Julie Hockett [Wed, 9 May 2018 18:27:33 +0000 (18:27 +0000)]
[clang] Adding CharacteristicKind to PPCallbacks::InclusionDirective
Adding a SrcMgr::CharacteristicKind parameter to the InclusionDirective
in PPCallbacks, and updating calls to that function. This will be useful
in https://reviews.llvm.org/D43778 to determine which includes are system
headers.
Alexey Bataev [Wed, 9 May 2018 14:15:18 +0000 (14:15 +0000)]
[OPENMP] Mark global tors/dtors as used.
If the global variables are marked as declare target and they need
ctors/dtors, these ctors/dtors are emitted and then invoked by the
offloading runtime library. They are not explicitly used in the emitted
code and thus can be optimized out. Patch marks these functions as used,
so the optimizer cannot remove these function during the optimization
phase.
Support the `__global`, `__local`, `__constant`, `__private`, and
`__generic` keywords in OpenCL C++. Leave the unprefixed address
space qualifiers such as global available, i.e., do not mark them as
reserved keywords in OpenCL C++. libclcxx provides explicit address
space pointer classes such as `global_ptr` and `global<T>` that are
implemented using the `__`-prefixed qualifiers.
Hans Wennborg [Wed, 9 May 2018 09:29:58 +0000 (09:29 +0000)]
Revert r331843 "[DebugInfo] Generate debug information for labels."
It broke the Chromium build (see reply on the review).
> Generate DILabel metadata and call llvm.dbg.label after label
> statement to associate the metadata with the label.
>
> Differential Revision: https://reviews.llvm.org/D45045
>
> Patch by Hsiangkai Wang.
This doesn't revert the change to backend-unsupported-error.ll
that seems to correspond to an llvm-side change.
Martin Storsjo [Wed, 9 May 2018 09:11:01 +0000 (09:11 +0000)]
Revert "[Driver] Use -fuse-line-directives by default in MSVC mode"
This reverts commit SVN r331666.
It was afterwards pointed out in https://reviews.llvm.org/D46520
that #line directives lose information about what parts come from a
system header. That means the result of -E usually won't compile,
since Windows headers are typically full of warnings and
default-error warnings.
[clang-format] Respect BreakBeforeClosingBrace while calculating length
Summary:
This patch makes `getLengthToMatchingParen` respect the `BreakBeforeClosingBrace`
ParenState for matching scope closers. In order to distinguish between paren states
introduced by real vs. fake parens, I've added the token opening the ParensState
to that struct.
Simon Atanasyan [Wed, 9 May 2018 08:42:30 +0000 (08:42 +0000)]
[driver] Add mips_Features_Group to Options to improve documentation sorting
Move all of the MIPS-only options into a new m_mips_Features_Group.
Nearly all other targets have most target-specific options grouped,
but MIPS does not.
The primary benefits are that the options will be listed together (and
thus identifiable as MIPS-specific even if they have no help string) in
the ClangCommandLineReference, and that Options.td is a bit more organized.
A secondary benefit is that a custom version of clang can more easily
hide/disable groups of options for unsupported targets.
JF Bastien [Wed, 9 May 2018 03:51:12 +0000 (03:51 +0000)]
_Atomic of empty struct shouldn't assert
Summary:
An _Atomic of an empty struct is pretty silly. In general we just widen empty
structs to hold a byte's worth of storage, and we represent size and alignment
as 0 internally and let LLVM figure out what to do. For _Atomic it's a bit
different: the memory model mandates concrete effects occur when atomic
operations occur, so in most cases actual instructions need to get emitted. It's
really not worth trying to optimize empty struct atomics by figuring out e.g.
that a fence would do, even though sane compilers should do optimize atomics.
Further, wg21.link/p0528 will fix C++20 atomics with padding bits so that
cmpxchg on them works, which means that we'll likely need to do the zero-init
song and dance for empty atomic structs anyways (and I think we shouldn't
special-case this behavior to C++20 because prior standards are just broken).
