Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111223 91177308-0d34-0410-b5e6-96231b3b80d8

Copy over some overridden MI wrappers for ARM fast-isel.  This is where
we're adding predicates and optional defs to the MachineInstrs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111222 91177308-0d34-0410-b5e6-96231b3b80d8

PHI elimination should not break back edge. It can cause some significant code placement issues. rdar://8263994

good:
LBB0_2:
  mov     r2, r0
  . . .
  mov     r1, r2
  bne     LBB0_2

bad:
LBB0_2:
  mov     r2, r0
  . . .
@ BB#3:
  mov     r1, r2
  b       LBB0_2

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111221 91177308-0d34-0410-b5e6-96231b3b80d8

Remove predicate workaround, we're going to require that predicate
and optional def operands are handled in the backend support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111220 91177308-0d34-0410-b5e6-96231b3b80d8

Make arm fast-isel possible to enable via command line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111219 91177308-0d34-0410-b5e6-96231b3b80d8

Try to silence a overeager GCC warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111214 91177308-0d34-0410-b5e6-96231b3b80d8

A round of minor cleanups for ELFObjectWriter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111213 91177308-0d34-0410-b5e6-96231b3b80d8

Add a testcase for svn 111208.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111212 91177308-0d34-0410-b5e6-96231b3b80d8

Fix another iterator invalidation that caused a *really* nasty miscompilation in 403.gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111210 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore undef shuffle indices when checking for a VTRN shuffle.  Radar 8290937.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111208 91177308-0d34-0410-b5e6-96231b3b80d8

tidy up. remove unused local.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111206 91177308-0d34-0410-b5e6-96231b3b80d8

Silence warnings and simplify code. Eliminate a 32/64 bit portability issue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111201 91177308-0d34-0410-b5e6-96231b3b80d8

The plural of analysis is analyses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111200 91177308-0d34-0410-b5e6-96231b3b80d8

Make dumpPassStructure be a PMDataManager abstraction, rather than
a Pass abstraction, since that's the level it's actually used at.
Rename Pass' dumpPassStructure to dumpPass.

This eliminates an awkward use of getAsPass() to convert a PMDataManager*
into a Pass* just to permit a dumpPassStructure call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111199 91177308-0d34-0410-b5e6-96231b3b80d8

Better handle alignment requirements for local objects in pre-regalloc frame
mapping. Have the local block track its alignment requirement, and then
apply that when the block itself is allocated. Previously, offsets could
get adjusted in PEI to be different, relative to one another, than the
block allocation thought they would be, which defeats the point of doing
the allocation this way. Continuing rdar://8277890

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111197 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize a pattern for PKHTB: an SRL of 16-31 bits will guarantee
that the high halfword is zero.  The shift need not be exactly 16 bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111196 91177308-0d34-0410-b5e6-96231b3b80d8

Convert test to FileCheck.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111195 91177308-0d34-0410-b5e6-96231b3b80d8

Make some of PMTopLevelManager's members non-public. In particular,
make its constructor protected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111193 91177308-0d34-0410-b5e6-96231b3b80d8

Remove redundant inline keywords.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111192 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate the TopLevelManagerType enum; instead, just make
PMTopLevelManager's constructor take a PMDataManager *, which already
provides the needed abstraction support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111189 91177308-0d34-0410-b5e6-96231b3b80d8

Comment out some broken/unused/useless instructions which mess up disassembly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111185 91177308-0d34-0410-b5e6-96231b3b80d8

Fixes for generation of ELF relocations.  Patch by Roman Divacky.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111183 91177308-0d34-0410-b5e6-96231b3b80d8

Don't attempt to SimplifyShortMoveForm in 64-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111182 91177308-0d34-0410-b5e6-96231b3b80d8

Until uleb/sleb are MC-ized, add a hack to make them work with ELF object
emission.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111177 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a few warnings in and detabify MCELFStreamer and ELFObjectWriter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111175 91177308-0d34-0410-b5e6-96231b3b80d8

Add ELF ObjectWriter and Streamer support.

I forgot to add these files in commit 111172.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111174 91177308-0d34-0410-b5e6-96231b3b80d8

Hookup ELF support for X86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111173 91177308-0d34-0410-b5e6-96231b3b80d8

Add ELF ObjectWriter and Streamer support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111172 91177308-0d34-0410-b5e6-96231b3b80d8

Layout helper function.

Introduce a helper method to add a section to the end of a layout. This
will be used by the ELF ObjectWriter code to add the metadata sections
(symbol table, etc) to the end of an object file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111171 91177308-0d34-0410-b5e6-96231b3b80d8

Record a symbol's size which is needed for ELF symbol tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111170 91177308-0d34-0410-b5e6-96231b3b80d8

ELF entry size support.

