llvm-svn: 373420

llvm-svn: 373419

We got confused and thought we might be pseudo-destroying the pointee
instead.

llvm-svn: 373418

llvm-svn: 373417

We currently just look for files named in the modulemap in its
associated source directory.  This means that we can't name
generated files, like TypeNodes.def now is, which means we can't
explicitly mark it as textual.  But fortunately that's okay
because (as I understand it) the most important purpose of naming
the header in the modulemap is to ensure that it's not treated as
public, and the search for public headers also only considers
files in the associated source directory.  This isn't an elegant
solution, since among other things it means that a build which
wrote the generated files directly into the source directory would
result in something that wouldn't build as a module, but that's
a problem for all our other generated files as well.

llvm-svn: 373416

In principle this should behave as any other constant. However
eliminateFrameIndex currently assumes a VALU use and uses a vector
shift. Work around this by selecting to VGPR for now until
eliminateFrameIndex is fixed.

llvm-svn: 373415

llvm-svn: 373414

This will be needed to support AGPR operations.

llvm-svn: 373413

llvm-svn: 373412

Account and report agprs separately on gfx908. Other targets
do not change the reporting.

Differential Revision: https://reviews.llvm.org/D68307

llvm-svn: 373411

llvm-svn: 373410

The original test was passing false to the append argument of
FindTypes (the only use of this feature!). This patch now replicates
that by passing a fresh TypeMap into the function where applicable.

llvm-svn: 373409

[X86] Add a DAG combine to shrink vXi64 gather/scatter indices that are constant with sufficient sign bits to fit in vXi32

The gather/scatter instructions can implicitly sign extend the indices. If we're operating on 32-bit data, an v16i64 index can force a v16i32 gather to be split in two since the index needs 2 registers. If we can shrink the index to the i32 we can avoid the split. It should always be safe to shrink the index regardless of the number of elements. We have gather/scatter instructions that can use v2i32 index stored in a v4i32 register with v2i64 data size.

I've limited this to before legalize types to avoid creating a v2i32 after type legalization. We could check for it, but we'd also need testing. I'm also only handling build_vectors with no bitcasts to be sure the truncate will constant fold.

Differential Revision: https://reviews.llvm.org/D68247

llvm-svn: 373408

The primary goal here is to make the type node hierarchy available to
other tblgen backends, although it should also make it easier to generate
more selective x-macros in the future.

Because tblgen doesn't seem to allow backends to preserve the source
order of defs, this is not NFC because it significantly re-orders IDs.
I've fixed the one (fortunately obvious) place where we relied on
the old order.  Unfortunately, I wasn't able to share code with the
existing AST-node x-macro generators because the x-macro schema we use
for types is different in a number of ways.  The main loss is that
subclasses aren't ordered together, which doesn't seem important for
types because the hierarchy is generally very shallow with little
clustering.

llvm-svn: 373407

redeclaration checking.  NFC.

llvm-svn: 373406

The main problem here is that `-*-version_min=` was not being passed to
the compiler when building test cases. This can cause problems when
testing on devices running older OSs because Clang would previously
assume the minimum deployment target is the the latest OS in the SDK
which could be much newer than what the device is running.

Previously the generated value looked like this:

`-arch arm64 -isysroot
<path_to_xcode>/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk`

With this change it now looks like:

`-arch arm64 -stdlib=libc++ -miphoneos-version-min=8.0 -isysroot
<path_to_xcode>/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk`

This mirrors the setting of config.target_cflags on macOS.

This change is made for ASan, LibFuzzer, TSan, and UBSan.

To implement this a new `get_test_cflags_for_apple_platform()` function
has been added that when given an Apple platform name and architecture
returns a string containing the C compiler flags to use when building
tests. This also calls a new helper function `is_valid_apple_platform()`
that validates Apple platform names.

This is the third attempt at landing the patch.

