[OPENMP][NVPTX]Run combined constructs with if clause in SPMD mode.
Combined constructs with parallel and if clauses without modifiers may
be executed in SPMD mode since if the condition is true for the target
region, it is also true for parallel region and the threads must be run
in parallel.
Hans Wennborg [Tue, 16 Apr 2019 12:13:25 +0000 (12:13 +0000)]
Re-commit r357452: SimplifyCFG SinkCommonCodeFromPredecessors: Also sink function calls without used results (PR41259)
The original commit caused false positives from AddressSanitizer's
use-after-scope checks, which have now been fixed in r358478.
> The code was previously checking that candidates for sinking had exactly
> one use or were a store instruction (which can't have uses). This meant
> we could sink call instructions only if they had a use.
>
> That limitation seemed a bit arbitrary, so this patch changes it to
> "instruction has zero or one use" which seems more natural and removes
> the need to special-case stores.
>
> Differential revision: https://reviews.llvm.org/D59936
David Blaikie [Tue, 16 Apr 2019 00:16:29 +0000 (00:16 +0000)]
DebugInfo: Default to standalone debug when tuning for LLDB
LLDB can't currently handle Clang's default (limit/no-standalone) DWARF,
so platforms that default to LLDB (Darwin) or anyone else manually
requesting LLDB tuning, should also get standalone DWARF.
That doesn't mean a user can't explicitly enable (because they have
other reasons to prefer standalone DWARF (such as that they're only
building half their application with debug info enabled, and half
without - or because they're tuning for GDB, but want to be able to use
it under LLDB too (this is the default on FreeBSD))) or disable (testing
LLDB fixes/improvements that handle no-standalone mode, building C code,
perhaps, which wouldn't have the LLDB<>no-standalone conflict, etc) the
feature regardless of the tuning.
[OPENMP][NVPTX]Run parallel regions with num_threads clauses in SPMD
mode.
After the previous patch with the more correct handling of the number of
threads in parallel regions, the parallel regions with num_threads
clauses can be executed in SPMD mode.
Reuben Thomas [Mon, 15 Apr 2019 20:13:20 +0000 (20:13 +0000)]
[clang-format] Fix -Wconversion-null warning in GCC
GCC -Wconversion-null warning appeared after 9a63380260860b657b72f07c4f0e61e382ab934a.
There was a similar problem already in the past:
http://lists.llvm.org/pipermail/cfe-commits/Week-of-Mon-20131230/096230.html
Louis Dionne [Mon, 15 Apr 2019 19:08:52 +0000 (19:08 +0000)]
Revert "[clang] Aligned allocation is actually supported in macosx 10.13"
This reverts r358409, which I think broke the bots in compiler-rt.
Since I'm having trouble reproducing the failure, I'm reverting this
until I can investigate locally.
Don Hinton [Mon, 15 Apr 2019 17:18:10 +0000 (17:18 +0000)]
[CommandLineParser] Add DefaultOption flag
Summary: Add DefaultOption flag to CommandLineParser which provides a
default option or alias, but allows users to override it for some
other purpose as needed.
Also, add `-h` as a default alias to `-help`, which can be seamlessly
overridden by applications like llvm-objdump and llvm-readobj which
use `-h` as an alias for other options.
(relanding after revert, r358414)
Added DefaultOptions.clear() to reset().
Louis Dionne [Mon, 15 Apr 2019 14:14:45 +0000 (14:14 +0000)]
[clang] Aligned allocation is actually supported in macosx 10.13
Summary:
In r350649, I changed aligned allocation from being available starting
in macosx10.13 to macosx10.14. However, aligned allocation is indeed
available starting with macosx10.13, my investigation had been based
on the wrong libc++abi dylib.
This means that Clang before the fix will be more stringent when it
comes to aligned allocation -- it will not allow it when back-deploying
to macosx 10.13, when it would actually be safe to do so.
Note that a companion change will be coming to fix the libc++ tests.
Eric Liu [Mon, 15 Apr 2019 08:46:34 +0000 (08:46 +0000)]
[Lookup] Invisible decls should not be ambiguous when renaming.
Summary:
For example, a renamed type in a header file can conflict with declaration in
a random file that includes the header, but we should not consider the decl ambiguous if
it's not visible at the rename location. This improves consistency of generated replacements
when header file is included in different TUs.
