Matthew Simpson [Sun, 7 Feb 2016 17:14:03 +0000 (17:14 +0000)]
Make -fno-math-builtin a cc1 option
This patch makes -fno-math-builtin a frontend only option instead of a driver
option. The appropriate test case was committed in r186899 when the flag was
introduced. This should fix PR26317.
Contributed-by: Frank Herrmann <fgh@4gh.tv>
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@260044 91177308-0d34-0410-b5e6-96231b3b80d8
Devin Coughlin [Sun, 7 Feb 2016 16:55:44 +0000 (16:55 +0000)]
[analyzer] Invalidate destination of std::copy() and std::copy_backward().
Now that the libcpp implementations of these methods has a branch that doesn't call
memmove(), the analyzer needs to invalidate the destination for these methods explicitly.
Adjust the driver to invoke the linker more similar to gcc. -dynamic-linker is
only passed if -static and -shared are not part of the compiler (driver)
invocation. Replicate the passing of -export-rdynamic as per the GCC link spec:
The ivar ref would be transformed by the Typo Correction TreeTransform, but not
be owned, resulting in the source location being invalid. This would eventually
lead to an assertion in findCapturingExpr. Prevent this assertion from
triggering.
Sema: handle typo correction with ARC'ed objc properties
We would previously assert in findCapturingExpr when performing a typo
correction resulting in an assignment of an ObjC property with a strong lifetype
specifier due to the expression not being rooted in the file (invalid SLoc)
during the retain cycle check on the typo-corrected expression. Handle the
expression type appropriately during the TreeTransform to ensure that we have a
source location associated with the expression.
Devin Coughlin [Sat, 6 Feb 2016 17:17:32 +0000 (17:17 +0000)]
[analyzer] DeallocChecker: Don't warn on release of readonly assign property in dealloc.
It is common for the ivars for read-only assign properties to always be stored retained,
so don't warn for a release in dealloc for the ivar backing these properties.
Samuel Antao [Sat, 6 Feb 2016 02:12:34 +0000 (02:12 +0000)]
[OpenMP] Reorganize code to allow specialized code generation for different devices.
Summary:
Different devices may in some cases require different code generation schemes in order to implement OpenMP. This is required not only for performance reasons, but also because it may not be possible to have the current (default) implementation working for these devices. E.g. GPU's cannot implement the same scheme a target such as powerpc or x86b would use, in the sense that it does not have the ability to fork threads, instead all the threads are always executing and need to be managed by the implementation.
This patch proposes a reorganization of the code in the OpenMP code generation to pave the way to have specialized implementation of OpenMP support. More than a "real" patch this is more a request for comments in order to understand if what is proposed is acceptable or if there are better/easier ways to do it.
In this patch part of the common OpenMP codegen infrastructure is moved to a new file under a new namespace (CGOpenMPCommon) so it can be shared between the default implementation and the specialized one. When CGOpenMPRuntime is created, an attempt to select a specialized implementation is done.
In the patch a specialization for nvptx targets is done which currently checks if the target is an OpenMP device and trap if it is not.
Adrian Prantl [Sat, 6 Feb 2016 01:59:09 +0000 (01:59 +0000)]
Fix a crash when emitting dbeug info for forward-declared scoped enums.
It is possible for enums to be created as part of their own
declcontext. We need to cache a placeholder to avoid the type being
created twice before hitting the cache.
Richard Trieu [Fri, 5 Feb 2016 23:02:38 +0000 (23:02 +0000)]
Exempt char array initializers from -Wconstant-converion.
Sometimes, char arrays are used as bit storage, with no difference made between
signed and unsigned char. Thus, it is reasonable to use 0 to 255 instead of
-128 to 127 and not trigger this warning.
Define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_[1248] macros on SystemZ.
This fixes a miscompile of GCC C++11 standard library headers
due to use of those macros in an ABI-changing manner.
See e.g. /usr/include/c++/4.8.5/ext/concurrence.h:
// Compile time constant that indicates prefered locking policy in
// the current configuration.
static const _Lock_policy __default_lock_policy =
#ifdef __GTHREADS
#if (defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_2) \
&& defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4))
_S_atomic;
#else
_S_mutex;
#endif
#else
_S_single;
#endif
A different choice of __default_lock_policy causes different
sizes of several of the C++11 data structures, which are then
incompatible when inlined in clang-compiled code with what the
(GCC-compiled) external library expects.
This in turn leads to various crashes when using std::thread
in code compiled with clang, as see e.g. via the ThreadPool
unit tests. See PR 26473 for an example.
[Parser] Perform CachedTokens update dependent on token consumption
In the context where we break one tok::greatergreater into two
tok::greater in order to correctly update the cached tokens; update the
CachedTokens with two tok::greater only if ParseGreaterThanInTemplateList
clients asks to consume the last token. Otherwise we only need to add
one because the second is already added later on, as a not yet cached token.
