Summary:
This patch introduce the parser for OpenACC 3.0 in Flang. It uses the same TableGen mechanism
than OpenMP.

Reviewers: nvdatian, sscalpone, tskeith, klausler, ichoyjx, jdoerfert, DavidTruby

Reviewed By: klausler

Subscribers: MaskRay, SouraVX, mgorny, hiraditya, jfb, sstefan1, llvm-commits

Tags: #llvm, #flang

Differential Revision: https://reviews.llvm.org/D83649

Currently custom code predicates can only really be used for
contextless checks tied to a single instruction (e.g. check the def
for hasOneUse). If you do want to inspect the input instructions in
the source pattern, you cannot without re-verifying the opcode and
type checks implied by the patterns, since this check was emitted
before any operand constraints. Really, these are pattern level
predicates that implicitly depend on the instruction and operand
checks.

Introduce a filtering function so the custom predicate is emitted
last. I'm not sure this is the most elegant solution. It seems like
this is really a different thing from the InstructionMatcher/IPM_
predicate kinds. I initially tried keeping this in a separate
predicate list, but that also seemed awkward.

This only half fixes the problem I'm trying to solve. The AMDGPU
pattern I'm attempting to port also uses the PredicateCodeUsesOperands
feature to allow checks on the source operands when the input pattern
is commuted. Really the emitter should reject the pattern since it
doesn't handle this case, but at this point it would be more
productive to just implement this.

<rdar://problem/65537147>

Differential Revision:  https://reviews.llvm.org/D83796

Add missing const - addresses clang tidy warnings.

Differential Revision: https://reviews.llvm.org/D83794

I mistyped the ubsan objc_cast handler names on the first try.

Testing:

./bin/llvm-lit projects/compiler-rt/test/asan/X86_64HDarwinConfig/TestCases/Darwin/interface_symbols_darwin.cpp

LazyReexportsManager instances use the trampoline pool, but they don't need to
own it. Keeping TrampolinePool ownership separate allows re-use of the
trampoline pool by other clients.

Summary:
The following preprocessor macros are implied when `-march=armv8.6a+sve`:

```
__ARM_FEATURE_SVE 1
__ARM_FEATURE_SVE_BF16 1
__ARM_FEATURE_SVE_MATMUL_FP32 1
__ARM_FEATURE_SVE_MATMUL_INT8 1
```

Reviewers: sdesmalen, efriedma, SjoerdMeijer, rengolin

Subscribers: tschuett, kristof.beyls, danielkiss, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83079

Use runBuildCommands, like all other builders, to raise a build-specific
error when the command fails.

 - Make the open more Pythonic.
 - Remove the unused `cleanup` Make target.
 - Remove commented-out/obvious/low-value comments.
 - Cleanup the forked process PID list.

As per discussion in D69207, have lit ignore UnicodeDecodeErrors
when running with python 2 in an ASCII shell.

Differential Revision: https://reviews.llvm.org/D82754

Summary: This is a follow up to https://reviews.llvm.org/D78306

Reviewers: DavidTruby, sscalpone, jdoerfert

Reviewed By: DavidTruby

Subscribers: tskeith, Meinersbur, ChinouneMehdi, richard.barton.arm, mehdi_amini, mgorny, llvm-commits

Tags: #llvm, #flang

Differential Revision: https://reviews.llvm.org/D81695

Skip TestProcessConnect.py on Windows and Android (the same platforms as
TestPlatformProcessConnect.py) and mark it as a NO_DEBUG_INFO test so we
don't run all the variants.

Summary:
We only support Build/Skip for now, but with 'Load' or similar as an
option for future (load existing shards but don't build new ones).

This requires creating the config for each TU on startup. In LLVM, this
is 4000 occurrences for a total of 800ms on my machine.
But together with caching from D83755 it is only 25ms.

Reviewers: kadircet

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83768

If we deduced the arity of a pack in two different ways, but didn't
deduce an element of the pack in either of those deductions, we'd merge
that element to produce a null template argument, which we'd incorrectly
interpret as the merge having failed.

Testcase based on one supplied by Hubert Tong.

This fixes warnings raised by Clang's new -Wsuggest-override, in preparation for enabling that warning in the LLVM build. This patch also removes the virtual keyword where redundant, but only in places where doing so improves consistency within a given file. It also removes a couple unnecessary virtual destructor declarations in derived classes where the destructor inherited from the base class is already virtual.

