This allows the following syntax:

  void baz(__attribute__((nonnull)) const char *str);

instead of:

  void baz(const char *str) __attribute__((nonnull(1)));

This also extends to Objective-C methods.

The checking logic in Sema is not as clean as I would like.  Effectively
now we need to check both the FunctionDecl/ObjCMethodDecl and the parameters,
so the point of truth is spread in two places, but the logic isn't that
cumbersome.

Implements <rdar://problem/14691443>.

llvm-svn: 199467

This is an alternate interface to the separate iterator interfaces
provided by FunctionDecl and ObjCMethodDecl, which precede ArrayRef.

Providing this new interface is more convenient for some uses, and
likely the old interfaces should be removed.  I have not undertaken
that effort in this change because this API introduction may
solicit commentary and that was a larger change than I wanted
to introduce all at once to serve a direct purpose.

llvm-svn: 199466

No functionality change.

llvm-svn: 199465

llvm-svn: 199464

Also add copy support for FPR16.
Also add a missing test case file belongs to commit r197361.

llvm-svn: 199463

[AArch64 NEON] Custom lower conversion between vector integer and vector floating point if element bit-width doesn't match.

llvm-svn: 199462

Also fix the problem can't select scalar_to_vector from f32 to v2f32/v4f32.

llvm-svn: 199461

llvm-svn: 199460

Unused since r199250.

llvm-svn: 199459

The driver wasn't doing anything with it. Also rephrase the help text.

llvm-svn: 199458

Patch thanks to Nikola Smiljanic!

llvm-svn: 199457

Matching up the argument order in r199455's two RUN lines better demonstrates
what's going on.

llvm-svn: 199456

'%clang -verify' will now issue an error instead of succeeding without
verification. This should catch flawed tests like r199347.

Followup to r199451.

llvm-svn: 199455

llvm-svn: 199454

handling C++11 default initializers. Without this, other parts of Sema (such as
lambda capture) would think the default initializer is part of the surrounding
function scope.

llvm-svn: 199453

pointer, since this invokes undefined behavior. Based on a patch by Artyom
Skrobov! Handling of dependent exception specifications and some additional
testcases by me.

llvm-svn: 199452

Instead of dual-purposing a single flag, rename the driver option to
--verify-debug-info.

The frontend -verify option that enables diagnostic verification remains
unchanged except that it's now a pure CC1Option.

Both have been given proper help text.

llvm-svn: 199451

llvm-svn: 199450

llvm-svn: 199449

llvm-svn: 199448

llvm-svn: 199447

Show the top-level pch file as the culprit, rather than the immediate
dependency when a pch file imports a pcm from a module. To clarify the
relationship, the pch import stack is printed as notes. The old behaviour was
misleading when a pch imported a pcm (from a module), since removing the pcm
would not fix the problem, whereas rebuilding the pch would.

llvm-svn: 199446

class and use it pervasively to restore debug locations.
Fixes an interaction between cleanup and EH that caused the location
to not be restored properly after emitting a landing pad.
rdar://problem/15208190

llvm-svn: 199444

attribute syntax. There's nothing generalized about this; it's one of
several first-class attribute syntaxes we support, all of which are
more-or-less equally general.

As discussed on cfe-commits, we may want to revisit this if we start allowing
this syntax as an extension in C (or if C adopts the syntax), but hopefully
this diagnostic wording will be crystal clear to everyone in the mean time.

llvm-svn: 199443

llvm-svn: 199442

The part that checks that certain functions are marked deprecated doesn't
seem that useful, and it has the bad effect that the test hard-coded the
locations of the notes from that test.

llvm-svn: 199441

The _cpuid() implementation is the same as in lib/Headers/cpuid.h
with the parameter names adjusted to match the interface.

_xgetbv just does what the Intel manual says.

Differential Revision: http://llvm-reviews.chandlerc.com/D2564

llvm-svn: 199439

lookup when declaring a variable template specialization.

llvm-svn: 199438

llvm-svn: 199437

issues, a few testcases, and kills fish.

llvm-svn: 199436

This makes things a lot easier, because we can now talk about the
"argument allocation", which allocates all the memory for the call in
one shot.

The only functional change is to the verifier for a feature that hasn't
shipped yet.

llvm-svn: 199434

llvm-svn: 199432

Using -mmacosx-version-min (etc.) on non-Darwin platforms should be a warning,
not a hard error. There is no reason to add a special check for these options
in the default toolchain. This just removes the special check and then we get
the usual -Wunused-command-line-argument warning if someone tries to use one
of these options for a target where they are not supported.
<rdar://problem/15569346>

llvm-svn: 199431

When registering a pass, a pass can now specify a second construct that takes as
argument a pointer to TargetMachine.
The PassInfo class has been updated to reflect that possibility.
If such a constructor exists opt will use it instead of the default constructor
when instantiating the pass.

Since such IR passes are supposed to be rare, no specific support has been
added to this commit to allow an easy registration of such a pass.
In other words, for such pass, the initialization function has to be
hand-written (see CodeGenPrepare for instance).

Now, codegenprepare can be tested using opt:
opt -codegenprepare -mtriple=mytriple input.ll

llvm-svn: 199430

Adding a doxygen comment for each bit of API to indicate at which
LTO_API_VERSION each was available, manually gleaned from successive
git-blames.  A few notes:

- LTO_API_VERSION was set to 3 at its introduction.
- I've indicated all the API introduced before LTO_API_VERSION was
  around as available "prior to LTO_API_VERSION=3".
- A number of API changes neglected to bump LTO_API_VERSION.  These I've
  indicated as available at the *next* bump of LTO_API_VERSION.

llvm-svn: 199429

llvm-svn: 199428

llvm-svn: 199427

Part 1 changes PlatformLinux::CreateInstance to always create with
is_host=false; that method is only used as the plug-in creator method
associated with the remote-linux platform plugin, and should therefore
always be remote.

Part 1 patch by Steve Pucci.

Part 2: fix a test break on linux.

In test/functionalities/thread/thread_specific_break, when using gcc,
either C99 mode would need to be enabled, or the code would need to
change.  I changed a couple loop variable definitions to conform
to pre-C99 to simplify testing the fix.  The second issue was
the necessity to include -lpthread as a linker option in the Makefile.

Any issues with that part are due to me (Todd Fiala).

llvm-svn: 199426

Fix an instance where we would drop fast math flags when performing an fdiv to reciprocal multiply transformation.

llvm-svn: 199425

Fix a bug in InstCombine where we failed to preserve fast math flags when optimizing an FMUL expression.

llvm-svn: 199424