[CodeGen]: don't treat structures returned in registers as memory inputs
Summary:
The "=r" output constraint for a structure variable passed to inline asm
shouldn't be converted to "=*r", as this changes the asm directive
semantics and prevents DSE optimizations.
Instead, preserve the constraints and return such structures as integers
of corresponding size, which are converted back to structures when
storing the result.
Dmitri Gribenko [Thu, 29 Aug 2019 10:23:29 +0000 (10:23 +0000)]
[Index] Create PP callbacks in the ASTConsumer
Doing so removes one reason to create a custom FrontendAction.
FrontendActions are not desirable because they are difficult to compose.
ASTConsumers are much easier to compose.
[OpenCL] Fix diagnosing enqueue_kernel call with too few args
The err_typecheck_call_too_few_args diagnostic takes arguments, but
none were provided causing clang to crash when attempting to diagnose
an enqueue_kernel call with too few arguments.
Adam Balogh [Thu, 29 Aug 2019 09:35:47 +0000 (09:35 +0000)]
[Analyzer] Iterator Checkers - Make range errors and invalidated access fatal
Range errors (dereferencing or incrementing the past-the-end iterator or
decrementing the iterator of the first element of the range) and access of
invalidated iterators lead to undefined behavior. There is no point to
continue the analysis after such an error on the same execution path, but
terminate it by a sink node (fatal error). This also improves the
performance and helps avoiding double reports (e.g. in case of nested
iterators).
Volodymyr Sapsai [Wed, 28 Aug 2019 23:31:32 +0000 (23:31 +0000)]
[Modules] Fix rebuilding an updated module for each of its consumers.
Marking a module for a rebuild when its signature differs from the
expected one causes redundant module rebuilds for incremental builds.
When a module is updated, its signature changes. But its consumers still
have the old signature and loading them will result in signature
mismatches. It will correctly cause the rebuilds for the consumers but
we don't need to rebuild the common module for each of them as it is
already up to date.
In practice this bug causes longer build times. We are doing more work
than required and only a single process can build a module, so parallel
builds degrade to a single-process mode where extra processes are just
waiting on a file lock.
Fix by not marking a module dependency for a rebuild on signature
mismatch. We'll check if it is up to date when we load it.
Erich Keane [Wed, 28 Aug 2019 20:59:25 +0000 (20:59 +0000)]
Fix always_inline 'target' compatibility check code for Lambdas
The previous version of this used CurFuncDecl in CodeGenFunction,
however this doesn't include lambdas. However, CurCodeDecl DOES. Switch
the check to use CurCodeDecl so that the actual function being emitted
gets checked, preventing an error in ISEL.
Artem Dergachev [Wed, 28 Aug 2019 18:44:35 +0000 (18:44 +0000)]
[analyzer] pr43036: Fix support for operator 'sizeof...'.
It was known to be a compile-time constant so it wasn't evaluated during
symbolic execution, but it wasn't evaluated as a compile-time constant either.
Artem Dergachev [Wed, 28 Aug 2019 18:44:32 +0000 (18:44 +0000)]
[analyzer] Trust global initializers when analyzing main().
If the global variable has an initializer, we'll ignore it because we're usually
not analyzing the program from the beginning, which means that the global
variable may have changed before we start our analysis.
However when we're analyzing main() as the top-level function, we can rely
on global initializers to still be valid. At least in C; in C++ we have global
constructors that can still break this logic.
This patch allows the Static Analyzer to load constant initializers from
global variables if the top-level function of the current analysis is main().
Alexey Bataev [Wed, 28 Aug 2019 14:55:08 +0000 (14:55 +0000)]
[OPENMP][Analysis] Add analysis of the map clauses.
Summary:
Added basic analysis of map clauses. Only map clauses with to and tofrom
map type must be analyzed since all other map types (alloc, delete, etc.) do not require to use the value of the initial variable, instead they create the new copy of the variable.
