]> granicus.if.org Git - clang/log
clang
9 years agoDon't try to get a CXXRecordDecl from a non-class TemplateSpecializationType.
Kaelyn Takata [Thu, 1 Oct 2015 22:38:51 +0000 (22:38 +0000)]
Don't try to get a CXXRecordDecl from a non-class TemplateSpecializationType.

With -fms-extensions it is possible to have a non-class record that is a
template specialization cause an assertion failure via the call to
Type::getAsCXXRecordDecl. Fixes PR 24246.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249090 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't adjust field offsets when using external record layout.
Zachary Turner [Thu, 1 Oct 2015 22:08:02 +0000 (22:08 +0000)]
Don't adjust field offsets when using external record layout.

This was already being done when injecting the VBPtr, but not
when injecting the VFPtr.  This fixes a number of tests in LLDB's
test suite.

Reviewed by: David Majnemer
Differential Revision: http://reviews.llvm.org/D13276

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249085 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPerform Objective-C lifetime adjustments before comparing deduced lambda result types.
Douglas Gregor [Thu, 1 Oct 2015 20:20:47 +0000 (20:20 +0000)]
Perform Objective-C lifetime adjustments before comparing deduced lambda result types.

Objective-C ARC lifetime qualifiers are dropped when canonicalizing
function types. Perform the same adjustment before comparing the
deduced result types of lambdas. Fixes rdar://problem/22344904.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249065 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[analyzer] Add TK_EntireMemSpace invalidation trait.
Devin Coughlin [Thu, 1 Oct 2015 20:09:11 +0000 (20:09 +0000)]
[analyzer] Add TK_EntireMemSpace invalidation trait.

This commit supports Sean Eveson's work on loop widening. It is NFC for now.
It adds a new TK_EntireMemSpace invalidation trait that, when applied to a
MemSpaceRegion, indicates that the entire memory space should be invalidated.

Clients can add this trait before invalidating. For example:

RegionAndSymbolInvalidationTraits ITraits;
ITraits.setTrait(MRMgr.getStackLocalsRegion(STC),
                 RegionAndSymbolInvalidationTraits::TK_EntireMemSpace);

This commit updates the existing logic invalidating global memspace regions for
calls to additionally handle arbitrary memspaces. When generating initial
clusters during cluster analysis we now add a cluster to the worklist if
the memspace for its base is marked with TK_EntireMemSpace.

This also moves the logic for invalidating globals from ClusterAnalysis to
invalidateRegionsWorker so that it is not shared with removeDeadBindingsWorker.

There are no explicit tests with this patch -- but when applied to Sean's patch
for loop widening in http://reviews.llvm.org/D12358 and after updating his code
to set the trait, the failing tests in that patch now pass.

Differential Revision: http://reviews.llvm.org/D12993

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249063 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSimplify Sema::DeduceFunctionTypeFromReturnExpr and eliminae a redundant check.
Douglas Gregor [Thu, 1 Oct 2015 19:52:44 +0000 (19:52 +0000)]
Simplify Sema::DeduceFunctionTypeFromReturnExpr and eliminae a redundant check.

NFC

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249060 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTeach -Wtautological-overlap-compare about enums
George Burgess IV [Thu, 1 Oct 2015 18:47:52 +0000 (18:47 +0000)]
Teach -Wtautological-overlap-compare about enums

Prior to this patch, -Wtautological-overlap-compare would only warn us
if there was a sketchy logical comparison between variables and
IntegerLiterals. This patch makes -Wtautological-overlap-compare aware
of EnumConstantDecls, so it can apply the same logic to them.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249053 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoModule debugging: Also emit Objective-C interfaces forward declarations
Adrian Prantl [Thu, 1 Oct 2015 16:57:02 +0000 (16:57 +0000)]
Module debugging: Also emit Objective-C interfaces forward declarations
in their module scope when building a clang module.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249031 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAllow a ToolChain to compute the path of a compiler-rt's component.
Vasileios Kalintiris [Thu, 1 Oct 2015 16:54:58 +0000 (16:54 +0000)]
Allow a ToolChain to compute the path of a compiler-rt's component.

Summary:
This patch moves getCompilerRT() from the clang::driver::tools namespace to
the ToolChain class. This is needed for multilib toolchains that need to
place their libraries in Clang's resource directory with a layout that is
different from the default one.

Reviewers: atanasyan, rsmith

Subscribers: tberghammer, danalbert, srhines, cfe-commits

Differential Revision: http://reviews.llvm.org/D13339

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249030 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[VFS] Remove unused setters. NFC.
Benjamin Kramer [Thu, 1 Oct 2015 15:02:15 +0000 (15:02 +0000)]
[VFS] Remove unused setters. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249019 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PowerPC] Define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_* macros on all PPC cores
Hal Finkel [Thu, 1 Oct 2015 13:39:49 +0000 (13:39 +0000)]
[PowerPC] Define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_* macros on all PPC cores

We support all __sync_val_compare_and_swap_* builtins (only 64-bit on 64-bit
targets) on all cores, and should define the corresponding
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_* macros, just as GCC does. As it turns out,
this is really important because they're needed to prevent a bad ODR violation
with libstdc++'s std::shared_ptr (this is well explained in PR12730).

We were doing this only for P8, but this is necessary on all PPC systems.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249009 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "Decorating virtual functions load with invariant.load" and fix
Renato Golin [Thu, 1 Oct 2015 12:58:41 +0000 (12:58 +0000)]
Revert "Decorating virtual functions load with invariant.load" and fix

This reverts commit r248982 as it was breaking the ARM buildbots and the fix didn't work.
This reverts commit r248984, the fix that didn't work.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249005 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoC++11 rangify for loops, NFC.
Yaron Keren [Thu, 1 Oct 2015 11:19:28 +0000 (11:19 +0000)]
C++11 rangify for loops, NFC.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@249001 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[clang-format] Add support of consecutive declarations alignment
Daniel Jasper [Thu, 1 Oct 2015 10:06:54 +0000 (10:06 +0000)]
[clang-format] Add support of consecutive declarations alignment

This allows clang-format to align identifiers in consecutive
declarations. This is useful for increasing the readability of the code
in the same way the alignment of assignations is.

The code is a slightly modified version of the consecutive assignment
alignment code. Currently only the identifiers are aligned, and there is
no support of alignment of the pointer star or reference symbol.

The patch also solve the issue of alignments not being possible due to
the ColumnLimit for both the existing AlignConsecutiveAligments and the
new AlignConsecutiveDeclarations.

Patch by Beren Minor, thank you.

