[OpenMP] Codegen for the 'target parallel' directive on the NVPTX device.
This patch adds codegen for the 'target parallel' directive on the NVPTX
device. We term offload OpenMP directives such as 'target parallel' and
'target teams distribute parallel for' as SPMD constructs. SPMD constructs,
in contrast to Generic ones like the plain 'target', can never contain
a serial region.
SPMD constructs can be handled more efficiently on the GPU and do not
require the Warp Loop of the Generic codegen scheme. This patch adds
SPMD codegen support for 'target parallel' on the NVPTX device and can
be reused for other SPMD constructs.
This rule permits the injected-class-name of a class template to be used as
both a template type argument and a template template argument, with no extra
syntax required to disambiguate.
[OpenMP] Codegen support for 'target parallel' on the host.
This patch adds support for codegen of 'target parallel' on the host.
It is also the first combined directive that requires two or more
captured statements. Support for this functionality is included in
the patch.
A combined directive such as 'target parallel' has two captured
statements, one for the 'target' and the other for the 'parallel'
region. Two captured statements are required because each has
different implicit parameters (see SemaOpenMP.cpp). For example,
the 'parallel' has 'global_tid' and 'bound_tid' while the 'target'
does not. The patch adds support for handling multiple captured
statements based on the combined directive.
When codegen'ing the 'target parallel' directive, the 'target'
outlined function is created using the outer captured statement
and the 'parallel' outlined function is created using the inner
captured statement.
Benjamin Kramer [Wed, 18 Jan 2017 15:50:26 +0000 (15:50 +0000)]
[Basic] Remove source manager references from diag state points.
This is just wasted space, we don't support state points from multiple
source managers. Validate that there's no state when resetting the
source manager and use the 'global' reference to the sourcemanager
instead of the ones in the diag state.
Jonathan Roelofs [Wed, 18 Jan 2017 15:31:11 +0000 (15:31 +0000)]
Warn when calling a non interrupt function from an interrupt on ARM
The idea for this originated from a really tricky bug: ISRs on ARM don't
automatically save off the VFP regs, so if say, memcpy gets interrupted and the
ISR itself calls memcpy, the regs are left clobbered when the ISR is done.
[OpenMP] Codegen support for 'target parallel' on the host.
This patch adds support for codegen of 'target parallel' on the host.
It is also the first combined directive that requires two or more
captured statements. Support for this functionality is included in
the patch.
A combined directive such as 'target parallel' has two captured
statements, one for the 'target' and the other for the 'parallel'
region. Two captured statements are required because each has
different implicit parameters (see SemaOpenMP.cpp). For example,
the 'parallel' has 'global_tid' and 'bound_tid' while the 'target'
does not. The patch adds support for handling multiple captured
statements based on the combined directive.
When codegen'ing the 'target parallel' directive, the 'target'
outlined function is created using the outer captured statement
and the 'parallel' outlined function is created using the inner
captured statement.
Akira Hatanaka [Tue, 17 Jan 2017 19:35:54 +0000 (19:35 +0000)]
[Sema] Fix bug in handling of designated initializer.
CheckDesignatedInitializer wasn't taking into account the base classes
when computing the index for the field in the derived class, which
caused the test case to crash during IRGen because of a malformed AST.
George Rimar [Tue, 17 Jan 2017 15:45:31 +0000 (15:45 +0000)]
[Clang] - Update code to match upcoming llvm::zlib API.
D28684 changed llvm::zlib to return Error instead of Status.
It was accepted and committed in r292214, but then reverted in r292217
because I missed that clang code also needs to be updated.
Frederic Riss [Tue, 17 Jan 2017 03:38:45 +0000 (03:38 +0000)]
Fix AArch64 global-merge backend option name.
-mglobal-merge is translated to the appropriate backend option in
the driver. r277322 changed the AArch64 option name in the backend,
but the driver was never updated.
Richard Smith [Tue, 17 Jan 2017 02:14:37 +0000 (02:14 +0000)]
Partial revert of r290511.
The rules around typechecking deduced template arguments during partial
ordering are not clear, and while the prior behavior does not seem to be
correct (it doesn't follow the general model of partial ordering where each
template parameter is replaced by a non-dependent but unique value), the new
behavior is also not clearly right and breaks some existing idioms.
The new behavior is retained for dealing with non-type template parameters
with 'auto' types, as without it even the most basic uses of that feature
don't work. We can revisit this once CWG has come to an agreement on how
partial ordering with 'auto' non-type template parameters is supposed to
work.
Antonio Maiorano [Tue, 17 Jan 2017 00:12:27 +0000 (00:12 +0000)]
clang-format: Make GetStyle return Expected<FormatStyle> instead of FormatStyle
Change the contract of GetStyle so that it returns an error when an error occurs
(i.e. when it writes to stderr), and only returns the fallback style when it
can't find a configuration file.
