David Chisnall [Mon, 16 Jan 2012 17:27:18 +0000 (17:27 +0000)]
Some improvements to the handling of C11 atomic types:
- Add atomic-to/from-nonatomic cast types
- Emit atomic operations for arithmetic on atomic types
- Emit non-atomic stores for initialisation of atomic types, but atomic stores and loads for every other store / load
- Add a __atomic_init() intrinsic which does a non-atomic store to an _Atomic() type. This is needed for the corresponding C11 stdatomic.h function.
- Enables the relevant __has_feature() checks. The feature isn't 100% complete yet, but it's done enough that we want people testing it.
Still to do:
- Make the arithmetic operations on atomic types (e.g. Atomic(int) foo = 1; foo++;) use the correct LLVM intrinsic if one exists, not a loop with a cmpxchg.
- Add a signal fence builtin
- Properly set the fenv state in atomic operations on floating point values
- Correctly handle things like _Atomic(_Complex double) which are too large for an atomic cmpxchg on some platforms (this requires working out what 'correctly' means in this context)
- Fix the many remaining corner cases
Douglas Gregor [Sun, 15 Jan 2012 18:17:48 +0000 (18:17 +0000)]
Now that deserializing a definition of a C++ class/Objective-C
class/Objective-C protocol suffices get all of the redeclarations of
that declaration wired to the definition, we no longer need to record
the identity of the definition in every declaration. Instead, just
record a bit to indicate whether a particular declaration is the
definition.
Douglas Gregor [Sun, 15 Jan 2012 18:08:05 +0000 (18:08 +0000)]
When deserializing the definition of a C++ class/ObjC class/ObjC
protocol, record the definition pointer in the canonical declaration
for that entity, and then propagate that definition pointer from the
canonical declaration to all other deserialized declarations. This
approach works well even when deserializing declarations that didn't
know about the original definition, which can occur with modules.
A nice bonus from this definition-deserialization approach is that we
no longer need update records when a definition is added, because the
redeclaration chains ensure that the if any declaration is loaded, the
definition will also get loaded.
Douglas Gregor [Sun, 15 Jan 2012 16:58:34 +0000 (16:58 +0000)]
Completely re-implement (de-)serialization of redeclaration
chains, again. The prior implementation was very linked-list oriented, and
the list-splicing logic was both fairly convoluted (when loading from
multiple modules) and failed to preserve a reasonable ordering for the
redeclaration chains.
This new implementation uses a simpler strategy, where we store the
ordered redeclaration chains in an array-like structure (indexed based
on the first declaration), and use that ordering to add individual
deserialized declarations to the end of the existing chain. That way,
the chain mimics the ordering from its modules, and a bug somewhere is
far less likely to result in a broken linked list.
Chandler Carruth [Sun, 15 Jan 2012 09:03:45 +0000 (09:03 +0000)]
Two variables had been added for an assert, but their values were
re-computed rather than the variables be re-used just after the assert.
Just use the variables since we have them already. Fixes an unused
variable warning.
Richard Smith [Sun, 15 Jan 2012 06:24:57 +0000 (06:24 +0000)]
decltype(e) is type-dependent if e is instantiation-dependent. Scary but true.
Don't consider decltype(e) for an instantiation-dependent, but not
type-dependent, e to be non-type-dependent but canonical(!).
Richard Smith [Sun, 15 Jan 2012 03:51:30 +0000 (03:51 +0000)]
Pedantic diagnostic correction: in C++, we have integral constant expressions,
not integer constant expressions. In passing, fix the 'folding is an extension'
diagnostic to not claim we're accepting the code, since that's not true in
-pedantic-errors mode, and add this diagnostic to -Wgnu.
Richard Smith [Sun, 15 Jan 2012 03:25:41 +0000 (03:25 +0000)]
constexpr: casts to void* are allowed in constant expressions, don't set the
designator invalid. (Since we can't read the value of such a pointer, this only
affects the quality of diagnostics.)
Eli Friedman [Sun, 15 Jan 2012 01:23:58 +0000 (01:23 +0000)]
Change linkage computation so it doesn't depend on FunctionDecl::isExternC or VarDecl::isExternC, and instead queries what it actually cares about: whether the given declaration is inside an extern "C" context. Fundamentally, figuring out whether a function/variable uses C linkage requires knowing the linkage, and the logic in FunctionDecl::isExternC and VarDecl::isExternC was getting it wrong. Given that, fix FunctionDecl::isExternC and VarDecl::isExternC to use much simpler implementations that depend on the fixed linkage computation.
Fixes a regression to test/SemaCXX/linkage.cpp caused by a new warning exposing the fact that the internal state was wrong.
Benjamin Kramer [Sat, 14 Jan 2012 19:31:39 +0000 (19:31 +0000)]
Give OverloadCandidateSet the responsibility for destroying the implicit conversion sequences so we don't get double frees when the vector reallocates.