Don Hinton [Sat, 13 Apr 2019 16:55:28 +0000 (16:55 +0000)]
[CommandLineParser] Add DefaultOption flag
Summary: Add DefaultOption flag to CommandLineParser which provides a
default option or alias, but allows users to override it for some
other purpose as needed.
Also, add `-h` as a default alias to `-help`, which can be seamlessly
overridden by applications like llvm-objdump and llvm-readobj which
use `-h` as an alias for other options.
Richard Smith [Sat, 13 Apr 2019 04:33:39 +0000 (04:33 +0000)]
[verify] Add support for location markers in directives.
A marker (matching /#[A-Za-z0-9_-]/) is specified by attaching a comment
containing the marker to the line at which the diagnostic is expected,
and then can be referenced from an expected-* directive after an @:
The intent is for markers to be used in situations where relative line
numbers are currently used, to avoid the need to renumber when the test
case is rearranged.
[analyzer] Escape pointers stored into top-level parameters with destructors.
Writing stuff into an argument variable is usually equivalent to writing stuff
to a local variable: it will have no effect outside of the function.
There's an important exception from this rule: if the argument variable has
a non-trivial destructor, the destructor would be invoked on
the parent stack frame, exposing contents of the otherwise dead
argument variable to the caller.
If such argument is the last place where a pointer is stored before the function
exits and the function is the one we've started our analysis from (i.e., we have
no caller context for it), we currently diagnose a leak. This is incorrect
because the destructor of the argument still has access to the pointer.
The destructor may deallocate the pointer or even pass it further.
Treat writes into such argument regions as "escapes" instead, suppressing
spurious memory leak reports but not messing with dead symbol removal.
Original summary:
Emit !heapallocsite in the metadata for calls to functions marked with
__declspec(allocator). Eventually this will be emitted as S_HEAPALLOCSITE debug
info in codeview.
Yaxun Liu [Fri, 12 Apr 2019 16:23:31 +0000 (16:23 +0000)]
[HIP] Use -mlink-builtin-bitcode to link device library
Use -mlink-builtin-bitcode instead of llvm-link to link
device library so that device library bitcode and user
device code can be compiled in a consistent way.
This is the same approach used by CUDA and OpenMP.
Bruno Ricci [Fri, 12 Apr 2019 15:36:02 +0000 (15:36 +0000)]
[AST][NFC] Add const children() accessors to all AST nodes
Systematically add the const-qualified version of children()
to all statement/expression nodes. Previously the const-qualified
variant was only defined for some nodes. NFC.
Bruno Ricci [Fri, 12 Apr 2019 13:26:55 +0000 (13:26 +0000)]
[AST] Forbid copy/move of statements/types
Statements, expressions and types are not supposed to be copied/moved,
and trying to do so is only going to result in tears. Someone tripped
on this a few days ago on the mailing list. NFC.
[Aarch64] Add v8.2-a half precision element extract intrinsics
Summary:
Implements the intrinsics define on the ACLE to extract half precision fp scalar elements from float16x4_t and float16x8_t vector types.
a.k.a:
vduph_lane_f16
vduph_laneq_f16
[clang-format] Use SpacesBeforeTrailingComments for "option" directive
Summary:
AnnotatingParser::next() is needed to implicitly set TT_BlockComment
versus TT_LineComment. On most other paths through
AnnotatingParser::parseLine(), all tokens are consumed to achieve that.
This change updates one place where this wasn't done.
Summary:
alloca isn’t auto-init’d right now because it’s a different path in clang that
all the other stuff we support (it’s a builtin, not an expression).
Interestingly, alloca doesn’t have a type (as opposed to even VLA) so we can
really only initialize it with memset.
Richard Smith [Thu, 11 Apr 2019 21:18:22 +0000 (21:18 +0000)]
Remove use of lookahead from _Pragma handling and from all other
internal lexing steps in the preprocessor.
It is not safe to use the preprocessor's token lookahead except when
operating on the final sequence of tokens that would be produced by
phase 4 of translation. Doing so corrupts the token lookahead cache used
by the parser. (See added testcase for an example.) Lookahead should
instead be viewed as a layer on top of the normal lexer.
Added assertions to catch any further incorrect uses of lookahead within
lexing actions.