Differential Revision: https://reviews.llvm.org/D83709

ClangAttrEmitter.cpp generates ParsedAttr derived classes with virtual overrides in them (which end up in AttrParsedAttrImpl.inc); this patch ensures these generated functions are marked override, and not (redundantly) virtual.

I hesitate to say NFC since this does of course affect the behavior of the generator code, but the generated code behaves the same as it did before, so it's NFC in that sense.

Differential Revision: https://reviews.llvm.org/D83616

- Arguments of the first block of a region are considered region arguments.
- Add API on Region class to deal with these arguments directly instead of
  using the front() block.
- Changed several instances of existing code that can use this API
- Fixes https://bugs.llvm.org/show_bug.cgi?id=46535

Differential Revision: https://reviews.llvm.org/D83599

Summary:
llvm-cov.test generates .gcov files and compared with target
sample files. Since the files do not contain any binary data
(files are plain ASCII texts), remove -a from diff. And this
fix will the error on AIX since the default diff tool on AIX
does not support -a option.

Reviewers: hubert.reinterpretcast, daltenty, stevewan

Subscribers: llvm-commits

Tags: #LLVM

Differential Revision: https://reviews.llvm.org/D83711

- Avoid generating "assert(resultTypes.size() >= 0u)" which is always true

Differential Revision: https://reviews.llvm.org/D83735

Loop metadata nodes do not adhere to the documented property:

(a) LoopIDs are not unique: Any pass that duplicates IR will do it
    including its metadata (e.g. LoopVersioning) such that multiple
    loops are linked with the same LoopID. There is even a test case
    (Transforms/LoopUnroll/unroll-pragmas-disabled.ll) for multiple
    loops with the same LoopID.

(b) LoopIDs are not persistent: Adding or removing an item from a LoopID
    can only be done by creating a new MDNode and assigning it to the
    loop's branch(es). Passes such as LoopUnroll (llvm.loop.unroll.disable)
    and LoopVectorize (llvm.loop.isvectorized) use this to mark loops to
    not be transformed multiple times or to avoid that a LoopVersioned
    original loop is transformed.

Update the documentation according to how llvm.loop is used in practice.

Differential Revision: https://reviews.llvm.org/D55290

This patch adds override to several overriding virtual functions that were missing the keyword within the clang/ directory. These were found by the new -Wsuggest-override.

The libc++ headers were included twice, which broke the #include_next
logic.

Summary:
This patch extends the ModuleSpec class to include a
DataBufferSP which contains the module data. If this
data is provided, LLDB won't try to hit the filesystem
to create the Module, but use only the data stored in
the ModuleSpec.

Reviewers: labath, espindola

Subscribers: emaste, MaskRay, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D83512

In synchronous mode, the process connect command and its aliases should
wait for the stop event before claiming the command is complete.
Currently, the stop event is always handled asynchronously by the
debugger.

The implementation takes the same approach as Process::ResumeSynchronous
which hijacks the event and handles it on the current thread. Similarly,
after this patch, the stop event is part of the command return object,
which is the property used by the test case.

Differential revision: https://reviews.llvm.org/D83728

Summary:
This is motivated by:
 - code completion: nice to do no i/o on the request path
 - background index: deciding whether to enqueue each file would stat the config
   file thousands of times in quick succession.

Currently it's applied uniformly to all requests though.

This gives up on performing stat() outside the lock, all this achieves is
letting multiple threads stat concurrently (and thus finish without contention
for nonexistent files).
The ability to finish without IO (just mutex lock + integer check) should
outweigh this, and is less sensitive to platform IO characteristics.

Reviewers: kadircet

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83755

This is another problem raised in:
http://bugs.llvm.org/PR46627

Summary: While here, add documentation to CompileFlags and CompileFlags.Add.

Reviewers: hokein

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83705

Summary:
This can be run after a ninja-based build, and analyzes the ninja
build files and dependency database to spot any missing dependencies
in the build scripts.

I wrote it in the course of investigating D82659, and it seems likely
to be useful again.