Review: http://reviews.llvm.org/D12362

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248999 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTest fix
Piotr Padlewski [Thu, 1 Oct 2015 04:19:45 +0000 (04:19 +0000)]
Test fix

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248984 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDecorating virtual functions load with invariant.load
Piotr Padlewski [Thu, 1 Oct 2015 03:50:41 +0000 (03:50 +0000)]
Decorating virtual functions load with invariant.load

http://reviews.llvm.org/D13279

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248982 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPatch over a really horrible bug in our vector builtins that showed up
Chandler Carruth [Thu, 1 Oct 2015 02:21:34 +0000 (02:21 +0000)]
Patch over a really horrible bug in our vector builtins that showed up
recently when we started using direct conversion to model sign
extension. The __v16qi type we use for SSE v16i8 vectors is defined in
terms of 'char' which may or may not be signed! This causes us to
generate pmovsx and pmovzx depending on the setting of -funsigned-char.

This patch just forms an explicitly signed type and uses that to
formulate the sign extension. While this gets the correct behavior
(which we now verify with the enhanced test) this is just the tip of the
ice berg. Now that I know what to look for, I have found errors of this
sort *throughout* our vector code. Fortunately, this is the only
specific place where I know of users actively having their code
miscompiled by Clang due to this, so I'm keeping the fix for those users
minimal and targeted.

I'll be sending a proper email for discussion of how to fix these
systematically, what the implications are, and just how widely broken
this is... From what I can tell, we have never shipped a correct set of
builtin headers for x86 when users rely on -funsigned-char. Oops.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248980 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agocreateUniqueFile() is documented to create the file in the temporary directory unless...
Cameron Esfahani [Thu, 1 Oct 2015 01:24:59 +0000 (01:24 +0000)]
createUniqueFile() is documented to create the file in the temporary directory unless it's supplied an absolute path.

Make sure the output filepath supplied to createUniqueFile() in HTMLDiagnostics::ReportDiag() is absolute.

Summary: Make sure the output filepath supplied to createUniqueFile() in HTMLDiagnostics::ReportDiag() is absolute.

Reviewers: rsmith, akyrtzi

Subscribers: cfe-commits

Differential Revision: http://reviews.llvm.org/D12774

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248977 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix printing of parameterized Objective-C interfaces.
Bob Wilson [Thu, 1 Oct 2015 00:53:13 +0000 (00:53 +0000)]
Fix printing of parameterized Objective-C interfaces.

This change was accidentally omitted from Doug's change in r241541.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248975 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoModule debugging: Emit Objective-C interfaces in their module scope when
Adrian Prantl [Thu, 1 Oct 2015 00:48:51 +0000 (00:48 +0000)]
Module debugging: Emit Objective-C interfaces in their module scope when
building a clang module.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248974 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Sema] Don't crash when friending an unqualified templated constructor
David Majnemer [Wed, 30 Sep 2015 22:07:43 +0000 (22:07 +0000)]
[Sema] Don't crash when friending an unqualified templated constructor

Unqualified templated constructors cannot be friended and our lack of a
diagnostic led to violated invariants.  Instead, raise a diagnostic when
processing the friend declaration.

This fixes PR20251.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248953 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CUDA] fix codegen for __nvvm_atom_cas_*
Jingyue Wu [Wed, 30 Sep 2015 21:49:32 +0000 (21:49 +0000)]
[CUDA] fix codegen for __nvvm_atom_cas_*

Summary: __nvvm_atom_cas_* returns the old value instead of whether the swap succeeds.

Reviewers: eliben, tra

Subscribers: jholewinski, llvm-commits

Differential Revision: http://reviews.llvm.org/D13306

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248951 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't inherit the "unavailable" attribute from an overridden superclass method.
Douglas Gregor [Wed, 30 Sep 2015 21:34:33 +0000 (21:34 +0000)]
Don't inherit the "unavailable" attribute from an overridden superclass method.

Fixes rdar://problem/22922259.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248950 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't inherit availability information when implementing a protocol requirement.
Douglas Gregor [Wed, 30 Sep 2015 21:27:42 +0000 (21:27 +0000)]
Don't inherit availability information when implementing a protocol requirement.

When an Objective-C method implements a protocol requirement, do not
inherit any availability information from the protocol
requirement. Rather, check that the implementation is not less
available than the protocol requirement, as we do when overriding a
method that has availability. Fixes rdar://problem/22734745.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248949 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "[DarwinDriver] Use -lto_library to specify the path for libLTO.dylib"
Bruno Cardoso Lopes [Wed, 30 Sep 2015 20:11:02 +0000 (20:11 +0000)]
Revert "[DarwinDriver] Use -lto_library to specify the path for libLTO.dylib"

Revert r248932. Bots complaining about new warnings where they shouldn't.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248935 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DarwinDriver] Use -lto_library to specify the path for libLTO.dylib
Bruno Cardoso Lopes [Wed, 30 Sep 2015 19:55:07 +0000 (19:55 +0000)]
[DarwinDriver] Use -lto_library to specify the path for libLTO.dylib

Usually, when using LTO with a clang installation newer than the
system's one, there's a libLTO.dylib version mismatch and LTO fails. One
solution to this is to make ld point to the right libLTO.dylib by
changing DYLD_LIBRARY_PATH.

However, ld64 supports specifying the complete path to the desired
libLTO.dylib through the -lto_library option. This commit adds support
for the clang driver to use this option whenever it's capable of finding
a libLTO.dylib in clang's installed library directory. This way, we
don't need to rely on DYLD_LIBRARY_PATH nor get caught by version
mismatches.

Differential Revision: http://reviews.llvm.org/D13117

rdar://problem/7363476

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248932 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang-format: Add clangRewrite dependency to fix cmake build.
Daniel Jasper [Wed, 30 Sep 2015 19:28:47 +0000 (19:28 +0000)]
clang-format: Add clangRewrite dependency to fix cmake build.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248929 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't correct non-class using declarations to class members.
Kaelyn Takata [Wed, 30 Sep 2015 18:23:35 +0000 (18:23 +0000)]
Don't correct non-class using declarations to class members.

Such declarations would be invalid anyway, and trying to make the
correction will lead to a crash. Fixes PR 24781.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248928 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Sema] Avoid crashing during this-> insertion recovery
Reid Kleckner [Wed, 30 Sep 2015 17:30:48 +0000 (17:30 +0000)]
[Sema] Avoid crashing during this-> insertion recovery

We get into this bad state when someone defines a new member function
for a class but forgets to add the declaration to the class body.
Calling the new member function from a member function template of the
class will crash during instantiation.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248925 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAlias "-ggdbN" to "-gN" options, fixing some incompatibilities.
Douglas Katzman [Wed, 30 Sep 2015 15:55:59 +0000 (15:55 +0000)]
Alias "-ggdbN" to "-gN" options, fixing some incompatibilities.

