Added first bit of support for the dwarf .file directive.  This patch collects
the info from the .file directive and makes file and directory tables that
will eventually be put out as part of the dwarf info in the output file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109651 91177308-0d34-0410-b5e6-96231b3b80d8

RegionInfo: Make sure to free cached nodes; Tobias, please check!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109650 91177308-0d34-0410-b5e6-96231b3b80d8

MC: Put back the MCFragment vtable, so subclasses are destroyed properly (duh).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109649 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109647 91177308-0d34-0410-b5e6-96231b3b80d8

Use a C++ compiler for the atomic builtin check since we'll
be using a C++ compiler to build.

Patch by Török Edwin!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109646 91177308-0d34-0410-b5e6-96231b3b80d8

Define a maximum supported alignment value for load, store, and
alloca instructions (constrained by their internal encoding),
and add error checking for it. Fix an instcombine bug which
generated huge alignment values (null is infinitely aligned).
This fixes undefined behavior noticed by John Regehr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109643 91177308-0d34-0410-b5e6-96231b3b80d8

Print out the regclass of any virtual registers used by a machine instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109608 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a warning from gcc-4.0 (from the ppc buildbot).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109605 91177308-0d34-0410-b5e6-96231b3b80d8

Instead of abusing swapProgramIn, just add a Module argument to
EmitProgressBitcode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109602 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak my CMake build, say you'll compile for me again...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109598 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this code to avoid decrementing an iterator past the beginning
of a std::vector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109597 91177308-0d34-0410-b5e6-96231b3b80d8

When user code intentionally dereferences null, the alignment of the
dereference is theoretically infinite. Put a cap on the computed
alignment to avoid overflow, noticed by John Regehr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109596 91177308-0d34-0410-b5e6-96231b3b80d8

Do GEP offset calculations with unsigned math rather than signed math
to avoid undefined behavior on overflow, noticed by John Regehr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109594 91177308-0d34-0410-b5e6-96231b3b80d8

Add some extra friend declarations to fix a gcc-4.0 compile error.
This is a temporary fix, until more elaborate changes are ready.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109593 91177308-0d34-0410-b5e6-96231b3b80d8

simplify

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109589 91177308-0d34-0410-b5e6-96231b3b80d8

Support: Add CrashRecoveryContext helper object.
 - Designed as a simple wrapper to allow clients to attempt to catch crashes
   (memory errors, assertion violations, etc.) and do some kind of recovery.

 - Currently doesn't actually attempt to catch crashes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109586 91177308-0d34-0410-b5e6-96231b3b80d8

simplify

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109585 91177308-0d34-0410-b5e6-96231b3b80d8

Update svn:ignore properties.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109584 91177308-0d34-0410-b5e6-96231b3b80d8

use Value* constructor of CallSite to create potentially improper site, and test that

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109581 91177308-0d34-0410-b5e6-96231b3b80d8

use Value* constructor of CallSite to create potentially improper site, and test that

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109580 91177308-0d34-0410-b5e6-96231b3b80d8

use Value* constructor of CallSite to create potentially improper site

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109579 91177308-0d34-0410-b5e6-96231b3b80d8

simplify

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109578 91177308-0d34-0410-b5e6-96231b3b80d8

simplify

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109577 91177308-0d34-0410-b5e6-96231b3b80d8

we are supposed to only create proper CallSites from an instruction (esp. CallInst and InvokeInst)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109576 91177308-0d34-0410-b5e6-96231b3b80d8

Filter out patterns that have PredicateOperands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109572 91177308-0d34-0410-b5e6-96231b3b80d8

Make SCEVCallbackVH::allUsesReplacedWith update the old SCEVUnknown
object, as it may still be referenced by SCEVs not cleaned up by the
use list traversal.