This patch therefore makes a minor change to r176658 "Promote atomic type sizes
up to a power of two": if the width of the atomic's value type is 0, just use 1
byte for width and leave alignment as-is (since it should never be zero, and
over-aligned zero-width structs are weird but fine).
This fixes an assertion:
(NumBits >= MIN_INT_BITS && "bitwidth too small"), function get, file ../lib/IR/Type.cpp, line 241.
It seems like this has run into other assertions before (namely the unreachable
Kind check in ImpCastExprToType), but I haven't reproduced that issue with
tip-of-tree.
Adrian Prantl [Wed, 9 May 2018 01:00:01 +0000 (01:00 +0000)]
Remove \brief commands from doxygen comments.
This is similar to the LLVM change https://reviews.llvm.org/D46290.
We've been running doxygen with the autobrief option for a couple of
years now. This makes the \brief markers into our comments
redundant. Since they are a visual distraction and we don't want to
encourage more \brief markers in new code either, this patch removes
them all.
Patch produced by
for i in $(git grep -l '\@brief'); do perl -pi -e 's/\@brief //g' $i & done
for i in $(git grep -l '\\brief'); do perl -pi -e 's/\\brief //g' $i & done
Petr Hosek [Wed, 9 May 2018 00:58:12 +0000 (00:58 +0000)]
Set CMAKE_BUILD_WITH_INSTALL_RPATH for Fuchsia runtimes
This doesn't make any difference since we don't use RPATH/RUNPATH
on Fuchsia but it avoids the CMake error when re-linking libraries
while building with Ninja.
Erich Keane [Tue, 8 May 2018 21:26:21 +0000 (21:26 +0000)]
Fix float->int conversion warnings when near barriers.
As Eli brought up here: https://reviews.llvm.org/D46535
I'd previously messed up this fix by missing conversions
that are just slightly outside the range. This patch fixes
this by no longer ignoring the return value of
convertToInteger. Additionally, one of the error messages
wasn't very sensical (mentioning out of range value, when it
really was not), so it was cleaned up as well.
Add a mno-outline flag to disable the MachineOutliner
Since we're working on turning the MachineOutliner by default under -Oz for
AArch64, it makes sense to have an -mno-outline flag available. This currently
doesn't do much (it basically just undoes -moutline).
When the MachineOutliner is on by default under AArch64, this flag should
set -mllvm -enable-machine-outliner=never.
Martin Storsjo [Tue, 8 May 2018 20:55:23 +0000 (20:55 +0000)]
[Driver] Don't add -dwarf-column-info when using -gcodeview on non-msvc targets
-dwarf-column-info is omitted if -gcodeview is specified for msvc
targets at the moment, but since -gcodeview is an option that can be
specified for any target, there's little reason to restrict this
handling to msvc targets.
This allows getting proper codeview debug info by passing -gcodeview
for e.g. MinGW targets as well.
Richard Smith [Mon, 7 May 2018 22:23:38 +0000 (22:23 +0000)]
PR37352: mangle numbering for decomposition declarations.
In order to match our mangling scheme, use a different set of numbers for
decomposition declarations, and consider all binding names when forming the
numbering. This does not yet affect any mangled names we produce, because
local decomposition declarations can't yet have linkage, but a C++ standard
proposal to change that is currently being processed.
Eric Fiselier [Mon, 7 May 2018 21:07:10 +0000 (21:07 +0000)]
[C++2a] Implement operator<=> CodeGen and ExprConstant
Summary:
This patch tackles long hanging fruit for the builtin operator<=> expressions. It is currently needs some cleanup before landing, but I want to get some initial feedback.
The main changes are:
* Lookup, build, and store the required standard library types and expressions in `ASTContext`. By storing them in ASTContext we don't need to store (and duplicate) the required expressions in the BinaryOperator AST nodes.
