This causes the LLVMgold plugin to segfault. More information on the
replies to r201608.

llvm-svn: 201669

    
If LLVM is built without X86 as a supported target then the test would
mysteriously fail.

llvm-svn: 201668

Just a wild stab in the dark really, but in the absence of any ability to
reproduce the problem...

llvm-svn: 201658

On x86, shifting a vector by a scalar is significantly cheaper than shifting a
vector by another fully general vector. Unfortunately, because SelectionDAG
operates on just one basic block at a time, the shufflevector instruction that
reveals whether the right-hand side of a shift *is* really a scalar is often
not visible to CodeGen when it's needed.

This adds another handler to CodeGenPrepare, to sink any useful shufflevector
instructions down to the basic block where they're used, predicated on a target
hook (since on other architectures, doing so will often just introduce extra
real work).

rdar://problem/16063505

llvm-svn: 201655

This change also removes CMAKE_LINK_FLAGS setting that seems to be ignored by cmake.

llvm-svn: 201654

llvm-svn: 201651

Remove special FP opcode maps and instead add enough MRM_XX formats to handle all the FP operations. This increases format by 1 bit, but decreases opcode map by 1 bit so the TSFlags size doesn't change.

llvm-svn: 201649

llvm-svn: 201646

llvm-svn: 201645

Remove A6/A7 opcode maps. They can all be handled with a TB map, opcode of 0xa6/0xa7, and adding MRM_C0/MRM_E0 forms. Removes 376K from the disassembler tables.

llvm-svn: 201641

Load Configuration Table may contain a pointer to SEH table. This patch is to
print the offset to the table. Printing SEH table contents is a TODO.

The layout of Layout Configuration Table is described in Microsoft PE/COFF
Object File Format Spec, but the table's offset/size descriptions seems to be
totally wrong, at least in revision 8.3 of the spec. I believe the table in
this patch is the correct one.

llvm-svn: 201638

llvm-svn: 201633

This enhances the macro parser to parse and handle parameter qualifications,
which is needed to support required formal parameters in macro definitions.  A
required parameter may not be defaulted (though providing a default value is
accepted with a warning).  This improves GAS compatibility.

Partially addresses PR9248.

llvm-svn: 201630

Rather than using std::pair, create a structure to represent the type.  This is
a preliminary refactoring to enable required parameter handling.  Additional
state is needed to indicate required parameters.  This has a minor side effect
of improving readability by providing more accurate names compared to first and
second.

llvm-svn: 201629

llvm-svn: 201625

Now that llvm's codegen knows to use an 'l' prefix when needed, we can just
use PrivateLinkage.

llvm-svn: 201624

When outputting an object we check its section to find its name, but when
looking for the section with -ffunction-section we look for the symbol name.

Break the loop by requesting a name with the private prefix when constructing
the section name. This matches the behavior before r201608.

llvm-svn: 201622

Some references to llvm-gcc were so crusty that I wasn't sure how to
proceed and so I've left them intact.

I also slipped in a quick peephole fix to use a :doc: link instead of
raw HTML link.

llvm-svn: 201619

From a cursory look it seems like all the described commandline options
and such apply to clang just fine, but I'd appreciate a second opinion.

llvm-svn: 201616

The IR
@foo = private constant i32 42

is valid, but before this patch we would produce an invalid MachO from it. It
was invalid because it would use an L label in a section where the liker needs
the labels in order to atomize it.

One way of fixing it would be to just reject this IR in the backend, but that
would not be very front end friendly.

What this patch does is use an 'l' prefix in sections that we know the linker
requires symbols for atomizing them. This allows frontends to just use
private and not worry about which sections they go to or how the linker handles
them.

One small issue with this strategy is that now a symbol name depends on the
section, which is not available before codegen. This is not a problem in
practice. The reason is that it only happens with private linkage, which will
be ignored by the non codegen users (llvm-nm and llvm-ar).

llvm-svn: 201608

This is quiet a bit less confusing now that TargetData was renamed DataLayout.

llvm-svn: 201606

findOrEmitSection).

Vaidas Gasiunas's patch, r201259, fixed one instance where we were always
allocating sections as text. This patch fixes the remaining buggy call sites.

No test case: This isn't breaking anything that I know of, it's just
inconsistent.

<rdar://problem/15943542>

llvm-svn: 201605

llvm-svn: 201601

Summary:
This fixes several test failures when building LLVM on a MIPS host.

The failures were:
    LLVM :: DebugInfo/enum.ll
    LLVM :: DebugInfo/inlined-arguments.ll
    LLVM :: DebugInfo/member-order.ll
    LLVM :: DebugInfo/namespace.ll
    LLVM :: DebugInfo/template-recursive-void.ll
    LLVM :: DebugInfo/tu-composite.ll
    LLVM :: DebugInfo/two-cus-from-same-file.ll
    LLVM :: Linker/type-unique-simple-a.ll
    LLVM :: Linker/type-unique-simple2.ll

Reviewers: jacksprat, matheusalmeida

Reviewed By: matheusalmeida

Differential Revision: http://llvm-reviews.chandlerc.com/D2721

llvm-svn: 201582

TargetData was renamed DataLayout back in r165242.

llvm-svn: 201581

llvm-svn: 201573

llvm-svn: 201563

BuildMI instructions were not including MachineMemOperand information.
This was discovered by 'SingleSource/Benchmarks/Stanford/Oscar' failing
due to a FrameIndex load incorrectly being hoisted by postra-machine-licm.
No other tests have been found to fail.

llvm-svn: 201562

llvm-svn: 201561

Modifying build_llvm to handle SDKROOT being the name of an SDK rather than a
path. This will still work if SDKROOT is a path.

rdar://problem/15162322

llvm-svn: 201560

It's rather odd to have the flag enabling and disabling this pass only affect a
single target.

llvm-svn: 201559

llvm-svn: 201558

In gcov, the -o flag can accept either a directory or a file name.
When given a directory, the gcda and gcno files are expected to be in
that directory. When given a file, the gcda and gcno files are
expected to be named based on the stem of that file. Non-existent
paths are treated as files.

This implements compatible behaviour.

llvm-svn: 201555

llvm-svn: 201551

Add a bunch of OpSize32 tags to 64-bit mode only instructions to match their 32-bit mode counterparts for cases where there is also a OpSize16 instruction.

llvm-svn: 201550

llvm-svn: 201546

llvm-svn: 201541

Add an x86 prefix encoding for instructions that would decode to a different instruction with 0xf2/f3/66 were in front of them, but don't themselves have a prefix. For now this doesn't change any bbehavior, but plan to use it to fix some bugs in the disassembler.

llvm-svn: 201538

Introducing llvm-profdata, a tool for merging profile data generated by
PGO instrumentation in clang.

- The name indicates a file extension of <name>.profdata.  Eventually
  profile data output by clang should be changed to that extension.

- llvm-profdata merges two profiles.  However, the name is more general,
  since it will likely pick up more tasks (such as summarizing a single
  profile).

- llvm-profdata parses the current text-based format, but will be
  updated once we settle on a binary format.

<rdar://problem/15949645>

llvm-svn: 201535

    ldrd r6, r7 [r2, #15]
simply gives an error and does not triggers an assertion.

As Jim points out, the diagnostic is really strange here,
but fixing that would be more complicated. The missing
comma results in the parser expecting a construct like r2[2],
which is the vector index thing the error message is talking
about. That's not what the user intended, though, and there's
nothing else in the instruction that looks at all like a vector.
Yet more fallout from not having a real parser here and trying
to do context-free generic matching for addressing modes.

rdar://15097243

llvm-svn: 201531