Aaron Smith [Thu, 11 Apr 2019 20:24:54 +0000 (20:24 +0000)]
[DebugInfo] Combine Trivial and NonTrivial flags
Summary:
These flags are used when emitting debug info and needed to initialize subprogram and member function attributes (function options) for Codeview. These function options are used to create an accurate compiler type for UDT symbols (class/struct/union) from PDBs.
The Trivial flag was introduced in https://reviews.llvm.org/D45122
It's been pointed out that Trivial and NonTrivial may imply each other and that seems to be the case in the current tests. This change combines them into a single flag -- NonTrivial -- and updates the corresponding unit tests. There is an additional change to llvm to update the flags.
[OpenCL] Re-fix invalid address space generation for clk_event_t arguments of enqueue_kernel builtin function
Summary:
https://reviews.llvm.org/D53809 fixed wrong address space(assert in debug build)
generated for event_ret argument. But exactly the same problem exists for
event_wait_list argument. This patch should fix both.
Erik Pilkington [Wed, 10 Apr 2019 21:18:21 +0000 (21:18 +0000)]
Fix a test, NFC
This test was duplicated, and the last declaration had some syntax errors since
the invalid attribute caused the @implementation to be skipped by the parser.
Dmitri Gribenko [Wed, 10 Apr 2019 20:25:07 +0000 (20:25 +0000)]
Check i < FD->getNumParams() before querying
Summary:
As was already stated in a previous comment, the parameter isn't
necessarily referring to one of the DeclContext's parameter. We
should check the index is within the range to avoid out-of-boundary
access.
Jan Korous [Wed, 10 Apr 2019 20:23:33 +0000 (20:23 +0000)]
[clang][ASTContext] Try to exit early before loading serialized comments from AST files
Loading external comments is expensive. This change probably doesn't apply to common cases but is almost for free and would save some work in case none of the declaration needs external comments to be loaded.
[OPENMP]Improve detection of number of teams, threads in target
regions.
Added more complex analysis for number of teams and number of threads in
the target regions, also merged related common code between CGOpenMPRuntime
and CGOpenMPRuntimeNVPTX classes.
[CodeGen][ObjC] Emit the retainRV marker as a module flag instead of
named metadata.
This fixes a bug where ARC contract wasn't inserting the retainRV
marker when LTO was enabled, which caused objects returned from a
function to be auto-released.
Tom Stellard [Tue, 9 Apr 2019 13:26:10 +0000 (13:26 +0000)]
Add support for detection of devtoolset-8
Summary:
The current llvm/clang et al. project can be built with the latest developer toolset (devtoolset-8) on RHEL, which provides GCC 8.2.1.
However, the result compiler will not identify this toolset itself when compiling programs, which is of course not desirable.
After the patch - which simply adds the name of the developer toolset to the existing list - it gets identified and selected, as shown below:
[bamboo@bamboo llvm-project]$ clang -v
clang version 9.0.0 (https://github.com/llvm/llvm-project.git e5ac385fb1ffa4bd3875ea6a4d24efdbd7814572)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /home/bamboo/llvm/bin
Found candidate GCC installation: /opt/rh/devtoolset-4/root/usr/lib/gcc/x86_64-redhat-linux/5.2.1
Found candidate GCC installation: /opt/rh/devtoolset-7/root/usr/lib/gcc/x86_64-redhat-linux/7
Found candidate GCC installation: /opt/rh/devtoolset-8/root/usr/lib/gcc/x86_64-redhat-linux/8
Found candidate GCC installation: /usr/lib/gcc/x86_64-redhat-linux/4.8.2
Found candidate GCC installation: /usr/lib/gcc/x86_64-redhat-linux/4.8.5
Selected GCC installation: /opt/rh/devtoolset-8/root/usr/lib/gcc/x86_64-redhat-linux/8
Candidate multilib: .;@m64
Candidate multilib: 32;@m32
Selected multilib: .;@m64
[ASTImporter] Fix in ASTImporter::Import_New(const Decl *)
Make sure ASTImporter::Import_New(const Decl *) returns
a Expected<const Decl *> and not Expected<Decl *> to
make the clang/unittests/AST/ASTImporterTest.cpp compile
without the warning
clang/unittests/AST/ASTImporterTest.cpp:117:12: error: no viable conversion from 'Expected<clang::Decl *>' to 'Expected<const clang::Decl *>'
return Imported;
Move the builtin headers to use the new license file header.
