r184166 added an X86_32 function in the middle of the SystemZ code.
The SystemZ port had been added only a couple of weeks earlier and
the original patch probably predated that.
Protect MMX specific declarations under a __MMX__ guard. This header can be
included on non-x86 architectures (e.g. ARM) which do not support the MMX ISA.
Use the preprocessor to prevent these declarations from being processed.
Although the functions are marked as always_inline, the compiler with which they
are used may not honour the extended attributes and emit them as functions. In
such a case, indicate that they should have extern "C" linkage and should not be
mangled in C++ style if used within C++.
[ASan] Completely remove sanitizer blacklist file from instrumentation pass.
All blacklisting logic is now moved to the frontend (Clang).
If a function (or source file it is in) is blacklisted, it doesn't
get sanitize_address attribute and is therefore not instrumented.
If a global variable (or source file it is in) is blacklisted, it is
reported to be blacklisted by the entry in llvm.asan.globals metadata,
and is not modified by the instrumentation.
The latter may lead to certain false positives - not all the globals
created by Clang are described in llvm.asan.globals metadata (e.g,
RTTI descriptors are not), so we may start reporting errors on them
even if "module" they appear in is blacklisted. We assume it's fine
to take such risk:
1) errors on these globals are rare and usually indicate wild memory access
2) we can lazily add descriptors for these globals into llvm.asan.globals
lazily.
[Sanitizer] Remove brittle cache variable and slightly simplify blacklisting code.
Now CodeGenFunction is responsible for looking at sanitizer blacklist
(in CodeGenFunction::StartFunction) and turning off instrumentation,
if necessary.
[Sanitizer] Reduce the usage of sanitizer blacklist in CodeGenModule
Get rid of cached CodeGenModule::SanOpts, which was used to turn off
sanitizer codegen options if current LLVM Module is blacklisted, and use
plain LangOpts.Sanitize instead.
1) Some codegen decisions (turning TBAA or writable strings on/off)
shouldn't depend on the contents of blacklist.
2) llvm.asan.globals should *always* be created, even if the module
is blacklisted - soon Clang's CodeGen where we read sanitizer
blacklist files, so we should properly report which globals are
blacklisted to the backend.
Alp Toker [Mon, 7 Jul 2014 22:42:03 +0000 (22:42 +0000)]
libclang: pass return code out argument by reference
r212427 formalized the message-passing pattern by making these argument
structures const. This commit changes output arguments to get passed by
reference so we can eliminate mutable fields.
Revert "Add lib/Codgen/[CGRTTI|MicrosoftRTTI].cpp to CMakeLists.txt (r212445)"
I misidentified the problem and did not test suffifiently. The files had
correctly been removed, but for some reason they still remained in my git
checkout. Not adding the files back to CMakeLists.txt, but rather removing
the stale files was the solution. Sorry for the unnecessary noise.
David Majnemer [Mon, 7 Jul 2014 08:09:15 +0000 (08:09 +0000)]
CodeGen: Cleanup MS ABI RTTI LLVM IR
The MS ABI RTTI emission code would choose names for IR types like
%"MSRTTITypeDescriptor\02". This name is undesirable because it
requires escaping; the underlying reason for this is that the name is
unprintable. Fix this by naming it %rtti.TypeDescriptor2.
While here, stop trying to do lookups in the LLVM Module's type table.
Instead, store the IR types in MicrosoftCXXABI. Lookups by name aren't
particularly fast.
Richard Smith [Mon, 7 Jul 2014 06:38:20 +0000 (06:38 +0000)]
Fix an iterator invalidation issue: deserializing a key function can write to
the key functions table. Don't hold references to anything within that table
across such an access.
David Majnemer [Mon, 7 Jul 2014 06:20:47 +0000 (06:20 +0000)]
CodeGen: Refactor RTTI emission
Let's not expose ABI specific minutia inside of CodeGenModule and Type.
Instead, let's abstract it through CXXABI.
This gets rid of:
CodeGenModule::getCompleteObjectLocator,
CodeGenModule::EmitFundamentalTypeDescriptor{s,},
CodeGenModule::getMSTypeDescriptor,
CodeGenModule::getMSCompleteObjectLocator,
CGCXXABI::shouldRTTIBeUnique,
CGCXXABI::classifyRTTIUniqueness.
CGRTTI was *almost* entirely centered around providing Itanium-style
RTTI information. Instead of providing interfaces that only it
consumes, move it to the ItaniumCXXABI implementation file. This allows
it to have access to Itanium-specific implementation details without
providing useless expansion points for the Microsoft ABI side.