It was added in r370129 with a .gitattributes file that means the file
always shows up as having a local diff in Git checkouts (at least on
Linux). Deleting it until we can figure out the right way to do this.
Sergey Dmitriev [Wed, 28 Aug 2019 01:26:13 +0000 (01:26 +0000)]
[Clang][Bundler] Do not require host triple for extracting device bundles
Bundler currently requires host triple to be provided no matter if you are performing bundling or unbundling, but for unbundling operation such requirement is too restrictive. You may for example want to examine device part of the object for a particular offload target, but you have to extract host part as well even though you do not need it. Host triple isn't really needed for unbundling, so this patch removes that requirement.
Volodymyr Sapsai [Wed, 28 Aug 2019 00:25:06 +0000 (00:25 +0000)]
[ObjC] Fix type checking for qualified id block parameters.
When checking if block types are compatible, we are checking for
compatibility their return types and parameters' types. As these types
have different variance, we need to check them in different order.
Sergey Dmitriev [Tue, 27 Aug 2019 21:47:52 +0000 (21:47 +0000)]
[Clang][Bundler] Fix for a hang when unbundling fat binary
clang-offload-bundler tool may hang under certain conditions when it extracts a subset of all available device bundles from the fat binary that is handled by the BinaryFileHandler. This patch fixes this problem.
Reid Kleckner [Tue, 27 Aug 2019 17:52:03 +0000 (17:52 +0000)]
Move EH spec mismatches under -fms-compatibility
-fms-extensions is intended to enable conforming language extensions and
-fms-compatibility is intended to language rule relaxations, so a user
could plausibly compile with -fno-ms-compatibility on Windows while
still using dllexport, for example. This exception specification
validation behavior has been handled as a warning since before
-fms-compatibility was added in 2011. I think it's just an oversight
that it hasn't been moved yet.
This will help users find conformance issues in their code such as those
found in _com_ptr_t as described in https://llvm.org/PR42842.
Alex Lorenz [Tue, 27 Aug 2019 17:32:42 +0000 (17:32 +0000)]
Use FileEntryRef for PPCallbacks::HasInclude
This fixes the issue where a filename dependendency was missing if the file that
was referenced with __has_include() was accessed through a symlink in an earlier run,
if the file manager was reused between runs.
Sam Elliott [Tue, 27 Aug 2019 15:41:16 +0000 (15:41 +0000)]
[RISCV] Set MaxAtomicInlineWidth and MaxAtomicPromoteWidth for RV32/RV64 targets with atomics
Summary: This ensures that libcalls aren't generated when the target supports atomics. Atomics aren't in the base RV32I/RV64I instruction sets, so MaxAtomicInlineWidth and MaxAtomicPromoteWidth are set only when the atomics extension is being targeted. This must be done in setMaxAtomicWidth, as this should be done after handleTargetFeatures has been called.
Joe Ranieri [Tue, 27 Aug 2019 14:43:54 +0000 (14:43 +0000)]
Update the SARIF exporter to SARIF 2.1
This updates the SARIF exporter to produce SARIF 2.1 output. The bulk of the diffs come from two changes to SARIF:
* https://github.com/oasis-tcs/sarif-spec/issues/309
* https://github.com/oasis-tcs/sarif-spec/issues/179
Gabor Marton [Tue, 27 Aug 2019 11:36:10 +0000 (11:36 +0000)]
[ASTImporter] Fix name conflict handling with different strategies
There are numorous flaws about the name conflict handling, this patch
attempts fixes them. Changes in details:
* HandleNameConflict return with a false DeclarationName
Hitherto we effectively never returned with a NameConflict error, even
if the preceding StructuralMatch indicated a conflict.
Because we just simply returned with the parameter `Name` in
HandleNameConflict and that name is almost always `true` when converted to
`bool`.
