Remove redundant code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85668 91177308-0d34-0410-b5e6-96231b3b80d8

Merge the enhancements from LoopUnroll's FoldBlockIntoPredecessor into
MergeBlockIntoPredecessor. This makes SimplifyCFG slightly more aggressive,
and makes it unnecessary for LoopUnroll to have its own copy of this code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85667 91177308-0d34-0410-b5e6-96231b3b80d8

Rename forgetLoopBackedgeTakenCount to forgetLoop, because it
clears out more information than just the stored backedge taken count.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85664 91177308-0d34-0410-b5e6-96231b3b80d8

Replace LoopUnrollPass.cpp's custom code-size estimation code using
the new common CodeMetrics code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85663 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify this code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85662 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an unnecessary #include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85661 91177308-0d34-0410-b5e6-96231b3b80d8

Update CMakeLists for recent renames.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85660 91177308-0d34-0410-b5e6-96231b3b80d8

Rename UnrollLoop.cpp to LoopUnroll.cpp, and LoopUnroll.cpp to
LoopUnrollPass.cpp, for consistency with other passes which are
similarly split.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85659 91177308-0d34-0410-b5e6-96231b3b80d8

Remove CodeGenLICM. It's largely obsoleted by MachineLICM's new ability
to unfold loop-invariant loads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85657 91177308-0d34-0410-b5e6-96231b3b80d8

Make ScalarEvolutionAliasAnalysis slightly more aggressive, by making an
underlying alias call even for non-identified-object values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85656 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r85634, with the bug fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85655 91177308-0d34-0410-b5e6-96231b3b80d8

When discarding SrcValue information, discard all of it so that code
that uses this information knows to behave conservatively.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85654 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80-column violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85653 91177308-0d34-0410-b5e6-96231b3b80d8

Fix warning with gcc-4.0 and signed/unsigned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85648 91177308-0d34-0410-b5e6-96231b3b80d8

It's safe to remat t2LDRpci; Add PseudoSourceValue to load / store's to enable more machine licm. More changes coming.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85643 91177308-0d34-0410-b5e6-96231b3b80d8

Revert 85634. It's breaking consumer-typeset (and others).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85641 91177308-0d34-0410-b5e6-96231b3b80d8

Add a target triple so that this test behaves consistently across hosts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85640 91177308-0d34-0410-b5e6-96231b3b80d8

Add assertion checks here to turn silent miscompiles into aborts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85639 91177308-0d34-0410-b5e6-96231b3b80d8

Don't mark registers dead here when processing nodes with MVT::Flag
results. This works around a problem affecting targets which rely on
MVT::Flag to handle physical register defs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85638 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the -mattr line for this test so that it passes on hosts that lack SSSE3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85637 91177308-0d34-0410-b5e6-96231b3b80d8

Add a testcase for the recent duplicate PHI elimination changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85636 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment about a missed opportunity.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85635 91177308-0d34-0410-b5e6-96231b3b80d8

Optimize around the fact that pred_iterator is slow: instead of sorting
PHI operands by the predecessor order, sort them by the order used by the
first PHI in the block. This is still suffucient to expose duplicates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85634 91177308-0d34-0410-b5e6-96231b3b80d8

Updates to the ARM target assembler for llvm-mc per review comments from
Daniel Dunbar.
- Reordered the fields in the ARMOperand Mem struct to make the struct smaller.
Making bool's into 1 bit fields and put the MCExpr* fields adjacent to each
other.
- Fixed a number of places in ARMAsmParser.cpp so they have doxygen comments.
- Change the name of ARMAsmParser::ParseRegister() to MaybeParseRegister and
added the bool ParseWriteBack parameter.
- Changed ARMAsmParser::ParseMemory() to call MaybeParseRegister().
- Added ARMAsmParser::ParseMemoryOffsetReg to factor out parsing the offset of a
memory operand.  And use it for both parsing both preindexed and post indexing
addressing forms in ARMAsmParser::ParseMemory.
- Changed the first argument to ParseShift() to a reference.
- Changed ParseShift() to check for Rrx first and return to reduce nesting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85632 91177308-0d34-0410-b5e6-96231b3b80d8

