Move getUniqueExitBlocks from LoopBase to Loop, since they depend on
LoopSimplify form, which is currently only available on Loops (and
not MachineLoops). Also, move the code out of the header file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80923 91177308-0d34-0410-b5e6-96231b3b80d8

Use IRReader.h in opt, to support reading of LLVM Assembly files directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80922 91177308-0d34-0410-b5e6-96231b3b80d8

Change PHINode::hasConstantValue to have a DominatorTree argument
instead of a bool argument, and to do the dominator check itself.
This makes it eaiser to use when DominatorTree information is
available.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80920 91177308-0d34-0410-b5e6-96231b3b80d8

Don't try to verify a LoopPass analysis if the loop has been deleted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80919 91177308-0d34-0410-b5e6-96231b3b80d8

Remove references to expression "handles", which are no longer used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80918 91177308-0d34-0410-b5e6-96231b3b80d8

CppBackend: avoid printing unnecessary whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80917 91177308-0d34-0410-b5e6-96231b3b80d8

Keep track of how many memmove calls were turned into
memcpy calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80915 91177308-0d34-0410-b5e6-96231b3b80d8

Fix build warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80912 91177308-0d34-0410-b5e6-96231b3b80d8

Code Cleanup.

Removed inverted flag form MaximumSpanningTree, also do not handle so much
information to MaximumSpanningTree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80911 91177308-0d34-0410-b5e6-96231b3b80d8

Filter out -fno-rtti from CXXFLAGS as well (in an expensive checks build).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80910 91177308-0d34-0410-b5e6-96231b3b80d8

Code Cleanup.
(See http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20090831/086139.html)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80909 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply profiling tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80908 91177308-0d34-0410-b5e6-96231b3b80d8

Remove undefined behavior when loading optimal edge profile info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80907 91177308-0d34-0410-b5e6-96231b3b80d8

don't call getOffset() on jump tables, this fixes three failing olden benchmarks
with the new asmprinter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80906 91177308-0d34-0410-b5e6-96231b3b80d8

Implement support for X86II::MO_GOT_ABSOLUTE_ADDRESS.  We get very
different formatting from the old asmprinter, but it should be
semantically the same.  We used to get:

popl %eax
addl $_GLOBAL_OFFSET_TABLE_ + [.-.Lllvm$6.$piclabel], %eax
...

Now we get:

popl %eax
.Lpicbaseref6:
addl $(_GLOBAL_OFFSET_TABLE_ + (.Lpicbaseref6 - .Lllvm$6.$piclabel)), %eax
...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80905 91177308-0d34-0410-b5e6-96231b3b80d8

Reference to hidden symbols do not have to go through non-lazy pointer in non-pic mode. rdar://7187172.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80904 91177308-0d34-0410-b5e6-96231b3b80d8

Remove VISIBILITY_HIDDEN from this file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80903 91177308-0d34-0410-b5e6-96231b3b80d8

merge all the basic linux/32 pic tests together into one test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80902 91177308-0d34-0410-b5e6-96231b3b80d8

rename test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80901 91177308-0d34-0410-b5e6-96231b3b80d8

use a darwin triple

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80900 91177308-0d34-0410-b5e6-96231b3b80d8

TAI -> MAI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80899 91177308-0d34-0410-b5e6-96231b3b80d8

adjust expected lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80898 91177308-0d34-0410-b5e6-96231b3b80d8

improve comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80897 91177308-0d34-0410-b5e6-96231b3b80d8

fix MCSymbol printing to exactly match the normal mangler rules so
we can diff .s files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80894 91177308-0d34-0410-b5e6-96231b3b80d8

remove extraneous hack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80893 91177308-0d34-0410-b5e6-96231b3b80d8

Make these functions static and local.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80892 91177308-0d34-0410-b5e6-96231b3b80d8

Tweak comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80891 91177308-0d34-0410-b5e6-96231b3b80d8

Thread an MCAsmInfo pointer through the various MC printing APIs,
and fix a few things using << on MCSymbols to use ->print(). No
functionality change other than unbreaking my previous patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80890 91177308-0d34-0410-b5e6-96231b3b80d8

just use dump()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80889 91177308-0d34-0410-b5e6-96231b3b80d8

inline insertion operators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80888 91177308-0d34-0410-b5e6-96231b3b80d8

In C++, code is not allowed to call main.  In C it is, this
simplifylibcalls optimization is thus valid for C++ but not C.
It's not important enough to worry about for C++ apps, so just
remove it.

rdar://7191924

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80887 91177308-0d34-0410-b5e6-96231b3b80d8

merge globaladdress symbol processing stuff into other stuff.  Now
all global variable operand flag processing stuff is shared between
different operand types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80886 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak x86_64 build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80885 91177308-0d34-0410-b5e6-96231b3b80d8

Split the "operand -> symbol" logic from the "get offset and other munging
from operand" logic.  GlobalAddress still todo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80884 91177308-0d34-0410-b5e6-96231b3b80d8

implement lowering support for constant pool index operands, this gets a bunch more
olden programs working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80881 91177308-0d34-0410-b5e6-96231b3b80d8

X86JITInfo::getLazyResolverFunction() should not read cpu id to determine whether sse is available. Just use consult subtarget.

