When modeling a call to a setter for a property that is synthesized to be
backed by an instance variable, don't invalidate the entire instance
but rather only the storage for the updated instance variable itself.
This still doesn't model the effect of the setter completely. It doesn't
bind the set value to the ivar storage location because doing so would cause
the set value to escape, removing valuable diagnostics about potential
leaks of the value from the retain count checker.
llvm-svn: 261243
Currently we return no results when completing inside of the brackets in
a 'char foo[]' declaration. Let the generic expression completion code
handle it instead. We could get fancier here (e.g. filter non-constant
expressions in contexts where VLAs are not allowed), but it's a strict
improvement over the existing version.
llvm-svn: 261217
Clang implements an enable_if attribute as an extension. Hook up `-Wpedantic`
to issue an extension usage warning when __enable_if__ is used.
llvm-svn: 261192
An optional nopartial can be placed after the platform name.
int bar() __attribute__((availability(macosx,nopartial,introduced=10.12))
When deploying back to a platform version prior to when the declaration was
introduced, with 'nopartial', Clang emits an error specifying that the function
is not introduced yet; without 'nopartial', the behavior stays the same: the
declaration is `weakly linked`.
A member is added to the end of AttributeList to save the location of the
'nopartial' keyword. A bool member is added to AvailabilityAttr.
The diagnostics for 'nopartial' not-yet-introduced is handled in the same way as
we handle unavailable cases.
Reviewed by Doug Gregor and Jordan Rose.
rdar://23791325
llvm-svn: 261163
thousands of modules, each of which declares the same namespace, linearly
scanning the redecl chain looking for a visible declaration (once for each leaf
module, for each use) performs very poorly. Namespace visibility can only
decrease when we leave a module during a module build step, and we never care
*which* visible declaration of a namespace we find, so we can cache this very
effectively.
This results in a 35x speedup on one of our internal build steps (2m -> 3.5s),
but is hard to unit test because it requires a very large number of modules.
Ideas for a test appreciated! No functionality change intended other than the
speedup.
llvm-svn: 261161
OpenCL Extension v1.2 s9.5 allows half precision floating point
type literals with suffices h or H when cl_khr_fp16 is enabled.
Example: half x = 1.0h;
Patch by Liu Yaxun (Sam)!
Differential Revision: http://reviews.llvm.org/D16865
llvm-svn: 261084
ptxas optimizations are disabled if we need to generate debug info
as ptxas does not accept '-g' otherwise.
Differential Revision: http://reviews.llvm.org/D17111
llvm-svn: 261018
Like cl.exe, clang-cl allows adding system include directories via the
INCLUDE env var. Having a driver flag for this functionality is useful,
so add this too.
(In the future, we probably also want to have a flag alternative to
VCINSTALLDIR as used in MSVCToolChain::getVisualStudioBinaries(), and
a way to override the registry accesses in MSVCToolChain::getWindowsSDKDir()
-- maybe -ivcroot= and -iwinsdkroot=?).
llvm-svn: 260990
Expressions inside 'schedule'|'dist_schedule' clause must be captured in
combined directives to avoid possible crash during codegen. Patch
improves handling of such constructs
llvm-svn: 260954
Enable it for USRs and names when indexing.
Forward references can have different template argument names; including them
makes USRs and names unstable, since the name depends on whether we saw a forward reference or not.
llvm-svn: 260866
This is a follow-up to PR26085. That was fixed in r257710 but the testcase
there was incomplete. There is a related issue where the overload resolution
for Objective-C incorrectly picks a method that is not valid without a
bridge cast. The call to Sema::CheckSingleAssignmentConstraints that was
added to SemaOverload.cpp's IsStandardConversion() function does not catch
that case and reports that the method is Compatible even when it is not.
The root cause here is that various Objective-C-related functions in Sema
do not consistently return a value to indicate whether there was an error.
This was fine in the past because they would report diagnostics when needed,
but r257710 changed them to suppress reporting diagnostics when checking
during overload resolution.
