assembly.  

By default, we don't emit IR for MS-style inline assembly (see r158833 as to 
why).  This is strictly for testing purposes and should not be enabled with the
expectation that things will work.  This is a temporary flag and will be removed
once MS-style inline assembly is fully supported.

llvm-svn: 160573

CI's name, and then used the StringRef pointing at its old name. I'm
fixing it by storing the name in a std::string, and hoisting the
renaming logic to happen always. This is nicer anyways as it will allow
the upgraded IR to have the same names as the input IR in more cases.

Another bug found by AddressSanitizer. Woot.

llvm-svn: 160572

PHIElimination splits critical edges when it predicts it can resolve
interference and eliminate copies. It doesn't split the edge if the
interference wouldn't be resolved anyway because the phi-use register is
live in the critical edge anyway.

Teach PHIElimination to split loop exiting edges with interference, even
if it wouldn't resolve the interference. This removes the necessary
copies from the loop, which is still an improvement from injecting the
copies into the loop.

The test case demonstrates the improvement. Before:

LBB0_1:
  cmpb  $0, (%rdx)
  leaq  1(%rdx), %rdx
  movl  %esi, %eax
  je  LBB0_1

After:

LBB0_1:
  cmpb  $0, (%rdx)
  leaq  1(%rdx), %rdx
  je  LBB0_1

  movl  %esi, %eax

llvm-svn: 160571

No longer assuming the number of prototype arguments is always less than the number of formal parameters for a variadic function call.

llvm-svn: 160570

The assertion was wrong in case we have a verbatim block without a closing
command.

Also add tests for closing command name in a verbatim block, since now it can
be empty in such cases.

llvm-svn: 160568

llvm-svn: 160567

llvm-svn: 160566

llvm-svn: 160565

llvm-svn: 160564

llvm-svn: 160563

llvm-svn: 160562

This time, make sure we don't try to print fixits with newline characters,
since they don't have a valid column width, and they don't look good anyway.

PR13417 (and originally <rdar://problem/11877454>)

llvm-svn: 160561

The CFG creates dummy DeclStmts with one Decl per statement, and it has
to do so from last to first in order to build the graph correctly.

llvm-svn: 160560

color.

llvm-svn: 160559

llvm-svn: 160558

llvm-svn: 160557

subprocesses.

llvm-svn: 160556

rdar://11842763

llvm-svn: 160554

llvm-svn: 160553

of c-functions nested in namespace in method implementations
by turning off its delayed parsing until a proper solution is 
figured out. pr13418

llvm-svn: 160552

llvm-svn: 160551

paragraph.

llvm-svn: 160550

llvm-svn: 160549

GetBestDestForJumpOnUndef() assumes there is at least 1 successor, which isn't
true if the block ends in an indirect branch with no successors. Fix this by
bailing out earlier in this case.

llvm-svn: 160546

[asan] make sure that the crash callbacks do not get merged (Chandler's idea: insert an empty InlineAsm). Change the order in which the new BBs are inserted: the slow path BB is insert between old BBs, the crash BB is inserted at the end. Don't create an empty BB (introduced by recent commits). Update the test. The experimental code that does manual crash callback merge will most likely be deleted later. 

llvm-svn: 160544

Don't use implicit register operands to calculate L-bit for AVX instructions. Needed because super reg defs and kills are added as implicit operands on 128-bit instructions. Fixes PR13349. Patch by Jose Fonseca.

llvm-svn: 160543

llvm-svn: 160542

Fixes PR13413, -Wunused-private-field now warns on unused fields initialized to NULL.

llvm-svn: 160541

Make RegisterOperand a subclass of DAGOperand so that RegisterOperands can be passed into multiclasses that take DAGOperands as multiclass parameters.

llvm-svn: 160540

char*'s - Greg removed the methods which accept char*'s earlier today.

llvm-svn: 160539

make it match the flag it tests.

llvm-svn: 160536

FileCheck.

This avoids copying files around needlessly during test runs.

llvm-svn: 160535

clients to default text files to 'eol-native'.

llvm-svn: 160534

llvm-svn: 160532

llvm-svn: 160531

same implementation for call evaluation.

llvm-svn: 160530

memory. This makes clang play nice with leak checkers.

llvm-svn: 160529

llvm-svn: 160528

previous ResetObjCLayout calls since this is now handled in Sema.

Part of rdar://11842763

llvm-svn: 160527

or implementation since we've now got a different layout.

Fixes rdar://11842763

llvm-svn: 160526