No functionality changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80880 91177308-0d34-0410-b5e6-96231b3b80d8

Set the 'cached digest' flag after computing the digest for an
ImutAVLTree.  This was accidentally left out, and essentially caused
digest caching to be ignored in ImmutableMap and ImmutableSet (this
bug was detected from shark traces that showed ComputeDigest was in
the hot path in the clang static analyzer).

This reduces the running time of the clang static analyzer on an
example benchmark by ~32% for both RegionStore (field-sensitivty) and
BasicStore (without field-sensitivity).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80877 91177308-0d34-0410-b5e6-96231b3b80d8

update test for alignment value in hex

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80876 91177308-0d34-0410-b5e6-96231b3b80d8

output alignment value in hex so that we get:
  .align 3, 0x90
instead of,
  .align 3, 144

suggested by eric.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80875 91177308-0d34-0410-b5e6-96231b3b80d8

simplify this by using SmallString::str(), much nicer!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80874 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed a test that ensures the LocalRewriter does not attempt to
avoid reloads by reusing clobbered registers.

This was causing issues in 256.bzip2 when compiled with PIC for
a while (starting at r78217), though the problem has since been masked.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80872 91177308-0d34-0410-b5e6-96231b3b80d8

back out my recent commit (r80858), it seems to break self-hosting buildbot's stage 2 configure

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80871 91177308-0d34-0410-b5e6-96231b3b80d8

Now Bitcode reader bug is fixed. Reapply 80839.

Use CallbackVH, instead of WeakVH, to hold MDNode elements.
Use FoldingSetNode to unique MDNodes in a context.
Use CallbackVH hooks to update context's MDNodeSet appropriately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80868 91177308-0d34-0410-b5e6-96231b3b80d8

Add new value for given index in MDValuePtrs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80867 91177308-0d34-0410-b5e6-96231b3b80d8

Improve llvm::getHostTriple for some cases where the LLVM_HOSTTRIPLE is not
reliable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80863 91177308-0d34-0410-b5e6-96231b3b80d8

re-commit r66920 (which has been backed out in r66953) I may have more luck this time. I'll back out if needed...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80858 91177308-0d34-0410-b5e6-96231b3b80d8

Added opaque 32-, 48-, and 80-bit memory operand types to the X86
instruction tables to support segmented addressing (and other objects
of obscure type).
Modified the X86 assembly printers to handle these new operand types.
Added JMP and CALL instructions that use segmented addresses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80857 91177308-0d34-0410-b5e6-96231b3b80d8

Show derived host triple in --version.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80855 91177308-0d34-0410-b5e6-96231b3b80d8

Improve support for cross-hosted builds of LLVM.
--build=triple and other configure options are passed
to the BuildTools/ sub-invocation more consistently

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80854 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak my CMake build. Say you'll link again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80842 91177308-0d34-0410-b5e6-96231b3b80d8

Revert 80839 for now. It causes test failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80841 91177308-0d34-0410-b5e6-96231b3b80d8

Use CallbackVH, instead of WeakVH, to hold MDNode elements.
Use FoldingSetNode to unique MDNodes in a context.
Use CallbackVH hooks to update context's MDNodeSet appropriately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80839 91177308-0d34-0410-b5e6-96231b3b80d8

More missed vdup patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80838 91177308-0d34-0410-b5e6-96231b3b80d8

Switch llc from ParseBitcodeFile to ParseIRFile. This lets llc
transparently read either LLVM Assembly or LLVM Bitcode files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80829 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment noting the memory ownership rules.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80827 91177308-0d34-0410-b5e6-96231b3b80d8

Add convenience functions for reading in LLVM IR that autodetect
and LLVM Assembly and LLVM Bitcode and automatically call the
corresponding reader.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80809 91177308-0d34-0410-b5e6-96231b3b80d8

switch from std::string to SmallString + raw_svector_ostream.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80807 91177308-0d34-0410-b5e6-96231b3b80d8

split mcinst lowering stuff out to its own file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80806 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the syntax of add/sub/mul nsw/nuw and sdiv exact.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80805 91177308-0d34-0410-b5e6-96231b3b80d8

Add const qualifiers for isBitcodeWrapper, and add new functions
isRawBitcode and isBitcode to allow clients to test whether a given
memory buffer holds a bitcode image.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80804 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor common code from ParseAssemblyString and ParseAssemblyFile,
to expose a low-level interface for parsing from an existing MemoryBuffer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80803 91177308-0d34-0410-b5e6-96231b3b80d8

Add const qualifiers to dominates' arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80801 91177308-0d34-0410-b5e6-96231b3b80d8

Removed temporarily because of breaking Darwin builds.
(See http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20090831/086214.html)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80799 91177308-0d34-0410-b5e6-96231b3b80d8

Rearrange code to eliminate redundancy and avoid gotos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80798 91177308-0d34-0410-b5e6-96231b3b80d8

plug another leak in LLParser::PerFunctionState::SetInstName()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80792 91177308-0d34-0410-b5e6-96231b3b80d8

