llvm-svn: 334645

llvm-svn: 334644

Previously ThreadPool could only queue async "jobs", i.e. work
that was done for its side effects and not for its result.  It's
useful occasionally to queue async work that returns a value.
From an API perspective, this is very intuitive.  The previous
API just returned a shared_future<void>, so all we need to do is
make it return a shared_future<T>, where T is the type of value
that the operation returns.

Making this work required a little magic, but ultimately it's not
too bad.  Instead of keeping a shared queue<packaged_task<void()>>
we just keep a shared queue<unique_ptr<TaskBase>>, where TaskBase
is a class with a pure virtual execute() method, then have a
templated derived class that stores a packaged_task<T()>.  Everything
else works out pretty cleanly.

Differential Revision: https://reviews.llvm.org/D48115

llvm-svn: 334643

Summary:
test_set_working_dir was testing two scenario: failure to set the working dir because of a non existent directory and succeeding to set the working directory. Since the negative case fails on both Linux and Windows, the positive case was never tested. I split the test into two which allows us to always run both the negative and positive cases. The positive case now succeeds on Linux and the negative case still fails.
During the investigation, it turned out that lldbtest.py will try to execute a process launch command up to 3 times if the command failed. This means that we could be covering up intermittent failures by running any test that does process launch multiple times without ever realizing it. I've changed the counter to 1 (though it can still be overwritten with the environment variable).
This change also fixes both the positive and negative cases on Windows. There were a few issues:
1) In ProcessLauncherWindows::LaunchProcess, the error was not retrieved until CloseHandle was possibly called. Since CloseHandle is also a system API, its success would overwrite any existing error that could be retrieved using GetLastError. So by the time the error was retrieved, it was now a success.
2) In DebuggerThread::StopDebugging TerminateProcess was called on the process handle regardless of whether it was a valid handle. This was causing the process to crash when the handle was LLDB_INVALID_PROCESS (0xFFFFFFFF).
3) In ProcessWindows::DoLaunch we need to check that the working directory exists before launching the process to have the same behavior as other platforms which first check the directory and then launch process. This way we also control the exact error string.

Reviewers: labath, zturner, asmith, jingham

Reviewed By: labath

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D48050

llvm-svn: 334642

Summary:
Any invocation of `clang -fuse-ld=lld` that results in a link command
on a macOS host currently fails, because the Darwin lld driver does not
recognize the `-lto_library` option that Clang passes it. Fix the error
by having the Darwin driver ignore the option.

The Clang driver's macOS toolchain is written such that it will always
pass the `-lto_library` option to the linker invocation on a macOS host.
And although the DarwinLdDriver is written to ignore any unknown arguments,
because `-lto_library` begins with `-l`, the DarwinLdDriver interprets it
as a library search command, for a library named "to_library". When the
DarwinLdDriver is unable to find a library specified via `-l`, it exits
with a hard error. This causes any invocation of `clang -fuse-ld=lld`
that results in a link command on a macOS host to fail with an error.

To fix the issue, I considered two alternatives:

1. Modify the Clang Darwin toolchain to only pass `-lto_library` if lld
   is *not* being used. lld doesn't support LTO on Darwin anyway, so it
   can't use the option. However, I opted against this because, if and
   when lld *does* support LTO on Darwin, I'll have to make another
   commit to Clang in order to get it to pass the option to lld again.
2. Modify the Darwin lld driver to ignore the `-lto_library` option.
   Just in case users may take this to mean LTO is supported, I also
   added a warning. If and when lld supports LTO on Darwin, the same
   commit that adds support for this option can remove the warning.

Option (2) seemed better to me, and is the rationale behind this commit.

Test Plan: check-lld

Reviewers: ruiu, smeenai, pcc

Reviewed By: smeenai

Subscribers: JDevlieghere, pcc, mehdi_amini, inglorion, steven_wu, llvm-commits

Differential Revision: https://reviews.llvm.org/D47994

llvm-svn: 334641

Differential Revision: https://reviews.llvm.org/D48133

llvm-svn: 334640

Summary: These intrinsics result in hint instructions. They are provided here for MSVC ARM64 compatibility.

Reviewers: mstorsjo, compnerd, javed.absar

Reviewed By: mstorsjo

Subscribers: kristof.beyls, chrib, cfe-commits

Differential Revision: https://reviews.llvm.org/D48132

llvm-svn: 334639

This patch adds a data formatter for NSDecimalNumber. The latter is a
Foundation object used for representing and performing arithmetic on
base-10 numbers that bridges to Decimal.