Also, in ScalarEvolution::forgetValue, only check for a SCEVUnknown
object for the original value, not for any value in the use list,
because other SCEVUnknown values aren't necessary obsolete at that
point.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109570 91177308-0d34-0410-b5e6-96231b3b80d8

Make SCEVCallbackVH::allUsesReplacedWith unconditionally delete
the old value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109567 91177308-0d34-0410-b5e6-96231b3b80d8

Implement a vectorized algorithm for <16 x i8> << <16 x i8>
This is about 4x faster and smaller than the existing scalarization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109566 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109565 91177308-0d34-0410-b5e6-96231b3b80d8

Rearrange several datastructures in LazyValueInfo to improve compile time.
This is still not perfect, but better than it was before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109563 91177308-0d34-0410-b5e6-96231b3b80d8

Fill out the interface of DenseSet a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109562 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for r109556.  Radar 8198362.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109557 91177308-0d34-0410-b5e6-96231b3b80d8

reintroduce original (asserting) semantics of CallSite(Instruction *II)
add instead a CallSite(Value* V) constructor that is consistent with ImmutableCallSize
and use that one in client code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109553 91177308-0d34-0410-b5e6-96231b3b80d8

~40% faster vector shl <4 x i32> on SSE 4.1  Larger improvements for smaller types coming in future patches.

For:

define <2 x i64> @shl(<4 x i32> %r, <4 x i32> %a) nounwind readnone ssp {
entry:
  %shl = shl <4 x i32> %r, %a                     ; <<4 x i32>> [#uses=1]
  %tmp2 = bitcast <4 x i32> %shl to <2 x i64>     ; <<2 x i64>> [#uses=1]
  ret <2 x i64> %tmp2
}

We get:

_shl:                                   ## @shl
pslld $23, %xmm1
paddd LCPI0_0, %xmm1
cvttps2dq %xmm1, %xmm1
pmulld %xmm1, %xmm0
ret

Instead of:

_shl:                                   ## @shl
pshufd $3, %xmm0, %xmm2
movd %xmm2, %eax
pshufd $3, %xmm1, %xmm2
movd %xmm2, %ecx
shll %cl, %eax
movd %eax, %xmm2
pshufd $1, %xmm0, %xmm3
movd %xmm3, %eax
pshufd $1, %xmm1, %xmm3
movd %xmm3, %ecx
shll %cl, %eax
movd %eax, %xmm3
punpckldq %xmm2, %xmm3
movd %xmm0, %eax
movd %xmm1, %ecx
shll %cl, %eax
movd %eax, %xmm2
movhlps %xmm0, %xmm0
movd %xmm0, %eax
movhlps %xmm1, %xmm1
movd %xmm1, %ecx
shll %cl, %eax
movd %eax, %xmm0
punpckldq %xmm0, %xmm2
movdqa %xmm2, %xmm0
punpckldq %xmm3, %xmm0
ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109549 91177308-0d34-0410-b5e6-96231b3b80d8

recommit simplification (originally r109504, backed out in r109508) now that problem in CallSiteBase is fixed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109547 91177308-0d34-0410-b5e6-96231b3b80d8

remove bogus assert, use static_cast for additional checking

left two new asserts commented out, because they would fire in clang, have to hunt those down first

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109544 91177308-0d34-0410-b5e6-96231b3b80d8

It is FE's responsibility to emit proper directory name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109538 91177308-0d34-0410-b5e6-96231b3b80d8

make lookup failures not fatal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109530 91177308-0d34-0410-b5e6-96231b3b80d8

Grammar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109525 91177308-0d34-0410-b5e6-96231b3b80d8

Update tests to not rely on input file's absolute path.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109521 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a crash in the dag combiner caused by ConstantFoldBIT_CONVERTofBUILD_VECTOR calling itself
recursively and returning a SCALAR_TO_VECTOR node, but assuming the input was always a BUILD_VECTOR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109519 91177308-0d34-0410-b5e6-96231b3b80d8

80 column

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109513 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109511 91177308-0d34-0410-b5e6-96231b3b80d8

recommit simplification (r109502, backed out r109509); seems to innocent

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109510 91177308-0d34-0410-b5e6-96231b3b80d8

back out this too to restore the bots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109509 91177308-0d34-0410-b5e6-96231b3b80d8

back out r109504, breaks the bots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109508 91177308-0d34-0410-b5e6-96231b3b80d8

simplify: CallSite::get --> CallSite constructor

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109506 91177308-0d34-0410-b5e6-96231b3b80d8

simplify

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109504 91177308-0d34-0410-b5e6-96231b3b80d8

use ImmutableCallSite for const-corrgoodness

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109503 91177308-0d34-0410-b5e6-96231b3b80d8

simplify

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109502 91177308-0d34-0410-b5e6-96231b3b80d8

Fix silent failure with no input files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109500 91177308-0d34-0410-b5e6-96231b3b80d8

Return -1 only on failure to execute a program.