* assembling from a .s file mistook -ggdb0 for -g
* -ggdb1 is supposed to mean basically -g1, not -gN for N>1

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248912 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OpenCL 2.0] Enable program scope variables, Section 6.5.1.
Anastasia Stulova [Wed, 30 Sep 2015 14:08:20 +0000 (14:08 +0000)]
[OpenCL 2.0] Enable program scope variables, Section 6.5.1.

 - Remove virtual SC_OpenCLWorkGroupLocal storage type specifier
as it conflicts with static local variables now and prevents
diagnosing static local address space variables correctly.

 - Allow static local and global variables (OpenCL2.0 s6.8 and s6.5.1).

 - Improve diagnostics of allowed ASes for variables in different scopes:
(i) Global or static local variables have to be in global
or constant ASes (OpenCL1.2 s6.5, OpenCL2.0 s6.5.1);
(ii) Non-kernel function variables can't be declared in local
or constant ASes (OpenCL1.1 s6.5.2 and s6.5.3).

http://reviews.llvm.org/D13105

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248906 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix FunctionParmPackExpr::Create() to take a ParmVarDecl* array.
James Y Knight [Wed, 30 Sep 2015 14:04:23 +0000 (14:04 +0000)]
Fix FunctionParmPackExpr::Create() to take a ParmVarDecl* array.

FunctionParmPackExpr actually stores an array of ParmVarDecl* (and
accessors return that). But, the FunctionParmPackExpr::Create()
constructor accepted an array of Decl *s instead.

It was easy for this mismatch to occur without any obvious sign of
something wrong, since both the store and the access used independent
'reinterpet_cast<XX>(this+1)' calls.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248905 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang-format: Use Rewriter again to write the output files.
Daniel Jasper [Wed, 30 Sep 2015 13:59:29 +0000 (13:59 +0000)]
clang-format: Use Rewriter again to write the output files.

This has two advantages:
1. Atomic writes.
2. Proper handling of line endings (hopefully solving llvm.org/PR24999

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248904 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OpenCL] Add missing OpenCL LangOpts in address space compatibility checks
Anastasia Stulova [Wed, 30 Sep 2015 13:49:55 +0000 (13:49 +0000)]
[OpenCL] Add missing OpenCL LangOpts in address space compatibility checks

and test checking broken (due to CL specific diagnostics) C functionality

M    test/Sema/address_spaces.c
M    lib/Sema/SemaExpr.cpp

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248902 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OpenCL 2.0] This change adds extra diagnostics to atomic types
Anastasia Stulova [Wed, 30 Sep 2015 13:18:52 +0000 (13:18 +0000)]
[OpenCL 2.0] This change adds extra diagnostics to atomic types

Applied restrictions from OpenCL v2.0 s6.13.11.8
that mainly disallow operations on atomic types (except for taking their address - &).

The patch is taken from SPIR2.0 provisional branch, contributed by Guy Benyei!

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248896 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OpenCL 2.0] Fix wrong atomic type detection in the diagnostics of allowed types
Anastasia Stulova [Wed, 30 Sep 2015 11:48:15 +0000 (11:48 +0000)]
[OpenCL 2.0] Fix wrong atomic type detection in the diagnostics of allowed types

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248891 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ARM][NEON] Use address space in vld([1234]|[234]lane) and vst([1234]|[234]lane)...
Jeroen Ketema [Wed, 30 Sep 2015 10:56:56 +0000 (10:56 +0000)]
[ARM][NEON] Use address space in vld([1234]|[234]lane) and vst([1234]|[234]lane) instructions

This is the clang commit associated with llvm r248887.

This commit changes the interface of the vld[1234], vld[234]lane, and vst[1234],
vst[234]lane ARM neon intrinsics and associates an address space with the
pointer that these intrinsics take. This changes, e.g.,

<2 x i32> @llvm.arm.neon.vld1.v2i32(i8*, i32)

to

<2 x i32> @llvm.arm.neon.vld1.v2i32.p0i8(i8*, i32)

This change ensures that address spaces are fully taken into account in the ARM
target during lowering of interleaved loads and stores.

Differential Revision: http://reviews.llvm.org/D13127

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248888 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OPENMP 4.1] Sema analysis for array sections in 'reduction' clause.
Alexey Bataev [Wed, 30 Sep 2015 09:22:36 +0000 (09:22 +0000)]
[OPENMP 4.1] Sema analysis for array sections in 'reduction' clause.
OpenMP 4.1 allows to use array sections|subscript expressions in 'reduction' clauses. Added sema analysis, updated tests.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248880 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix the sample profile format that breaks in test http://lab.llvm.org:8011/builders...
Dehao Chen [Wed, 30 Sep 2015 01:03:10 +0000 (01:03 +0000)]
Fix the sample profile format that breaks in test http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast/builds/505

http://reviews.llvm.org/D13145

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248869 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPromote a warning on ill-formed code (redeclaration missing an exception
Richard Smith [Wed, 30 Sep 2015 00:48:50 +0000 (00:48 +0000)]
Promote a warning on ill-formed code (redeclaration missing an exception
specification) to an error. No compiler other than Clang seems to allow this,
and it doesn't seem like a useful thing to accept as an extension in general.

The current behavior was added for PR5957, where the problem was specifically
related to mismatches of the exception specification on the implicitly-declared
global operator new and delete. To retain that workaround, we downgrade the
error to an ExtWarn when the declaration is of a replaceable global allocation
function.

Now that this is an error, stop trying (and failing) to recover from a missing
computed noexcept specification. That recovery didn't work, and led to crashes
in code like the added testcase.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248867 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't crash when a reserved global placement operator new is paired
John McCall [Tue, 29 Sep 2015 23:55:17 +0000 (23:55 +0000)]
Don't crash when a reserved global placement operator new is paired
with a non-reserved operator delete in a new-expression.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248862 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMove functions declared in ExprObjC.h into ExprObjC.cpp.
James Y Knight [Tue, 29 Sep 2015 22:28:44 +0000 (22:28 +0000)]
Move functions declared in ExprObjC.h into ExprObjC.cpp.

r51703 back in 2008 split out all the ObjC Expr subclasses from Expr.h
to a new ExprObjC.h file, but failed to also split the implementation
from Expr.cpp to ExprObjC.cpp. Do so, finally, for readability's sake.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248836 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix Clang-tidy modernize-use-nullptr warnings in headers and generated files; other...
Hans Wennborg [Tue, 29 Sep 2015 20:56:43 +0000 (20:56 +0000)]
Fix Clang-tidy modernize-use-nullptr warnings in headers and generated files; other minor cleanups.