Some ELF sections contain fixed-sized entries. Provide a way to record
the entry size of a section.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111169 91177308-0d34-0410-b5e6-96231b3b80d8

Rename sat_shift operand to shift_imm, in preparation for using it for other
instructions besides saturate instructions.  No functional changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111168 91177308-0d34-0410-b5e6-96231b3b80d8

Partially revert r111155. It looks like MSVC is calling an operator<() that
clang says is unused.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111167 91177308-0d34-0410-b5e6-96231b3b80d8

track local frame size in MFI, not local to the pass, since PEI needs it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111164 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r111031. The way LLVM defines loop invariance, the property of an
expression being loop invariant is not equivalent to the property of
properly dominating the loop header.

Other optimizations have also made this optimization less important.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111160 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111156 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111155 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111154 91177308-0d34-0410-b5e6-96231b3b80d8

Convert a test to use FileCheck.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111153 91177308-0d34-0410-b5e6-96231b3b80d8

Placate overzealous compiler warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111152 91177308-0d34-0410-b5e6-96231b3b80d8

To create a copy of a SmallVector with an element removed from the
middle, copy the elements in two groups, rather than copying all the
elements and then doing an erase on the middle of the result. These
are SmallVectors, so we shouldn't expect to hit dynamic allocation
in the common case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111151 91177308-0d34-0410-b5e6-96231b3b80d8

Install Embedded builds into usr/local.  Radar 8313723.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111149 91177308-0d34-0410-b5e6-96231b3b80d8

silence a vc2010 warning: " result of 32-bit shift implicitly converted to
64 bits (was 64-bit shift intended?)", pointed out by 'nobled' on llvmdev

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111148 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111147 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111145 91177308-0d34-0410-b5e6-96231b3b80d8

Use const_iterator in a few places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111144 91177308-0d34-0410-b5e6-96231b3b80d8

Use iterators instead of indices in a few more places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111143 91177308-0d34-0410-b5e6-96231b3b80d8

Micro-optimize SCEVConstant comparison.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111142 91177308-0d34-0410-b5e6-96231b3b80d8

Move SCEVNAryExpr's virtual member functions out of line, and convert
them to iterators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111140 91177308-0d34-0410-b5e6-96231b3b80d8

Use iterators instead of indices in simple cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111138 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid gratuitous inefficiency in ifndef NDEBUG code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111137 91177308-0d34-0410-b5e6-96231b3b80d8

Make one getAddExpr call when analyzing a+b+c+d+e+... instead of one
for each add instruction. Ditto for Mul.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111136 91177308-0d34-0410-b5e6-96231b3b80d8

Delete an unused function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111135 91177308-0d34-0410-b5e6-96231b3b80d8

Instead of having CollectSubexpr's categorize operands as interesting or
uninteresting, just put all the operands on one list and make
GenerateReassociations make the decision about what's interesting.
This is simpler, and it avoids an extra ScalarEvolution::getAddExpr call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111133 91177308-0d34-0410-b5e6-96231b3b80d8

Put add operands in ScalarEvolution-canonical order, when convenient.
This isn't necessary, because ScalarEvolution sorts them anyway,
but it's tidier this way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111132 91177308-0d34-0410-b5e6-96231b3b80d8

Specialize FoldingSetTrait<SCEV>, providing implementations of node
comparison and hash computation which don't require constructing
temporary ID values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111131 91177308-0d34-0410-b5e6-96231b3b80d8

Add hooks to FoldingSetTrait to allow specializations to provide
implementations of equality comparison and hash computation. This
can be used to optimize node lookup by avoiding creating lots of
temporary ID values just for hashing and comparison purposes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111130 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up whitespace in comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111129 91177308-0d34-0410-b5e6-96231b3b80d8

Constify FoldingSetNodeIDRef's Data.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111128 91177308-0d34-0410-b5e6-96231b3b80d8

Reverse the order of GetNodeProfile's arguments, for consistency
with FoldingSetTrait::Profile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111127 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indentation in example code in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111125 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid #include <ScalarEvolution.h> in LoopSimplify.cpp, which doesn't
actually use ScalarEvolution.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111124 91177308-0d34-0410-b5e6-96231b3b80d8

Instead, teach SimplifyCFG to trim non-address-taken blocks from
indirectbr destination lists.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111122 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r111058, the lint check for indirectbr successors that aren't
address-taken. This can occur normally, if the code which took the
address got DCEd.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111121 91177308-0d34-0410-b5e6-96231b3b80d8

Test expects SSE, give him SSE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111115 91177308-0d34-0410-b5e6-96231b3b80d8

Updated LLVMLibDeps.cmake

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111112 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Improved COMMENT on a custom command

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111111 91177308-0d34-0410-b5e6-96231b3b80d8

Restore arch on these test, they fail on arm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111109 91177308-0d34-0410-b5e6-96231b3b80d8

Mark as XFAIL on darwin 8.  PR 7886.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111108 91177308-0d34-0410-b5e6-96231b3b80d8

Updated the GenLibDeps -> LLVMLibDeps.cmake transformation example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111104 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r111082. No warnings for this common pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111102 91177308-0d34-0410-b5e6-96231b3b80d8

Update tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111096 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111095 91177308-0d34-0410-b5e6-96231b3b80d8

llvmc: remove dynamic plugins.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111094 91177308-0d34-0410-b5e6-96231b3b80d8

Rework how the non-sse2 memory barrier is lowered so that the
encoding is correct for the built-in assembler.