Reviewers: thakis, chandlerc, theraven

Reviewed By: thakis

Subscribers: riccibruno, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83032

Instead of having complex logic around how to include the libc++ headers
and __config_site, handle that by defining cxx-headers as an INTERFACE
library and linking against it. After this patch, linking against cxx-headers
is sufficient to get the right __config_site include and include paths
for libc++.

Differential Revision: https://reviews.llvm.org/D82702

Summary: Currently forward decls are included with all the op classes. But there are cases (say when splitting up headers) where one wants the forward decls but not all the classes. Add an option to enable this. This does not change any current behavior (some further refactoring is probably due here).

Differential Revision: https://reviews.llvm.org/D83727

Summary:
Names of local variables have no linkage (see C++20 [basic.link] p8).

Names of variables in unnamed namespace have internal linkage (see C++20
[basic.link] p4).

Reviewers: aaron.ballman, rsmith, ymandel

Reviewed By: ymandel

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83700

Summary:
This is designed for tweaking compile commands by specifying flags to add/remove
in a config file. Something like:
  CompileFlags: { Remove: -fcolor-diagnostics }

Having users tweak raw argv (e.g. with a regex) is going to end in tears: bugs
around clang-cl, xclang, aliases, joined-vs-separate args etc are inevitable.

This isn't in tooling because of the performance choices: build a big table
up-front to make subsequent actions fast. Maybe it should be though.

Reviewers: adamcz, hokein

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D81958

The _ExtInt type allows custom width integers, but the atomic memory
access's operand must have a power-of-two size. _ExtInts with
non-power-of-two size should not be allowed for atomic intrinsic.

Before this change:

$ cat test.c

typedef unsigned _ExtInt(42) dtype;
void verify_binary_op_nand(dtype* pval1, dtype val2)
{    __sync_nand_and_fetch(pval1, val2); }

$ clang test.c

clang-11:
/home/ubuntu/llvm_workspace/llvm/clang/lib/CodeGen/CGBuiltin.cpp:117:
llvm::Value*
EmitToInt(clang::CodeGen::CodeGenFunction&, llvm::Value*,
clang::QualType, llvm::IntegerType*): Assertion `V->getType() ==
IntType' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the
crash backtrace, preprocessed source, and associated run script.

After this change:

$ clang test.c

test.c:3:30: error: Atomic memory operand must have a power-of-two size
{    __sync_nand_and_fetch(pval1, val2); }
^

List of the atomic intrinsics that have this
problem:

__sync_fetch_and_add
__sync_fetch_and_sub
__sync_fetch_and_or
__sync_fetch_and_and
__sync_fetch_and_xor
__sync_fetch_and_nand
__sync_nand_and_fetch
__sync_and_and_fetch
__sync_add_and_fetch
__sync_sub_and_fetch
__sync_or_and_fetch
__sync_xor_and_fetch
__sync_fetch_and_min
__sync_fetch_and_max
__sync_fetch_and_umin
__sync_fetch_and_umax
__sync_val_compare_and_swap
__sync_bool_compare_and_swap

Differential Revision: https://reviews.llvm.org/D83340

Because of the layout of stores (that don't have a destination operand)
this check is exactly the same as the one in
RISCVInstrInfo::isLoadFromStackSlot.

Differential Revision: https://reviews.llvm.org/D81805

New test to show the changes after D81805 is committed.

Differential Revision: https://reviews.llvm.org/D83750

Summary:
This patch implements lowering of OpenMP barrier construct from
pft to OpenMPDialect.

Patch is carved out of following merged PR's from fir-dev branch
of https://github.com/flang-compiler/f18-llvm-project/

PR's:
https://github.com/flang-compiler/f18-llvm-project/pull/248
https://github.com/flang-compiler/f18-llvm-project/pull/251

Unfortunately primary tool `bbc` for functional validation is not
yet upstreamed. So this patch includes a unittest for lowering
`!OMP barrier` construct.

Some part of the these PR's still remains downstream(functional test
and dialect registration to legalizer) for obvious reasons.
Will upstream them when the dependencies are upstreamed.

Reviewed By: schweitz, kiranchandramohan

Differential Revision: https://reviews.llvm.org/D83659

This reverts commit f1efb8bb.

Reverted because it doesn't correctly update the pass return status, see

http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-debian/builds/9441/steps/test-check-all/logs/FAIL%3A%20LLVM%3A%3Awiden-i32-i8ptr.ll