[OpenMP] Refactor code that calls codegen for target regions on the device.
This patch refactors code that calls codegen for target regions. Currently
the codebase only supports the 'target' directive. The patch pulls out
common target processing code into a static function that can be called
by codegen for any target directive.
Martin Probst [Mon, 16 Jan 2017 09:52:40 +0000 (09:52 +0000)]
clang-format: [JS] revert over-eager ASI check.
Summary: Change r291428 introduced ASI detection after closing curly braces. That would generally be correct, however this breaks indentation for structural statements. What happens is that CompoundStatementIndenter increases indentation for the current line, then after reading ASI creates a new line (with the increased line level), and only after the structural parser sees e.g. the if/then/else branch closed, line level is reduced. That leads to the new line started by ASI having a level too high.
Daniel Jasper [Fri, 13 Jan 2017 23:18:16 +0000 (23:18 +0000)]
clang-format: Fix bug in making line break decisions.
Here, the optimization to not line wrap when it would not lead to a
reduction in columns was overwriting and enforced break that we want to
do no matter what.
Eric Fiselier [Fri, 13 Jan 2017 22:11:40 +0000 (22:11 +0000)]
[clang] Emit `diagnose_if` warnings from system headers
Summary: In order for libc++ to meaningfully use `diagnose_if` warnings they need to be emitted from system headers by default. This patch changes the `diagnose_if` warning diagnostic to be shown in system headers.
Reid Kleckner [Fri, 13 Jan 2017 17:18:19 +0000 (17:18 +0000)]
Use less byval on 32-bit Windows x86 for classes with bases
This comes up in V8, which has a Handle template class that wraps a
typed pointer, and is frequently passed by value. The pointer is stored
in the base, HandleBase. This change allows us to pass the struct as a
pointer instead of using byval. This avoids creating tons of temporary
allocas that we copy from during call lowering.
Eventually, it would be good to use FCAs here instead.
Richard Smith [Fri, 13 Jan 2017 02:22:01 +0000 (02:22 +0000)]
Implement DR1265 (wg21.link/cwg1265).
Diasllow a declaration using the 'auto' type specifier from using two different
meanings of it at once, or from declaring multiple functions with deduced
return types or introducing multiple trailing return types.
The standard does not technically disallow the multiple trailing return types
case if all the declarators declare variables (such as function pointers with
trailing return types), but we disallow that too, following the clear intent.
In the case where the template class itself is already `dllexport`, the
implicit instantiation will have already emitted all members. When we
check the explicit instantiation definition, the `Specialization` will
have inherited the `dllexport` attribute, so we'll attempt to emit all
members for a second time, which causes an assertion failure. Restrict
the exporting to when the `dllexport` attribute is newly introduced by
the explicit instantiation definition.
Richard Smith [Fri, 13 Jan 2017 00:57:54 +0000 (00:57 +0000)]
Update C++ status pages for Clang 4 branch:
* Update version number in DR tests from 4.0 to 4
* Teach make_cxx_dr_status script about version numbers that don't contain a
period.
* Update cxx_status.html and cxx_dr_status.html to list Clang 4 features as
"Clang 4" rather than "SVN"
Clang 4 features are still listed in yellow rather than green until release.
Dehao Chen [Fri, 13 Jan 2017 00:51:55 +0000 (00:51 +0000)]
Pass -fprofile-sample-use to lto backends.
Summary: LTO backend will not invoke SampleProfileLoader pass even if -fprofile-sample-use is specified. This patch passes the flag down so that pass manager can add the SampleProfileLoader pass correctly.
Anna Zaks [Fri, 13 Jan 2017 00:50:50 +0000 (00:50 +0000)]
[tsan] Do not report errors in __destroy_helper_block_
There is a synchronization point between the reference count of a block dropping to zero and it's destruction, which TSan does not observe. Do not report errors in the compiler-emitted block destroy method and everything called from it.
This is similar to https://reviews.llvm.org/D25857
Anna Zaks [Fri, 13 Jan 2017 00:50:41 +0000 (00:50 +0000)]
[analyzer] Fix false positives in Keychain API checker
The checker has several false positives that this patch addresses:
- Do not check if the return status has been compared to error (or no error) at the time when leaks are reported since the status symbol might no longer be alive. Instead, pattern match on the assume and stop tracking allocated symbols on error paths.
- The checker used to report error when an unknown symbol was freed. This could lead to false positives, let's not repot those. This leads to loss of coverage in double frees.
- Do not enforce that we should only call free if we are sure that error was not returned and the pointer is not null. That warning is too noisy and we received several false positive reports about it. (I removed: "Only call free if a valid (non-NULL) buffer was returned")
- Use !isDead instead of isLive in leak reporting. Otherwise, we report leaks for objects we loose track of. This change triggered change #1.
This also adds checker specific dump to the state.
