Add support for empty metadata nodes: !{}.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108259 91177308-0d34-0410-b5e6-96231b3b80d8

Extend the r107852 optimization which turns some fp compare to code sequence using only i32 operations. It now optimize some f64 compares when fp compare is exceptionally slow (e.g. cortex-a8). It also catches comparison against 0.0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108258 91177308-0d34-0410-b5e6-96231b3b80d8

Add an ARM "feature". Cortex-a8 fp comparison is very slow (> 20 cycles).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108256 91177308-0d34-0410-b5e6-96231b3b80d8

Patched Makefile.rules to allow objective C and objective C++ compilation.
This allows LLDB (llvm/tools/lldb) to build on Mac OS X.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108255 91177308-0d34-0410-b5e6-96231b3b80d8

-enable-unsafe-fp-math should not imply -enable-finite-only-fp-math.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108254 91177308-0d34-0410-b5e6-96231b3b80d8

Grammar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108252 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR number.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108251 91177308-0d34-0410-b5e6-96231b3b80d8

Switch from EXPECT_EQ({true,false, ...) to the more canonical
EXPECT_{TRUE,FALSE}(...) macros. This also prevents suprious warnings about
bool-to-pointer conversion that occurs withit EXPECT_EQ.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108248 91177308-0d34-0410-b5e6-96231b3b80d8

Fix compile unit in example. Thanks Renato Golin!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108246 91177308-0d34-0410-b5e6-96231b3b80d8

Handle the case of a tail recursion in which the tail call is followed
by a return that returns a constant, while elsewhere in the function
another return instruction returns a different constant.  This is a
special case of accumulator recursion, so just generalize the existing
logic a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108241 91177308-0d34-0410-b5e6-96231b3b80d8

rotate CallInst operands

with this commit the callee moves to the end of
the operand array (from the start) and the call
arguments now start at index 0 (formerly 1)

this ordering is now consistent with InvokeInst

this commit only flips the switch,
functionally it is equivalent to
r101465

I intend to commit several cleanups after a few
days of soak period

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108240 91177308-0d34-0410-b5e6-96231b3b80d8

HTML cleanup and validation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108239 91177308-0d34-0410-b5e6-96231b3b80d8

When doing sext/zext of constants, the target type must be strictly
wider than the source type.  Correct LangRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108238 91177308-0d34-0410-b5e6-96231b3b80d8

use a typedef instead of its expansion, patch by Michael Spencer

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108229 91177308-0d34-0410-b5e6-96231b3b80d8

80-columns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108228 91177308-0d34-0410-b5e6-96231b3b80d8

Move NEON "modified immediate" encode/decode into ARMAddressingModes.h to
avoid replicated code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108227 91177308-0d34-0410-b5e6-96231b3b80d8

my work on adding segment registers to LEA missed the
disassembler.  Remove some code from the disassembler to
compensate, unbreaking disassembly of lea's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108226 91177308-0d34-0410-b5e6-96231b3b80d8

Add AVX 256-bit packed logical forms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108224 91177308-0d34-0410-b5e6-96231b3b80d8

Add AVX 256-bit unop arithmetic instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108223 91177308-0d34-0410-b5e6-96231b3b80d8

Since AVX is a superset of all SSE versions, only use HasAVX for AVX instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108222 91177308-0d34-0410-b5e6-96231b3b80d8

Bring ELF64 on par with ELF32, add a few magic constants and support ".dynamic"
section entries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108221 91177308-0d34-0410-b5e6-96231b3b80d8

Don't add memory operands to storeRegToStackSlot / loadRegFromStackSlot results,
they already have one.

This fixes the himenobmtxpa miscompilation on ARM.

The PostRA scheduler got confused by the double memoperand and hoisted a stack
slot load above a store to the same slot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108219 91177308-0d34-0410-b5e6-96231b3b80d8

Move some SIMD fragment code into X86InstrFragmentsSIMD so that the
utility classes can be used from multiple files.  This will aid
transitioning to a new refactored x86 SIMD specification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108213 91177308-0d34-0410-b5e6-96231b3b80d8

Add AVX 256 binary arithmetic instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108207 91177308-0d34-0410-b5e6-96231b3b80d8

More refactoring of basic SSE arith instructions. Open room for 256-bit instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108204 91177308-0d34-0410-b5e6-96231b3b80d8

Simple format/style fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108203 91177308-0d34-0410-b5e6-96231b3b80d8

Test commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108197 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move ELF specific parser to ELFAsmParser.cpp.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108196 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move Darwin specific parse to DarwinAsmParser.cpp.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108193 91177308-0d34-0410-b5e6-96231b3b80d8

Add convenience method FoldingSetImpl::InsertNode(Node *N) that asserts if the node is already inserted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108192 91177308-0d34-0410-b5e6-96231b3b80d8

Apply the SSE dependence idiom for SSE unary operations to
SD instructions too, in addition to SS instructions. And
add a comment about it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108191 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move .section parsing to Darwin specific parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108190 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmLexer: Raise LexUntilEndOfStatement to MCAsmLexer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108189 91177308-0d34-0410-b5e6-96231b3b80d8

getOrInsertLeaderValue cannot be const because it calls insert.
Patch by Xi Wang!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108188 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move special section directive parsing to Darwin specific parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108187 91177308-0d34-0410-b5e6-96231b3b80d8

Remove some code that doesn't appear to do anything.  All the ARM call
instructions already have implicit defs of LR.  The comment suggests that
this is intended to fix something like pr6111, but it doesn't really do
that either.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108186 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Add a basic ELFAsmParser extension.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108185 91177308-0d34-0410-b5e6-96231b3b80d8