Richard Smith [Fri, 5 Feb 2016 19:03:40 +0000 (19:03 +0000)]
[modules] Separately track whether an identifier's preprocessor information and
name lookup information have changed since deserialization. For a C++ modules
build, we do not need to re-emit the identifier into the serialized identifier
table if only the name lookup information has changed (and in all cases, we
don't need to re-emit the macro information if only the name lookup information
has changed).
[ASTMatchers] Allow hasName() to look through inline namespaces
Summary:
Allow hasName() to look through inline namespaces.
This will fix the interaction between some clang-tidy checks and libc++.
libc++ defines names in an inline namespace named std::<version_#>.
When we try to match a name using hasName("std::xxx") it fails to match and the clang-tidy check does not work.
Because the Decl is explicitly passed as nullptr further up the call chain, it
is possible to invoke isa on a nullptr, which will assert. Guard against the
nullptr.
Take the opportunity to reuse the helper method rather than re-implementing this
logic.
Chris Bieneman [Fri, 5 Feb 2016 03:02:40 +0000 (03:02 +0000)]
[CMake] One more try to fix this.
This change will catch any bots that generated the order file that GNU ld doesn't like and delete it before trying to generate one that I think GNU ld will deal with.
Richard Smith [Fri, 5 Feb 2016 01:40:54 +0000 (01:40 +0000)]
[modules] Factor out common code to mark identifier being "from AST", and add a
call in one more place to reduce the size of identifier tables in non-leaf
modules. No behavior change.
Chris Bieneman [Fri, 5 Feb 2016 01:22:03 +0000 (01:22 +0000)]
[CMake] Improve the clang order-file generation workflow
Summary:
With this change generating clang order files using dtrace uses the following workflow:
cmake <whatever options you want>
ninja generate-order-file
ninja clang
This patch works by setting a default path to the order file (which can be overridden by the user). If the order file doesn't exist during configuration CMake will create an empty one.
CMake then ties up the dependencies between the clang link job and the order file, and generate-order-file overwrites CLANG_ORDER_FILE with the new order file.
Ben Langmuir [Fri, 5 Feb 2016 01:10:05 +0000 (01:10 +0000)]
Don't synthesize an ImportDecl for a module named in -fmodule-implementation-of
When building a PCH with modules enabled this import would assert in the
ASTWriter and (if assertions were disabled) sometimes crash the compiler
that loaded the resulting PCH when trying to lookup the submodule ID.
Richard Smith [Thu, 4 Feb 2016 22:54:41 +0000 (22:54 +0000)]
PR25271: When attaching default template arguments to redeclarations of a
template, keep looking for default arguments if we see a template parameter
pack. There may be default arguments preceding a pack with no default argument.
Rong Xu [Thu, 4 Feb 2016 18:39:09 +0000 (18:39 +0000)]
[PGO] cc1 option name change for profile instrumentation
This patch changes cc1 option -fprofile-instr-generate to an enum option
-fprofile-instrument={clang|none}. It also changes cc1 options
-fprofile-instr-generate= to -fprofile-instrument-path=.
The driver level option -fprofile-instr-generate and -fprofile-instr-generate=
remain intact. This change will pave the way to integrate new PGO
instrumentation in IR level.
Alexey Bataev [Thu, 4 Feb 2016 11:27:03 +0000 (11:27 +0000)]
[OPENMP 4.0] Fixed support of array sections/array subscripts.
Codegen for array sections/array subscripts worked only for expressions with arrays as base. Patch fixes codegen for bases with pointer/reference types.
Samuel Antao [Thu, 4 Feb 2016 08:13:16 +0000 (08:13 +0000)]
Add -nocudainc option to CUDA preprocessor test.
If include files are used in the CUDA preprocessor tests it will cause a
failure due to a missing header file in hosts that do not match the triple
in the test. E.g. powerpc64le have CUDA support but the include files
cannot be used for an x86 target.
Alexey Bataev [Thu, 4 Feb 2016 04:22:09 +0000 (04:22 +0000)]
PR23057: fix use-after-free due to local token buffer in ParseCXXAmbiguousParenExpression, by Dmitry Polukhin
Differential Revision: http://reviews.llvm.org/D16572
A test/Parser/cxx-ambig-paren-expr-asan.cpp
M lib/Parse/ParseExprCXX.cpp
Ben Langmuir [Thu, 4 Feb 2016 00:55:24 +0000 (00:55 +0000)]
Fix predefine for __NSConstantString struct type
Per review feedback the name was wrong and it can be used outside
Objective-C.
Unfortunately, making the internal struct visible broke some ASTMatchers
tests that assumed that the first record decl would be from user code,
rather than a builtin type. I'm worried that this will also affect
users' code. So this patch adds a typedef to wrap the internal struct
and only makes the typedef visible to namelookup. This is sufficient to
allow the ASTReader to merge the decls we need without making the struct
itself visible.
