As shown by bug 48540, GCC vector types would cause a crash when the
declaration hada ParenType. This was because the walking of the
declaration would try to expand the 'inner' type, but there was no
ability to get it from the vector type. This patch adds that element
type access to the vector type loc objects.
Differential Revision: https://reviews.llvm.org/D93483
Remove continuation line in code snippet to prevent GCC warning about
multiline comments (-Wcomment) when building a project using libclang
with GCC.
Reviewed By: rjmccall
Differential Revision: https://reviews.llvm.org/D92409
When introducing support for @llvm.experimental.noalias.scope.decl, this tests started failing because it checks
(for no good reason) for a function attribute id of '#8' which now becomes '#9'
Reviewed By: pratlucas
Differential Revision: https://reviews.llvm.org/D94233
the conversion of LLVM IR dialect. These attributes are used in FIR to
support the lowering of Fortran using target-specific calling
conventions.
Add roundtrip tests. Add changes per review comments/concerns.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D94052
Incorrect usage of NDEBUG to guard ABI changes can prevent clients
from enabling assertions for their C++ code while having assertions in
LLVM turned off. So we use LLVM_ENABLE_ABI_BREAKING_CHECKS instead, as
described in llvm/docs/ProgrammersManual.rst. Most types already use this
macro, however, there were a couple of stragglers in ValueHandle.h, which
are fixed by this revision.
Reviewed By: dblaikie, dexonsmith
Differential Revision: https://reviews.llvm.org/D93433
Adds some tests to check the formatting of the dot
file produced when using -dot-ddg.
Reviewed By: Meinersbur
Differential Revision: https://reviews.llvm.org/D93949
There are various hacks working around limitations in
handleAssignments, and the logical split between different parts isn't
correct. Start separating the type legalization to satisfy going
through the DAG infrastructure from the code required to split into
register types. The type splitting should be moved to generic code.
When creating pi-blocks we try to avoid creating duplicate edges
between outside nodes and the pi-block when an edge is of the
same kind and direction as another one that has already been
created. We do this by keeping track of the edges in an
enumerated array called EdgeAlreadyCreated. The problem is that
this array is declared local to the loop that iterates over the
nodes in the pi-block, so the information gets lost every time a
new inside-node is iterated over. The fix is to move the
declaration to the outer loop.
Reviewed By: Meinersbur
Differential Revision: https://reviews.llvm.org/D94094
I have added it in d15d81c because it *seemed* correct, was holding
for all the tests so far, and was validating the fix added in the same
commit, but as David Major is pointing out (with a reproducer),
the assertion isn't really correct after all. So remove it.
Note that the d15d81c still fine.
Much like with ReduceFunctionBodies delta pass,
we need to remove comdat and set linkage to external,
else verifier will complain, and our deltas are invalid.
Summary:
Currently SplitEdge does not support passing in parameter which allows you to
name the newly created BasicBlock.
This patch updates the function such that the name of the block can be passed
in, if users of this utility decide to do so.
Reviewed By: Whitney, bmahjour, asbirlea, jamieschmeiser
Differential Revision: https://reviews.llvm.org/D94176
Add semantic check for most of the restrictions for the declare directive.
Reviewed By: kiranktp
Differential Revision: https://reviews.llvm.org/D92741
Don't directly dereference a dyn_cast<> - use cast<> so we assert for the correct type.
Also, simplify the for loop to a range loop.
Fixes clang static analyzer warning.
This patch fixes a bug introduced in the patch:
https://reviews.llvm.org/D93030
This patch pulls the test for scalable vector to be the first instruction
to be checked. This avoids the Gather and Scatter cost model for AArch64 to
compute the number of vector elements for something that is not a vector and
therefore crashing.
This is necessary for a future patch, where we start using this macro in another function.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D93702
This patch moves `parseSimpleArgs` closer to `ParseDiagnosticArgs` so that sharing the parsing macro between them can be done more locally in a future patch.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D94172
Before this patch, ParseDiagnosticArgs can be called with a nullptr DiagnosticsEngine *. This happens early on in the compilation process, where no proper DiagnosticEngine exists, because the diagnostic options (passed through command line) are not known yet.
This patch ensures nullptr is replaced by an ignoring DiagnosticEngine in ParseDiagnosticArgs, which allows to switch from pointer to a reference in some utility functions.
Besides simplifying the code, this patch enables a future patch (D84673) that ports diagnostic options to the new marshalling infrastructure.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D93701
After merging the shuffles, we cannot rely on the previous shuffle
anymore and need to shrink the final shuffle, if it is required.
Reported in D92668
Differential Revision: https://reviews.llvm.org/D93967
Currently some clients watch for CDB changes and restart clangd, now that we
can reload compile_commands.json ourselves this is counterproductive.