Summary:
These all had somewhat custom file headers with different text from the
ones I searched for previously, and so I missed them. Thanks to Hal and
Kristina and others who prompted me to fix this, and sorry it took so
long.
[ASTImporter] Call to HandleNameConflict in VisitEnumDecl mistakeningly using Name instead of SearchName
Summary:
https://reviews.llvm.org/D51633 added error handling to the ASTNodeImporter::VisitEnumDecl(...) for the conflicting names case. This could lead to erroneous return of an error in that case since we should have been using SearchName. Name may be empty in the case where we find the name via getTypedefNameForAnonDecl(...).
David Goldman [Mon, 8 Apr 2019 19:52:45 +0000 (19:52 +0000)]
Clean up ObjCPropertyDecl printing
Summary:
- `@property(attr, attr2)` instead of `@property ( attr,attr2 )`.
- Change priority of attributes (see code/comments inline).
- Support for printing weak and unsafe_unretained attributes.
Summary:
Emit !heapallocsite in the metadata for calls to functions marked with
__declspec(allocator). Eventually this will be emitted as S_HEAPALLOCSITE debug
info in codeview.
[OPENMP][NVPTX]Fixed processing of memory management directives.
Added special processing of the memory management directives/clauses for
NVPTX target. For private locals, omp_default_mem_alloc and
omp_thread_mem_alloc result in allocation in local memory.
omp_const_mem_alloc allocates const memory, omp_teams_mem_alloc
allocates shared memory, and omp_cgroup_mem_alloc and
omp_large_cap_mem_alloc allocate global memory.
The assertion prevents it from applying fixes when used along with compilation
databases with relative paths. Added a test that demonstrates the assertion
failure.
Simon Pilgrim [Mon, 8 Apr 2019 15:49:19 +0000 (15:49 +0000)]
Revert rL357915 from cfe/trunk: Remove a useless assertion in clang-check.
The assertion prevents it from applying fixes when used along with compilation
databases with relative paths. Added a test that demonstrates the assertion
failure.
----------------------------------------------------------------------
Ran 126 tests in 2.123s
FAILED (failures=2, skipped=6)
Unfortunately, this aborts the rest of `make check-all`, even with `-k`, so
this patch disables the test as is already done on a couple of other
targets.
This allowed the `sparc-sun-solaris2.11` test to finish.
The assertion prevents it from applying fixes when used along with compilation
databases with relative paths. Added a test that demonstrates the assertion
failure.
[Lexer] NFC: Fix an off-by-one bug in getAsCharRange().
As the unit test demonstrates, subtracting 1 from the offset was unnecessary.
The only user of this function was the plist file emitter (in Static Analyzer
and ARCMigrator). It means that a lot of Static Analyzer's plist arrows
are in fact off by one character. The patch carefully preserves this
completely incorrect behavior and causes no functional change,
i.e. no plist format breakage.
[analyzer] NoStoreFuncVisitor: Suppress reports with no-store in system headers.
The idea behind this heuristic is that normally the visitor is there to
inform the user that a certain function may fail to initialize a certain
out-parameter. For system header functions this is usually dictated by the
contract, and it's unlikely that the header function has accidentally
forgot to put the value into the out-parameter; it's more likely
that the user has intentionally skipped the error check.
Warnings on skipped error checks are more like security warnings;
they aren't necessarily useful for all users, and they should instead
be introduced on a per-API basis.
[CUDA][Windows] Last fix for the clang Bug 38811 "Clang fails to compile with CUDA-9.x on Windows" (https://bugs.llvm.org/show_bug.cgi?id=38811).
[IMPORTANT]
With that last fix, CUDA has just started being compiling by clang on Windows after nearly a year and two clang’s major releases (7 and 8).
As long as the last LLVM release, in which clang was compiling CUDA on Windows successfully, was 6.0.1, this fix and two previous have to be included into upcoming 7.1.0 and 8.0.1 releases.