Changed profiling-tool-chain.ll test to use optimal-edge-profiling instead of
edge-profiling, this is more useful since the loading of the
optimal-edge-profiling is more complicated.
The edge-profiling is tested in edge-profiling.ll where only the
instrumentation is tested.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80791 91177308-0d34-0410-b5e6-96231b3b80d8

plug memory leak in LLParser::PerFunctionState::SetInstName() by deleting a value after replacing it

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80790 91177308-0d34-0410-b5e6-96231b3b80d8

Sort edges in MaximumSpanningTree more stable in case of equal weight.
(See http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20090824/085890.html)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80789 91177308-0d34-0410-b5e6-96231b3b80d8

Changed set of BlocksToInstrument to set of InsertedBlocks that do not have to
be instrumented.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80788 91177308-0d34-0410-b5e6-96231b3b80d8

Code cleanups and added comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80781 91177308-0d34-0410-b5e6-96231b3b80d8

Opaque types didn't work if llvm_is_multithreaded().
AlwaysOpaqueTy is always NULL at this point, and it causes an assertion failure.
Fix it by using the just constructed tmp instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80780 91177308-0d34-0410-b5e6-96231b3b80d8

set svn:ignore

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80779 91177308-0d34-0410-b5e6-96231b3b80d8

Fix DbgStopPointInst->getFileName/getDirectory, broken by the MDNodification in
r80406, and readd a -print-dbginfo test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80778 91177308-0d34-0410-b5e6-96231b3b80d8

Retype from unsigned to CallingConv::ID accordingly. Approved by Bob Wilson.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80773 91177308-0d34-0410-b5e6-96231b3b80d8

Fix month.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80769 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR4815: some cases where DeleteDeadInstruction can delete
the instruction BBI points to.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80768 91177308-0d34-0410-b5e6-96231b3b80d8

clean up this code a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80767 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate VISIBILITY_HIDDEN from Transforms/Scalar.  PR4861

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80766 91177308-0d34-0410-b5e6-96231b3b80d8

refactor select 'sched insertion' out to its own method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80764 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for modeling whether or not the processor has support for
conditional moves as a subtarget feature.  This is the easy part of
PR4841.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80763 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR4848 an infinite loop when indexing down through a recursive gep
and we get the original pointer type.  This doesn't mean that we're
at the first pointer being indexed.  Correct the predicate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80762 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR4837, some bugs folding vector compares.  These
return a vector of i1, not i1 itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80761 91177308-0d34-0410-b5e6-96231b3b80d8

revert my patch, duncan points out what is wrong with my logic.  Add
a comment so that I don't change this in the future :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80760 91177308-0d34-0410-b5e6-96231b3b80d8

one more try at making this simpler, hopefully it won't break everything :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80759 91177308-0d34-0410-b5e6-96231b3b80d8

Complicate Chris's simplification, avoiding complaints
about singular iterators when building with expensive
checks turned on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80757 91177308-0d34-0410-b5e6-96231b3b80d8

Don't force the triple or data layout in this test. We just have to get them
from the host and hope that works.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80751 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80749 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed the asmstrings for 8-bit, 16-bit, and 32-bit ADD %rAX, imm instructions.
Added a 64-bit ADD %RAX, imm32 instruction.
Added all 4 forms for AND %rAX, imm and CMP %rAX, imm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80746 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR4845: r77946 completely broke x86_64 Darwin (or any situation where the
desired triplet is a sub-target, e.g. thumbv7 vs. arm host). Reverting the
patch isn't quite right either since the previous behavior does not allow the
triplet to be overridden with -march.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80742 91177308-0d34-0410-b5e6-96231b3b80d8

Disable uniqueness test for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80741 91177308-0d34-0410-b5e6-96231b3b80d8

For now disable MDNode uniquing. This fixes llvm-gcc bootstrap failure on certain Mac OS X 10.5. I am working on a proper fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80738 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid calling removeVirtualRegisterKilled which iterates over the operands
to find the kill, since we already have the operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80736 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor some code into separate functions.  No functional changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80733 91177308-0d34-0410-b5e6-96231b3b80d8

Move use of LV inside condition that guards for null LV.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80731 91177308-0d34-0410-b5e6-96231b3b80d8

Fix build warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80730 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80729 91177308-0d34-0410-b5e6-96231b3b80d8

Fix what I believe is a copy-n-pasto introduced in r78129.
 - Bruno, please check!!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80728 91177308-0d34-0410-b5e6-96231b3b80d8

X86/Encoding: Support ExternalSymbol operands in emitDisplacementField (for consistency).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80727 91177308-0d34-0410-b5e6-96231b3b80d8

Remove Offset from ExternalSybmol MachineOperands, this is unused (and at least partly unsupported, in X86 encoding at least).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80726 91177308-0d34-0410-b5e6-96231b3b80d8

debug intrinsics do not go in the callgraph, this fixes a couple
clang regtest failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80724 91177308-0d34-0410-b5e6-96231b3b80d8