By Eugene Zelenko!

Differential Revision: http://reviews.llvm.org/D13187

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248828 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCGDebugInfo: Don't reuse a reference into a DenseMap if the DenseMap may
Adrian Prantl [Tue, 29 Sep 2015 20:44:46 +0000 (20:44 +0000)]
CGDebugInfo: Don't reuse a reference into a DenseMap if the DenseMap may
be modified in between. (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248826 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoForgot to remove a FIXME that has been fixed. NFC.
Nemanja Ivanovic [Tue, 29 Sep 2015 18:20:59 +0000 (18:20 +0000)]
Forgot to remove a FIXME that has been fixed. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248815 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAddition of interfaces the FE to conform to Table A-2 of ELF V2 ABI V1.1
Nemanja Ivanovic [Tue, 29 Sep 2015 18:13:34 +0000 (18:13 +0000)]
Addition of interfaces the FE to conform to Table A-2 of ELF V2 ABI V1.1

This patch corresponds to review:
http://reviews.llvm.org/D13190

Implemented the following interfaces to conform to ELF V2 ABI version 1.1.

vector signed __int128 vec_adde (vector signed __int128, vector signed __int128, vector signed __int128);
vector unsigned __int128 vec_adde (vector unsigned __int128, vector unsigned __int128, vector unsigned __int128);
vector signed __int128 vec_addec (vector signed __int128, vector signed __int128, vector signed __int128);
vector unsigned __int128 vec_addec (vector unsigned __int128, vector unsigned __int128, vector unsigned __int128);
vector signed int vec_addc(vector signed int __a, vector signed int __b);
vector bool char vec_cmpge (vector signed char __a, vector signed char __b);
vector bool char vec_cmpge (vector unsigned char __a, vector unsigned char __b);
vector bool short vec_cmpge (vector signed short __a, vector signed short __b);
vector bool short vec_cmpge (vector unsigned short __a, vector unsigned short __b);
vector bool int vec_cmpge (vector signed int __a, vector signed int __b);
vector bool int vec_cmpge (vector unsigned int __a, vector unsigned int __b);
vector bool char vec_cmple (vector signed char __a, vector signed char __b);
vector bool char vec_cmple (vector unsigned char __a, vector unsigned char __b);
vector bool short vec_cmple (vector signed short __a, vector signed short __b);
vector bool short vec_cmple (vector unsigned short __a, vector unsigned short __b);
vector bool int vec_cmple (vector signed int __a, vector signed int __b);
vector bool int vec_cmple (vector unsigned int __a, vector unsigned int __b);
vector double vec_double (vector signed long long __a);
vector double vec_double (vector unsigned long long __a);
vector bool char vec_eqv(vector bool char __a, vector bool char __b);
vector bool short vec_eqv(vector bool short __a, vector bool short __b);
vector bool int vec_eqv(vector bool int __a, vector bool int __b);
vector bool long long vec_eqv(vector bool long long __a, vector bool long long __b);
vector signed short vec_madd(vector signed short __a, vector signed short __b, vector signed short __c);
vector signed short vec_madd(vector signed short __a, vector unsigned short __b, vector unsigned short __c);
vector signed short vec_madd(vector unsigned short __a, vector signed short __b, vector signed short __c);
vector unsigned short vec_madd(vector unsigned short __a, vector unsigned short __b, vector unsigned short __c);
vector bool long long vec_mergeh(vector bool long long __a, vector bool long long __b);
vector bool long long vec_mergel(vector bool long long __a, vector bool long long __b);
vector bool char vec_nand(vector bool char __a, vector bool char __b);
vector bool short vec_nand(vector bool short __a, vector bool short __b);
vector bool int vec_nand(vector bool int __a, vector bool int __b);
vector bool long long vec_nand(vector bool long long __a, vector bool long long __b);
vector bool char vec_orc(vector bool char __a, vector bool char __b);
vector bool short vec_orc(vector bool short __a, vector bool short __b);
vector bool int vec_orc(vector bool int __a, vector bool int __b);
vector bool long long vec_orc(vector bool long long __a, vector bool long long __b);
vector signed long long vec_sub(vector signed long long __a, vector signed long long __b);
vector signed long long vec_sub(vector bool long long __a, vector signed long long __b);
vector signed long long vec_sub(vector signed long long __a, vector bool long long __b);
vector unsigned long long vec_sub(vector unsigned long long __a, vector unsigned long long __b);
vector unsigned long long vec_sub(vector bool long long __a, vector unsigned long long __b);
vector unsigned long long vec_sub(vector unsigned long long __V2 ABI V1.1

