, EDG was wrong: the
direct-initialization following a user-defined conversion can select
any constructor; it just can't employ any user-defined
conversions. So we ban those conversions and classify the constructor
call based on the relationship between the "from" and "to" types in
the conversion.
Douglas Gregor [Mon, 2 Feb 2009 21:35:47 +0000 (21:35 +0000)]
Add iterators to LookupResult, allowing one to iterate over the
non-ambiguous name lookup results without allocating any memory, e.g.,
for sets of overloaded functions.
Douglas Gregor [Mon, 2 Feb 2009 18:25:48 +0000 (18:25 +0000)]
Split specific_decl_iterator, which had a run-time field for
determining what decls are acceptable, into specific_decl_iterator
(in which all decls matching the SpecificDecl type requirements are
acceptable) and filtered_decl_iterator (which also does a run-time
check via a member pointer non-type template parameter). This saves
some space in the iterators.
Steve Naroff [Mon, 2 Feb 2009 18:24:29 +0000 (18:24 +0000)]
Change the ObjC type encoding for block pointer types to "@?" (for consistency with GCC).
This fixes <rdar://problem/6538564> clang ObjC rewriter: Wrong encoding emitted for methods with Block parameters.
Steve Naroff [Mon, 2 Feb 2009 17:19:26 +0000 (17:19 +0000)]
RewriteObjC::RewriteBlockDeclRefExpr(): Add parens to enforce precedence. This fixes <rdar://problem/6529468> clang ObjC rewriter: Need parenthesis around dereferences in rewritten Blocks.
Also changed RewriteObjC::SynthesizeBlockFunc() to declare a pointer to the block argument even when there are no user-supplied arguments to the block.
Daniel Dunbar [Sat, 31 Jan 2009 00:06:58 +0000 (00:06 +0000)]
x86_64 ABI: Retool classification to compute lo & hi classifications
in terms of where the type resides in the containing object. This is a
more clear embodiement of the spec & fixes a merging issue with
unions. Down to 3/1000 failures.
Douglas Gregor [Fri, 30 Jan 2009 23:27:23 +0000 (23:27 +0000)]
Improve our handling of the second step in a user-defined conversion
sequence. Previously, we weren't permitting the second step to call
copy constructors, which left user-defined conversion sequences
surprisingly broken.
Now, we perform overload resolution among all of the constructors, but
only accept the result if it makes the conversion a standard
conversion. Note that this behavior is different from both GCC and EDG
(which don't agree with each other, either); I've submitted a core
issue on the matter.
Daniel Dunbar [Fri, 30 Jan 2009 22:40:15 +0000 (22:40 +0000)]
x86_64 ABI: Fix more thinkos, straddling computation for complex was
computing in bytes not bits. We are now down to 22/1000 failures on
the return types tests, and 18 of those are gcc bugs I believe.
Douglas Gregor [Fri, 30 Jan 2009 22:26:29 +0000 (22:26 +0000)]
Upgrade the "excess elements in array initializer" warning to an
error, since both C99 and C++ consider it an error. For reference, GCC
makes this a warning while G++ makes it an error.
Douglas Gregor [Fri, 30 Jan 2009 22:09:00 +0000 (22:09 +0000)]
Implement and test aggregate initialization in C++. Major changes:
- Support initialization of reference members; complain if any
reference members are left uninitialized.
- Use C++ copy-initialization for initializing each element (falls
back to constraint checking in C)
- Make sure we diagnose when one tries to provide an initializer
list for a non-aggregate.
- Don't complain about empty initializers in C++ (they are permitted)
- Unrelated but necessary: don't bother trying to convert the
decl-specifier-seq to a type when we're dealing with a C++
constructor, destructor, or conversion operator; it results in
spurious warnings.
Ted Kremenek [Fri, 30 Jan 2009 21:35:30 +0000 (21:35 +0000)]
Fix horrible non-termination bug in LiveVariables. The issue was that
the liveness state of block-level expressions could oscillate because
of two issues:
- The initial value before a merge was not always set to "Top"
- The set of live block-level expressions is a union, not an intersection
Chris Lattner [Fri, 30 Jan 2009 17:41:53 +0000 (17:41 +0000)]
" Attached is a patch for TextDiagnosticPrinter that adds an optional
parameter that allows users to omit the printing of the source
location on a diagnostic. So basically it would omit the "abc.c:5:1: "
at the beginning of the line."
Douglas Gregor [Fri, 30 Jan 2009 17:31:00 +0000 (17:31 +0000)]
Switch Type::isAggregateType to use the C++ definition of "aggregate
type" rather than the C definition. We do this because both C99 and
Clang always use "aggregate type" as "aggregate or union type", and
the C++ definition includes union types.
Douglas Gregor [Fri, 30 Jan 2009 01:04:22 +0000 (01:04 +0000)]
Eliminated LookupCriteria, whose creation was causing a bottleneck for
LookupName et al. Instead, use an enum and a bool to describe its
contents.
Optimized the C/Objective-C path through LookupName, eliminating any
unnecessarily C++isms. Simplify IdentifierResolver::iterator, removing
some code and arguments that are no longer used.
Eliminated LookupDeclInScope/LookupDeclInContext, moving all callers
over to LookupName, LookupQualifiedName, or LookupParsedName, as
appropriate.
All together, I'm seeing a 0.2% speedup on Cocoa.h with PTH and
-disable-free. Plus, we're down to three name-lookup routines.
Ted Kremenek [Fri, 30 Jan 2009 00:08:43 +0000 (00:08 +0000)]
Fix a couple bugs:
- NonLoc::MakeVal() would use sizeof(unsigned) (literally) instead of consulting
ASTContext for the size (in bits) of 'int'. While it worked, it was a
conflation of concepts and using ASTContext.IntTy is 100% correct.
- RegionStore::getSizeInElements() no longer assumes that a VarRegion has the
type "ConstantArray", and handles the case when uses use ordinary variables
as if they were arrays.
- Fixed ElementRegion::getRValueType() to just return the rvalue type of its
"array region" in the case the array didn't have ArrayType.
- All of this fixes <rdar://problem/6541136>