* Add tests which indicate wrong NameConflict handling
* Add to ConflictingDecls only if decl kind is different
Note, we might not indicate an ODR error when there is an existing record decl
and a enum is imported with same name. But there are other cases. E.g. think
about the case when we import a FunctionTemplateDecl with name f and we found a
simple FunctionDecl with name f. They overload. Or in case of a
ClassTemplateDecl and CXXRecordDecl, the CXXRecordDecl could be the 'templated'
class, so it would be false to report error. So I think we should report a
name conflict error only when we are 100% sure of that. That is why I think it
should be a general pattern to report the error only if the kind is the same.
* Fix failing ctu test with EnumConstandDecl
In ctu-main.c we have the enum class 'A' which brings in the enum
constant 'x' with value 0 into the global namespace.
In ctu-other.c we had the enum class 'B' which brought in the same name
('x') as an enum constant but with a different enum value (42). This is clearly
an ODR violation in the global namespace. The solution was to rename the
second enum constant.
Bruno Ricci [Tue, 27 Aug 2019 11:35:49 +0000 (11:35 +0000)]
[clang] Ensure that statements, expressions and types are trivially destructible
Since statements, expressions and types are allocated with the BumpPtrAllocator
from ASTContext their destructor is not executed. Two classes are currently
exempted from the check : InitListExpr due to its ASTVector and
ConstantArrayType due to its APInt.
Ilya Biryukov [Tue, 27 Aug 2019 10:02:18 +0000 (10:02 +0000)]
[Driver] Add an option for createInvocationFromCommandLine to recover on errors
Summary:
Previously, it would always return nullptr on any error.
This change adds a parameter, controlling whether the function should
attempt to return a non-null result even if unknown arguments (or other
errors were encountered).
The new behavior is only used in clangd.
Considered an alternative of changing the return value instead of adding
a new parameter, but that would require updating all callsites. Settled
with the parameter to minimize the code changes.
Richard Smith [Tue, 27 Aug 2019 01:06:23 +0000 (01:06 +0000)]
Revert "[clang-scan-deps] Minimizer: Correctly handle multi-line content with CR+LF line endings"
This reverts commit r369986.
This change added a dependency on the 'dos2unix' tool, which is not one
of our accepted test dependencies and may not exist on all machines that
build Clang.
Alex Lorenz [Tue, 27 Aug 2019 01:03:25 +0000 (01:03 +0000)]
Use FileEntryRef for PPCallbacks::FileSkipped
This fixes the issue where a filename dependendency was missing if the file that
was skipped was included through a symlink in an earlier run, if the file
manager was reused between runs.
Richard Smith [Mon, 26 Aug 2019 22:51:28 +0000 (22:51 +0000)]
PR42587: diagnose unexpanded uses of a pack parameter of a generic
lambda from within the lambda-declarator.
Instead of trying to reconstruct whether a parameter pack was declared
inside a lambda (which we can't do correctly in general because we might
not have attached parameters to their declaration contexts yet), track
the set of parameter packs introduced in each live lambda scope, and
require only those parameters to be immediately expanded when they
appear inside that lambda.
In passing, fix incorrect disambiguation of a lambda-expression starting
with an init-capture pack in a braced-init-list. We previously
incorrectly parsed that as a designated initializer.
ContentCache: Drop getBuffer's dependency on SourceManager
Refactor ContentCache::IsSystemFile to IsFileVolatile, checking
SourceManager::userFilesAreVolatile at construction time. This is a
step toward lowering ContentCache down from SourceManager to
FileManager.
Sergey Dmitriev [Mon, 26 Aug 2019 19:48:43 +0000 (19:48 +0000)]
[Clang][Bundler] Use llvm-objcopy for creating fat object files
clang-offload-bundler currently uses partial linking for creating fat object files, but such technique cannot be used on Windows due to the absence of partial linking support in the linker. This patch changes implementation to use llvm-objcopy for merging device and host objects instead of doing partial linking. This is one step forward towards enabling OpenMP offload on Windows.