If string field is empty then return NULL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85630 91177308-0d34-0410-b5e6-96231b3b80d8

if basic blocks are destroyed while there are *just* BlockAddress' hanging
around, then zap them.  This is analogous to dangling constantexprs hanging
off functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85627 91177308-0d34-0410-b5e6-96231b3b80d8

Teach SimplifyCFG how to eliminate duplicate PHI nodes within a block.
This reduces codesize on a variety of codes by 1-2% on x86-64. It also
helps clean up after SSAUpdater.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85626 91177308-0d34-0410-b5e6-96231b3b80d8

make hasAddressTaken() constant time by storing a refcount in BB's subclass data.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85625 91177308-0d34-0410-b5e6-96231b3b80d8

Add a note about Robert Muth's alternate jump table implementation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85624 91177308-0d34-0410-b5e6-96231b3b80d8

Sort the incoming values in PHI nodes to match the predecessor order.
This helps expose duplicate PHIs, which will make it easier for them
to be eliminated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85623 91177308-0d34-0410-b5e6-96231b3b80d8

Fix MachineLICM to use the correct virtual register class when
unfolding loads for hoisting.  getOpcodeAfterMemoryUnfold returns the
opcode of the original operation without the load, not the load
itself, MachineLICM needs to know the operand index in order to get
the correct register class. Extend getOpcodeAfterMemoryUnfold to
return this information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85622 91177308-0d34-0410-b5e6-96231b3b80d8

it isn't valid to take the address of the entry block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85621 91177308-0d34-0410-b5e6-96231b3b80d8

If a type is derived from a derived type then calculate size appropriately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85619 91177308-0d34-0410-b5e6-96231b3b80d8

Build in ARM mode explicitly when on ARM Darwin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85615 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing substitution for %llvmgcc_only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85614 91177308-0d34-0410-b5e6-96231b3b80d8

Allow cross target build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85611 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85610 91177308-0d34-0410-b5e6-96231b3b80d8

Add option to createGVNPass to disable PRE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85609 91177308-0d34-0410-b5e6-96231b3b80d8

I forgot to commit this test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85608 91177308-0d34-0410-b5e6-96231b3b80d8

When cross-building, the CFLAGS and CXXFLAGS are for the target, and don't
apply to the build tools. If we want to allow build tool flags input, we
should have separate inputs (BUILD_CFLAGS and BUILD_CXXFLAGS, perhaps).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85607 91177308-0d34-0410-b5e6-96231b3b80d8

Remove extraneous comment line

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85606 91177308-0d34-0410-b5e6-96231b3b80d8

update name check for Apple style builds to be more permissive

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85605 91177308-0d34-0410-b5e6-96231b3b80d8

Stop the iterator in ValueLiveAt from potentially running off the end of the interval.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85599 91177308-0d34-0410-b5e6-96231b3b80d8

This fixes functions like

void f (int a1, int a2, int a3, int a4, int a5,...)

In ARMTargetLowering::LowerFormalArguments if the function has 4 or
more regular arguments we used to set VarArgsFrameIndex using an
offset of 0, which is only correct if the function has exactly 4
regular arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85590 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: install .def files from source `include/llvm' directory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85587 91177308-0d34-0410-b5e6-96231b3b80d8

Rather than having llvm-gcc changing the meaning of OptimizeSize, just make sure loop unswitch is conservative when optimization level is < 3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85581 91177308-0d34-0410-b5e6-96231b3b80d8

Add ARM codegen for indirect branches.
clang/test/CodeGen/indirect-goto.c runs! (unoptimized)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85577 91177308-0d34-0410-b5e6-96231b3b80d8

