Andrey Bokhanko [Mon, 15 Feb 2016 10:39:04 +0000 (10:39 +0000)]
PR26449: Fixes for bugs in __builtin_classify_type implementation
This patch fixes the following bugs in __builtin_classify_type implementation:
1) Support for member functions and fields
2) Same behavior as GCC in C mode (specifically, return integer_type_class for
enums and pointer_type_class for function pointers and arrays). Behavior in
C++ mode didn't changed.
Also, it refactors the whole implementation, by replacing a sequence of
if-else-if with a couple of switches.
Alexey Bataev [Mon, 15 Feb 2016 08:07:17 +0000 (08:07 +0000)]
[OPENMP] Remove extra sync barriers for 'firstprivate' clause.
Sync barrier will be emitted after generation of firstprivate variables
only if one of the firstprivate vars is used in lastprivate clause.
In the case that the array indexing itself is within a type dependent context,
bail out of the evaluation. We would previously try to symbolically evaluate
the expression which would then try to evaluate a non-address expression as an
address, triggering an assertion in Asserts builds.
We only need to consider the array subscript expression itself as in the case
that the base itself being type dependent is handled appropriately in EvalAddr.
[AST/index] Introduce an option 'SuppressTemplateArgsInCXXConstructors' in printing policy.
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.
These codepaths would generate warnings with GCC on linux even though the switch
was covered. Add llvm_unreachable markers to indicate that the switch should be
covered. NFC.
Dimitry Andric [Sun, 14 Feb 2016 16:08:20 +0000 (16:08 +0000)]
As reported in https://llvm.org/bugs/show_bug.cgi?id=25496, on FreeBSD,
C++ programs compiled for profiling (using `-pg`) should be linked with
`-lc++_p` (or `-lstdc++_p`, depending on the `-stdlib=` setting), not
with the regular C++ libraries.
Add a `FreeBSD::AddCXXStdlibLibArgs()` override to handle this, and add
a test case for it. While here, extend the test case for the proper
passing of -lm and -lm_p.
Samuel Antao [Sat, 13 Feb 2016 23:35:10 +0000 (23:35 +0000)]
[OpenMP] Rename the offload entry points.
Summary:
Unlike other outlined regions in OpenMP, offloading entry points have to have be visible (external linkage) for the device side. Using dots in the names of the entries can be therefore problematic for some toolchains, e.g. NVPTX.
Also the patch drops the column information in the unique name of the entry points. The parsing of directives ignore unknown tokens, preventing several target regions to be implemented in the same line. Therefore, the line information is sufficient for the name to be unique. Also, the preprocessor printer does not preserve the column information, causing offloading-entry detection issues if the host uses an integrated preprocessor and the target doesn't (or vice versa).
Benjamin Kramer [Sat, 13 Feb 2016 18:11:49 +0000 (18:11 +0000)]
Fix a leak in the generated code for attributes with strings.
Storing std::strings in attributes simply doesn't work, we never call
the destructor. Use an array of StringRefs instead of std::strings and
copy the data into memory taken from the ASTContext.
Bob Wilson [Sat, 13 Feb 2016 01:41:41 +0000 (01:41 +0000)]
[Sema] More changes to fix Objective-C fallout from r249995.
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.
[RecursiveASTVisitor] Introduce dataTraverseStmtPre()/dataTraverseStmtPost() to allow clients to do before/after actions during data recursive visitation.
This should fix the asan bot that hits stack overflow in a couple of test/Index tests.
Richard Trieu [Sat, 13 Feb 2016 00:58:53 +0000 (00:58 +0000)]
Make -Wnull-conversion more useful.
When a null constant is used in a macro, walk through the macro stack to
determine where the null constant is written and where the context is located.
Only warn if both locations are within the same macro expansion. This helps
function-like macros which involve pointers be treated as if they were
functions.
Chris Bieneman [Fri, 12 Feb 2016 21:36:55 +0000 (21:36 +0000)]
[CMake] Improve the clang order-file generation workflow
Summary:
This commit re-lands r259862. The underlying cause of the build breakage was an incorrectly written capabilities test. In tools/Driver/CMakeLists.txt I was attempting to check if a linker flag worked, the test was passing it to the compiler, not the linker. CMake doesn't have a linker test, so we have a hand-rolled one.