Douglas Gregor [Sat, 14 Jan 2012 16:38:05 +0000 (16:38 +0000)]
De-virtualize getPreviousDecl() and getMostRecentDecl() when we know
we have a redeclarable type, and only use the new virtual versions
(getPreviousDeclImpl() and getMostRecentDeclImpl()) when we don't have
that type information. This keeps us from penalizing users with strict
type information (and is the moral equivalent of a "final" method).
Plus, settle on the names getPreviousDecl() and getMostRecentDecl()
throughout.
Benjamin Kramer [Sat, 14 Jan 2012 16:32:05 +0000 (16:32 +0000)]
Pool allocate ImplicitConversionSequences.
To avoid malloc thrashing give OverloadCandidateSet an inline capacity for conversion sequences.
We use the fact that OverloadCandidates never outlive the OverloadCandidateSet and have a fixed
amount of conversion sequences.
This eliminates the oversized SmallVector from OverloadCandidate shrinking it from 752 to 208 bytes.
On the test case from the "Why is CLANG++ so freaking slow" thread on llvmdev this avoids one gig
of vector reallocation (including memcpy) which translates into 5-10% speedup on Lion/x86_64.
Overload candidate computation is still the biggest malloc contributor when compiling templated
c++ code.
Douglas Gregor [Sat, 14 Jan 2012 15:55:47 +0000 (15:55 +0000)]
Introduce Decl::getPreviousDecl() and Decl::getMostRecentDecl(),
virtual functions that provide previous/most recent redeclaration
information for any declaration. Use this to eliminate the redundant,
less efficient getPreviousDecl() functions.
Douglas Gregor [Sat, 14 Jan 2012 15:13:49 +0000 (15:13 +0000)]
Reimplement RedeclarableTemplateDecl in terms of
Redeclarable<RedeclarableTemplateDecl>, eliminating a bunch of
redeclaration-chain logic both in RedeclarableTemplateDecl and
especially in its (de-)serialization.
As part of this, eliminate the RedeclarableTemplate<> class template,
which was an abstraction that didn't actually save anything.
Richard Smith [Sat, 14 Jan 2012 04:30:29 +0000 (04:30 +0000)]
constexpr irgen: Add irgen support for APValue::Struct, APValue::Union,
APValue::Array and APValue::MemberPointer. All APValue values can now be emitted
as constants.
Add new CGCXXABI entry point for emitting an APValue MemberPointer. The other
entrypoints dealing with constant member pointers are no longer necessary and
will be removed in a later change.
Switch codegen from using EvaluateAsRValue/EvaluateAsLValue to
VarDecl::evaluateValue. This performs caching and deals with the nasty cases in
C++11 where a non-const object's initializer can refer indirectly to
previously-initialized fields within the same object.
Building the intermediate APValue object incurs a measurable performance hit on
pathological testcases with huge initializer lists, so we continue to build IR
directly from the Expr nodes for array and record types outside of C++11.
Rafael Espindola [Sat, 14 Jan 2012 00:30:36 +0000 (00:30 +0000)]
Remember if a type has its visibility set explicitly or implicitly.
With that, centralize the way we merge visibility, always preferring explicit over
implicit and then picking the most restrictive one.
Fixes pr10113 and pr11690.
Douglas Gregor [Fri, 13 Jan 2012 23:49:34 +0000 (23:49 +0000)]
(Implicit) parameters deserialized as part of a function type must not
get added to the identifier chains as part of deserialization, because
they should not be visible to name lookup.
Eli Friedman [Fri, 13 Jan 2012 23:41:25 +0000 (23:41 +0000)]
Progress towards making isUsed() reflect whether a declaration is odr-used; don't set isUsed for local variables which are referenced in unevaluated contexts. Make other code use isReferenced() (which basically indicates that a declaration isn't dead) where appropriate.
I was forced to change test/SemaCXX/linkage.cpp because we aren't actually modeling extern "C" in the AST the way that testcase expects; we were not printing a warning only because we skipped the relevant check. Someone who actually understands the semantics here should fix that.
Kaelyn Uhrain [Fri, 13 Jan 2012 23:10:36 +0000 (23:10 +0000)]
Convert SemaTemplate*.cpp to pass a callback object to CorrectTypo.
The change to SemaTemplateVariadic.cpp improves the typo correction
results in certain situations, while the change to SemaTemplate.cpp
does not change existing behavior.
Douglas Gregor [Fri, 13 Jan 2012 23:06:53 +0000 (23:06 +0000)]
Make sure to consider non-DeclContext scopes properly when finding
multiple name lookup results in C/Objective-C. Fixes a regression a
caused in r147533, found by Enea Zaffanella!
Douglas Gregor [Fri, 13 Jan 2012 22:31:52 +0000 (22:31 +0000)]
When inferring a module for a framework, first determine whether that
framework is actually a subframework within a top-level framework. If
so, only infer a module for the top-level framework and then dig out
the appropriate submodule.