The capability allows this behavior to be phased out.
This is going to be a mild regression, as we do not actually watch for files on
disk and so new diagnostics need to wait until a rebuild is requested e.g. due
to file change (and the internal caches have expired).
However this is still a better tradeoff (and if it's important, we can request
the client to watch files for us in the future).
Differential Revision: https://reviews.llvm.org/D94222
The original implementation of the OpenMP dialect to LLVM IR translation has
been relying on a stack of insertion points for delayed insertion of branch
instructions that correspond to terminator ops. This is an intrusive into
ModuleTranslation and makes the translation non-local. A recent addition of the
WsLoop translation exercised another approach where the parent op is
responsible for converting terminators of all blocks in its regions. Use this
approach for other OpenMP dialect operations with regions, remove the stack and
deduplicate the code for converting such regions.
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D94086
* The only exception is that the flag -std=c++2a is still used not to break compatibility with older compilers (clang <= 9, gcc <= 9).
* Bump _LIBCPP_STD_VER for C++20 to 20 and use 21 for the future standard (C++2b).
That's a preparation step to add c++2b support to libc++.
Reviewed By: ldionne, #libc
Differential Revision: https://reviews.llvm.org/D93383
The nullptr_t_integral_cast.pass.cpp test is currently xfailed for
C++03, but actually, it only fails with the first version of libc++
ABI.
This patch changes XFAIL to UNSUPPORTED to avoid unexpected passes
with ABI v2 or later.
Reviewed By: ldionne, #libc
Differential Revision: https://reviews.llvm.org/D93941
Attempt to simplify all/any-of style patterns that concatenate 2 smaller integers together into an and(x,y)/or(x,y) + icmp 0/-1 instead.
This is mainly to help some bool predicate reduction patterns where we end up concatenating bool vectors that have been bitcasted to integers.
Differential Revision: https://reviews.llvm.org/D93599
This patch adds a frontend action for emitting object files. While Flang
does not support code-generation, this action remains a placeholder.
This patch simply provides glue-code to connect the compiler driver
with the appropriate frontend action.
The new action is triggered with the `-c` compiler driver flag, i.e.
`flang-new -c`. This is then translated to `flang-new -fc1 -emit-obj`,
so `-emit-obj` has to be marked as supported as well.
As code-generation is not available yet, `flang-new -c` results in a
driver error:
```
error: code-generation is not available yet
```
Hopefully this will help communicating the level of available
functionality within Flang.
The definition of `emit-obj` is updated so that it can be shared between
Clang and Flang. As the original definition was enclosed within a
Clang-specific TableGen `let` statement, it is extracted into a new `let`
statement. That felt like the cleanest option.
I also commented out `-triple` in Flang::ConstructJob and updated some
comments there. This is similar to https://reviews.llvm.org/D93027. I
wanted to make sure that it's clear that we can't support `-triple`
until we have code-generation. However, once code-generation is
available we _will need_ `-triple`.
As this patch adds `-emit-obj`, the emit-obj.f90 becomes irrelevant and
is deleted. Instead, phases.f90 is added to demonstrate that users can
control compilation phases (indeed, `-c` is a phase control flag).
Reviewed By: SouraVX, clementval
Differential Revision: https://reviews.llvm.org/D93301
Add optional memory tagging extension on AArch64.
Use isAArch64() instead of listing the AArch64 triples,
which fixes us not recognising aarch64_be.
Reviewed By: omjavaid
Differential Revision: https://reviews.llvm.org/D94084
The assertion can happen if ASTImporter imports a CXXRecordDecl in a template
and then imports another redeclaration of this declaration, while the first import is in progress.
The process of first import did not set the "described template" yet
and the second import finds the first declaration at setting the injected types.
Setting the injected type requires in the assertion that the described template is set.
The exact assertion was:
clang/lib/AST/ASTContext.cpp:4411:
clang::QualType clang::ASTContext::getInjectedClassNameType(clang::CXXRecordDecl*, clang::QualType) const:
Assertion `NeedsInjectedClassNameType(Decl)' failed.
Reviewed By: shafik
Differential Revision: https://reviews.llvm.org/D94067
The patterns that want to use 'vnot' use a custom PatFrag. This is
because 'vnot' uses immAllOnesV which implicitly uses BUILD_VECTOR
rather than SPLAT_VECTOR.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D94078
This allows us to verify that we don't emit options multiple times.
In most cases, that would be benign, but for options with `MarshallingInfoVectorString`, emitting wrong number of arguments might change the semantics.
Reviewed By: Bigcheese
Differential Revision: https://reviews.llvm.org/D93636
Mircea Trofin