Differential revision: https://reviews.llvm.org/D48114

llvm-svn: 334638

Differential Revision: https://reviews.llvm.org/D47791

llvm-svn: 334637

rdar://39593879
https://bugs.llvm.org/show_bug.cgi?id=37142

Differential Revision: https://reviews.llvm.org/D48139

llvm-svn: 334636

Multiplication followed by addition
(https://en.wikipedia.org/wiki/Multiply–accumulate_operation) is a
sufficiently common use-case to warrant a separate helper.

Differential Revision: https://reviews.llvm.org/D48138

llvm-svn: 334635

Returning optional is much safer.
The previous API had potential to cause use of undefined variables, if
the value passed by pointer was accidentally read afterwards.

Differential Revision: https://reviews.llvm.org/D48137

llvm-svn: 334634

Fixes PR37790.

In some (very rare) cases, the LSUnit (Load/Store unit) was wrongly marking a
load (or store) as "ready to execute" effectively bypassing older memory barrier
instructions.

To reproduce this bug, the memory barrier must be the first instruction in the
input assembly sequence, and it doesn't have to perform any register writes.

llvm-svn: 334633

This can happen on macOS if the user's Xcode is at a path with spaces in it.

llvm-svn: 334632

Replace const char pointers with llvm::StringRef and use its equality
operator for string comparisons.

llvm-svn: 334631

On Windows we've observed that if you open a file, write to it, map it into
memory and close the file handle, the contents of the memory mapping can
sometimes be incorrect. That was what we did when adding an entry to the
ThinLTO cache using the TempFile and MemoryBuffer classes, and it was causing
intermittent build failures on Chromium's ThinLTO bots on Windows. More
details are in the associated Chromium bug (crbug.com/786127).

We can prevent this from happening by keeping a handle to the file open while
the mapping is active. So this patch changes the mapped_file_region class to
duplicate the file handle when mapping the file and close it upon unmapping it.

One gotcha is that the file handle that we keep open must not have been
created with FILE_FLAG_DELETE_ON_CLOSE, as otherwise the operating system
will prevent other processes from opening the file. We can achieve this
by avoiding the use of FILE_FLAG_DELETE_ON_CLOSE altogether.  Instead,
we use SetFileInformationByHandle with FileDispositionInfo to manage the
delete-on-close bit. This lets us remove the hack that we used to use to
clear the delete-on-close bit on a file opened with FILE_FLAG_DELETE_ON_CLOSE.

A downside of using SetFileInformationByHandle/FileDispositionInfo as
opposed to FILE_FLAG_DELETE_ON_CLOSE is that it prevents us from using
CreateFile to open the file while the flag is set, even within the same
process. This doesn't seem to matter for almost every client of TempFile,
except for LockFileManager, which calls sys::fs::create_link to create a
hard link from the lock file, and in the process of doing so tries to open
the file. To prevent this change from breaking LockFileManager I changed it
to stop using TempFile by effectively reverting r318550.

Differential Revision: https://reviews.llvm.org/D48051

llvm-svn: 334630

spatel noticed it was missing in D47993.

llvm-svn: 334629

https://bugs.llvm.org/show_bug.cgi?id=37778
...shows a miscompile resulting from marking nan builtins as 'const'.

The nan libcalls/builtins take a pointer argument:
http://www.cplusplus.com/reference/cmath/nan-function/
...and the chars dereferenced by that arg are used to fill in the NaN constant payload bits.

"const" means that the pointer argument isn't dereferenced. That's translated to "readnone" in LLVM.
"pure" means that the pointer argument may be dereferenced. That's translated to "readonly" in LLVM.

This change prevents the IR optimizer from killing the lead-up to the nan call here:

double a() {
  char buf[4];
  buf[0] = buf[1] = buf[2] = '9';
  buf[3] = '\0';
  return __builtin_nan(buf);
}

...the optimizer isn't currently able to simplify this to a constant as we might hope, 
but this patch should solve the miscompile.

Differential Revision: https://reviews.llvm.org/D48134

llvm-svn: 334628

Use `\brief` instead of `\Brief`.  NFC.

llvm-svn: 334627

It seems to be passing, at least on the bots I've seen.

llvm-svn: 334626

Currently the handle type is a global pointer which holds 8 bytes.
We need a larger type which hold 16 bytes, therefore change it
to [i64 x 2].