http://ror float vec_sub(vector float __a, vector float __b);
unsigned char vec_extract(vector bool char __a, int __b);
signed short vec_extract(vector signed short __a, int __b);
unsigned short vec_extract(vector bool short __a, int __b);
signed int vec_extract(vector signed int __a, int __b);
unsigned int vec_extract(vector bool int __a, int __b);
signed long long vec_extract(vector signed long long __a, int __b);
unsigned long long vec_extract(vector unsigned long long __a, int __b);
unsigned long long vec_extract(vector bool long long __a, int __b);
double vec_extract(vector double __a, int __b);
vector bool char vec_insert(unsigned char __a, vector bool char __b, int __c);
vector signed short vec_insert(signed short __a, vector signed short __b, int __c);
vector bool short vec_insert(unsigned short __a, vector bool short __b, int __c);
vector signed int vec_insert(signed int __a, vector signed int __b, int __c);
vector bool int vec_insert(unsigned int __a, vector bool int __b, int __c);
vector signed long long vec_insert(signed long long __a, vector signed long long __b, int __c);
vector unsigned long long vec_insert(unsigned long long __a, vector unsigned long long __b, int __c);
vector bool long long vec_insert(unsigned long long __a, vector bool long long __b, int __c);
vector double vec_insert(double __a, vector double __b, int __c);
vector signed long long vec_splats(signed long long __a);
vector unsigned long long vec_splats(unsigned long long __a);
vector signed __int128 vec_splats(signed __int128 __a);
vector unsigned __int128 vec_splats(unsigned __int128 __a);
vector double vec_splats(double __a);
int vec_all_eq(vector double __a, vector double __b);
int vec_all_ge(vector double __a, vector double __b);
int vec_all_gt(vector double __a, vector double __b);
int vec_all_le(vector double __a, vector double __b);
int vec_all_lt(vector double __a, vector double __b);
int vec_all_nan(vector double __a);
int vec_all_ne(vector double __a, vector double __b);
int vec_all_nge(vector double __a, vector double __b);
int vec_all_ngt(vector double __a, vector double __b);
int vec_any_eq(vector double __a, vector double __b);
int vec_any_ge(vector double __a, vector double __b);
int vec_any_gt(vector double __a, vector double __b);
int vec_any_le(vector double __a, vector double __b);
int vec_any_lt(vector double __a, vector double __b);
int vec_any_ne(vector double __a, vector double __b);
vector unsigned char vec_sbox_be (vector unsigned char);
vector unsigned char vec_cipher_be (vector unsigned char, vector unsigned char);
vector unsigned char vec_cipherlast_be (vector unsigned char, vector unsigned char);
vector unsigned char vec_ncipher_be (vector unsigned char, vector unsigned char);
vector unsigned char vec_ncipherlast_be (vector unsigned char, vector unsigned char);
vector unsigned int vec_shasigma_be (vector unsigned int, const int, const int);
vector unsigned long long vec_shasigma_be (vector unsigned long long, const int, const int);
vector unsigned short vec_pmsum_be (vector unsigned char, vector unsigned char);
vector unsigned int vec_pmsum_be (vector unsigned short, vector unsigned short);
vector unsigned long long vec_pmsum_be (vector unsigned int, vector unsigned int);
vector unsigned __int128 vec_pmsum_be (vector unsigned long long, vector unsigned long long);
vector unsigned char vec_gb (vector unsigned char);
vector unsigned long long vec_bperm (vector unsigned __int128 __a, vector unsigned char __b);

Removed the folowing interfaces either because their signatures have changed
in version 1.1 of the ABI or because they were implemented for ELF V2 ABI but
have actually been deprecated in version 1.1.

vector signed char vec_eqv(vector bool char __a, vector signed char __b);
vector signed char vec_eqv(vector signed char __a, vector bool char __b);
vector unsigned char vec_eqv(vector bool char __a, vector unsigned char __b);
vector unsigned char vec_eqv(vector unsigned char __a, vector bool char __b);
vector signed short vec_eqv(vector bool short __a, vector signed short __b);
vector signed short vec_eqv(vector signed short __a, vector bool short __b);
vector unsigned short vec_eqv(vector bool short __a, vector unsigned short __b);
vector unsigned short vec_eqv(vector unsigned short __a, vector bool short __b);
vector signed int vec_eqv(vector bool int __a, vector signed int __b);
vector signed int vec_eqv(vector signed int __a, vector bool int __b);
vector unsigned int vec_eqv(vector bool int __a, vector unsigned int __b);
vector unsigned int vec_eqv(vector unsigned int __a, vector bool int __b);
vector signed long long vec_eqv(vector bool long long __a, vector signed long long __b);
vector signed long long vec_eqv(vector signed long long __a, vector bool long long __b);
vector unsigned long long vec_eqv(vector bool long long __a, vector unsigned long long __b);
vector unsigned long long vec_eqv(vector unsigned long long __a, vector bool long long __b);
vector float vec_eqv(vector bool int __a, vector float __b);
vector float vec_eqv(vector float __a, vector bool int __b);
vector double vec_eqv(vector bool long long __a, vector double __b);
vector double vec_eqv(vector double __a, vector bool long long __b);
vector unsigned short vec_nand(vector bool short __a, vector unsigned short __b);

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248813 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdded MSVC natvis visualizers for Type and QualType. These could probably be improved...
Aaron Ballman [Tue, 29 Sep 2015 17:32:36 +0000 (17:32 +0000)]
Added MSVC natvis visualizers for Type and QualType. These could probably be improved, but anything is better than staring at hex values in the debugger.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248808 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix typo.
Yaron Keren [Tue, 29 Sep 2015 16:51:08 +0000 (16:51 +0000)]
Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248807 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoThread Safety Analysis: allow capability attribute on unions.
DeLesley Hutchins [Tue, 29 Sep 2015 16:24:18 +0000 (16:24 +0000)]
Thread Safety Analysis: allow capability attribute on unions.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248805 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoThread Safety Analysis: fix before/after checks so that they work on global
DeLesley Hutchins [Tue, 29 Sep 2015 15:25:51 +0000 (15:25 +0000)]
Thread Safety Analysis:  fix before/after checks so that they work on global
variables as well member variables.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248803 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang-format: Add a new brace style "custom" as well as flags to
Daniel Jasper [Tue, 29 Sep 2015 14:57:55 +0000 (14:57 +0000)]
clang-format: Add a new brace style "custom" as well as flags to
control the individual braces. The existing choices for brace wrapping
are now merely presets for the different flags that get expanded upon
calling the reformat function.

All presets have been chose to keep the existing formatting, so there
shouldn't be any difference in formatting behavior.

Also change the dump_format_style.py to properly document the nested
structs that are used to keep these flags discoverable among all the
configuration flags.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248802 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang-format: Extend #include sorting functionality
Daniel Jasper [Tue, 29 Sep 2015 07:53:08 +0000 (07:53 +0000)]
clang-format: Extend #include sorting functionality

Recognize the main module header as well as different #include categories.
This should now mimic the behavior of llvm/utils/sort_includes.py as
well as clang-tools-extra/clang-tidy/llvm/IncludeOrderCheck.cpp very
closely.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248782 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert part of r248776. One of the changes worked on my local build, but not on the...
Craig Topper [Tue, 29 Sep 2015 05:08:30 +0000 (05:08 +0000)]
Revert part of r248776. One of the changes worked on my local build, but not on the bots.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248778 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSimplify or remove calls to makeArrayRef based on feedback from David Blaikie. NFC.
Craig Topper [Tue, 29 Sep 2015 04:53:28 +0000 (04:53 +0000)]
Simplify or remove calls to makeArrayRef based on feedback from David Blaikie. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248776 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoHonor the casted-to alignment of an explicit cast even when
John McCall [Tue, 29 Sep 2015 04:37:40 +0000 (04:37 +0000)]
Honor the casted-to alignment of an explicit cast even when
Sema thinks the cast is a no-op, as it does when (e.g.) the
only thing that changes is an alignment attribute.