Reapply r259624, it is likely not the commit causing the bot failures.
Original message:
Make CF constant string decl visible to name lookup to fix module errors
The return type of the __builtin___*StringMakeConstantString functions
is a pointer to a struct, so we need that struct to be visible to name
lookup so that we will correctly merge multiple declarations of that
type if they come from different modules.
Incidentally, to make this visible to name lookup we need to rename the
type to __NSConstantString, since the real NSConstantString is an
Objective-C interface type. This shouldn't affect anyone outside the
compiler since users of the constant string builtins cast the result
immediately to CFStringRef.
Since this struct type is otherwise implicitly created by the AST
context and cannot access namelookup, we make this a predefined type
and initialize it in Sema.
Note: this issue of builtins that refer to types not visible to name
lookup technically also affects other builtins (e.g. objc_msgSendSuper),
but in all other cases the builtin is a library builtin and the issue
goes away if you include the library that defines the types it uses,
unlike for these constant string builtins.
[cfi] Safe handling of unaddressable vtable pointers (clang).
Avoid crashing when printing diagnostics for vtable-related CFI
errors. In diagnostic mode, the frontend does an additional check of
the vtable pointer against the set of all known vtable addresses and
lets the runtime handler know if it is safe to inspect the vtable.
Richard Smith [Wed, 3 Feb 2016 20:40:30 +0000 (20:40 +0000)]
Ensure that we substitute into the declaration of a template parameter pack
(that is not a pack expansion) during template argument deduction, even if we
deduced that the pack would be empty.
Douglas Gregor [Wed, 3 Feb 2016 19:13:08 +0000 (19:13 +0000)]
[Sema debugger support] Require non-void types to be complete in unknown-anytype casts.
When performing a cast from an __unknown_anytype function call to a
non-void type, we need to make sure that type is complete. Fixes
rdar://problem/23959960.
Richard Smith [Wed, 3 Feb 2016 18:48:43 +0000 (18:48 +0000)]
Fix regression from r259622: the operand of an increment that is the operand of
a cast expression is not the operand of a cast expression itself, so if it's
parenthesized we need to form a ParenExpr not a ParenListExpr.
[OpenCL] Adding reserved operator logical xor for OpenCL
This patch adds the reserved operator ^^ when compiling for OpenCL (spec v1.1 s6.3.g),
which results in a more meaningful error message.
Patch by Neil Hickey!
Review: http://reviews.llvm.org/D13280
M test/SemaOpenCL/unsupported.cl
M include/clang/Basic/TokenKinds.def
M include/clang/Basic/DiagnosticParseKinds.td
M lib/Basic/OperatorPrecedence.cpp
M lib/Lex/Lexer.cpp
M lib/Parse/ParseExpr.cpp
Yury Gribov [Wed, 3 Feb 2016 13:35:33 +0000 (13:35 +0000)]
[analyzer] AnalysisConsumer: print fully-qualified function name while displaying progress
-analyzer-display progress option prints only function names which may be ambiguous. This patch forces AnalysisConsumer to print fully-qualified function names.
Patch by Alex Sidorin!
Marina Yatsina [Wed, 3 Feb 2016 11:32:08 +0000 (11:32 +0000)]
-inline-asm][X86] Add ability to use AVX512 in MS inline asm
Defined the new AVX512 registers in clang inline asm.
Fixed a bug in the MC subtarget info creation during the parsing of MS asm statement - now it receives the actual CPU and target features information.
Ben Langmuir [Wed, 3 Feb 2016 03:26:19 +0000 (03:26 +0000)]
Make CF constant string decl visible to name lookup to fix module errors
The return type of the __builtin___*StringMakeConstantString functions
is a pointer to a struct, so we need that struct to be visible to name
lookup so that we will correctly merge multiple declarations of that
type if they come from different modules.
Incidentally, to make this visible to name lookup we need to rename the
type to __NSConstantString, since the real NSConstantString is an
Objective-C interface type. This shouldn't affect anyone outside the
compiler since users of the constant string builtins cast the result
immediately to CFStringRef.
Since this struct type is otherwise implicitly created by the AST
context and cannot access namelookup, we make this a predefined type
and initialize it in Sema.
Note: this issue of builtins that refer to types not visible to name
lookup technically also affects other builtins (e.g. objc_msgSendSuper),
but in all other cases the builtin is a library builtin and the issue
goes away if you include the library that defines the types it uses,
unlike for these constant string builtins.
Richard Smith [Wed, 3 Feb 2016 02:58:20 +0000 (02:58 +0000)]
Fix miscompile and rejects-valids when disambiguating after an ambiguous
C-style-cast to function/array type or parenthesized function-style cast/array
indexing.