Differential Revision: https://reviews.llvm.org/D48094

llvm-svn: 334625

Summary:
Move madvise(MADV_NOHUGEPAGE) for the meta shadow memory after the meta
shadow memory is mapped (currently it silently fails with ENOMEM).

Add a diagnostic message to detect similar problems in the future.

Reviewers: dvyukov

Subscribers: kubamracek, delcypher, #sanitizers, llvm-commits

Differential Revision: https://reviews.llvm.org/D48097

llvm-svn: 334624

[CostModel][X86] Test showing failure to recognise REVERSE shuffle mask if the elements come from the second src

llvm-svn: 334623

See bug 37566: https://bugs.llvm.org/show_bug.cgi?id=37566

Reviewers: artem.tamazov, arsenm, nhaehnle

Differential Revision: https://reviews.llvm.org/D47884

llvm-svn: 334622

llvm-svn: 334620

Just use field initializers that don't suffer from this problem

llvm-svn: 334619

As noted by Pavel on lldb-commits, we don't need the temp path, we can
just pass the filename directly into extension() and path().

llvm-svn: 334618

Not sure why, but it breaks buildbot clang-cmake-armv8-full.
It causes a failure in TEST 'Xray-armhf-linux :: TestCases/Posix/profiling-single-threaded.cc'.

llvm-svn: 334617

[CostModel][X86] Test showing failure to recognise BROADCAST shuffle mask if the elements come from the second src

llvm-svn: 334616

With the recent changes in FileSpec to use LLVM's path style, it is
possible to delegate a bunch of common path operations to LLVM's path
helpers. This means we only have to maintain a single implementation and
at the same time can benefit from the efforts made by the rest of the
LLVM community.

This is part one of a set of patches. There was no obvious way to split
this so I just worked from top to bottom.

Differential revision: https://reviews.llvm.org/D48084

llvm-svn: 334615

llvm-svn: 334614

Diasble the use of the type __float128 for PPC machines older
than Power9.

The use of -mfloat128 for PPC machine older than Power9 will result
in an error.

Differential Revision: https://reviews.llvm.org/D48088

llvm-svn: 334613

Fixes a build bot breakage caused by r334593.

llvm-svn: 334612

Summary:
This patch adds a modulemap which allows compiling the lldb headers into C++ modules
(for example in builds with LLVM_ENABLE_MODULES=On).

Even though most of the affected code has been cleaned up to work with the more strict
C++ module semantics, there are still some workarounds left in the current modulemap
(the most obvious one is the big `lldb` wrapper module).

It also moves the Obj-C++ files in lldb to their own subdirectories. This was necessary
because we need to filter out the modules flags for this code.

Note: With the latest clang and libstdc++ it seems necessary to have a STL C++ module
to get a working LLVM_ENABLE_MODULES build for lldb. Otherwise clang will falsely
detect ODR violations in the textually included STL code inside the lldb modules.

Reviewers: aprantl, bruno

Reviewed By: aprantl, bruno

Subscribers: mgorny, yamaguchi, v.g.vassilev, lldb-commits

Differential Revision: https://reviews.llvm.org/D47929

llvm-svn: 334611

llvm-svn: 334610

See bug 37653: https://bugs.llvm.org/show_bug.cgi?id=37653

Reviewers: artem.tamazov, arsenm

Differential Revision: https://reviews.llvm.org/D47885

llvm-svn: 334609

We're constant folding here, so we shouldn't check uses. This matches
the IR optimizer behavior.

The x86 test shows the expected win. The AArch64 test shows something
else. This only seems to happen if the "generic" AArch64 CPU model is 
used by MachineCombiner, so I'll file a bug report to follow-up.

llvm-svn: 334608

Summary:
The code that handles ISD:Register and ISD::CopyFromReg assumes
the target is amdgcn, so this is broken on r600.  We don't
need this analysis on r600 anyway so we can safely move
it to SITargetLowering.

Reviewers: alex-t, arsenm, nhaehnle

Reviewed By: arsenm

Subscribers: msearles, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits

Differential Revision: https://reviews.llvm.org/D46298

llvm-svn: 334607

http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast/builds/31500

llvm-svn: 334606

The equivalent AArch64 test added at rL334556 isn't showing
the expected output from the DAGCombiner code change that 
would fix this example. That's a machine combiner bug from 
what I see.

llvm-svn: 334605