Richard Smith [Fri, 13 Jan 2017 00:43:31 +0000 (00:43 +0000)]
Improve handling of instantiated thread_local variables in Itanium C++ ABI.
* Do not initialize these variables when initializing the rest of the
thread_locals in the TU; they have unordered initialization so they can be
initialized by themselves.
This fixes a rejects-valid bug: we would make the per-variable initializer
function internal, but put it in a comdat keyed off the variable, resulting
in link errors when the comdat is selected from a different TU (as the per
TU TLS init function tries to call an init function that does not exist).
* On Darwin, when we decide that we're not going to emit a thread wrapper
function at all, demote its linkage to External. Fixes a verifier failure
on explicit instantiation of a thread_local variable on Darwin.
Daniel Jasper [Thu, 12 Jan 2017 19:35:26 +0000 (19:35 +0000)]
clang-format: Treat braced lists like other complex parameters.
Specifically, wrap before them if they are multi-line so that we don't
create long hanging indents. This prevents having a lot of code
indented a lot in some cases.
Dehao Chen [Thu, 12 Jan 2017 16:29:25 +0000 (16:29 +0000)]
Pass -fprofile-sample-use to lto backends.
Summary: LTO backend will not invoke SampleProfileLoader pass even if -fprofile-sample-use is specified. This patch passes the flag down so that pass manager can add the SampleProfileLoader pass correctly.
Malcolm Parsons [Thu, 12 Jan 2017 16:11:28 +0000 (16:11 +0000)]
Tracking exception specification source locations
Summary:
We do not currently track the source locations for exception specifications such
that their source range can be queried through the AST. This leads to trying to
write more complex code to determine the source range for uses like FixItHints
(see D18575 for an example). In addition to use within tools like clang-tidy, I
think this information may become more important to track as exception
specifications become more integrated into the type system.
Alex Lorenz [Thu, 12 Jan 2017 10:48:03 +0000 (10:48 +0000)]
Avoid multiple -Wunreachable-code diagnostics that are triggered by
the same source range and use the unary operator fixit only when it
actually silences the warning.
Artem Dergachev [Thu, 12 Jan 2017 09:46:16 +0000 (09:46 +0000)]
[analyzer] Avoid a crash in DereferenceChecker on string literal initializers.
A hotfix for pr31592 that fixes the crash but not the root cause of the problem.
We need to update the analyzer engine further to account for AST changes
introduced in r289618. At the moment we're erroneously performing a redundant
lvalue-to-rvalue cast in this scenario, and squashing the rvalue of the object
bound to the reference into the reference itself.
Vassil Vassilev [Thu, 12 Jan 2017 09:16:26 +0000 (09:16 +0000)]
PR31469: Don't add friend template class decls to redecl chain in dependent contexts.
Fixes a crash in modules where the template class decl becomes the most recent
decl in the redeclaration chain and forcing the template instantiator try to
instantiate the friend declaration, rather than the template definition.
In practice, A::list<int> produces a TemplateSpecializationType
A::__1::list<int, allocator<type-parameter-0-0> >' failing to replace to
subsitute the default argument to allocator<int>.
Richard Smith [Thu, 12 Jan 2017 02:27:38 +0000 (02:27 +0000)]
Remove redundant passing around of a "ContainsAutoType" flag.
This flag serves no purpose other than to prevent us walking through a type to
check whether it contains an 'auto' specifier; this duplication of information
is error-prone, does not appear to provide any performance benefit, and will
become less practical once we support C++1z deduced class template types and
eventually constrained types from the Concepts TS.
Manman Ren [Wed, 11 Jan 2017 18:32:30 +0000 (18:32 +0000)]
This reverts r291628. As suggested by Richard, we can simply
filter out the implicilty imported modules at CodeGen instead of removing the
implicit ImportDecl when an implementation TU of a module imports a header of
that same module.
Devin Coughlin [Wed, 11 Jan 2017 01:02:34 +0000 (01:02 +0000)]
[analyzer] Fix crash in body farm for getter without implicit self.
Fix a crash in body farm when synthesizing a getter for a property
synthesized for a property declared in a protocol on a class extension
that shadows a declaration of the property in a category.
In this case, Sema doesn't fill in the implicit 'self' parameter for the getter
in the category, which leads to a crash when trying to synthesize the getter
for it.
To avoid the crash, skip getter synthesis in body farm if the self parameter is
not filled int.
Summary:
openSuse has AArch64 support, with images running on the Raspberry Pi 3.
The libraries and headers live under the aarch64-suse-linux subdirectory,
which is currently not in the AArch64 triples list. Address this by adding
the corresponding string to AArch64Triples.
Richard Smith [Tue, 10 Jan 2017 20:52:50 +0000 (20:52 +0000)]
Don't try to check implicit conversion sequences for an object argument if
there is no object argument, when early checking of implicit conversion
sequences for a function template fails.