* Implement [expr.spaceship] checking, including diagnosing narrowing conversions.
* Implement `ExprConstant` for builtin spaceship operators.
* Implement builitin operator<=> support in `CodeGenAgg`. Initially I emitted the required comparisons using `ScalarExprEmitter::VisitBinaryOperator`, but this caused the operand expressions to be emitted once for every required cmp.
* Implement [builtin.over] with modifications to support the intent of P0946R0. See the note on `BuiltinOperatorOverloadBuilder::addThreeWayArithmeticOverloads` for more information about the workaround.
[CFI] Force LLVM to die if the implicit blacklist files cannot be found.
Currently LLVM CFI tries to use an implicit blacklist file, currently
in /usr/lib64/clang/<version>/share. If the file is not there, LLVM
happily continues, which causes CFI to add checks to files/functions
that are known to fail, generating binaries that fail. This CL causes
LLVM to die (I hope) if it can't find these implicit blacklist files.
Erich Keane [Mon, 7 May 2018 20:52:56 +0000 (20:52 +0000)]
Correct warning on Float->Integer conversions.
As identified and briefly discussed here:
https://bugs.llvm.org/show_bug.cgi?id=37305
Converting a floating point number to an integer type when
the integral part is out of the range of the integer type is
undefined behavior in C. Additionally, CodeGen emits an undef
in this situation.
HOWEVER, we've been giving a warning that says that the value is
changed. This patch corrects the warning to list that it is actually
undefined behavior.
Erich Keane [Mon, 7 May 2018 17:05:20 +0000 (17:05 +0000)]
Fix explicit template parameter reporting for narrowing conversions
I found that explicit template parameters that caused a
narrowing integer conversion resulted in the incorrect parameter
being mentioned in the note (see test attached). This is because
the argument checking code doesn't check to see if it caused
SFINAE errors when checking the arguments, so instead of giving
up on the first error, it continues through the list. This
makes the error reporting pick up the last template param every time.
This patch checks these parameters on each argument and gives up
if there is an error. The result is that only the required amount
of arguments are checked, and that the 'Converted' array contains
only the successful arguments before the first failure, as the
calls seem to all expect.
In order to disable PIC and to match GCC behaviour, -mno-abicalls
option is neccessary. When -fno-[pic/PIC] is used witout -mno-abicalls,
warning is reported. An error is reported when -fno-pic or -fno-PIC is
used in combination with -mabicalls.
In order to disable PIC and to match GCC behaviour, -mno-abicalls
option is neccessary. When -fno-[pic/PIC] is used witout -mno-abicalls,
warning is reported. An error is reported when -fno-pic or -fno-PIC is
used in combination with -mabicalls.
Richard Smith [Mon, 7 May 2018 06:43:30 +0000 (06:43 +0000)]
Non-zero-length bit-fields make a class non-empty.
This implements the rule intended by the standard (see LWG 2358)
and the rule intended by the Itanium C++ ABI (see
https://github.com/itanium-cxx-abi/cxx-abi/pull/51), and makes
Clang match the behavior of GCC, ICC, and MSVC.
A pedantic reading of both the standard and the ABI indicate that Clang
is currently technically correct, but that's not worth much when it's
clear that the wording is wrong in both those places.
This is an ABI break for classes that derive from a class that is empty
other than one or more unnamed non-zero-length bit-fields. Such cases
are expected to be rare, but -fclang-abi-compat=6 restores the old
behavior just in case.
Teresa Johnson [Sat, 5 May 2018 16:37:31 +0000 (16:37 +0000)]
Add -target to address errors in test from r331592
The error turns out to be:
Assertion failed: (Target.isCompatibleDataLayout(getDataLayout()) && "Can't create a MachineFunction using a Module with a " "Target-incompatible DataLayout attached\n"), function init, file /Users/buildslave/jenkins/workspace/clang-stage1-cmake-RA-incremental/llvm/lib/CodeGen/MachineFunction.cpp, line 180.
