Marking remote mcjit tests as XFAIL for cygwin (hopefully only temporarily).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168226 91177308-0d34-0410-b5e6-96231b3b80d8

Claim ownership of MCJIT

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168225 91177308-0d34-0410-b5e6-96231b3b80d8

Add some basic support for the fission addr forms to DebugInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168223 91177308-0d34-0410-b5e6-96231b3b80d8

Marking remote mcjit tests as XFAIL for mingw32 (hopefully only temporarily).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168221 91177308-0d34-0410-b5e6-96231b3b80d8

Add constant definitions for fission dwarf attributes, forms, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168218 91177308-0d34-0410-b5e6-96231b3b80d8

Claim the driver.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168215 91177308-0d34-0410-b5e6-96231b3b80d8

Remove default public copy ctors.

They are just useless and prevent SmallVector from picking an optimized codepath
for memcpyable elements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168211 91177308-0d34-0410-b5e6-96231b3b80d8

Marking remote mcjit tests as XFAIL for ARM (hopefully only temporarily).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168210 91177308-0d34-0410-b5e6-96231b3b80d8

Remove trailing spaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168208 91177308-0d34-0410-b5e6-96231b3b80d8

Remove hard coded registers in ARM ldrexd and strexd instructions

This patch replaces the hard coded GPR pair [R0, R1] of
Intrinsic:arm_ldrexd and [R2, R3] of Intrinsic:arm_strexd with
even/odd GPRPair reg class.
Similar to the lowering of atomic_64 operation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168207 91177308-0d34-0410-b5e6-96231b3b80d8

Use array_pod_sort instead of std::sort.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168203 91177308-0d34-0410-b5e6-96231b3b80d8

typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168202 91177308-0d34-0410-b5e6-96231b3b80d8

Add myself as owner of XCore Backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168201 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure FABS on v2f32 and v4f32 is legal on ARM NEON
This fixes PR14359

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168200 91177308-0d34-0410-b5e6-96231b3b80d8

Fix handling of aliases to functions.

An alias to a function should use pc relative addressing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168199 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Order global variables in def-use order before emiting them in the final assembly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168198 91177308-0d34-0410-b5e6-96231b3b80d8

Preserve address space of forward-referenced global variables in the LL parser

Before, the parser would assert on the following code:

@a2 = global i8 addrspace(1)* @a
@a = addrspace(1) global i8 0

because the type of @a was "i8*" instead of "i8 addrspace(1)*" when parsing
the initializer for @a2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168197 91177308-0d34-0410-b5e6-96231b3b80d8

Make this easier to understand, as suggested by Chandler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168196 91177308-0d34-0410-b5e6-96231b3b80d8

Added program header emission

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168195 91177308-0d34-0410-b5e6-96231b3b80d8

Remove conditions from 'else if' that were guaranteed by preceding 'if'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168191 91177308-0d34-0410-b5e6-96231b3b80d8

ReleaseNotes.html updates reflecting PPC64 Elf Linux efforts and progress.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168189 91177308-0d34-0410-b5e6-96231b3b80d8

InstructionSimplify should be able to simplify A+B==B+A to 'true'
but wasn't due to the same logic bug that caused PR14361.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168186 91177308-0d34-0410-b5e6-96231b3b80d8

Using const cast to alleviate a warning.

A PR is being filed to address some code issues here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168185 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out the final FADD that's common to multiple code paths in the visitLog* functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168183 91177308-0d34-0410-b5e6-96231b3b80d8

move irrelevant attribution.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168182 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR14361: wrong simplification of A+B==B+A.  You may think that the old logic
replaced by this patch is equivalent to the new logic, but you'd be wrong, and
that's exactly where the bug was.  There's a similar bug in instsimplify which
manifests itself as instsimplify failing to simplify this, rather than doing it
wrong, see next commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168181 91177308-0d34-0410-b5e6-96231b3b80d8

Adding new tests to test lli's pseudo-remote feature (-remote-mcjit).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168180 91177308-0d34-0410-b5e6-96231b3b80d8

Finally add myself to the credits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168178 91177308-0d34-0410-b5e6-96231b3b80d8

SimplifyCFG: Don't assume non-null ScalarTargetTransformInfo.