The first attempt (r359305) had to be reverted (r359327) due to a buildbot
failure. The problem was that calling `get_test_cflags_for_apple_platform()`
can trigger a CMake error if the provided architecture is not supported by the
current CMake configuration. Previously, this could be triggered by passing
`-DCOMPILER_RT_ENABLE_IOS=OFF` to CMake. The root cause is that we were
generating test configurations for a list of architectures without checking if
the relevant Sanitizer actually supported that architecture. We now intersect
the list of architectures for an Apple platform with
`<SANITIZER>_SUPPORTED_ARCH` (where `<SANITIZER>` is a Sanitizer name) to
iterate through the correct list of architectures.

The second attempt (r363633) had to be reverted (r363779) due to a build
failure. The failed build was using a modified Apple toolchain where the iOS
simulator SDK was missing. This exposed a bug in the existing UBSan test
generation code where it was assumed that `COMPILER_RT_ENABLE_IOS` implied that
the toolchain supported both iOS and the iOS simulator. This is not true. This
has been fixed by using the list `SANITIZER_COMMON_SUPPORTED_OS` for the list
of supported Apple platforms for UBSan. For consistency with the other
Sanitizers we also now intersect the list of architectures with
UBSAN_SUPPORTED_ARCH.

rdar://problem/50124489

Differential Revision: https://reviews.llvm.org/D61242

llvm-svn: 373405

Reviewers:
  arsenm

Differential Revision:
  https://reviews.llvm.org/D67574

llvm-svn: 373404

Summary: Lazy counters haven't improved performance for large fuzz targets.

Reviewers: kcc

Reviewed By: kcc

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67476

llvm-svn: 373403

Determine endianness at a time when it doesn't inadvertantly clear gdb's wrap_buffer via gdb.execute.

Summary:
I haven't managed a small reproduction for this bug, it involves
complicated and deeply nested data structures with a wide variety
of pretty printers. But in general, we shouldn't be combining
gdb's command line interface (via gdb.execute) with pretty-printers.

Subscribers: christof, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D68306

llvm-svn: 373402

This seems to be causing some performance regresions that I'm
trying to investigate.

One thing that stands out is that this transform can increase
the live range of the operands of the earlier logic op. This
can be bad for register allocation. If there are two logic
op inputs we should really combine the one that is closest, but
SelectionDAG doesn't have a good way to do that. Maybe we need
to do this as a basic block transform in Machine IR.

llvm-svn: 373401

llvm-svn: 373400

llvm-svn: 373399

The static analyzer is warning about potential null dereferences, but we should be able to use castAs<FunctionProtoType> directly and if not assert will fire for us.

llvm-svn: 373398

[X86] convertToThreeAddress, make sure second operand of SUB32ri is really an immediate before calling getImm().

It might be a symbol instead. We can't fold those since we can't
negate them.

Similar for other SUB with immediates.

Fixes PR43529.

llvm-svn: 373397

Remove duplicate getAs<> call, avoiding a clang static analyzer null dereference warning.

llvm-svn: 373396

Summary:
Most of the class definition in llvm/include/llvm/Support/FileCheck.h
are actually implementation details that should not be relied upon. This
commit moves all of it in a new header file under
llvm/lib/Support/FileCheck. It also takes advantage of the code movement
to put the code into a new llvm::filecheck namespace.

Reviewers: jhenderson, chandlerc, jdenny, probinson, grimar, arichardson, rnk

Subscribers: hiraditya, llvm-commits, probinson, dblaikie, grimar, arichardson, tra, rnk, kristina, hfinkel, rogfer01, JonChesterfield

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67649

llvm-svn: 373395

https://bugs.llvm.org/show_bug.cgi?id=43514

llvm-svn: 373394

llvm-svn: 373393

Summary:
OSSpinLock* are Apple/Darwin functions, but were previously located with ObjC checks as those were most closely tied to Apple platforms before.

Now that there's a specific Darwin module, relocating the check there.

This change was prepared by running rename_check.py.

