llvm-svn: 72712

in merge_types. It is incomplete. We probably want to issue 
a warning if user attempts to change the attribute from __weak to
__strong or vice-vera. It also assumes that a __weak/__strong
attribute can not be specified with other (currently one) type
attriute. 

llvm-svn: 72711

Avoid infinite looping in AllGlobalLoadUsesSimpleEnoughForHeapSRA(). This can happen when PHI uses are recursively dependent on each other.

llvm-svn: 72710

llvm-svn: 72709

llvm-svn: 72708

ADDC/ADDE use MVT::i1 (later, whatever it gets legalized to)
instead of MVT::Flag.  Remove CARRY_FALSE in favor of 0; adjust
all target-independent code to use this format.

Most targets will still produce a Flag-setting target-dependent
version when selection is done.  X86 is converted to use i32
instead, which means TableGen needs to produce different code
in xxxGenDAGISel.inc.  This keys off the new supportsHasI1 bit
in xxxInstrInfo, currently set only for X86; in principle this
is temporary and should go away when all other targets have
been converted.  All relevant X86 instruction patterns are
modified to represent setting and using EFLAGS explicitly.  The
same can be done on other targets.

The immediate behavior change is that an ADC/ADD pair are no
longer tightly coupled in the X86 scheduler; they can be
separated by instructions that don't clobber the flags (MOV).
I will soon add some peephole optimizations based on using
other instructions that set the flags to feed into ADC.

llvm-svn: 72707

llvm-svn: 72706

llvm-svn: 72705

directory exist.

llvm-svn: 72704

for the Next runtime.

llvm-svn: 72703

as necessary.

llvm-svn: 72702

Parse __OBJC data structures and synthesize magic .objc_ symbols.
Also, alter mangler so that objc method names are readable.

llvm-svn: 72700

llvm-svn: 72699

llvm-svn: 72698

llvm-svn: 72697

llvm-svn: 72696

not runaable.

llvm-svn: 72695

llvm-svn: 72694

llvm-svn: 72691

K&R-style definition.

llvm-svn: 72690

definition variadic.  I'm not completely sure it's legal, but the 
standard can be interpreted as making it legal, and gcc seems to think 
it's legal, so I didn't add an extension warning.

llvm-svn: 72689

RewriteStoreUserOfWholeAlloca deal with tail padding because 
isSafeUseOfBitCastedAllocation expects them to.  Otherwise, we crash 
trying to erase the bitcast.

llvm-svn: 72688

llvm-svn: 72687

preprocessor.

llvm-svn: 72686

metadata references non-Constant values such as instructions.

llvm-svn: 72685

llvm-svn: 72684

properties at the moment:

1. It allows stuff like "__strong id x; __weak id x;".

2. For constructs like "__strong id x; id x;", subsequent references to 
x lose the objc_gc attribute.

3. This incorrectly allows merges involving the address_space attribute.

4. Constructs like "id x; /* some code using x */ __weak id x;" don't 
apply the objc_gc attribute consistently to all uses of x.

The first three can probably be fixed relatively easily; the fourth 
would be extremely difficult to fix.

llvm-svn: 72683

llvm-svn: 72682

llvm-svn: 72681

After spending 15 minutes trying to figure out why this was crashing, I realized it was a bug in the test case :)

llvm-svn: 72680

llvm-svn: 72679

llvm-svn: 72678

llvm-svn: 72677

llvm-svn: 72676

llvm-svn: 72675

llvm-svn: 72674

Make sure to copy back arguments that can be changed by FindAllocationOverload. This fixes placement new. (Sebastian, please review).

llvm-svn: 72673

llvm-svn: 72672

Don't try to call getFileCharacteristic if the function declaration has an invalid source location (as is the case for the global allocation functions.

llvm-svn: 72671

Fix an off by one error when trying to perform copy initialization of operator new and operator delete arguments. Sebastian, please review.

llvm-svn: 72670