Patch by Pekka Jääskeläinen!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168176 91177308-0d34-0410-b5e6-96231b3b80d8

Work around a layering violation from Target to CodeGen.

Technically this is still a layering violation but it's header-only which makes
it less harmful. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168173 91177308-0d34-0410-b5e6-96231b3b80d8

llvm/test/CodeGen/X86/hipe-cc*.ll: Add explicit -mcpu, or they don't expect to pass on Atom.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168171 91177308-0d34-0410-b5e6-96231b3b80d8

Add the Erlang/HiPE calling convention, patch by Yiannis Tsiouris.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168166 91177308-0d34-0410-b5e6-96231b3b80d8

Update credits file regarding SROA work... Also include my other primary
email address for completeness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168165 91177308-0d34-0410-b5e6-96231b3b80d8

Add myself as the code owner for a bunch of stuff.

Notably, I'm reviewing this stuff already, and this makes it clear I'm
on the hook for it.

I'm relying on Eric Christopher and Michael Spencer to help with Support
when there are platform specific issues, but Chris wants a single point
of contact, and I'm happy with that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168164 91177308-0d34-0410-b5e6-96231b3b80d8

Add MCJIT test case for running global constructors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168149 91177308-0d34-0410-b5e6-96231b3b80d8

Add R_ARM_TARGET1 relocation to MCJIT and ensure that the runtime dyld reads
all symbols during object loading, not just global ones.

This fixes JIT execution of code using llvm.global_ctors with internal
linkage constructors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168148 91177308-0d34-0410-b5e6-96231b3b80d8

Constant::IsThreadDependent(): Use dyn_cast<Constant> instead of cast

It turns out that the operands of a Constant are not always themselves
Constant. For example, one of the operands of BlockAddress is
BasicBlock, which is not a Constant.

This should fix the dragonegg-x86_64-linux-gcc-4.6-test build which
broke in r168037.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168147 91177308-0d34-0410-b5e6-96231b3b80d8

[DebugInfo] Generate address ranges for compile units even if .debug_aranges is present: it is often the case that .debug_aranges section contains ranges only for a small subset of compile units. Test cases will be added in separate commits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168144 91177308-0d34-0410-b5e6-96231b3b80d8

Factor some common code to reduce compile size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168143 91177308-0d34-0410-b5e6-96231b3b80d8

LoopVectorize: Division reductions generate incorrect code. Remove the part of the code that deals with divs.
Thanks to Paul Redmond for catching this while reviewing the code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168142 91177308-0d34-0410-b5e6-96231b3b80d8

Use roundps/pd for llvm.ceil, llvm.trunc, llvm.rint, and llvm.nearbyint of vector types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168141 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168138 91177308-0d34-0410-b5e6-96231b3b80d8

[mips] Fix delay slot filler so that instructions with register operand $1 are
allowed in branch delay slot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168131 91177308-0d34-0410-b5e6-96231b3b80d8

Remove trailing whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168121 91177308-0d34-0410-b5e6-96231b3b80d8

Interface changes to allow RuntimeDyld memory managers to set memory permissions after an object has been loaded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168114 91177308-0d34-0410-b5e6-96231b3b80d8

Add some tests for the FileCheck utility.

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20121112/156007.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168113 91177308-0d34-0410-b5e6-96231b3b80d8

Added myself as owner of LLVM Buildbot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168111 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168110 91177308-0d34-0410-b5e6-96231b3b80d8

Claim ownership of everyone's favorite part of the backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168109 91177308-0d34-0410-b5e6-96231b3b80d8

Mark FP_ROUND for converting NEON v2f64 to v2f32 as expand. Add a missing
case to vector legalization so this actually works.

