llvm-svn: 123643

llvm-svn: 123623

'rotq*' and 'shlq*' instructions go to the odd pipeline,
wheras the inter-vector equivalents 'rot*', 'shl*' go 
to the even.

llvm-svn: 123622

This shaves off 4 popcounts from the hacked 186.crafty source.

This is enabled even when a native popcount instruction is available. The
combined code is one operation longer but it should be faster nevertheless.

llvm-svn: 123621

llvm-svn: 123620

        movw    r0, :lower16:(L_foo$non_lazy_ptr-(LPC0_0+4))
        movt    r0, :upper16:(L_foo$non_lazy_ptr-(LPC0_0+4))
LPC0_0:
        add     r0, pc, r0

It's not yet enabled by default as some tests are failing. I suspect bugs in
down stream tools.

llvm-svn: 123619

llvm-svn: 123618

llvm-svn: 123617

long strings in "char *" (with any combo if qualifiers).

llvm-svn: 123616

V2 and we only have gdb_class_getClass, then make sure the isa isn't NULL 
before trying to call gdb_class_getClass otherwise we end up deadlocking the
objective C runtime.

llvm-svn: 123615

llvm-svn: 123614

the way LLDB lazily gets complete definitions for types within the debug info.
When we run across a class/struct/union definition in the DWARF, we will only
parse the full definition if we need to. This works fine for top level types
that are assigned directly to variables and arguments, but when we have a 
variable with a class, lets say "A" for this example, that has a member:
"B *m_b". Initially we don't need to hunt down a definition for this class
unless we are ever asked to do something with it ("expr m_b->getDecl()" for
example). With my previous approach to lazy type completion, we would be able
to take a "A *a" and get a complete type for it, but we wouldn't be able to
then do an "a->m_b->getDecl()" unless we always expanded all types within a
class prior to handing out the type. Expanding everything is very costly and
it would be great if there were a better way.

A few months ago I worked with the llvm/clang folks to have the 
ExternalASTSource class be able to complete classes if there weren't completed
yet:

class ExternalASTSource {
....

    virtual void
    CompleteType (clang::TagDecl *Tag);
    
    virtual void 
    CompleteType (clang::ObjCInterfaceDecl *Class);
};

This was great, because we can now have the class that is producing the AST
(SymbolFileDWARF and SymbolFileDWARFDebugMap) sign up as external AST sources
and the object that creates the forward declaration types can now also
complete them anywhere within the clang type system.

This patch makes a few major changes:
- lldb_private::Module classes now own the AST context. Previously the TypeList
  objects did.
- The DWARF parsers now sign up as an external AST sources so they can complete
  types.
- All of the pure clang type system wrapper code we have in LLDB (ClangASTContext,
  ClangASTType, and more) can now be iterating through children of any type,
  and if a class/union/struct type (clang::RecordType or ObjC interface) 
  is found that is incomplete, we can ask the AST to get the definition. 
- The SymbolFileDWARFDebugMap class now will create and use a single AST that
  all child SymbolFileDWARF classes will share (much like what happens when
  we have a complete linked DWARF for an executable).
  
We will need to modify some of the ClangUserExpression code to take more 
advantage of this completion ability in the near future. Meanwhile we should
be better off now that we can be accessing any children of variables through
pointers and always be able to resolve the clang type if needed.

llvm-svn: 123613

llvm-svn: 123612

Enforce C++[class.mem]p8:
A virt-specifier-seq shall contain at most one of each virt-specifier.

llvm-svn: 123611

llvm-svn: 123610

eliminating a potentially quadratic data structure, this also gives a 17%
speedup when running -scalarrepl on test-suite + SPEC2000 + SPEC2006. My initial
experiment gave a greater speedup around 25%, but I moved the dominator tree
level computation from dominator tree construction to PromoteMemToReg.

Since this approach to computing IDFs has a much lower overhead than the old
code using precomputed DFs, it is worth looking at using this new code for the
second scalarrepl pass as well.

llvm-svn: 123609

llvm-svn: 123608

llvm-svn: 123607

http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2010/n3206.htm

This lands support for parsing virt-specifier-seq after member functions, including the
contextual keywords 'final', and 'override'. The keywords are not yet used for anything.

llvm-svn: 123606

UnRevert "Revert "Archive: Replace all internal uses of PathV1 with PathV2. The external API still uses PathV1.""

llvm-svn: 123605

llvm-svn: 123604

Fix a bug where the -Wmissing-noreturn would always treat constructors with base or member initializers as noreturn.

llvm-svn: 123603

llvm-svn: 123602

llvm-svn: 123601

Downgrade ext_enumerator_too_large from ExtWarn to Extension in Microsoft mode. Otherwise you can warnings flooding trying to selfhost clang with fms-extensions because of "unsigned int" -> "signed int" enumerator conversion.

llvm-svn: 123600

llvm-svn: 123599

This fixes PR8987

llvm-svn: 123598

llvm-svn: 123597

Teach DAE to look for functions whose arguments are unused, and change all callers to pass in an undefvalue instead.

llvm-svn: 123596

This reverts commit dd103021a889a986a181ce36ed7b0e8dc9b645e1.

llvm-svn: 123595

it shouldn't, but I have no way to verify that it doesn't actually exist on the
buildbot.

llvm-svn: 123594

llvm-svn: 123593

This patch is enough to have shared objects recognized by LLDB.  We can handle
position independent executables.  We can handle dynamically loaded modules
brought in via dlopen.

The DYLDRendezvous class provides an interface to a structure present in the
address space of ELF-based processes.  This structure provides the address of a
function which is called by the linker each time a shared object is loaded and
unloaded (thus a breakpoint at that address will let LLDB intercept such
events), a list of entries describing the currently loaded shared objects, plus
a few other things.

On Linux, processes are brought up with an auxiliary vector on the stack.  One
element in this vector contains the (possibly dynamic) entry address of the
process.  One does not need to walk the stack to find this information as it is
also available under /proc/<pid>/auxv.  The new AuxVector class provides a
convenient read-only view of this auxiliary vector information.  We use the
dynamic entry address and the address as specified in the object file to compute
the actual load address of the inferior image.  This strategy works for both
normal executables and PIE's.

llvm-svn: 123592

llvm-svn: 123591

llvm-svn: 123590

llvm-svn: 123585

This fixes the original testcase in PR8927. It also causes a clang
binary built with a patched clang to increase in size by 0.21%.

We can probably get some of the size back by writing a pass that
detects that a global never has its pointer compared and adds
unnamed_addr to it (maybe extend global opt). It is also possible that
there are some other cases clang could add unnamed_addr to.

I will investigate extending globalopt next.

llvm-svn: 123584

llvm-svn: 123583

llvm-svn: 123582

non-variadic function template over a variadic one. This matches GCC
and the intent of the C++0x wording, in a way that I think is likely
to be acceptable to the committee.

llvm-svn: 123581