This patch adds a new ACR_error result to the ARCConversionResult enum and
updates Sema::CheckObjCARCConversion to return that value when there is an
error. Most of the calls to that function do not check the return value,
so adding this new result does not affect them. The one exception is in
SemaCast.cpp where it specifically checks for ACR_unbridged, so that is
also OK. The call in Sema::CheckSingleAssignmentConstraints can then check
for an ACR_okay result and identify assignments as Incompatible. To
preserve the existing behavior, it only changes the return value to
Incompatible when the new Diagnose argument (from r257710) is false.
Similarly, the CheckObjCBridgeRelatedConversions and
ConversionToObjCStringLiteralCheck need to identify when an assignment is
Incompatible. Those functions already return appropriate values but they
need some fixes related to the new Diagnose argument.
llvm-svn: 260787
This is an artefact of split-mode CUDA compilation that we need to
mimic. HD functions are sometimes allowed to call H or D functions. Due
to split compilation mode device-side compilation will not see host-only
function and thus they will not be considered at all. For clang both H
and D variants will become function overloads visible to
compiler. Normally target attribute is considered only if C++ rules can
not determine which function is better. However in this case we need to
ignore functions that would not be present during current compilation
phase before we apply normal overload resolution rules.
Changes:
* introduced another level of call preference to better describe
possible call combinations.
* removed WrongSide functions from consideration if the set contains
SameSide function.
* disabled H->D, D->H and G->H calls. These combinations are
not allowed by CUDA and we were reluctantly allowing them to work
around device-side calls to math functions in std namespace.
We no longer need it after r258880.
Differential Revision: http://reviews.llvm.org/D16870
llvm-svn: 260697
With this option one can optionally override the architecture dependent
default library to use if no -stdlib= is provided on compiler invocation.
Differential Revision: http://reviews.llvm.org/D15920
llvm-svn: 260662
OMPCapturedExprDecl allows caopturing not only of fielddecls, but also
other expressions. It also allows to simplify codegen for several
clauses.
llvm-svn: 260492
Summary:
Actions don't depend on the toolchain; they get bound to a particular
toolchain via BindArch.
No functional changes.
Reviewers: echristo
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D17100
llvm-svn: 260478
that type from the global scope.
Patch by Sterling Augustine, derived (with permission) from code from Cling by
Vassil Vassilev and Philippe Canal.
llvm-svn: 260278
While this won't help fix things like the bug that r260219 addressed, it
seems like good tidy up to have anyway.
(it might be nice if "makeArrayRef" always produced a MutableArrayRef &
let it decay to an ArrayRef when needed - then I'd use that for the
MutableArrayRefs in this patch)
If we had std::dynarray I'd use that instead of unique_ptr+size_t,
ideally (but then it'd have to be threaded down through the Preprocessor
all the way - no idea how painful that would be)
llvm-svn: 260246
being called with the wrong size: convert CGFunctionInfo to use TrailingObjects
and ask TrailingObjects to provide a working 'operator delete' for us.
llvm-svn: 260181
OpenMP 4.5 introduces privatization of non-static data members of current class in non-static member functions.
To correctly handle such kind of privatization a new (pseudo)declaration VarDecl-based node is added. It allows to reuse an existing code for capturing variables in Lambdas/Block/Captured blocks of code for correct privatization and codegen.
llvm-svn: 260077
diagnosing when 'concept' is specified on a function or template
specialization.
Since a concept can only be applied to a function or variable template,
the concept bit is stored in TemplateDecl as a PointerIntPair.
Reviewers: rsmith, faisalv, aaron.ballman, hubert.reinterpretcast
Differential Revision: http://reviews.llvm.org/D13357
llvm-svn: 260074
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>
llvm-svn: 260044
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.
rdar://problem/23575656
llvm-svn: 260043
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.
Let me know comments suggestions you may have.
Reviewers: hfinkel, carlo.bertolli, arpith-jacob, kkwli0, ABataev
Subscribers: Hahnfeld, cfe-commits, fraggamuffin, caomhin, jholewinski
Differential Revision: http://reviews.llvm.org/D16784
llvm-svn: 259977
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).
llvm-svn: 259901
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.