Patch by Pete Couperus.  Fixes PR12540.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168107 91177308-0d34-0410-b5e6-96231b3b80d8

More ownership, no one likes these things anyways.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168104 91177308-0d34-0410-b5e6-96231b3b80d8

Remove trailing whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168103 91177308-0d34-0410-b5e6-96231b3b80d8

Update the code ownership.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168102 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168091 91177308-0d34-0410-b5e6-96231b3b80d8

[mips] Add predicate HasFPIdx for floating-point indexed load instruction
support and use it in place of HasMips32r2Or64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168089 91177308-0d34-0410-b5e6-96231b3b80d8

Add description of how to build docs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168088 91177308-0d34-0410-b5e6-96231b3b80d8

PowerPC: Lowering floor intrinsic for Altivec

This patch lowers the llvm.floor, llvm.ceil, llvm.trunc, and
llvm.nearbyint to Altivec instruction when using 4 single-precision
float vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168086 91177308-0d34-0410-b5e6-96231b3b80d8

PathV2: Fix a possible infinite loop.

 - The code could infinite loop trying to create unique files, if the directory
   containing the unique file exists, but open() calls on non-existent files in
   the path return ENOENT. This is true on the /dev/fd filesystem, for example.

 - Will add a clang side test case for this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168081 91177308-0d34-0410-b5e6-96231b3b80d8

Add assertions in MipsLongBranch which check the size of basic blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168078 91177308-0d34-0410-b5e6-96231b3b80d8

Return 0 instead of false.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168076 91177308-0d34-0410-b5e6-96231b3b80d8

Use std::stable_sort instead of std::sort when sorting stack slots
to guarantee deterministic code generation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168074 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify my code ownership

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168073 91177308-0d34-0410-b5e6-96231b3b80d8

Add myself as owner of VLIW Instruction Scheduling and Packetization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168072 91177308-0d34-0410-b5e6-96231b3b80d8

Add ARM EABI to my list of responsibilities.
Update the email address (old works too, but less reliable) while there

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168065 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168064 91177308-0d34-0410-b5e6-96231b3b80d8

misspell

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168058 91177308-0d34-0410-b5e6-96231b3b80d8

whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168057 91177308-0d34-0410-b5e6-96231b3b80d8

[reg scavenger] Fix the isUsed/isAliasUsed functions so as to not report a false
positive.

In this particular case, R6 was being spilled by the register scavenger when it
was in fact dead.  The isUsed function reported R6 as used because the R6_R7
alias was reserved (due to the fact that we've reserved R7 as the FP).  The
solution is to only check if the original register (i.e., R6) isReserved and
not the aliases.  The aliases are only checked to make sure they're available.

The test case is derived from one of the nightly tester benchmarks and is rather
intractable and difficult to reproduce, so I haven't included it.
rdar://12592448

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168054 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indeterminism in MI scheduler DAG construction.
Similarly to several recent fixes throughout the code replace std::map use with the MapVector.
Add find() method to the MapVector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168051 91177308-0d34-0410-b5e6-96231b3b80d8

Use empty parens for empty function parameter list instead of '(void)'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168049 91177308-0d34-0410-b5e6-96231b3b80d8

FileCheck: remove useless 'continue' at the end of a 'while(){}' loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168048 91177308-0d34-0410-b5e6-96231b3b80d8

Do not handle void types in DataLayout.  Patch by Patrick Hägglund.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168042 91177308-0d34-0410-b5e6-96231b3b80d8

Make GlobalOpt be conservative with TLS variables (PR14309)

For global variables that get the same value stored into them
everywhere, GlobalOpt will replace them with a constant. The problem is
that a thread-local GlobalVariable looks like one value (the address of
the TLS var), but is different between threads.

This patch introduces Constant::isThreadDependent() which returns true
for thread-local variables and constants which depend on them (e.g. a GEP
into a thread-local array), and teaches GlobalOpt not to track such
values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168037 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for SPIR64 target - the 64bit counterpart of SPIR.

