Launching a kernel from the host code does not generate code for the
kernel itself. This fixes an issue with clang erroneously reporting
an error for a HD->D call from within the kernel.
Differential Revision: https://reviews.llvm.org/D44837
llvm-svn: 328362
Summary: Porting HWASan to Linux x86-64, the third of the three patches, clang part.
Reviewers: eugenis
Subscribers: cryptoad, cfe-commits
Differential Revision: https://reviews.llvm.org/D44745
llvm-svn: 328361
Need to override convertConstraint to recognise amdgpu specific register names.
Differential Revision: https://reviews.llvm.org/D44533
llvm-svn: 328359
Add two additional implicit arguments for OpenCL for the AMDGPU target using the AMDHSA runtime to support device enqueue.
Differential Revision: https://reviews.llvm.org/D44696
llvm-svn: 328350
- Remove use of the opencl and amdopencl environment member of the target triple for the AMDGPU target.
- Use a function attribute to communicate to the AMDGPU backend.
Differential Revision: https://reviews.llvm.org/D43735
llvm-svn: 328347
This make -ivfsoverlay behave more like other fatal errors (e.g. missing
-include file) by skipping the missing file instead of bailing out of
the whole compilation. This makes it possible for libclang to still
provide some functionallity as well as to correctly produce the fatal
error diagnostic (previously we lost the diagnostic in libclang since
there was no TU to tie it to).
rdar://33385423
llvm-svn: 328337
The issues was that we were setting hidden visibility if, when
processing a hidden class, we found out that we needed to emit a
reference to a vtable provided by the standard library.
Original message:
Set dso_local on vtables.
llvm-svn: 328288
Changes the analyzer to believe that methods annotated with _Nonnull
from system frameworks indeed return non null objects.
Local methods with such annotation are still distrusted.
rdar://24291919
Differential Revision: https://reviews.llvm.org/D44341
llvm-svn: 328282
Putting back the code in commit r327189 that was reverted in r322737. The code is being committed in three stages and this one is the last stage: 1) r327455 fp16 feature flags, 2) r327836 pass half type or i16 based on FullFP16, and 3) the code here which the front-end fp16 vector intrinsic for ARM.
Differential revision https://reviews.llvm.org/D43650
llvm-svn: 328277
When a temporary is constructed with a proper construction context, it should
be safe to inline the destructor. We have added suppressions for some of the
common false positives caused by such inlining, so there should be - and from my
observations there indeed is - more benefit than harm from enabling destructor
inlining.
Differential Revision: https://reviews.llvm.org/D44721
llvm-svn: 328258
CXXCtorInitializer-based constructors are also affected by the C++17 mandatory
copy elision, like variable constructors and return value constructors.
Extend r328248 to support those.
Differential Revision: https://reviews.llvm.org/D44763
llvm-svn: 328255
Function return values can be constructed directly in variables or passed
directly into return statements, without even an elidable copy in between.
This is how the C++17 mandatory copy elision AST behaves. The behavior we'll
have in such cases is the "old" behavior that we've had before we've
implemented destructor inlining and proper lifetime extension support.
Differential Revision: https://reviews.llvm.org/D44755
llvm-svn: 328253
In C++17 copy elision is mandatory for variable and return value constructors
(as long as it doesn't involve type conversion) which results in AST that does
not contain elidable constructors in their usual places. In order to provide
construction contexts in this scenario we need to cover more AST patterns.
This patch makes the CFG prepared for these scenarios by:
- Fork VariableConstructionContext and ReturnedValueConstructionContext into
two different sub-classes (each) one of which indicates the C++17 case and
contains a reference to an extra CXXBindTemporaryExpr.
- Allow CFGCXXRecordTypedCall element to accept VariableConstructionContext and
ReturnedValueConstructionContext as its context.
Differential Revision: https://reviews.llvm.org/D44597
llvm-svn: 328248
If a memory region (or an SVal that represents a pointer to that memory region)
is a (direct or indirect, not necessarily proper) sub-region of a SymbolicRegion
then it is said to have a symbolic base.
For now SVal::symbol_iterator explores the symbol within a symbolic region
only when the SVal represents a pointer to the symbolic region itself,
not to any of its sub-regions.