Add -target to address this. Also re-enable the test I had temporarily
commented, and move it further down in case there is still a failure
(since it pipes stderr to FileCheck).
Teresa Johnson [Sat, 5 May 2018 15:54:57 +0000 (15:54 +0000)]
Skip part of test added in r331592 to help debug bot failures
Trying to debug why/where a few bots getting exit code 256 e.g.
http://green.lab.llvm.org/green/job/clang-stage1-cmake-RA-incremental/48471/testReport/Clang/CodeGen/thinlto_diagnostic_handler_remarks_with_hotness_ll/
and a few windows bots getting no output from that RUN line e.g.
http://lab.llvm.org:8011/builders/clang-x86-windows-msvc2015/builds/11865/steps/ninja%20check%201/logs/FAIL%3A%20Clang%3A%3Athinlto-diagnostic-handler-remarks-with-hotness.ll
Teresa Johnson [Sat, 5 May 2018 14:37:29 +0000 (14:37 +0000)]
[ThinLTO] Support opt remarks options with distributed ThinLTO backends
Summary:
Passes down the necessary code ge options to the LTO Config to enable
-fdiagnostics-show-hotness and -fsave-optimization-record in the ThinLTO
backend for a distributed build.
Also, remove warning about not having PGO when the input is IR.
Eli Friedman [Sat, 5 May 2018 00:09:51 +0000 (00:09 +0000)]
Add warning flag -Wordered-compare-function-pointers.
The C standard doesn't allow comparisons like "f1 < f2" (where f1 and f2
are function pointers), but we allow them as an extension. Add a
warning flag to control this warning.
Artem Dergachev [Fri, 4 May 2018 22:19:32 +0000 (22:19 +0000)]
[analyzer] Invalidate union regions properly. Don't hesitate to load later.
We weren't invalidating our unions correctly. The previous behavior in
invalidateRegionsWorker::VisitCluster() was to direct-bind an UnknownVal
to the union (at offset 0).
For that reason we were never actually loading default bindings from our unions,
because there never was any default binding to load, and the value
that is presumed when there's no default binding to load
is usually completely incorrect (eg. UndefinedVal for stack unions).
The new behavior is to default-bind a conjured symbol (of irrelevant type)
to the union that's being invalidated, similarly to what we do for structures
and classes. Then it becomes safe to load the value properly.
Artem Dergachev [Fri, 4 May 2018 22:11:12 +0000 (22:11 +0000)]
[analyzer] pr36458: Fix retrieved value cast for symbolic void pointers.
C allows us to write any bytes into any memory region. When loading weird bytes
from memory regions of known types, the analyzer is required to make sure that
the loaded value makes sense by casting it to an appropriate type.
Fix such cast for loading values that represent void pointers from non-void
pointer type places.
Artem Dergachev [Fri, 4 May 2018 21:56:51 +0000 (21:56 +0000)]
[analyzer] pr18953: Split C++ zero-initialization from default initialization.
The bindDefault() API of the ProgramState allows setting a default value
for reads from memory regions that were not preceded by writes.
It was used for implementing C++ zeroing constructors (i.e. default constructors
that boil down to setting all fields of the object to 0).
Because differences between zeroing consturctors and other forms of default
initialization have been piling up (in particular, zeroing constructors can be
called multiple times over the same object, probably even at the same offset,
requiring a careful and potentially slow cleanup of previous bindings in the
RegionStore), we split the API in two: bindDefaultInitial() for modeling
initial values and bindDefaultZero() for modeling zeroing constructors.
This fixes a few assertion failures from which the investigation originated.
The imperfect protection from both inability of the RegionStore to support
binding extents and lack of information in ASTRecordLayout has been loosened
because it's, well, imperfect, and it is unclear if it fixing more than it
was breaking.