Fixed PR24944.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248775 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove 'const' from some ArrayRef arguments since they're passed by value anyway...
Craig Topper [Tue, 29 Sep 2015 04:30:07 +0000 (04:30 +0000)]
Remove 'const' from some ArrayRef arguments since they're passed by value anyway. NFC

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248774 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPass ArrayRef by value. NFC.
Craig Topper [Tue, 29 Sep 2015 04:30:05 +0000 (04:30 +0000)]
Pass ArrayRef by value. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248773 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OPENMP 4.1] Codegen for ‘simd’ clause in ‘ordered’ directive.
Alexey Bataev [Tue, 29 Sep 2015 03:48:57 +0000 (03:48 +0000)]
[OPENMP 4.1] Codegen for â€˜simd’ clause in â€˜ordered’ directive.
Description.
If the simd clause is specified, the ordered regions encountered by any thread will use only a single SIMD lane to execute the ordered regions in the order of the loop iterations.
Restrictions.
An ordered construct with the simd clause is the only OpenMP construct that can appear in the simd region.

An ordered directive with â€˜simd’ clause is generated as an outlined function and corresponding function call to prevent this part of code from vectorization later in backend.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248772 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUnique-pointerify these pointers an plug a memory leak.
Adrian Prantl [Mon, 28 Sep 2015 23:56:10 +0000 (23:56 +0000)]
Unique-pointerify these pointers an plug a memory leak.
Thanks to echristo for noticing!

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248762 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove the only use of LookupResult's implicit copy ctor
David Blaikie [Mon, 28 Sep 2015 23:48:55 +0000 (23:48 +0000)]
Remove the only use of LookupResult's implicit copy ctor

LookupResult should not be copyable, it's not readily copyable and can
only be copied when it's in specific states (in a query state, without
any results, basically). Instead, just extract the /query/ state and
pass that across the copy boundary, then build a new LookupResult on the
other side.

I wonder if a better API (one in which the query state is separate from
the result state - essentialyl making QueryState a first class part of
the Lookup API - pass a QueryState, get a LookupResult, rather than
mutating the LookupResult in place (LookupResult could contain a
QueryState if it's particularly helpful to be able to observe the query
parameters while also examining the result)) might be a good idea here.

Future patches will probably make LookupResult actually non-copyable
(transition the CXXBasePaths to unique_ptr, for example) and hopefully
we'll enable -Wdeprecated in LLVM soon to avoid issues like this.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248761 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd an explicitly defaulted copy ctor (and FIXME) to Command since its copy ctor...
David Blaikie [Mon, 28 Sep 2015 23:48:52 +0000 (23:48 +0000)]
Add an explicitly defaulted copy ctor (and FIXME) to Command since its copy ctor is currently used (and actually slices derived objects... ) in some diagnostics handling.

Justin mentioned he'd look into this.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248760 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove trivial dtor that was making uses of the copy ctor -Wdeprecated unclean
David Blaikie [Mon, 28 Sep 2015 23:48:49 +0000 (23:48 +0000)]
Remove trivial dtor that was making uses of the copy ctor -Wdeprecated unclean

And also remove an unused variable that's now diagnosed by
-Wunused-variable since the dtor is now trivial.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248759 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CUDA] 32-bit NVPTX should have 32-bit long type.
Artem Belevich [Mon, 28 Sep 2015 22:54:08 +0000 (22:54 +0000)]
[CUDA] 32-bit NVPTX should have 32-bit long type.

Currently it's 64-bit which will lead to mismatch between host and
device code if we compile for i386.

Differential Revision: http://reviews.llvm.org/D13181

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248753 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoGenerate assume loads only with -fstrict-vtable-pointers
Piotr Padlewski [Mon, 28 Sep 2015 20:30:22 +0000 (20:30 +0000)]
Generate assume loads only with -fstrict-vtable-pointers

Temporary fix till InstCombine and other possible passes will be
efficient to handle multiple assumes.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248734 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang-format: [JS] Support pseudo-keywords
Daniel Jasper [Mon, 28 Sep 2015 14:29:45 +0000 (14:29 +0000)]
clang-format: [JS] Support pseudo-keywords

JavaScript allows keywords to appear in IdenfierName positions, e.g.
fields, or object literal members, but not as plain identifiers.

Patch by Martin Probst. Thank you!

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248714 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang-format: [JS] handle let (ES6)
Daniel Jasper [Mon, 28 Sep 2015 14:28:08 +0000 (14:28 +0000)]
clang-format: [JS] handle let (ES6)

Patch by Martin Probst. Thank you!

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248713 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OPENMP 4.1] Add 'simd' clause for 'ordered' directive.
Alexey Bataev [Mon, 28 Sep 2015 06:39:35 +0000 (06:39 +0000)]
[OPENMP 4.1] Add 'simd' clause for 'ordered' directive.
Parsing and sema analysis for 'simd' clause in 'ordered' directive.
Description
If the simd clause is specified, the ordered regions encountered by any thread will use only a single SIMD lane to execute the ordered
regions in the order of the loop iterations.
Restrictions
An ordered construct with the simd clause is the only OpenMP construct that can appear in the simd region

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248696 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse llvm::makeArrayRef. NFC.
Craig Topper [Sun, 27 Sep 2015 03:44:08 +0000 (03:44 +0000)]
Use llvm::makeArrayRef. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248678 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang/test/Driver/arm-compiler-rt.c: Relax expressions for dos path.
NAKAMURA Takumi [Sat, 26 Sep 2015 11:17:32 +0000 (11:17 +0000)]
clang/test/Driver/arm-compiler-rt.c: Relax expressions for dos path.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248666 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert part of r248660 as using None didn't work with the ArrayRef being returned...
Craig Topper [Sat, 26 Sep 2015 05:42:17 +0000 (05:42 +0000)]
Revert part of r248660 as using None didn't work with the ArrayRef being returned as an Optional.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248661 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse None to avoid re-mentioning the ArrayRef type to call the default constructor.
Craig Topper [Sat, 26 Sep 2015 05:22:17 +0000 (05:22 +0000)]
Use None to avoid re-mentioning the ArrayRef type to call the default constructor.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248660 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove move constructor and assignment operator from SourceMappingRegion. The types...
Craig Topper [Sat, 26 Sep 2015 05:10:16 +0000 (05:10 +0000)]
Remove move constructor and assignment operator from SourceMappingRegion. The types of the fields are trivially copyable. NFC

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248659 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix a couple other places that were returning SourceLocation by reference. NFC
Craig Topper [Sat, 26 Sep 2015 05:10:14 +0000 (05:10 +0000)]
Fix a couple other places that were returning SourceLocation by reference. NFC