Original Patch Review: http://reviews.llvm.org/D16896
Original 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.
Hubert Tong [Fri, 12 Feb 2016 19:24:36 +0000 (19:24 +0000)]
test/Headers/float.c: fix theoretical edge values
For *_MANT_DIG, *_MAX_EXP and *_MIN_EXP, the C Standard does not list
the least requirements directly. This patch adjusts the test values with
refined ones.
Artem Belevich [Fri, 12 Feb 2016 18:29:18 +0000 (18:29 +0000)]
[CUDA] Tweak attribute-based overload resolution to match nvcc behavior.
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.
[Hexagon] Specify vector alignment in DataLayout string
The DataLayout can calculate alignment of vectors based on the alignment
of the element type and the number of elements. In fact, it is the product
of these two values. The problem is that for vectors of N x i1, this will
return the alignment of N bytes, since the alignment of i1 is 8 bits. The
vector types of vNi1 should be aligned to N bits instead. Provide explicit
alignment for HVX vectors to avoid such complications.
Now that Sparc/Sparc64 backend is mostly usable, provide the same
linking defaults as other NetBSD targets, i.e. compiler_rt-in-libc and
libc++ as STL.
Devin Coughlin [Thu, 11 Feb 2016 22:13:20 +0000 (22:13 +0000)]
[analyzer] Improve pattern matching in ObjCDealloc checker.
Look through PseudoObjectExpr and OpaqueValueExprs when scanning for
release-like operations. This commit also adds additional tests in anticipation
of re-writing this as a path-sensitive checker.
Tim Northover [Thu, 11 Feb 2016 21:31:51 +0000 (21:31 +0000)]
Relax recently added clang version checks.
You can override the value of these during CMake, and we often use sentinels
with more than one digit (not to mention our actual Clang being 700.whatever).
Manman Ren [Thu, 11 Feb 2016 19:19:18 +0000 (19:19 +0000)]
[PR26550] Use a different TBAA root for C++ vs C.
This commit changes the root from "Simple C/C++ TBAA" to "Simple C++ TBAA" for
C++.
The problem is that the type name in the TBAA nodes is generated differently
for C vs C++. If we link an IR file for C with an IR file for C++, since they
have the same root and the type names are different, accesses to the two type
nodes will be considered no-alias, even though the two type nodes are from
the same type in a header file.
The fix is to use different roots for C and C++. Types from C will be treated
conservatively in respect to types from C++.
Follow-up commits will change the C root to "Simple C TBAA" plus some mangling
change for C types to make it a little more aggresive.
Ben Langmuir [Thu, 11 Feb 2016 18:54:02 +0000 (18:54 +0000)]
[Modules] Early-exit if ReadOptionsBlock fails to avoid crashing
If we didn't tell ReadOptionsBlock to allow failures then we can't
assume that the stream is not in the middle of a block if it returns
out-of-date. This was causing a crash when we tried to continue reading.
Also, it's just generally a good idea to early-exit if we're doing
implicit module builds, since we will want to immediately rebuild this
module anyway and there's no reason to waste time continuing after
failure.
David Majnemer [Thu, 11 Feb 2016 17:49:28 +0000 (17:49 +0000)]
Revert "Revert r260388 "[MS ABI] Never reference dllimport'd vtables""
This reverts commit r260449.
We would supress our emission of vftable definitions if we thought
another translation unit would provide the definition because we saw an
explicit instantiation declaration. This is not the case with
dllimport, we want to synthesize a definition of the vftable regardless.
Ben Langmuir [Thu, 11 Feb 2016 17:04:42 +0000 (17:04 +0000)]
[Modules] Don't infinite recurse on implicit import of circular modules in preamble
Update the Preprocessor's VisibleModuleSet when typo-correction creates
an implicit module import so that we won't accidentally write an invalid
SourceLocation into the preamble AST. This would later lead to infinite
recursion when loading the preamble AST because we use the value in
ImportLocs to prevent visiting a module twice.