The new OpenCL SPIR extension spec will define separate SPIR for 32 and 64 bit architectures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168036 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a crash observed by Shuxin Yang.  The issue here is that LinearizeExprTree,
the utility for extracting a chain of operations from the IR, thought that it
might as well combine any constants it came across (rather than just returning
them along with everything else).  On the other hand, the factorization code
would like to see the individual constants (this is quite reasonable: it is
much easier to pull a factor of 3 out of 2*3 than it is to pull it out of 6;
you may think 6/3 isn't so hard, but due to overflow it's not as easy to undo
multiplications of constants as it may at first appear).  This patch therefore
makes LinearizeExprTree stupider: it now leaves optimizing to the optimization
part of reassociate, and sticks to just analysing the IR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168035 91177308-0d34-0410-b5e6-96231b3b80d8

Revert changing FNEG of v4f32 to Expand. It's legal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168030 91177308-0d34-0410-b5e6-96231b3b80d8

Make FNEG and FABS of v4f32 Expand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168029 91177308-0d34-0410-b5e6-96231b3b80d8

Make a bunch of floating point operations on vectors Expand so that instruction selection won't fail.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168028 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing documentation for llvm.exp2, llvm.log10, and llvm.log2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168026 91177308-0d34-0410-b5e6-96231b3b80d8

Add llvm.ceil, llvm.trunc, llvm.rint, llvm.nearbyint intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168025 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an obvious merge bug in -join-globalcopies (disabled).

Jakub Staszak spotted this in review. I don't notice these things
until I manually rerun benchmarks. But reducing unit tests is a very
high priority.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168021 91177308-0d34-0410-b5e6-96231b3b80d8

Add some release notes that say we removed the CellSPU port.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168014 91177308-0d34-0410-b5e6-96231b3b80d8

InstCombineAndOrXor.cpp: Escape bracket in doxygen description. [-Wdocumentation]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168013 91177308-0d34-0410-b5e6-96231b3b80d8

Use LLVM_DELETED_FUNCTION.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168009 91177308-0d34-0410-b5e6-96231b3b80d8

Add doInitialization and doFinalization methods to ModulePass's, to allow them to be re-initialized and reused on multiple Module's.

Patch by Pedro Artigas.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168008 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unneeded #includes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168006 91177308-0d34-0410-b5e6-96231b3b80d8

NVPTXISelLowering.cpp: Fix warnings. [-Wunused-variable]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168001 91177308-0d34-0410-b5e6-96231b3b80d8

This patch is in preparation for adding medium code model support to the
PPC64 target.  The five tests modified herein test code generation that is
sensitive to the code model selected.  So I've added -code-model=small to
the RUN commands for each.

Since small code model is the default, this has no effect for now; but this
prepares us for eventually changing the default to medium code model for PPC64.

Test changes verified with small and medium code model as default on
powerpc64-unknown-linux-gnu.  All tests continue to pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167999 91177308-0d34-0410-b5e6-96231b3b80d8

docs: Improve typographical correctness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167998 91177308-0d34-0410-b5e6-96231b3b80d8

docs: Sphinxify TestSuiteMakefileGuide

Some small related fixups to TestingGuide too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167996 91177308-0d34-0410-b5e6-96231b3b80d8

Use reserve() to avoid vector reallocation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167991 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure to not get AVX code on an AVX-capable host. Revealed in r167967.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167989 91177308-0d34-0410-b5e6-96231b3b80d8

test/CodeGen/Hexagon/postinc-load.ll: Suppress it for now. It triggered the failure on i686 hosts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167988 91177308-0d34-0410-b5e6-96231b3b80d8

Few more small CellSPU removals.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167987 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CMake build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167986 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the CellSPU port.

Approved by Chris Lattner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167984 91177308-0d34-0410-b5e6-96231b3b80d8

docs: Sphinxify TestingGuide

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167979 91177308-0d34-0410-b5e6-96231b3b80d8

Support for [[@LINE]], [[@LINE+<offset>]], [[@LINE-<offset>]] expressions in
FileCheck.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167978 91177308-0d34-0410-b5e6-96231b3b80d8

Fix invalid asserts, use llvm_unreachable instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167976 91177308-0d34-0410-b5e6-96231b3b80d8

llvm/test/CodeGen/X86/memset.ll: FileCheck-ize, and add another case on +avx.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167975 91177308-0d34-0410-b5e6-96231b3b80d8