debugging some leaks (PR5770 in particular).

llvm-svn: 91713

bootstrap.  This also replaces the WeakVH references that Chris objected to
with normal Value references.

llvm-svn: 91711

Enhance GRExprEngine::VisitCallExpr() to be used in an lvalue context.  Uncovered a new failing test case along the way, but we're making progress on handling C++ references in the analyzer.

llvm-svn: 91710

llvm-svn: 91709

- an MDNode is designated as function-local when created, and continues to be even if its operands are modified not to refer to function-local IR
- function-localness is designated via lowest bit in SubclassData
- getLocalFunction() descends MDNode tree to see if it is consistently function-local

Add verification of MDNodes to checks that MDNodes are consistently function-local.
Update AsmWriter to use isFunctionLocal().

llvm-svn: 91708

llvm-svn: 91707

llvm-svn: 91706

__USING_SJLJ_EXCEPTIONS__ on Darwin.

llvm-svn: 91705

llvm-svn: 91702

associative either.

llvm-svn: 91701

 - This should be done leveraging the backend, but I'm a little refactored
   out. I'll fix it one day, I promise.

llvm-svn: 91700

with "declared at" rather than "previous declaration is here".

llvm-svn: 91699

llvm-svn: 91698

llvm-svn: 91696

llvm-svn: 91695

dirs on my system. Which is wrong, since it was missing the x86_64-pc-linux-gnu
suffix.

llvm-svn: 91694

MI was not being used but it was also not being deleted, so it was kept in the garbage list. The memory itself was freed once the function code gen was done.

Once in a while the codegen of another function would create an instruction on the same address. Adding it to the garbage group would work once, but when another pointer was added it would cause an assert as "Cache" was about to be pushed to Ts.

For a patch that make us detect problems like this earlier, take a look at

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20091214/092758.html

With that patch we assert as soon and the new instruction is added to the garbage set.

llvm-svn: 91691

the heap, so that clients are not forced to copy the results during
the initial iteration. A separate clang_disposeCodeCompleteResults
function frees the returned results.

llvm-svn: 91690

new opaque value type, CharUnits. This will help us avoid accidentally mixing 
quantities that are in bit and character units.

llvm-svn: 91689

and getTypeSizeInChars() to reflect their basis in character type units, not
that of a possibly independent architecture-specific byte.

llvm-svn: 91688

llvm-svn: 91687

llvm-svn: 91686

llvm-svn: 91685

llvm-svn: 91684

sizes, offsets, and alignments in character units.

llvm-svn: 91683

Patch contributed by Ken Werner of IBM!

llvm-svn: 91681

Patch contributed by Ken Werner of IBM!

llvm-svn: 91680

This works now: (set_option "list_opt", ["val_1", "val_2", "val_3"])

llvm-svn: 91679

llvm-svn: 91678

nested-name specifier name.

I accidentally checked in the test case for this in the last commit ---
fortunately, that refactor was inspired by having debugged this problem already,
so I can fix the bug quick (though probably not fast enough for the buildbots).

llvm-svn: 91677

different functions and pick the function at lookup initialization time.
In theory we could actually divide the criteria functions into N different
functions for the N cases, but it's so not worth it.

Among other things, lets us invoke LookupQualifiedName without recomputing
IDNS info every time.

Do some refactoring in SemaDecl to avoid an awkward special case in LQN
that was only necessary for redeclaration testing for anonymous structs/unions ---
which could be done more efficiently with a scoped lookup anyway.

llvm-svn: 91676

multiple uses.  (The construct in question was found in gcc.)

llvm-svn: 91675

be non-optimal. To be precise, we should avoid folding loads if the instructions
only update part of the destination register, and the non-updated part is not
needed. e.g. cvtss2sd, sqrtss. Unfolding the load from these instructions breaks
the partial register dependency and it can improve performance. e.g.

movss (%rdi), %xmm0
cvtss2sd %xmm0, %xmm0

instead of
cvtss2sd (%rdi), %xmm0

An alternative method to break dependency is to clear the register first. e.g.
xorps %xmm0, %xmm0
cvtss2sd (%rdi), %xmm0

llvm-svn: 91672

clang -cc1: Rename -mcpu to -target-cpu to match other target options and not alias driver/backend option.

llvm-svn: 91671

llvm-svn: 91670

new InitializationSequence. This fixes some bugs (e.g., PR5808),
changed some diagnostics, and caused more churn than expected. What's
new:

  - InitializationSequence now has a "C conversion sequence" category
    and step kind, which falls back to
  - Changed the diagnostics for returns to always have the result type
    of the function first and the type of the expression second.
    CheckSingleAssignmentConstraints to peform checking in C. 
  - Improved ASTs for initialization of return values. The ASTs now
    capture all of the temporaries we need to create, but
    intentionally do not bind the tempoary that is actually returned,
    so that it won't get destroyed twice.
  - Make sure to perform an (elidable!) copy of the class object that
    is returned from a class.
  - Fix copy elision in CodeGen to properly see through the
    subexpressions that occur with elidable copies.
  - Give "new" its own entity kind; as with return values and thrown
    objects, we don't bind the expression so we don't call a
    destructor for it.

Note that, with this patch, I've broken returning move-only types in
C++0x. We'll fix it later, when we tackle NRVO.

llvm-svn: 91669

is not reassociative.

llvm-svn: 91667

llvm-svn: 91666

llvm-svn: 91664

Integer negation only overflows with INT_MIN, but that's an important case.

llvm-svn: 91662