It's the last include-stack-root we tried to walk up. Use a better name
and better doxyments for it.

llvm-svn: 142071

TextDiagnosticPrinter argument to the TextDiagnostic helper class. This
cements the proper ordering of things: TextDiagnostic is now a viable
stand-alone class for emitting pretty-printed textual diagnostics to
a terminal.

llvm-svn: 142070

utility. This is a particularly nice win because it removes a pile of
parameters from these routines. Also name them a bit better. I'm trying
to follow the pattern of 'emit' routines writing directly to what is
expected to be the final output, while 'print' routines take a output
stream argument and can be used to build up intermediate buffers, etc.

Also, fix a bug I spotted by inspection from my last commit where
'LastLoc' and 'LastNonNoteLoc' were reversed. It's really scary that
this didn't trigger a single test failure. Will be working on tests for
more of this functionality now.

llvm-svn: 142069

across emissions.

1) The include stack printing is conditioned on non-note diagnostics,
   not just on warning diagnostics.
2) Those should be full source locations as they're tied to a source
   manager.
3) We should pass in the prior state to the TextDiagnostic constructor,
   allow it to mutate as diagnostics are emitted, and then cache the
   final state before tearing it down.

Some of this remains incomplete, specifically #3 isn't finished for the
non-note location. That'll come when the include stack printing sinks
down a level.

This also highlights how *completely* bug-ridden this code is. For
example, we currently do all these comparisons of a FullSourceLoc and
a SourceLocation... which silently does a SourceLocation to
SourceLocation comparison, completely disregarding the source manager
from whence one of the arguments came. Oops! Good thing in practice this
wasn't important, but it could in theory be suppressing caret
diagnostics in a second TU on a single clang invocation. I'm hoping to
hammer these bugs out as the refactorings occur, although for so many of
them it's really unlikely I can dream up a test case that would show the
potentially buggy behavior.

llvm-svn: 142067

use can't be dominated, saving one domtree lookup.

llvm-svn: 142066

interface now that its only caller is there.

llvm-svn: 142065

consumer. The TextDiagnostic interface now has a generic entry point for
emitting a diagnostic which uses a minimal interface that should be
compatible with StoredDiagnostics such as are available in libclang etc.

Some unfortunate shuffling of static functions as things get relocated.
Also some unfortunate public interface points added to
TextDiagnosticPrinter, but those are the next bits to get moved so they
won't last long.

llvm-svn: 142064

to operate directly on the source location and ranges associated with
a diagnostic rather than digging them out of the diagnostic. This had
a side benefit of cleaning up its code a tiny bit by using the ArrayRef
interface.

No functionality changed.

llvm-svn: 142063

This will simplify a number of calls in the diagnostic printing going
forward.

llvm-svn: 142062

Based on Horspool's simplified version of Boyer-Moore. We use a constant-sized table of
uint8_ts to keep cache thrashing low, needles bigger than 255 bytes are uncommon anyways.

The worst case is still O(n*m) but we do a lot better on the average case now.

llvm-svn: 142061

cases we need the second type-legalization pass in order to support all cases.

llvm-svn: 142060

Just because we're dealing with a GEP doesn't mean we can assert the
SCEV has a pointer type. The fix is simply to ignore the SCEV pointer
type, which we really didn't need.
Fixes PR11138 webkit crash.

llvm-svn: 142058

llvm-svn: 142057

llvm-svn: 142056

It extremely breaks builds when optimization is enabled. Twine should not hold temporary objects.

By the way, I might feel sad if I saw "i786-" "i1586-" or "iF86-".

llvm-svn: 142055

llvm-svn: 142054

llvm-svn: 142053

Convert stray doxygen comments to regular comments.  They logically represent the fields in FunctionProtoType.

llvm-svn: 142052

John convinced me this wasn't the right thing to do.  I'll follow up with a second
commit to make these non-doxygen comments.

llvm-svn: 142051

llvm-svn: 142050

llvm-svn: 142049

'final', and don't accept (then silently discard) braced init lists in C++98
new-expressions.

llvm-svn: 142048

llvm-svn: 142047

from config.h, it discovers the triple based on the execution environment.

Patch by Aaron Ballman <aaron@aaronballman.com>

llvm-svn: 142046

llvm-svn: 142044

Windows/Memory.inc: Support the ability to allocate memory "near" another block of memory on Win32. It has fixed FIXME.

Thanks to Aaron Ballman!

llvm-svn: 142039

I rewrote the algorithm a while back so it doesn't require map lookup,
but neglected to change the data structure. This was caught by
llvm-gcc self host, not because there's anything special about
llvm-gcc, but because it is the only test for nondeterminism we
currently have. Unit tests don't work well for everything; we should
always try to have a nondeterminism stress test running.

Fixes PR11133: llvm-gcc self host .o mismatch after enable-iv-rewrite=false

llvm-svn: 142036

Also note that it is actually doing much more than it should. This paves
the way for building a more generic 'Emit' routine that is the real
entry point here.

llvm-svn: 142035

This also adds a -Wc++98-compat-pedantic for warning on constructs which would
be diagnosed by -std=c++98 -pedantic (that is, it warns even on C++11 features
which we enable by default, with no warning, in C++98 mode).

llvm-svn: 142034

llvm-svn: 142033

This isn't put into the 'clear()' method because the information needs to stick
around (at least for a little bit) after the selection DAG is built.

llvm-svn: 142032

llvm-svn: 142031

It really doesn't, but when r141929 removed the hasSideEffects flag from
this instruction, it caused miscompilations.  I am guessing that it got
moved across a stack pointer update.

Also clear isRematerializable after checking that this instruction is
in fact never rematerialized in the nightly test suite.

llvm-svn: 142030

llvm-svn: 142029

rdar://10288916 is tracking this fix.

In the past, instcombine and other passes were promoting alloca alignment past
the natural alignment, resulting in dynamic stack realignment.  Lang's work now
prevents this from happening (LLVM commit r141599).  Now that this really 
shouldn't happen report a fatal error rather than silently generate bad code.

llvm-svn: 142028

llvm-svn: 142027

Make the step range plans capable of supporting multiple ranges.  Also make their constructors public, there isn't any good reason why you shouldn't be able to make these plans.

llvm-svn: 142026

llvm-svn: 142025

llvm-svn: 142024

avoids loading data from an external source, since those lookups were
causing some "interesting" recursion in LLDB.

This code is not efficient. I plan to remedy this inefficiency in a
follow-up commit.

llvm-svn: 142023