Based on a patch from Chris.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111083 91177308-0d34-0410-b5e6-96231b3b80d8

Add ATTRIBUTE_UNUSED to methods that are not supposed to be used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111082 91177308-0d34-0410-b5e6-96231b3b80d8

improve indentation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111073 91177308-0d34-0410-b5e6-96231b3b80d8

T2I_rbin_irs rr variant is for disassembly only, so don't provide a pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111068 91177308-0d34-0410-b5e6-96231b3b80d8

Update CMake build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111063 91177308-0d34-0410-b5e6-96231b3b80d8

LoopSimplify shouldn't split loop backedges that use indirectbr. PR7867.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111061 91177308-0d34-0410-b5e6-96231b3b80d8

Teach SimplifyCFG how to simplify indirectbr instructions.
 - Eliminate redundant successors.
 - Convert an indirectbr with one successor into a direct branch.

Also, generalize SimplifyCFG to be able to be run on a function entry block.
It knows quite a few simplifications which are applicable to the entry
block, and it only needs a few checks to avoid trouble with the entry block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111060 91177308-0d34-0410-b5e6-96231b3b80d8

Add a local stack object block allocation pass. This is still an
experimental pass that allocates locals relative to one another before
register allocation and then assigns them to actual stack slots as a block
later in PEI. This will eventually allow targets with limited index offset
range to allocate additional base registers (not just FP and SP) to
more efficiently reference locals, as well as handle situations where
locals cannot be referenced via SP or FP at all (dynamic stack realignment
together with variable sized objects, for example). It's currently
incomplete and almost certainly buggy. Work in progress.

Disabled by default and gated via the -enable-local-stack-alloc command
line option.

rdar://8277890

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111059 91177308-0d34-0410-b5e6-96231b3b80d8

Add a lint check for an indirectbr destination which has not
had its address taken.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111058 91177308-0d34-0410-b5e6-96231b3b80d8

Add a Thumb2 t2RSBrr instruction for disassembly only.
This fixes another part of PR7792.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111057 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up the Spiller.h interface.

The earliestStart argument is entirely specific to linear scan allocation, and
can be easily calculated by RegAllocLinearScan.

Replace std::vector with SmallVector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111055 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily disable tail calls on ARM to work around some linker problems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111050 91177308-0d34-0410-b5e6-96231b3b80d8

Move the Thumb2 SSAT and USAT optional shift operator out of the
instruction opcode.  This fixes part of PR7792.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111047 91177308-0d34-0410-b5e6-96231b3b80d8

Various optimizations. Don't compare two loops' depths
when they are the same loop. Don't compare two instructions'
loop depths when they are in the same block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111045 91177308-0d34-0410-b5e6-96231b3b80d8

Implement splitting inside a single block.

When a live range is contained a single block, we can split it around
instruction clusters. The current approach is very primitive, splitting before
and after the largest gap between uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111043 91177308-0d34-0410-b5e6-96231b3b80d8

Fix LSR's ExtractImmediate and ExtractSymbol to avoid calling
ScalarEvolution::getAddExpr, which can be pretty expensive, when nothing
has changed, which is pretty common.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111042 91177308-0d34-0410-b5e6-96231b3b80d8

Add comments to some pattern fragments in x86

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111041 91177308-0d34-0410-b5e6-96231b3b80d8

tidy up comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111040 91177308-0d34-0410-b5e6-96231b3b80d8

When testing whether one loop contains another, test this directly
rather than testing whether the loop contains the other's header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111039 91177308-0d34-0410-b5e6-96231b3b80d8

Add a const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111038 91177308-0d34-0410-b5e6-96231b3b80d8

When creating a symmetric SCEV with a constant operand, put
the constant operand on the left, as that's where ScalarEvolution
will end up canonicalizing to.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111037 91177308-0d34-0410-b5e6-96231b3b80d8

An add recurrence is loop-invariant in any loop inside of its
associated loop. This avoids potentially expensive traversals
of the add recurrence's operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111034 91177308-0d34-0410-b5e6-96231b3b80d8

tidy up 80 column and whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111033 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r111007.

Apparently, this is now fixed in Clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111032 91177308-0d34-0410-b5e6-96231b3b80d8

Implement hasComputableLoopEvolution for Add, Mul, and Trunc operators,
since they can support trivial implementations. This avoids potentially
expensive traversals of the operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111031 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor the code for disassembling Thumb2 saturate instructions along the
same lines as the change I made for ARM saturate instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111029 91177308-0d34-0410-b5e6-96231b3b80d8