Richard Smith [Wed, 3 Feb 2016 01:43:59 +0000 (01:43 +0000)]
Fix typo in OpenCL type mangling. This is still bogus (we should either use the
actual source name of the typedef or a vendor mangling) but at least it stands
a chance of demangling now.
We would also benefit from some test coverage of this (OpenCL +
__attribute__((overloadable)) is probably required to reach this).
Richard Smith [Wed, 3 Feb 2016 01:32:42 +0000 (01:32 +0000)]
Fix Itanium RTTI emission so that we emit fundamental type information into the
C++ ABI library for the same set of types for which we expect the C++ ABI
library to provide the RTTI.
Specifically:
1) __int128 and unsigned __int128 are now emitted into the ABI library. We
always expected them to be there but never actually made sure to emit them.
2) Do not expect OpenCL builtin types to have type info in the C++ ABI library.
Neither libc++abi nor libstdc++ puts them there when built with either GCC
or Clang.
Richard Smith [Tue, 2 Feb 2016 23:58:56 +0000 (23:58 +0000)]
PR24989: Stop trying to use the C++11 rules for lambda return type inference in
C++14 generic lambdas. It conflicts with the C++14 return type deduction
mechanism, and results in us failing to actually deduce the lambda's return
type in some cases.
Richard Smith [Tue, 2 Feb 2016 23:11:49 +0000 (23:11 +0000)]
Work around build failure due to GCC 4.8.1 bug. We don't completely understand
the details of the bug, but avoiding overloading llvm::cast with another
function template sidesteps it.
See gcc.gnu.org/PR58022 for details of the bug, and llvm.org/PR26362 for more
backgound on how it manifested in Clang. Patch by Igor Sugak!
Artem Belevich [Tue, 2 Feb 2016 22:29:48 +0000 (22:29 +0000)]
[CUDA] Do not allow dynamic initialization of global device side variables.
In general CUDA does not allow dynamic initialization of
global device-side variables. One exception is that CUDA allows
records with empty constructors as described in section E2.2.1 of
CUDA 7.5 Programming guide.
This patch applies initializer checks for all device-side variables.
Empty constructors are accepted, but no code is generated for them.
Manman Ren [Tue, 2 Feb 2016 22:23:03 +0000 (22:23 +0000)]
ObjCXX: fix a crash during typo correction.
For ObjCXX, we can create a CastExpr with Kind being CK_UserDefinedConversion
and SubExpr being BlockExpr. Specifically one can return BlockExpr from
BuildCXXMemberCallExpr and the result can be used to build a CastExpr.
Fix the assumption in CastExpr::getSubExprAsWritten that SubExpr can only
be CXXMemberCallExpr.
Tim Northover [Tue, 2 Feb 2016 18:02:10 +0000 (18:02 +0000)]
ARM: allow both vfma and vfms intrinsics on v7.
The main purpose here is that vfma/vfms should be symmetric, and they are
supported on most v7 cores.
The new ArchGuard is suggested by ACLE but prophylactic for us. Almost all CPUs
with NEON *will* have vfma, and the few exceptions I know of (e.g. Cortex-A8)
are incorrectly modelled by Clang so can't trigger a test.
Fortunately, they're getting rarer. But if we ever do support them properly
arm_neon.h should now do the right thing.
Denis Zobnin [Tue, 2 Feb 2016 17:33:09 +0000 (17:33 +0000)]
PR23057: Fix assertion `Val && "isa<> used on a null pointer"' on invalid for-range expression.
Fix the issue discovered by fuzzing (PR23057, comment 18) by handling nullptr in Sema::ActOnCXXForRangeDecl
and correct delayed typos in for-range expression before calling Sema::ActOnCXXForRangeStmt. Also fixes PR26288.
Oliver Stannard [Tue, 2 Feb 2016 13:52:52 +0000 (13:52 +0000)]
Add backend dignostic printer for unsupported features
Re-commit of r258950 after fixing layering violation.
The related LLVM patch adds a backend diagnostic type for reporting
unsupported features, this adds a printer for them to clang.
In the case where debug location information is not available, I've
changed the printer to report the location as the first line of the
function, rather than the closing brace, as the latter does not give the
user any information. This also affects optimisation remarks.
[OpenCL] Eliminate warning when declaring OpenCL builtin functions.
OpenCL builtin functions are usually declared in header files.
Currently clang emits warning for OpenCL builtin functions
which have the same name as standard C library functions.
This commit eliminates such warnings by not adding the C standard
includes following the restriction from OpenCL v1.2 s6.9.f.
[OpenMP] Prevent nesting of target constructs within target code execution regions.
Summary:
This patch enhances Sema to check for the following restriction:
OpenMP 4.5 [2.17 Nesting of Regions]
If a target, target update, target data, target enter data, or
target exit data construct is encountered during execution of a
target region, the behavior is unspecified.