Also fix some comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109499 91177308-0d34-0410-b5e6-96231b3b80d8

In commit 91421, isPod was changed from false to true for these value handles.
Change it back again: destructors and constructors need to be run.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109498 91177308-0d34-0410-b5e6-96231b3b80d8

Make coff-dump.py executable and add python as executable for this script.

This fixes the MC/COFF/basic-coff.ll test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109497 91177308-0d34-0410-b5e6-96231b3b80d8

RegionInfo: Add getMaxRegionExit()

getMaxRegionExit returns the exit of the maximal refined region starting
at a specific basic block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109496 91177308-0d34-0410-b5e6-96231b3b80d8

After updating value handles for RAUW, check that no weak or tracking handles
are still on the list.  This might happen if a CallbackVH created some new value
handles for the old value when doing RAUW.  Barf if it occurs, since it is almost
certainly a mistake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109495 91177308-0d34-0410-b5e6-96231b3b80d8

Make MC use Windows COFF on Windows and add tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109494 91177308-0d34-0410-b5e6-96231b3b80d8

Add function to query RegionInfo about loops.

* contains(Loop), * getOutermostLoop()
* Improve getNameStr() to return a sensible name, if basic blocks are not named.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109490 91177308-0d34-0410-b5e6-96231b3b80d8

The isLoadFromStackSlot and isStoreToStackSlot have no way of reporting
subregister operands like this:

%reg1040:sub_32bit<def> = MOV32rm <fi#-2>, 1, %reg0, 0, %reg0, %reg1040<imp-def>; mem:LD4[FixedStack-2](align=8)

Make them return false when subreg operands are present. VirtRegRewriter is
making bad assumptions otherwise.

This fixes PR7713.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109489 91177308-0d34-0410-b5e6-96231b3b80d8

Add assertions that expose the PR7713 miscompilation: Accessing a stack slot
with a too-big register class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109488 91177308-0d34-0410-b5e6-96231b3b80d8

It's better to have the arrays, which would trigger the creation of stack
protectors, to be near the stack protectors on the stack. Accomplish this by
tagging the stack object with a predicate that indicates that it would trigger
this. In the prolog-epilog inserter, assign these objects to the stack after the
stack protector but before the other objects.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109481 91177308-0d34-0410-b5e6-96231b3b80d8

Use AssertingVH for InsertedValues and InsertedPostIncValues, to verify
that the values they refer to aren't being deleted underneath them.

Make sure these containters get cleared by clear(), which IndVarSimplify
and LSR both use before deleting instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109478 91177308-0d34-0410-b5e6-96231b3b80d8

(Hopefully) One last test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109473 91177308-0d34-0410-b5e6-96231b3b80d8

Another test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109472 91177308-0d34-0410-b5e6-96231b3b80d8

Another test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109471 91177308-0d34-0410-b5e6-96231b3b80d8

Add SplitEditor to SplitKit. This class will be used to edit live intervals and
rewrite instructions for live range splitting.

Still work in progress.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109469 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a use-after-free.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109468 91177308-0d34-0410-b5e6-96231b3b80d8

Using llvm.eh.catch.all.value instead of .llvm.eh.catch.all.value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109462 91177308-0d34-0410-b5e6-96231b3b80d8

And a bit more non-ASCII stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109458 91177308-0d34-0410-b5e6-96231b3b80d8

Drop some non-ascii stuff

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109456 91177308-0d34-0410-b5e6-96231b3b80d8

Don't call __register_frame from the JIT on systems that use setjmp/longjmp
exception handling.  Also fix an extra underscore typo in one instance of
"__ARM_EABI__".  Radar 8236264.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109451 91177308-0d34-0410-b5e6-96231b3b80d8

On x86, f32 / f64 nodes share the same registers as 128-bit vector values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109450 91177308-0d34-0410-b5e6-96231b3b80d8

The "excess register pressure" returned by HighRegPressure() is not accurate enough to factor into scheduling priority. Eliminate it and add early exits to speed up scheduling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109449 91177308-0d34-0410-b5e6-96231b3b80d8

Add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109448 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109435 91177308-0d34-0410-b5e6-96231b3b80d8