Artem Dergachev [Fri, 4 May 2018 20:52:39 +0000 (20:52 +0000)]
[analyzer] Treat more const variables and fields as known contants.
When loading from a variable or a field that is declared as constant,
the analyzer will try to inspect its initializer and constant-fold it.
Upon success, the analyzer would skip normal load and return the respective
constant.
The new behavior also applies to fields/elements of brace-initialized structures
and arrays.
Eric Liu [Fri, 4 May 2018 17:55:13 +0000 (17:55 +0000)]
[clang-format] Refactor #include insertion/deletion functionality into a class.
Summary:
The class will be moved into libToolingCore as followup.
The new behaviors in this patch:
- New #include is inserted in the right position in a #include block to
preserver sorted #includes. This is best effort - only works when the
block is already sorted.
- When inserting multiple #includes to the end of a file which doesn't
end with a "\n" character, a "\n" will be prepended to each #include.
This is a special and rare case that was previously handled. This is now
relaxed to avoid complexity as it's rare in practice.
Brian Gesiak [Fri, 4 May 2018 14:02:37 +0000 (14:02 +0000)]
[Coroutines] Catch exceptions in await_resume
Summary:
http://wg21.link/P0664r2 section "Evolution/Core Issues 24" describes a
proposed change to Coroutines TS that would have any exceptions thrown
after the initial suspend point of a coroutine be caught by the handler
specified by the promise type's 'unhandled_exception' member function.
This commit provides a sample implementation of the specified behavior.
Aaron Ballman [Thu, 3 May 2018 15:33:50 +0000 (15:33 +0000)]
Allow writing calling convention attributes on function types.
Calling convention attributes notionally appertain to the function type -- they modify the mangling of the function, change the behavior of assignment operations, etc. This commit allows the calling convention attributes to be written in the type position as well as the declaration position.
Piotr Padlewski [Thu, 3 May 2018 11:03:01 +0000 (11:03 +0000)]
Rename invariant.group.barrier to launder.invariant.group
Summary:
This is one of the initial commit of "RFC: Devirtualization v2" proposal:
https://docs.google.com/document/d/16GVtCpzK8sIHNc2qZz6RN8amICNBtvjWUod2SujZVEo/edit?usp=sharing
Richard Smith [Thu, 3 May 2018 03:58:32 +0000 (03:58 +0000)]
Track the result of evaluating a computed noexcept specification on the
FunctionProtoType.
We previously re-evaluated the expression each time we wanted to know whether
the type is noexcept or not. We now evaluate the expression exactly once.
This is not quite "no functional change": it fixes a crasher bug during AST
deserialization where we would try to evaluate the noexcept specification in a
situation where we have not deserialized sufficient portions of the AST to
permit such evaluation.
Lei Liu [Thu, 3 May 2018 01:43:23 +0000 (01:43 +0000)]
[Sema] Do not match function type with const T in template argument deduction
From http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_active.html#1584,
function type should not match cv-qualified type in template argument
deduction. This also matches what GCC and EDG do in template argument
deduction.
Alex Lorenz [Thu, 3 May 2018 01:12:06 +0000 (01:12 +0000)]
[ObjC] Supress the 'implementing unavailable method' warning when
the method declaration is unavailable for an app extension platform
Rationale:
Classes are often shared between an app extension code and
non-app extension code. There's no way to remove the implementation
using preprocessor when building the app extension, so we should not warn here.
[gcov] Make the CLang side coverage test work with the new
instrumentation codegeneration strategy of using a data structure and
a loop. Required some finesse to get the critical things being tested to
surface in a nice way for FileCheck but I think this preserves the
original intent of the test.
Alex Lorenz [Wed, 2 May 2018 22:40:19 +0000 (22:40 +0000)]
[ObjC] The absence of ownership qualifiers on an ambiguous property leads
to synthesis of a valid property even when the selected protocol property
has ownership qualifiers