Alexey Bataev [Mon, 26 Aug 2019 19:07:48 +0000 (19:07 +0000)]
[OPENMP][NVPTX]Fix critical region codegen.
Summary:
Previously critical regions were emitted with the barrier making it a
worksharing construct though it is not. Also, it leads to incorrect
behavior in Cuda9+. Patch fixes this problem.
FileManager: Use llvm::Expected in new getFileRef API
`FileManager::getFileRef` is a modern API which we expect to convert to
over time. We should modernize the error handling as well, using
`llvm::Expected` instead of `llvm::ErrorOr`, to help clients that care
about errors to ensure nothing is missed.
However, not all clients care. I've also added another path for those
that don't:
- `FileEntryRef` is now copy- and move-assignable (using a pointer
instead of a reference).
- `FileManager::getOptionalFileRef` returns an `llvm::Optional` instead
of `llvm::Expected`.
- Added an `llvm::expectedToOptional` utility in case this is useful
elsewhere.
Richard Smith [Mon, 26 Aug 2019 18:18:07 +0000 (18:18 +0000)]
Improve behavior in the case of stack exhaustion.
Summary:
Clang performs various recursive operations (such as template instantiation),
and may use non-trivial amounts of stack space in each recursive step (for
instance, due to recursive AST walks). While we try to keep the stack space
used by such steps to a minimum and we have explicit limits on the number of
such steps we perform, it's impractical to guarantee that we won't blow out the
stack on deeply recursive template instantiations on complex ASTs, even with
only a moderately high instantiation depth limit.
The user experience in these cases is generally terrible: we crash with
no hint of what went wrong. Under this patch, we attempt to do better:
* Detect when the stack is nearly exhausted, and produce a warning with a
nice template instantiation backtrace, telling the user that we might
run slowly or crash.
* For cases where we're forced to trigger recursive template
instantiation in arbitrarily-deeply-nested contexts, check whether
we're nearly out of stack space and allocate a new stack (by spawning
a new thread) after producing the warning.
Alex Lorenz [Mon, 26 Aug 2019 17:59:41 +0000 (17:59 +0000)]
[driver] add a new option `-gen-cdb-fragment-path` to emit
a fragment of a compilation database for each compilation
This patch adds a new option called -gen-cdb-fragment-path to the driver,
which can be used to specify a directory path to which clang can emit a fragment
of a CDB for each compilation it needs to invoke.
This option emits the same CDB contents as -MJ, and will be ignored if -MJ is specified.
Erich Keane [Mon, 26 Aug 2019 17:00:13 +0000 (17:00 +0000)]
Fix -dA flag, it is not a preprocessor flag.
-dA was in the d_group, which is a preprocessor state dumping group.
However -dA is a debug flag to cause a verbose asm. It was already
implemented to do the same thing as -fverbose-asm, so make it just be an
alias.
Richard Smith [Sat, 24 Aug 2019 02:30:00 +0000 (02:30 +0000)]
PR42513: Enter the proper DeclContext before substituting into an
default template argument expression.
We already did this for type template parameters and template template
parameters, but apparently forgot to do so for non-type template
parameters. This causes the substituted default argument expression to
be substituted in the proper context, and in particular to properly mark
its subexpressions as odr-used.
Alexey Bataev [Fri, 23 Aug 2019 19:52:05 +0000 (19:52 +0000)]
[OPENMP5]Use nonmonotonic modifier by default for non-static and
non-ordered loops.
According to OpenMP 5.0, 2.9.2 Worksharing-Loop Construct, Desription, If the static schedule kind is specified or if the ordered clause is specified, and if the nonmonotonic modifier is not specified, the effect is as if the monotonic modifier is specified. Otherwise, unless the monotonic modifier is specified, the effect is as if the nonmonotonic modifier is specified.
The first part of this requirement is implemented in runtime. Patch adds
support for the second, nonmonotonic, part of this requirement.
projects / clang / log