Temporary hack to let codegen assert or generate poor code in case
we are using AVX and no AVX version of the desired intruction is present,
this is better for incremental dev (without fallbacks it's easier to spot
what's missing). Not sure this is the best hack thought (we can also disable
all HasSSE* predicates by dinamically marking them 'false' if AVX is present)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109434 91177308-0d34-0410-b5e6-96231b3b80d8

Testing some more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109433 91177308-0d34-0410-b5e6-96231b3b80d8

Test some more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109431 91177308-0d34-0410-b5e6-96231b3b80d8

Fix library build messages.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109430 91177308-0d34-0410-b5e6-96231b3b80d8

Add svn:ignore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109429 91177308-0d34-0410-b5e6-96231b3b80d8

re-commit r109220, the compile error has already been fixed

Simplifying use_iterators by dereferencing
is not a good idea. The codebase does not depend
in this any more, and it may introduce hidden
runtime cost. If you get compile errors, please
dereference your iterator before passing to cast<>
(and friends).

Also: please consider caching the result of
operator* and reusing that instead of dereferencing
many times.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109425 91177308-0d34-0410-b5e6-96231b3b80d8

Add an initial implementation of LazyValueInfo updating for JumpThreading.  Disabled for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109424 91177308-0d34-0410-b5e6-96231b3b80d8

Currently EH lowering code expects typeinfo to be global only.
This assumption is not satisfied due to global mergeing.
Workaround the issue by temporary disablinge mergeing of const globals.
Also, ignore LLVM "special" globals. This fixes PR7716

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109423 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a test with malformed IR.  Not sure why this didn't fail before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109422 91177308-0d34-0410-b5e6-96231b3b80d8

ARM fastisel isn't ready.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109421 91177308-0d34-0410-b5e6-96231b3b80d8

One more test commit...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109420 91177308-0d34-0410-b5e6-96231b3b80d8

Fix SCEVExpander::visitAddRecExpr so that it remembers the induction variable
it inserted rather than using LoopInfo::getCanonicalInductionVariable to
rediscover it, since that doesn't work on non-canonical loops. This fixes
infinite recurrsion on such loops; PR7562.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109419 91177308-0d34-0410-b5e6-96231b3b80d8

Handle Values with no value in getCopyFromRegs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109415 91177308-0d34-0410-b5e6-96231b3b80d8

Hopefully the last test commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109414 91177308-0d34-0410-b5e6-96231b3b80d8

Remove LCSSA's bogus dependence on LoopSimplify and LoopSimplify's bogus
dependence on DominanceFrontier. Instead, add an explicit DominanceFrontier
pass in StandardPasses.h to ensure that it gets scheduled at the right
time.

Declare that loop unrolling preserves ScalarEvolution, and shuffle some
getAnalysisUsages.

This eliminates one LoopSimplify and one LCCSA run in the standard
compile opts sequence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109413 91177308-0d34-0410-b5e6-96231b3b80d8

Preserve ScalarEvolution in the loop unroller.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109412 91177308-0d34-0410-b5e6-96231b3b80d8

Another test commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109411 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid depending on LCSSA implicitly pulling in LoopSimplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109410 91177308-0d34-0410-b5e6-96231b3b80d8

Test commit for new post-commit hooks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109409 91177308-0d34-0410-b5e6-96231b3b80d8

Fix (at least) quadratic worst-case complexity in DominanceFrontier::splitBlock:
don't visit all blocks in the function, and don't iterate over the split blocks'
predecessor lists for each block visited.

Also, remove the special-case test for the entry block. Splitting the entry
block isn't common enough to make this worthwhile.

This fixes a major compile-time bottleneck which is exposed now that
LoopSimplify isn't being redundantly run both before and after
DominanceFrontier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109408 91177308-0d34-0410-b5e6-96231b3b80d8

80-col.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109407 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid copying and recopying a std::set.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109405 91177308-0d34-0410-b5e6-96231b3b80d8

Exit a search loop when the search goal is found.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109404 91177308-0d34-0410-b5e6-96231b3b80d8

Fix formatting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109403 91177308-0d34-0410-b5e6-96231b3b80d8

A block dominates itself, by definition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109402 91177308-0d34-0410-b5e6-96231b3b80d8