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248658 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDriver: support ARM/HF on a single toolchain
Saleem Abdulrasool [Sat, 26 Sep 2015 03:26:44 +0000 (03:26 +0000)]
Driver: support ARM/HF on a single toolchain

ARM EABI adds target attributes to the object file.  Amongst the attributes that
are emitted is the VFP argument passing (Hard vs Soft).  The linker is
responsible for checking these attributes and erroring on mismatches.  This
causes problems for the compiler-rt builtins when targeting both hard and
soft.  Because both of these options name the builtins compiler-rt component
the same (libclang_rt.builtins-arm.a or libclang_rt.builtins-arm-android).  GCC
is able to get away with this as it does one target per toolchain.  This
changes the naming convention for the ARM compiler-rt builtins to differentiate
between HF and Soft.  Although this means that compiler-rt may be duplicated, it
enables supporting both variants from a single toolchain.  A similar approach is
taken by the Darwin toolchain, naming the library to differentiate between the
calling conventions.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248649 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReplace a loop and temporary string copying with llvm::join.
Eric Christopher [Sat, 26 Sep 2015 01:25:08 +0000 (01:25 +0000)]
Replace a loop and temporary string copying with llvm::join.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248645 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoChange arc-cxx11-init-list.mm to work with upcoming SCEV changes.
Sanjoy Das [Fri, 25 Sep 2015 23:07:11 +0000 (23:07 +0000)]
Change arc-cxx11-init-list.mm to work with upcoming SCEV changes.

Summary:
The store being checked for in arc-cxx11-init-list.mm is a store to an
unescaped alloca.  After an uncoming change to ScalarEvolution, LLVM is
able to elide the store, so adjust the test accordingly.

Reviewers: compnerd

Subscribers: cfe-commits

Differential Revision: http://reviews.llvm.org/D13183

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248632 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRelax test to allow for __attribute__((thiscall)). Under Win32 c-index-test output is
Yaron Keren [Fri, 25 Sep 2015 22:09:07 +0000 (22:09 +0000)]
Relax test to allow for __attribute__((thiscall)). Under Win32 c-index-test output is

CXXMethod=foo:4:7 (unavailable) [type=int () __attribute__((thiscall))] [typekind=FunctionProto] [resulttype=int] [resulttypekind=Int] [isPOD=0]
CXXConstructor=Foo:5:3 (unavailable) [type=void () __attribute__((thiscall))] [typekind=FunctionProto] [resulttype=void] [resulttypekind=Void] [isPOD=0]

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248626 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix bug on reporting availability of deleted methods in libclang.
Manuel Klimek [Fri, 25 Sep 2015 17:53:16 +0000 (17:53 +0000)]
Fix bug on reporting availability of deleted methods in libclang.

Patch by Sergey Kalinichev.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248596 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMake incomplete type errors better with enable_if
George Burgess IV [Fri, 25 Sep 2015 17:53:16 +0000 (17:53 +0000)]
Make incomplete type errors better with enable_if

This patch fixes the order in which we evaluate the different ways that
a function call could be disallowed. Now, if you call a non-overloaded
function with an incomplete type and failing enable_if, we'll prioritize
reporting the more obvious error (use of incomplete type) over reporting
the failing enable_if.

Thanks to Ettore Speziale for the patch!

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248595 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoThe Clang gcc-compatible driver (clang/tools/driver/driver.cpp) has some
Eric Christopher [Fri, 25 Sep 2015 17:44:31 +0000 (17:44 +0000)]
The Clang gcc-compatible driver (clang/tools/driver/driver.cpp) has some
logic to select an alternate target based on the executable it was
called as. For instance, if you symlink i686-linux-android-gcc to clang
and invoke it, the driver will act as though it were called with another
argument ("-target i686-linux-android"). This leads to visible effects
even in syntax-only compilations (like the ANDROID preprocessor symbol
being defined).

This behavior is not replicated for tool invocations--for instance,
clang::createInvocationFromCommandLine will not choose an alternate
target based on ArgList[0]. This means that configurations stored in
compilation databases aren't accurately replayed.

This patch separates the logic for selecting a mode flag and target from
the executable name into a new member function on ToolChain. It should
have no functional effects (but will allow other code to reuse the
target/mode selection logic).

Patch by Luke Zarko!

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248592 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "This patch adds missing pieces to clang, including the PS4 toolchain definiti...
Greg Bedwell [Fri, 25 Sep 2015 16:11:00 +0000 (16:11 +0000)]
Revert "This patch adds missing pieces to clang, including the PS4 toolchain definition, added warnings, PS4 defaults, and Driver changes needed for our compiler."

This reverts commit r248546 to get our bot green again while we discuss the best way forward.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248578 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OPENMP 4.1] Add 'threads' clause for '#pragma omp ordered'.
Alexey Bataev [Fri, 25 Sep 2015 10:37:12 +0000 (10:37 +0000)]
[OPENMP 4.1] Add 'threads' clause for '#pragma omp ordered'.
OpenMP 4.1 extends format of '#pragma omp ordered'. It adds 3 additional clauses: 'threads', 'simd' and 'depend'.
If no clause is specified, the ordered construct behaves as if the threads clause had been specified. If the threads clause is specified, the threads in the team executing the loop region execute ordered regions sequentially in the order of the loop iterations.
The loop region to which an ordered region without any clause or with a threads clause binds must have an ordered clause without the parameter specified on the corresponding loop directive.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248569 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoNew test ps4-linker-win.c (added in r48546) is causing problems for Windows bots...
Ekaterina Romanova [Fri, 25 Sep 2015 09:10:33 +0000 (09:10 +0000)]
New test ps4-linker-win.c (added in r48546) is causing problems for Windows bots. I'm deleting this file during investigation.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248567 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoThis patch adds missing pieces to clang, including the PS4 toolchain
Ekaterina Romanova [Thu, 24 Sep 2015 22:06:52 +0000 (22:06 +0000)]
This patch adds missing pieces to clang, including the PS4 toolchain
definition, added warnings, PS4 defaults, and Driver changes needed for
our compiler.

A patch by Filipe Cabecinhas, Pierre Gousseau and Katya Romanova!