Alp Toker [Mon, 7 Jul 2014 06:05:00 +0000 (06:05 +0000)]
Fix layering of file remapping and header search initialization
These two functions initialize the source manager and header search objects and
shouldn't be in InitPreprocessor which is concerned with priming the
preprocessor itself and predefining macros.
StmtDumper.cpp is called ASTDumper.cpp these days, and usually works well
for new AST nodes without changes. There's now DataRecursiveASTVisitor
in addition to RecursiveASTVisitor, and serialization wasn't mentioned.
Sema: Check that __leave is contained in a __try block.
Give scope a SEHTryScope bit, set that in ParseSEHTry(), and let Sema
walk the scope chain to find the SEHTry parent on __leave statements.
(They are rare enough that it seems better to do the walk instead of
giving Scope a SEHTryParent pointer -- this is similar to AtCatchScope.)
The MicrosoftExtensions.c test file should not rely on -x objective-c++. Removed that from the RUN line, fixed obvious C++isms in the code, and moved some C++ tests into MicrosoftExtensions.cpp instead. As a drive-by, changed the line endings for MicrosoftExtensions.c to be LF instead of CRLF.
Alp Toker [Sun, 6 Jul 2014 06:24:00 +0000 (06:24 +0000)]
libclang: make darwin pthread hacks conditional on LLVM_ENABLE_THREADS
Although these aren't strictly related to LLVM's core threading, it's
reasonable to avoid pthread usage in clang when building with
LLVM_ENABLE_THREADS disabled.
CodeGen: add support for a few MSVC ARM intrinsics
This adds support for simple MSVC compatibility mode intrinsics. These
intrinsics are simple in that they are either directly passed through to the
annotated MSBuiltin intrinsic or they mirror existing GCC builtins.
This completes the infrastructure for the new MSBuiltin aliases in the
instruction definitions. These behave similar to the GCCBuiltin in that they
can be implicitly constructed without special handling unless needed.
With this change it is possible to annotate an LLVM intrinsic in the backend
instruction definitions and indicate it as a builtin in the Builtin*.def files
in clang via LANGBUILTIN. That will automatically pass through the instruction
much as a GCCBuiltin.
Note that there is no need for the special handling for ensuring that the
compatibility flag is enabled since the filtering on the LANGBUILTIN will
automatically prevent the intrinsic from bleeding into non-MS compatible
compiler invocations.
Robert Lytton [Thu, 3 Jul 2014 09:30:33 +0000 (09:30 +0000)]
Move the calling of emitTargetMD() later.
Summary:
Because a global created by GetOrCreateLLVMGlobal() is not finalised until later viz:
extern char a[];
char f(){ return a[5];}
char a[10];
Change MangledDeclNames to use a MapVector rather than a DenseMap so that the
Metadata is output in order of original declaration, so to make deterministic
and improve human readablity.
David Majnemer [Thu, 3 Jul 2014 05:51:27 +0000 (05:51 +0000)]
MS ABI: Get linkage of RTTI data correct
The Itanium rules are not appropriate for the MS ABI. RTTI data is
_never_ imported and thus is never available_externally. It is either
internal (if the type's linkage is internal) or linkonce_odr.
This also means that classes which inherit from dllimport'd bases force
their translation unit to duplicate the entirety of the RTTI data of
that base.
Interestingly, the complete object locator can never be referenced by
translation units which import the class.
Enable clang to continue to parse libstdc++4.6 and stlport after r210091.
r210091 made initialization checking more strict in c++11 mode. LWG2193 is
about changing standard libraries to still be valid under these new rules,
but older libstdc++ (e.g. libstdc++4.6 in -D_GLIBCXX_DEBUG=1 mode, or stlport)
do not implement that yet. So fall back to the C++03 semantics for container
classes in system headers below the std namespace.
Fix linkage bug that miscompiled variable templates instantiated from similarly named local types. In essence, this bug ensures that the x<Foo> specialization in function foo() defined as follows differs between two distinct translation units.
static int &foo() {
struct Foo { };
return x<Foo>;
}
This extends the target builtin support to allow language specific annotations
(i.e. LANGBUILTIN). This is to allow MSVC compatibility whilst retaining the
ability to have EABI targets use a __builtin_ prefix. This is merely to allow
uniformity in the EABI case where the unprefixed name is provided as an alias in
the header.