Contributed By: mwyman

Reviewers: stephanemoore, dmaclach

Reviewed By: stephanemoore

Subscribers: Eugene.Zelenko, mgorny, xazax.hun, cfe-commits

Tags: #clang-tools-extra, #clang, #llvm

Differential Revision: https://reviews.llvm.org/D68148

llvm-svn: 373392

I submitted that patch after I got the LGTM, but the comments didn't
appear until after I submitted the change. This adds `const` to the
constructor argument and makes it a pointer.

llvm-svn: 373391

llvm-svn: 373390

PR42924 points out that copying the GlobalsMetadata type during
construction of AddressSanitizer can result in exteremely lengthened
build times for translation units that have many globals. This can be addressed
by just making the GlobalsMD member in AddressSanitizer a reference to
avoid the copy. The GlobalsMetadata type is already passed to the
constructor as a reference anyway.

Differential Revision: https://reviews.llvm.org/D68287

llvm-svn: 373389

[clang-format] [PR43372] - clang-format shows replacements in DOS files when no replacement is needed

Summary:
This is a patch to fix PR43372 (https://bugs.llvm.org/show_bug.cgi?id=43372) - clang-format can't format file with includes, ( which really keep providing replacements for already sorted headers.)

A similar issue was addressed by @krasimir in {D60199}, however, this seemingly only prevented the issue when the files being formatted did not contain windows line endings (\r\n)

It's possible this is related to https://twitter.com/StephanTLavavej/status/1176722938243895296 given who @STL_MSFT  works for!

As people often used the existence of replacements to determine if a file needs clang-formatting, this is probably pretty important for windows users

There may be a better way of comparing 2 strings and ignoring \r (which appear in both Results and Code), I couldn't choose between this idiom or the copy_if approach, but I'm happy to change it to whatever people consider more performant.

Reviewers: krasimir, klimek, owenpan, ioeric

Reviewed By: krasimir

Subscribers: cfe-commits, STL_MSFT, krasimir

Tags: #clang-format, #clang, #clang-tools-extra

Differential Revision: https://reviews.llvm.org/D68227

llvm-svn: 373388

Initial implementation of global aliases emission for the declare
variant pragma with implementation vendor context selector set.

llvm-svn: 373387

Summary:
This patch adds Root Node to the DDG. The purpose of the root node is to create a single entry node that allows graph walk iterators to iterate through all nodes of the graph, making sure that no node is left unvisited during a graph walk (eg. SCC or DFS). Once the DDG is fully constructed it will have exactly one root node. Every node in the graph is reachable from the root. The algorithm for connecting the root node is based on depth-first-search that keeps track of visited nodes to try to avoid creating unnecessary edges.

Authored By: bmahjour

Reviewer: Meinersbur, fhahn, myhsu, xtian, dmgreen, kbarton, jdoerfert

Reviewed By: Meinersbur

Subscribers: ychen, arphaman, simoll, a.elovikov, mgorny, hiraditya, jfb, wuzish, llvm-commits, jsji, Whitney, etiotto, ppc-slack

Tag: #llvm

Differential Revision: https://reviews.llvm.org/D67970

llvm-svn: 373386

In reality, this workaround is for the fact that LIBCXX_CXX_ABI=libcxxabi
can't be specified on Linux, since libc++abi isn't shipped with the system.
Since the build bots explicitly specify LIBCXX_CXX_ABI=libcxxabi, they fail
unless we apply the workaround.

llvm-svn: 373385

I tried applying D63883 three times and could never get around to
making it work. I'm giving up on that for now, but soon this should
be irrelevant anyway since all builds will move to the monorepo
(where we're always using the in-tree libc++abi unless explicitly
specified otherwise).

llvm-svn: 373384

If a single predecessor is found, still check if the block is
unreachable. The test that found this had a self loop unreachable block.
Resolves PR43493.

llvm-svn: 373383

llvm-svn: 373382

This reduces the (circular) dependency of libc++abi on a C++ standard
library. Outside of the demangler which uses fancier C++ features, the
only C++ headers now required by libc++abi are pretty much <new> and
<exception>, and that's because libc++abi defines some types that are
declared in those headers.

llvm-svn: 373381