Most stack straces don't need 3 digits worth of levels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85575 91177308-0d34-0410-b5e6-96231b3b80d8

Don't delete blocks which have their address taken.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85572 91177308-0d34-0410-b5e6-96231b3b80d8

Mention if a block has its address taken in debug output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85571 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify this code and avoid an extra space character in the output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85568 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for BlockAddress static initializers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85562 91177308-0d34-0410-b5e6-96231b3b80d8

Add a FIXME comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85559 91177308-0d34-0410-b5e6-96231b3b80d8

Add some comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85558 91177308-0d34-0410-b5e6-96231b3b80d8

Initial x86 support for BlockAddresses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85557 91177308-0d34-0410-b5e6-96231b3b80d8

Initial target-independent CodeGen support for BlockAddresses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85556 91177308-0d34-0410-b5e6-96231b3b80d8

Remove dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85551 91177308-0d34-0410-b5e6-96231b3b80d8

Add a BlockAddress MachineOperand kind.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85549 91177308-0d34-0410-b5e6-96231b3b80d8

Add svn:ignore properties.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85548 91177308-0d34-0410-b5e6-96231b3b80d8

Remove a redundant copy constructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85547 91177308-0d34-0410-b5e6-96231b3b80d8

Dial back the realignment a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85546 91177308-0d34-0410-b5e6-96231b3b80d8

Between scheduling regions, correctly maintain anti-dep breaking state so that we don't incorrectly rename registers that span these regions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85537 91177308-0d34-0410-b5e6-96231b3b80d8

Remove some unnecessary spaces in debug output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85536 91177308-0d34-0410-b5e6-96231b3b80d8

Move some code from being emitted as boilerplate duplicated in every
*ISelDAGToDAG.cpp to being regular code in SelectionDAGISel.cpp.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85530 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a couple of bugs in aggressive anti-dep breaking.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85522 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor complicated predicate into a separate function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85519 91177308-0d34-0410-b5e6-96231b3b80d8

First bitcase use may not lead to a dbg.declare intrinsic. Iterate uses until one find's dbg.declare intrinsic.
Patch by Sunae Seo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85518 91177308-0d34-0410-b5e6-96231b3b80d8

Rename usesCustomDAGSchedInserter to usesCustomInserter, and update a
bunch of associated comments, because it doesn't have anything to do
with DAGs or scheduling. This is another step in decoupling MachineInstr
emitting from scheduling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85517 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor the code for unfolding a load into a separate function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85515 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r85338.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85514 91177308-0d34-0410-b5e6-96231b3b80d8

Fix MSVC build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85505 91177308-0d34-0410-b5e6-96231b3b80d8

Apply some cleanups. No functionality changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85498 91177308-0d34-0410-b5e6-96231b3b80d8

add sanity check for indbr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85496 91177308-0d34-0410-b5e6-96231b3b80d8

just for the hell of it, allow globalopt to statically evaluate
static constructors with indirect gotos :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85495 91177308-0d34-0410-b5e6-96231b3b80d8

add interpreter support for indirect goto / blockaddress.  The interpreter
now correctly runs clang's test/CodeGen/indirect-goto.c.  The JIT will abort
on it until someone feels compelled to implement this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85488 91177308-0d34-0410-b5e6-96231b3b80d8

add newline to make cl.exe happy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85483 91177308-0d34-0410-b5e6-96231b3b80d8

fix 80-col.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85480 91177308-0d34-0410-b5e6-96231b3b80d8

greatly improve the LLVM IR bitcode encoding documentation,
patch by Peter Housel!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85479 91177308-0d34-0410-b5e6-96231b3b80d8

Explicitly convert to double to suppress Visual C++ 2008 build error C2668 pow is ambiguous call to overloaded function

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85478 91177308-0d34-0410-b5e6-96231b3b80d8