This behavior is not indended by any user of symbol_iterator; all users who
cared about such behavior were expecting the iterator to descend into the
symbolic base of an arbitrary region, find the parent symbol of the symbolic
base region, and iterate over that symbol. Lack of such behavior resulted in
bugs demonstarted by the test cases.
Hence the decision to change the API to behave more intuitively.
Differential Revision: https://reviews.llvm.org/D44347
llvm-svn: 328247
The difference between CreateRuntimeFunction and CreateBuiltinFunction
is that CreateBuiltinFunction would not set dllimport or dso_local.
To keep the current semantics, just forward to CreateRuntimeFunction
with Local=true so it doesn't add dllimport.
llvm-svn: 328224
Summary: The workers also need to initialize the global stack. The call to the initialization function needs to happen after the kernel_init() function is called by the master. This ensures that the per-team data structures of the runtime have been initialized.
Reviewers: ABataev, grokos, carlo.bertolli, caomhin
Reviewed By: ABataev
Subscribers: jholewinski, guansong, cfe-commits
Differential Revision: https://reviews.llvm.org/D44749
llvm-svn: 328219
Now that LLVM has support for emitting calling conventions in DWARF (see
r328191) have clang emit them.
Patch by: Adrien Guinet
Differential revision: https://reviews.llvm.org/D42351
llvm-svn: 328196
Summary: Rewrites -Winfinite-recursion to remove the state dictionary and explore paths in loops - especially infinite loops. The new check now detects recursion in loop bodies dominated by a recursive call.
Reviewers: rsmith, rtrieu
Reviewed By: rtrieu
Subscribers: lebedev.ri, cfe-commits
Differential Revision: https://reviews.llvm.org/D43737
llvm-svn: 328173
This fixes host-side LTO during CUDA compilation. Before, LTO
pipeline construction was clashing with CUDA pipeline construction.
At the moment there's no point doing LTO on device side as each
device-side TU is a complete program. We will need to figure out
compilation pipeline construction for the device-side LTO when we
have working support for multi-TU device-side CUDA compilation.
Differential Revision: https://reviews.llvm.org/D44691
llvm-svn: 328161
This is needed for the upcoming implementation of the
new 8x32x16 and 32x8x16 variants of WMMA instructions
introduced in CUDA 9.1.
Differential Revision: https://reviews.llvm.org/D44719
llvm-svn: 328158
During reading C++ definition data for lambda we can access
CXXRecordDecl representing lambda before we finished reading the
definition data. This can happen by reading a captured variable which is
VarDecl, then reading its decl context which is CXXMethodDecl `operator()`,
then trying to merge redeclarable methods and accessing
enclosing CXXRecordDecl. The call stack looks roughly like
VisitCXXRecordDecl
ReadCXXRecordDefinition
VisitVarDecl
VisitCXXMethodDecl
mergeRedeclarable
getPrimaryContextForMerging
If we add fake definition data at this point, later we'll hit the assertion
Assertion failed: (!DD.IsLambda && !MergeDD.IsLambda && "faked up lambda definition?"), function MergeDefinitionData, file clang/lib/Serialization/ASTReaderDecl.cpp, line 1675.
The fix is to assign definition data before reading it. Fixes PR32556.
rdar://problem/37461072
Reviewers: rsmith, bruno
Reviewed By: rsmith
Subscribers: cfe-commits, jkorous-apple, aprantl
Differential Revision: https://reviews.llvm.org/D43494
llvm-svn: 328153
Summary:
Libc++'s default allocator uses `__builtin_operator_new` and `__builtin_operator_delete` in order to allow the calls to new/delete to be ellided. However, libc++ now needs to support over-aligned types in the default allocator. In order to support this without disabling the existing optimization Clang needs to support calling the aligned new overloads from the builtins.
See llvm.org/PR22634 for more information about the libc++ bug.
This patch changes `__builtin_operator_new`/`__builtin_operator_delete` to call any usual `operator new`/`operator delete` function. It does this by performing overload resolution with the arguments passed to the builtin to determine which allocation function to call. If the selected function is not a usual allocation function a diagnostic is issued.
One open issue is if the `align_val_t` overloads should be considered "usual" when `LangOpts::AlignedAllocation` is disabled.