Reviewers: klimek
Subscribers: klimek, cfe-commits
Differential Revision: http://reviews.llvm.org/D15506
llvm-svn: 259898
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.
Review: http://reviews.llvm.org/D16730
llvm-svn: 259811
Codegen for array sections/array subscripts worked only for expressions with arrays as base. Patch fixes codegen for bases with pointer/reference types.
llvm-svn: 259776
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.
rdar://problem/24425801
llvm-svn: 259734
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.
rdar://problem/24425801
llvm-svn: 259721
Summary:
This patch adds parsing + sema for the target parallel for directive along with testcases.
Reviewers: ABataev
Differential Revision: http://reviews.llvm.org/D16759
llvm-svn: 259654
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
llvm-svn: 259651
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.
rdar://problem/24425801
llvm-svn: 259624
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.
Differential Revision: http://reviews.llvm.org/D15305
llvm-svn: 259592
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.
llvm-svn: 259537
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.
llvm-svn: 259499
Allow "mode" attribute for enum types, except for vector modes, for compatibility with GCC.
Support "mode" attribute with dependent types.
Differential Revision: http://reviews.llvm.org/D16219
llvm-svn: 259497
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.
Reviewers: ABataev
Differential Revision: http://reviews.llvm.org/D16758
llvm-svn: 259464
Patch by H.J. Lu
```
typedef unsigned int gcc_word __attribute__((mode(word)));
```
and
```
typedef unsigned int gcc_unwind_word __attribute__((mode(unwind_word)));
```
define the largest unsigned integer types which can be stored in a
general purpose register, which may not be the pointer type. For x32,
they aren't pointer nor unsigned long. We should
1. Make getUnwindWordWidth and getRegisterWidth virtual,
2. Override them for x32, similar to hasInt128Type.
3. Use getRegisterWidth for __attribute__((mode(word)));
This fixes PR 24706.
Reviewers: rnk
Differential Revision: http://reviews.llvm.org/D16779
llvm-svn: 259383
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.
Reviewers: ABataev
Differential Revision: http://reviews.llvm.org/D16758
llvm-svn: 259366
Avoids unexpected overflows while performing pointer arithmetics in 64-bit code.
Moreover, neither PointerDiffType nor 'int' can be used as a common array index
type because arrays may have size (and indexes) more than PTRDIFF_MAX but less
than SIZE_MAX.
Patch by Aleksei Sidorin!
Differential Revision: http://reviews.llvm.org/D16063
llvm-svn: 259345
Consider the following ObjC++ snippet:
--
@protocol PA;
@protocol PB;
@class NSArray<ObjectType>;
typedef int some_t;
id<PA> FA(NSArray<id<PB>> *h, some_t group);
--
This would hit an assertion in the parser after generating an annotation token
while trying to update the token cache:
Assertion failed: (CachedTokens[CachedLexPos-1].getLastLoc() == Tok.getAnnotationEndLoc() && "The annotation should be until the most recent cached token")
...
7 clang::Preprocessor::AnnotatePreviousCachedTokens(clang::Token const&) + 494
8 clang::Parser::TryAnnotateTypeOrScopeTokenAfterScopeSpec(bool, bool, clang::CXXScopeSpec&, bool) + 1163
9 clang::Parser::TryAnnotateTypeOrScopeToken(bool, bool) + 361
10 clang::Parser::isCXXDeclarationSpecifier(clang::Parser::TPResult, bool*) + 598
...
The cached preprocessor token in this case is:
greatergreater '>>' Loc=<testcase.mm:7:24>
while the annotation ("NSArray<id<PB>>") ends at "testcase.mm:7:25", hence the
assertion.
Properly update the CachedTokens during template parsing to contain
two greater tokens instead of a greatergreater.
Differential Revision: http://reviews.llvm.org/D15173
rdar://problem/23494277
llvm-svn: 259311
- Include the position of the argument on which the nullability is violated
- Differentiate between a 'method' and a 'function' in the message wording
- Test for the error message text in the tests
- Fix a bug with setting 'IsDirectDereference' which resulted in regular dereferences assumed to have call context.
llvm-svn: 259221
Devin Coughlin