To get more thorough testing from llc-beta nightly runs, do dynamic stack
realignment regardless of whether it's strictly necessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85476 91177308-0d34-0410-b5e6-96231b3b80d8

When the function is doing dynamic stack realignment, the spill slot will be
indexed via the stack pointer, even if a frame pointer is present. Update the
heuristic to place it nearest the stack pointer in that case, rather than
nearest the frame pointer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85474 91177308-0d34-0410-b5e6-96231b3b80d8

Sorry to break the build.
I was trying to check the WIP file to some local repository, but ended up
checking in the llvm repository.  Oops!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85470 91177308-0d34-0410-b5e6-96231b3b80d8

Minor tweak to forgo the the curly braces for most case blocks, except when
declaring local variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85467 91177308-0d34-0410-b5e6-96231b3b80d8

teach various passes about blockaddress.  We no longer
crash on any clang tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85465 91177308-0d34-0410-b5e6-96231b3b80d8

When there is a 2-instruction spill sequence, record
the second (store) instruction in SpillSlotToUsesMap
consistently.  I don't think this matters functionally,
but it's cleaner and Evan wants it this way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85463 91177308-0d34-0410-b5e6-96231b3b80d8

Don't put in these EH changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85460 91177308-0d34-0410-b5e6-96231b3b80d8

A switch-on-string-literal construct that is a nice alternative to
cascading "ifs" of strcmps/memcmps.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85459 91177308-0d34-0410-b5e6-96231b3b80d8

teach ValueMapper about BlockAddress', making bugpoint a lot more useful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85458 91177308-0d34-0410-b5e6-96231b3b80d8

unindent massive blocks, no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85457 91177308-0d34-0410-b5e6-96231b3b80d8

Reverting r85338 for now. It's causing a bootstrap failure on PPC darwin9.

--- Reverse-merging r85338 into '.':
U    lib/CodeGen/SimpleRegisterCoalescing.cpp
U    lib/CodeGen/SimpleRegisterCoalescing.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85454 91177308-0d34-0410-b5e6-96231b3b80d8

Add indirectbr and blockaddress to the vim syntax highlighting file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85451 91177308-0d34-0410-b5e6-96231b3b80d8

Add a hasAddressTaken for BasicBlock.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85449 91177308-0d34-0410-b5e6-96231b3b80d8

add IRBuilder support for IndirectBr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85445 91177308-0d34-0410-b5e6-96231b3b80d8

Reimplement BranchFolding change to avoid tail merging for a 1 instruction
common tail, except when the OptimizeForSize function attribute is present.
Radar 7338114.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85441 91177308-0d34-0410-b5e6-96231b3b80d8

When we generate spill code, then decide we don't need
to spill after all, we weren't handling 2-instruction
spill sequences correctly (PPC Altivec).  We need to
remove the store in this case.  Removing the other
instruction(s) would be goodness but is not needed for
correctness, and isn't done here.  7331562.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85437 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure we return the right sized type here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85436 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r85346 change to control tail merging by CodeGenOpt::Level.
I'm going to redo this using the OptimizeForSize function attribute.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85426 91177308-0d34-0410-b5e6-96231b3b80d8

Extend getMallocArraySize() to determine the array size if the malloc argument is:
ArraySize * ElementSize
ElementSize * ArraySize
ArraySize << log2(ElementSize)
ElementSize << log2(ArraySize)

Refactor isArrayMallocHelper and delete isSafeToGetMallocArraySize, so that there is only 1 copy of the malloc array determining logic.
Update users of getMallocArraySize() to not bother calling isArrayMalloc() as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85421 91177308-0d34-0410-b5e6-96231b3b80d8

Fix to pass options from Gold plugin to LTO codegen

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85419 91177308-0d34-0410-b5e6-96231b3b80d8

Teach cmake that mk[sd]temp is defined in stdlib.h on some systems.
This fixes parallel build with clang on glibc platforms.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85414 91177308-0d34-0410-b5e6-96231b3b80d8