James Y Knight [Thu, 11 Feb 2016 16:51:29 +0000 (16:51 +0000)]
Use new --match-full-lines FileCheck feature for Preprocessor/init.c.
This required fixing a few check lines which had omitted trailing
characters, and were passing incorrectly (e.g., asserting that
__UINT64_C_SUFFIX__ is "UL" instead of the "ULL" that it actually is set
to). All were obviously broken tests, not broken code.
Oliver Stannard [Thu, 11 Feb 2016 16:05:52 +0000 (16:05 +0000)]
[ARM] Add command-line options for ARMv8.2-A
This allows ARMv8.2-A to be targeted either by using "armv8.2a" in the
triple, or by using -march=armv8.2-a (or the alias -march=armv8.2a).
The FP16 extension can be enabled with the "+fp16" suffix to the -march
or -mcpu option. This is consistent with the AArch64 option, rather than
the usual ARM option of -mfpu. We have agreed with the team which will
be upstreaming this to GCC that we want to use this new option format
for new architecture extensions for both ARM and AArch64.
Most of the work for this was done by the TargetParser patch in llvm.
Andrey Bokhanko [Thu, 11 Feb 2016 13:27:02 +0000 (13:27 +0000)]
Partial revert of rL260506.
After some experiments I discovered that clang doesn't support static
initialization of flexible array members in full, so restored this paragraph in
"GCC extensions not implemented yet" list.
Andrey Bokhanko [Thu, 11 Feb 2016 10:36:06 +0000 (10:36 +0000)]
Update of "GCC extensions not implemented yet" in Clang User's Manual
#pragma weak, global register variables and static initialization of flexible
array members are supported now, so they are removed from "GCC extensions not
implemented yet" list.
Akira Hatanaka [Thu, 11 Feb 2016 06:36:35 +0000 (06:36 +0000)]
[Objective-c] Stop attaching section "datacoal_nt" to global variables.
The current macho linker just copies symbols in section datacoal_nt to
section data, so it doesn't really matter whether or not section
"datacoal_nt" is attached to the global variable.
This is a follow-up to r250370, which made changes in llvm to stop
putting functions and data in the *coal* sections.
Alexey Bataev [Thu, 11 Feb 2016 05:35:55 +0000 (05:35 +0000)]
[OPENMP] Rename OMPCapturedFieldDecl to OMPCapturedExprDecl, NFC.
OMPCapturedExprDecl allows caopturing not only of fielddecls, but also
other expressions. It also allows to simplify codegen for several
clauses.
Justin Lebar [Wed, 10 Feb 2016 22:17:00 +0000 (22:17 +0000)]
Bail on compilation as soon as a job fails.
Previously we attempted to be smart; if one job failed, we'd run all
jobs that didn't depend on the failing job.
Problem is, this doesn't work well for e.g. CUDA compilation without
-save-temps. In this case, the device-side and host-side Assemble
actions (which actually are responsible for preprocess, compile,
backend, and assemble, since we're not saving temps) are necessarily
distinct. So our clever heuristic doesn't help us, and we repeat every
error message once for host and once for each device arch.
The main effect of this change, other than fixing CUDA, is that if you
pass multiple cc files to one instance of clang and you get a compile
error, we'll stop when the first cc1 job fails.
Reid Kleckner [Wed, 10 Feb 2016 21:28:38 +0000 (21:28 +0000)]
[clang-cl] /Z7 now generates normal debug info, not just line info
Previously LLVM could not process any debug info we produced, so it
didn't make sense to spend time generating it. Now that it has primitive
support for local variable info, it does make sense to generate normal
debug info.
David Majnemer [Wed, 10 Feb 2016 17:40:47 +0000 (17:40 +0000)]
[MS ABI] Never reference dllimport'd vtables
Referencing a dllimported vtable is impossible in a constexpr
constructor. It would be friendlier to C++ programmers if we
synthesized a copy of the vftable which referenced imported virtual
functions. This would let us initialize the object in a way which
preserves both the intent to import functionality from another DLL while
also making constexpr work.
projects / clang / log