In order to allow libc++ to detect this new behavior the value for `__has_builtin(__builtin_operator_new)` has been updated to `201802`.
Reviewers: rsmith, majnemer, aaron.ballman, erik.pilkington, bogner, ahatanak
Reviewed By: rsmith
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D43047
llvm-svn: 328134
r326249 wasn't quite enough because we often run out of inlining stack depth
limit and for that reason fail to see the atomics we're looking for.
Add a more straightforward false positive suppression that is based on the name
of the class. I.e. if we're releasing a pointer in a destructor of a "something
shared/intrusive/reference/counting something ptr/pointer something", then any
use-after-free or double-free that occurs later would likely be a false
positive.
rdar://problem/38013606
Differential Revision: https://reviews.llvm.org/D44281
llvm-svn: 328066
When skipping building the module for a private framework module,
LangOpts.CurrentModule isn't enough for implict modules builds; for
instance, in case a private module is built while building a public one,
LangOpts.CurrentModule doesn't reflect the -fmodule-name being passed
down, but instead the module name which triggered the build.
Store the actual -fmodule-name in LangOpts.ModuleName and actually
check a name was provided during compiler invocation in order to
skip building the private module.
rdar://problem/38434694
llvm-svn: 328053
constructs in generic mode.
Fixed codegen for distribute parallel combined constructs. We have to
pass and read the shared lower and upper bound from the distribute
region in the inner parallel region. Patch is for generic mode.
llvm-svn: 327990
If the generic codegen is enabled and private copy of the original
variable escapes the declaration context, this private copy should be
globalized just like it was the original variable.
llvm-svn: 327985
When the loop has a null terminator statement and sets 'widen-loops=true', 'invalidateRegions' will constructs the 'SymbolConjured' with null 'Stmt'. And this will lead to a crash in 'IteratorChecker.cpp'. This patch use 'dyn_cast_or_null<>' instead of 'dyn_cast<>' in IteratorChecker.cpp.
Differential Revision: https://reviews.llvm.org/D44606
llvm-svn: 327962
This allows users to turn off warnings about this pragma specifically,
while still receiving warnings about other ignored pragmas.
Differential Revision: https://reviews.llvm.org/D44630
llvm-svn: 327959
Digging through commit logs, it appears the checks in this block predate
`inline` class variables. With them, we fail to emit dynamic
initializers for members that don't have an explicit initializer, and we
won't go out of our way to instantiate the class denoted by
`Var->getType()`.
Fixes PR35599.
llvm-svn: 327945
source expressions when iterating over a PseudoObjectExpr's semantic
subexpression list.
Previously the loop in emitPseudoObjectExpr would emit the IR for each
OpaqueValueExpr that was in a PseudoObjectExpr's semantic-form
expression list and use the result when the OpaqueValueExpr later
appeared in other expressions. This caused an assertion failure when
AggExprEmitter tried to copy the result of an OpaqueValueExpr and the
copied type didn't have trivial copy/move constructors or assignment
operators.
This patch adds flag IsUnique to OpaqueValueExpr which indicates it is a
unique reference to its source expression (it is not used in multiple
places). The loop in emitPseudoObjectExpr ignores OpaqueValueExprs that
are unique and CodeGen visitors simply traverse the source expressions
of such OpaqueValueExprs.
rdar://problem/34363596
Differential Revision: https://reviews.llvm.org/D39562
llvm-svn: 327939
This is needed to avoid the test failure in case when compiler-rt
is set as the default runtime library for Clang.
Differential Revision: https://reviews.llvm.org/D44655
llvm-svn: 327932
Also use the opportunity to clean up the code and remove unnecessary duplication.
rdar://37625895
Differential Revision: https://reviews.llvm.org/D44594
llvm-svn: 327926
The inline assembly generated for the ARC autorelease elision marker
must have a funclet token if it's emitted inside a funclet, otherwise
the inline assembly (and all subsequent code in the funclet) will be
marked unreachable. r324689 fixed this issue for regular inline assembly
blocks.
Note that clang only emits the marker at -O0, so this only fixes that
case. The optimizations case (where the marker is emitted by the
backend) will be fixed in a separate change.
Differential Revision: https://reviews.llvm.org/D44640
llvm-svn: 327892
Richard Trieu