[Output]
In file included from C:\GIT\LLVM\trunk-for-submits\llvm-64-release-vs2017-15.9.9\dist\lib\clang\9.0.0\include\__clang_cuda_runtime_wrapper.h:327:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0/include\crt/math_functions.hpp:390:11: error: no matching function for call to '__isinfl'
return (__isinfl(a) != 0);
^~~~~~~~
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0/include\crt/math_functions.hpp:2662:14: note: candidate function not viable: call to __host__ function from __device__ function
__func__(int __isinfl(long double a))
^
In file included from <built-in>:1:
In file included from C:\GIT\LLVM\trunk-for-submits\llvm-64-release-vs2017-15.9.9\dist\lib\clang\9.0.0\include\__clang_cuda_runtime_wrapper.h:327:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0/include\crt/math_functions.hpp:438:11: error: no matching function for call to '__isnanl'
return (__isnanl(a) != 0);
^~~~~~~~
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0/include\crt/math_functions.hpp:2672:14: note: candidate function not viable: call to __host__ function from __device__ function
__func__(int __isnanl(long double a))
^
In file included from <built-in>:1:
In file included from C:\GIT\LLVM\trunk-for-submits\llvm-64-release-vs2017-15.9.9\dist\lib\clang\9.0.0\include\__clang_cuda_runtime_wrapper.h:327:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0/include\crt/math_functions.hpp:486:11: error: no matching function for call to '__finitel'
return (__finitel(a) != 0);
^~~~~~~~~
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0/include\crt/math_functions.hpp:2652:14: note: candidate function not viable: call to __host__ function from __device__ function
__func__(int __finitel(long double a))
^
3 errors generated when compiling for sm_50.
[Solution]
Add missing long double device functions' declarations. Provide only declarations to prevent any use of long double on the device side, because CUDA does not support long double on the device side.
Sam McCall [Fri, 5 Apr 2019 15:22:20 +0000 (15:22 +0000)]
[Tooling] add a Heuristic field indicating that a CompileCommand was guessed.
Summary:
Use cases:
- a tool that dumps the heuristic used for each header in a project can
be used to evaluate changes to the heuristic
- we want to expose this information to users in clangd as it affects
accuracy/reliability of editor features
- express interpolation tests more directly
[LibTooling] Add Transformer, a library for source-to-source transformations.
Summary: Adds a basic version of Transformer, a library supporting the concise specification of clang-based source-to-source transformations. A full discussion of the end goal can be found on the cfe-dev list with subject "[RFC] Easier source-to-source transformations with clang tooling".
[LibTooling] Add "SourceCode" library for functions relating to source-code manipulation.
Summary:
Introduces a utility library in Refactoring/ to collect routines related to
source-code manipulation. In this change, we move "extended-range" functions
from the FixIt library (in clangTooling) to this new library.
We need to use this functionality in Refactoring/ and cannot access it if it
resides in Tooling/, because that would cause clangToolingRefactor to depend on
clangTooling, which would be a circular dependency.
Alex Lorenz [Fri, 5 Apr 2019 01:48:11 +0000 (01:48 +0000)]
[test] Specify an explicit darwin version in a triple in
`test/Driver/debug-options.c` to ensure that the driver
selects the DWARF 2 version as intended by the test.
Fixes the `test/Driver/debug-options.c` test regression on GreenDragon
on macOS that started failing after r357713.
Make SourceManager::createFileID(UnownedTag, ...) take a const llvm::MemoryBuffer*
Requires making the llvm::MemoryBuffer* stored by SourceManager const,
which in turn requires making the accessors for that return const
llvm::MemoryBuffer*s and updating all call sites.
The original motivation for this was to use it and fix the TODO in
CodeGenAction.cpp's ConvertBackendLocation() by using the UnownedTag
version of createFileID, and since llvm::SourceMgr* hands out a const
llvm::MemoryBuffer* this is required. I'm not sure if fixing the TODO
this way actually works, but this seems like a good change on its own
anyways.
Stephen Hines [Thu, 4 Apr 2019 18:17:46 +0000 (18:17 +0000)]
Verify that Android targets generate DWARF 4 by default.
Summary:
In the future, Android releases will support DWARF 5, but we need to
ensure that older targets only have DWARF 4 generated for them. This
patch inserts that verification for all Android releases now. The patch
also fixes 2 minor mistakes (a mistakenly moved RUN line, and the
missing G_DWARF2 check label).
projects / clang / log