Differential Revision: http://reviews.llvm.org/D11279

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248546 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMove the darwin define static function to be close to the OS define.
Eric Christopher [Thu, 24 Sep 2015 21:17:04 +0000 (21:17 +0000)]
Move the darwin define static function to be close to the OS define.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248539 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse just one larger anonymous namespace instead of a lot of smaller ones.
Eric Christopher [Thu, 24 Sep 2015 21:17:02 +0000 (21:17 +0000)]
Use just one larger anonymous namespace instead of a lot of smaller ones.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248538 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agofix typos; NFC
Sanjay Patel [Thu, 24 Sep 2015 21:11:52 +0000 (21:11 +0000)]
fix typos; NFC

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248537 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRecommit r248154: [ARM] Handle DSP feature as an ArchExtKind
Artyom Skrobov [Thu, 24 Sep 2015 17:34:05 +0000 (17:34 +0000)]
Recommit r248154: [ARM] Handle DSP feature as an ArchExtKind

Currently, the availability of DSP instructions (ACLE 6.4.7) is handled in
a hand-rolled tricky condition block in lib/Basic/Targets.cpp, with a FIXME:
attached.

http://reviews.llvm.org/D12937 moved the handling of the DSP feature over to
ARMTargetParser.def in LLVM, to be in line with other architecture extensions.

This is the corresponding patch to clang, to clear the FIXME: and update
the tests.

Differential Revision: http://reviews.llvm.org/D12938

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248521 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoEnable SafeStack on all Linux platforms.
Evgeniy Stepanov [Thu, 24 Sep 2015 17:22:46 +0000 (17:22 +0000)]
Enable SafeStack on all Linux platforms.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248518 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[analyzer] When memcpy'ing into a fixed-size array, do not invalidate entire region.
Devin Coughlin [Thu, 24 Sep 2015 16:52:56 +0000 (16:52 +0000)]
[analyzer] When memcpy'ing into a fixed-size array, do not invalidate entire region.

Change the analyzer's modeling of memcpy to be more precise when copying into fixed-size
array fields. With this change, instead of invalidating the entire containing region the
analyzer now invalidates only offsets for the array itself when it can show that the
memcpy stays within the bounds of the array.

This addresses false positive memory leak warnings of the kind reported by
krzysztof in https://llvm.org/bugs/show_bug.cgi?id=22954

(This is the second attempt, now with assertion failures resolved.)

A patch by Pierre Gousseau!

Differential Revision: http://reviews.llvm.org/D12571

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248516 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoModule Debugging: Emit submodules as nested DW_TAG_modules.
Adrian Prantl [Thu, 24 Sep 2015 16:10:10 +0000 (16:10 +0000)]
Module Debugging: Emit submodules as nested DW_TAG_modules.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248511 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDebug Info: Use the module pointer as key for the module cache.
Adrian Prantl [Thu, 24 Sep 2015 16:10:04 +0000 (16:10 +0000)]
Debug Info: Use the module pointer as key for the module cache.
This way we don't need to rebuild the full module name for every decl.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248510 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRefactor ASTSourceDescriptor to not store copies of all strings. (NFC)
Adrian Prantl [Thu, 24 Sep 2015 16:10:00 +0000 (16:10 +0000)]
Refactor ASTSourceDescriptor to not store copies of all strings. (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248509 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDrop useless const in for-range loops.
Benjamin Kramer [Thu, 24 Sep 2015 14:48:49 +0000 (14:48 +0000)]
Drop useless const in for-range loops.

StringRefs always point to immutable memory so the const doesn't add value
here. Also quiets clang's -Wrange-loop-analysis which warns about the implicit
copying.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248496 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Driver] Don't implicitly copy std::strings in for-range loop.
Benjamin Kramer [Thu, 24 Sep 2015 14:48:37 +0000 (14:48 +0000)]
[Driver] Don't implicitly copy std::strings in for-range loop.

Found by -Wrange-loop-analysis.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248495 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Relax -mnan=2008 acceptance to permit MIPS32R2 and MIPS64R2.
Daniel Sanders [Thu, 24 Sep 2015 10:22:17 +0000 (10:22 +0000)]
[mips] Relax -mnan=2008 acceptance to permit MIPS32R2 and MIPS64R2.

Summary:
Strictly speaking, the MIPS*R2 ISA's should not permit -mnan=2008 since this
feature was added in MIPS*R3. However, other toolchains permit this and we
should do the same.

Reviewers: atanasyan

Subscribers: cfe-commits

Differential Revision: http://reviews.llvm.org/D13057

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248481 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ARM] Follow-up to fix crash "-target arm -mcpu=generic", without "-march="
Vladimir Sukharev [Thu, 24 Sep 2015 10:06:44 +0000 (10:06 +0000)]
[ARM] Follow-up to fix crash "-target arm -mcpu=generic", without "-march="

Fix of dangling StringRef after temporary std::string is destroyed

Follow-up to: http://reviews.llvm.org/rL248479

Reviewers: alexfh

Subscribers: cfe-commits

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248480 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ARM] Follow-up to fix crash "-target arm -mcpu=generic", without "-march="
Vladimir Sukharev [Thu, 24 Sep 2015 09:55:08 +0000 (09:55 +0000)]
[ARM] Follow-up to fix crash "-target arm -mcpu=generic", without "-march="

Fix of dangling StringRef after temporary std::string is destroyed

Follow-up to: http://reviews.llvm.org/rL248370

Reviewers: alexfh

Subscribers: cfe-commits

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248479 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Driver] Add support for Windows 10 SDK
Igor Kudrin [Thu, 24 Sep 2015 05:16:36 +0000 (05:16 +0000)]
[Driver] Add support for Windows 10 SDK

Summary:
With Windows 10 SDK, Include and Lib directories now contain an additional subfolder with the name that corresponds to the full version of the SDK, for example:
  - C:\Program Files (x86)\Windows Kits\10\Include\10.0.10240.0\um
  - C:\Program Files (x86)\Windows Kits\10\Lib\10.0.10240.0\um\x64

Reviewers: ruiu, rnk

Subscribers: majnemer, cfe-commits

Differential Revision: http://reviews.llvm.org/D12832

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248463 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agocuda-path: StringRef cannot be used here.
NAKAMURA Takumi [Thu, 24 Sep 2015 03:15:44 +0000 (03:15 +0000)]
cuda-path: StringRef cannot be used here.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248459 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang/test/Analysis/malloc-overflow2.c: Appease 32-bit targets.
NAKAMURA Takumi [Thu, 24 Sep 2015 02:49:00 +0000 (02:49 +0000)]
clang/test/Analysis/malloc-overflow2.c: Appease 32-bit targets.

size_t is not unsigned long for targeting i686 (and Windows x64).

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@248458 91177308-0d34-0410-b5e6-96231b3b80d8