* Significantly improve DEBUG output
* Aggressively fold calls together if we inlined a graph that provides
  call nodes.
* Add a bailout if the current graph has over 200 call nodes in it, this is
  a really whacky case that should never happen.

llvm-svn: 4675

This slows stuff down a bit, but it should get much better before it gets any
worse.

llvm-svn: 4666

llvm-svn: 4665

llvm-svn: 4664

llvm-svn: 4663

llvm-svn: 4661

llvm-svn: 4660

llvm-svn: 4659

llvm-svn: 4657

llvm-svn: 4656

llvm-svn: 4655

llvm-svn: 4654

llvm-svn: 4649

llvm-svn: 4648

for example: 197.parser (64M->14M), 164.gzip (14M->2.7M).  The actual graphs
represented should not change at all.

llvm-svn: 4643

llvm-svn: 4640

the aux call sites should be copied

llvm-svn: 4639

llvm-svn: 4638

llvm-svn: 4637

llvm-svn: 4635

llvm-svn: 4634

  are marked DEAD.  This helped track down some bugs
- Fix a bunch of bugs where we were doing work on nodes after they became dead
- Add support for aux function call sites
- Add support for not cloning call sites

llvm-svn: 4633

llvm-svn: 4632

llvm-svn: 4628

llvm-svn: 4622

  - This seems to dramatically improve many benchmarks, only slowing down
    gzip significantly.

llvm-svn: 4621

llvm-svn: 4619

llvm-svn: 4602

explanitory enum instead.

llvm-svn: 4600

the BU class.

This will be used by the IPModRef class to do stuff, eventually perhaps the
TD pass will use it also.

Speaking of the TD pass, this also eliminates the self recursive case, which
was broken, and couldn't occur anyway.

llvm-svn: 4599

llvm-svn: 4596

llvm-svn: 4595

The problem was merging two unsized types like function's, ie int (int*) with
int (uint *)

llvm-svn: 4594

llvm-svn: 4592

llvm-svn: 4588

statistics for skipped functions

llvm-svn: 4587

implement the mod/ref bit masking

llvm-svn: 4578

llvm-svn: 4577

llvm-svn: 4576

llvm-svn: 4575