llvm-svn: 150677

llvm-svn: 150676

llvm-svn: 150675

llvm-svn: 150674

llvm-svn: 150673

llvm-svn: 150672

llvm-svn: 150671

llvm-svn: 150670

Modify the algorithm when traversing the DAGCombiner's worklist to be O(log N) for all operations. This fixes a horrible worst case with lots of nodes where 99% of the time was being spent in std::remove.

llvm-svn: 150669

llvm-svn: 150668

With MSVCRT, prior checker missed emission of #INF and #NAN.

FIXME: Checking should be simpler.
llvm-svn: 150667

llvm-svn: 150666

Use this to allow the lldb Driver to emit notifications for breakpoint modifications.
<rdar://problem/10619974>

llvm-svn: 150665

Replace all instances of dg.exp file with lit.local.cfg, since all tests are run with LIT now and now Dejagnu. dg.exp is no longer needed.

Patch reviewed by Daniel Dunbar. It will be followed by additional cleanup patches.

llvm-svn: 150664

Shift Microsoft enum extensions from -fms-extensions to -fms-compatibility, so -fms-extensions doesn't affect enum semantics in incompatible ways. <rdar://problem/10657186>.

llvm-svn: 150663

Template specialize SmallVector::push_back based on POD-ness of the type.  Reduces clang binary by 188KB

llvm-svn: 150662

atof(3) might behave differently among platforms.

llvm-svn: 150661

llvm-svn: 150660

piece can always be generated.

The default end of diagnostic path piece was failing to generate on a
BlockEdge that was outgoing from a basic block without a terminator,
resulting in a very simple diagnostic being rendered (ex: no path
highlighting or custom visitors). Reuse another function, which is
essentially doing the same thing and correct it not to fail when a block
has no terminator.

llvm-svn: 150659

We are not properly handling the memory regions that escape into struct
fields, which led to a bunch of false positives. Be conservative here
and give up when a pointer escapes into a struct.

llvm-svn: 150658

  * Fix bug when determining whether && / || are potential constant expressions
  * Try harder when determining whether ?: is a potential constant expression
  * Produce a diagnostic on sizeof(VLA) to provide a better source location

llvm-svn: 150657

llvm-svn: 150656

llvm-svn: 150655

llvm-svn: 150654

llvm-svn: 150653

that take formats or sizes.

Also document that scalar expression results can be used in any command using
expressions inside backticks.

llvm-svn: 150652

llvm-svn: 150651

llvm-svn: 150650

llvm-svn: 150649

The garbage collection metadata needs to be merged "intelligently", when two or
more modules are linked together, and not merely appended. (Appending creates a
section which is too large.) The module flags metadata method is the way to do
this.
<rdar://problem/8198537>

llvm-svn: 150648

llvm-svn: 150647

The rule governing the flags is this:

  no-gc + no-gc   = no-gc
  no-gc +    gc   = no-gc
  no-gc + gc-only = error
     gc +    gc   = gc
     gc + gc-only = gc-only
gc-only + gc-only = gc-only

llvm-svn: 150646

pointers and block pointers). We use dummy definitions to keep the
invariant that an implicit, used definition has a body; IR generation
will substitute the actual contents, since they can't be represented
as C++. 

For the block pointer case, compute the copy-initialization needed to
capture the lambda object in the block, which IR generation will need
later.

llvm-svn: 150645

-fno-objc-arc-exceptions. This will allow the optimizer to perform
optimizations which are only safe under that flag.

This is a part of rdar://10803830.

llvm-svn: 150644

and move it out of -Wgnu so that -Wno-gnu leaves it enabled. As requested by
Eli.

llvm-svn: 150643

loop-rotate shouldn't hoist alloca instructions out of a loop.  Patch by Patrik Hägglund, with slightly modified test.  Issue reported by Patrik Hägglund on llvmdev.

llvm-svn: 150642

llvm-svn: 150641

and unions, and C++11 generalized constant expressions.

llvm-svn: 150640

as unscanned. // rdar://10832643

llvm-svn: 150639

Call instructions no longer have a list of 43 call-clobbered registers.
Instead, they get a single register mask operand with a bit vector of
call-preserved registers.

This saves a lot of memory, 42 x 32 bytes = 1344 bytes per call
instruction, and it speeds up building call instructions because those
43 imp-def operands no longer need to be added to use-def lists. (And
removed and shifted and re-added for every explicit call operand).

Passes like LiveVariables, LiveIntervals, RAGreedy, PEI, and
BranchFolding are significantly faster because they can deal with call
clobbers in bulk.

Overall, clang -O2 is between 0% and 8% faster, uniformly distributed
depending on call density in the compiled code.  Debug builds using
clang -O0 are 0% - 3% faster.

I have verified that this patch doesn't change the assembly generated
for the LLVM nightly test suite when building with -disable-copyprop
and -disable-branch-fold.

Branch folding behaves slightly differently in a few cases because call
instructions have different hash values now.

Copy propagation flushes its data structures when it crosses a register
mask operand. This causes it to leave a few dead copies behind, on the
order of 20 instruction across the entire nightly test suite, including
SPEC. Fixing this properly would require the pass to use different data
structures.

llvm-svn: 150638