This helps us cope with an amusing subframeworks anti-pattern, where
one uses -F <framework>/Frameworks to get direct include access to the
subframeworks of a framework (which otherwise would not be
permitted).
Anna Zaks [Fri, 13 Jan 2012 21:52:01 +0000 (21:52 +0000)]
Move identification of memory setting and copying functions (memset,
memcmp, strncmp,..) out of Sema and into FunctionDecl so that the logic
could be reused in the analyzer.
Sebastian Pop [Fri, 13 Jan 2012 20:37:10 +0000 (20:37 +0000)]
remove assertions in the Hexagon backend specific clang driver
Patch from Jyotsna Verma:
I have made the changes to remove assertions in the Hexagon backend
specific clang driver. Instead of asserting on invalid arch name, it has
been modified to use the default value.
I have changed the implementation of the CPU flag validation for the
Hexagon backend. Earlier, the clang driver performed the check and
asserted on invalid inputs. In the new implementation, the driver passes
the last CPU flag (or sets to "v4" if not specified) to the compiler (and
also to the assembler and linker which perform their own check) instead of
asserting on incorrect values. This patch changes the setCPU function for
the Hexagon backend in clang/lib/Basic/Targets.cpp which causes the
compiler to error out on incorrect CPU flag values.
Eli Friedman [Fri, 13 Jan 2012 02:20:01 +0000 (02:20 +0000)]
A few minor improvements to error recovery trying to access member of a function. In particular, this restores the cool error recovery for the example from http://blog.llvm.org/2010/04/amazing-feats-of-clang-error-recovery.html , which regressed a few months back.
Kaelyn Uhrain [Fri, 13 Jan 2012 01:32:50 +0000 (01:32 +0000)]
Fix up the calls to CorrectTypo in Sema*ObjC.cpp to use callback
objects, and add a basic CorrectionCandidateCallback template class
to simplify the fixups.
Douglas Gregor [Fri, 13 Jan 2012 01:20:43 +0000 (01:20 +0000)]
When inferring a module map for a framework, add the 'private'
requirement to headers under PrivateHeaders. We don't want to build
them as part of the module (yet).
Richard Smith [Thu, 12 Jan 2012 23:53:29 +0000 (23:53 +0000)]
Improve 0-argument -Wvexing-parse diagnostic by adding notes with fix-its:
- If the declarator is at the start of a line, and the previous line contained
another declarator and ended with a comma, then that comma was probably a
typo for a semicolon:
int n = 0, m = 1, l = 2, // k = 5;
myImportantFunctionCall(); // oops!
- If removing the parentheses would correctly initialize the object, then
produce a note suggesting that fix.
- Otherwise, if there is a simple initializer we can suggest which performs
value-initialization, then provide a note suggesting a correction to that
initializer.
Sema::Declarator now tracks the location of the comma prior to the declarator in
the declaration, if there is one, to facilitate providing the note. The code to
determine an appropriate initializer from the -Wuninitialized warning has been
factored out to allow use in both that and -Wvexing-parse.
Ted Kremenek [Thu, 12 Jan 2012 19:25:46 +0000 (19:25 +0000)]
[analyzer] fix inlining's handling of mapping actual to formal arguments and limit the call stack depth. The analyzer can now accurately simulate factorial for limited depths.
Richard Smith [Thu, 12 Jan 2012 18:54:33 +0000 (18:54 +0000)]
constexpr: initialization of a union from an empty initializer-list should
zero-initialize the first union member. Also fix a bug where initializing an
array of types compatible with wchar_t from a wide string literal failed in C,
and fortify the C++ tests in this area. This part can't be tested without a code
change to enable array evaluation in C (where an existing test fails).
Douglas Gregor [Thu, 12 Jan 2012 16:11:24 +0000 (16:11 +0000)]
In Objective-C++, actually compute the base type of a member access
expression for an Objective-C object or pointer type, so that we don't
attempt to treat the member name as a template. Fixes
<rdar://problem/10672501>.
Add IsImplicit field in ObjCMessageExpr that is true when the message
was constructed, e.g. for a property access.
This allows the selector identifier locations machinery for ObjCMessageExpr
to function correctly, in that there are not real locations to handle/report for
such a message.
[arcmt] The migrator tests for the buildbot in http://lab.llvm.org:8011/builders/clang-native-mingw32-win7/
are messed up, XFAIL does not help. Waiting until DISABLE is supported..
Anna Zaks [Thu, 12 Jan 2012 02:22:34 +0000 (02:22 +0000)]
[analyzer] Add taint transfer by strcpy & others (part 1).
To simplify the process:
Refactor taint generation checker to simplify passing the
information on which arguments need to be tainted from pre to post
visit.
Todo: We need to factor out the code that sema is using to identify the
string and memcpy functions and use it here and in the CString checker.