Add AVX 256-bit MOVMSK forms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108184 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Inline AsmParser::CreateSymbol into callers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108183 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move .tbss and .zerofill parsing to Darwin specific parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108180 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move .desc parsing to Darwin specific parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108179 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move .lsym parsing to Darwin specific parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108176 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Move some misc. Darwin directive handling to DarwinAsmParser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108174 91177308-0d34-0410-b5e6-96231b3b80d8

MCAsmParser: Add getSourceManager().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108171 91177308-0d34-0410-b5e6-96231b3b80d8

Silence a warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108169 91177308-0d34-0410-b5e6-96231b3b80d8

Check begin!=end, rather than !begin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108167 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Add a DarwinAsmParser extension.
 - Currently initialization is a bit of a hack, but harmless. We need to rework
   various parts of target initialization to clean this up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108165 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo and fit in 80 columns. Found by Bob Wilson.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108164 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Switch a bunch of directive parsing to use accessors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108163 91177308-0d34-0410-b5e6-96231b3b80d8

Add a lint check for mismatched return types, inspired by PR6944.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108162 91177308-0d34-0410-b5e6-96231b3b80d8

MCAsmParser: Pull some directive handling out into a helper class, and change
DirectiveMap to be based on MCAsmParserExtension.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108161 91177308-0d34-0410-b5e6-96231b3b80d8

MC/AsmParser: Switch some directive parsing to use accessor methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108160 91177308-0d34-0410-b5e6-96231b3b80d8

MC: Add MCAsmParserExtension, a base class for all the target/object specific
classes which want to extend the basic asm parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108158 91177308-0d34-0410-b5e6-96231b3b80d8

MC: Move AsmParser::TokError to MCAsmParser().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108155 91177308-0d34-0410-b5e6-96231b3b80d8

MC: Move getLoc() to MCAsmLexer().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108154 91177308-0d34-0410-b5e6-96231b3b80d8

Nope, still breaks the release selfhost bots :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108153 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply the "or" half of r108136, which seems to be less problematic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108152 91177308-0d34-0410-b5e6-96231b3b80d8

cache result of operator*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108150 91177308-0d34-0410-b5e6-96231b3b80d8

Don't fast-isel an x87 comparison opcode, as fast-isel doesn't
support branching on x87 comparisons yet. This fixes PR7624.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108149 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r108141 again, sigh.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108148 91177308-0d34-0410-b5e6-96231b3b80d8

cache result of operator*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108147 91177308-0d34-0410-b5e6-96231b3b80d8

cache result of operator*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108146 91177308-0d34-0410-b5e6-96231b3b80d8

cache result of operator*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108145 91177308-0d34-0410-b5e6-96231b3b80d8

cache result of operator*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108144 91177308-0d34-0410-b5e6-96231b3b80d8

cache results of operator*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108143 91177308-0d34-0410-b5e6-96231b3b80d8

cache results of operator*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108142 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply 108136 with an ugly pasto fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108141 91177308-0d34-0410-b5e6-96231b3b80d8

Move optimization to avoid redundant matching.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108140 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r108136 until I figure out why it broke selfhost.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108139 91177308-0d34-0410-b5e6-96231b3b80d8

cache dereferenced iterators

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108138 91177308-0d34-0410-b5e6-96231b3b80d8

recommit r108131 (hich has been backed out in r108135) with a fix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108137 91177308-0d34-0410-b5e6-96231b3b80d8

instcombine: fold (x & y) | (~x & z) and (x & y) ^ (~x & z) into ((y ^ z) & x) ^ z which is one instruction shorter. (PR6773)

before:
  %and = and i32 %y, %x
  %neg = xor i32 %x, -1
  %and4 = and i32 %z, %neg
  %xor = xor i32 %and4, %and

after:
  %xor1 = xor i32 %z, %y
  %and2 = and i32 %xor1, %x
  %xor = xor i32 %and2, %z

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108136 91177308-0d34-0410-b5e6-96231b3b80d8

back out r108131 (of TailDuplication.cpp) for now, it causes a buildbot failure

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108135 91177308-0d34-0410-b5e6-96231b3b80d8

cache dereferenced iterators

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108134 91177308-0d34-0410-b5e6-96231b3b80d8

cache dereferenced iterators

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108133 91177308-0d34-0410-b5e6-96231b3b80d8

cache dereferenced iterators

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108132 91177308-0d34-0410-b5e6-96231b3b80d8

cache dereferenced iterators

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108131 91177308-0d34-0410-b5e6-96231b3b80d8

Convert some tab stops into spaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108130 91177308-0d34-0410-b5e6-96231b3b80d8

Add parentheses around an || to correct the logic. Also silences a GCC warning
that was actually useful here.

Chris, please double check that this is the correct interpretation. I was
pretty sure, and ran it by Nick as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108129 91177308-0d34-0410-b5e6-96231b3b80d8

Remove tab characters and 80-col.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108127 91177308-0d34-0410-b5e6-96231b3b80d8

80 columns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108126 91177308-0d34-0410-b5e6-96231b3b80d8

Path::isRootDirectory is unimplemented on Unix and not used,
remove it, fixing PR6909.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108125 91177308-0d34-0410-b5e6-96231b3b80d8

Convert getLoadStoreRegOpcode to use a switch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108123 91177308-0d34-0410-b5e6-96231b3b80d8

Convert the last use of getPhysicalRegisterRegClass and remove it.

AggressiveAntiDepBreaker should not be using getPhysicalRegisterRegClass. An
instruction might be using a register that can only be replaced with one from
a subclass of getPhysicalRegisterRegClass.

With this patch we use getMinimalPhysRegClass. This is correct, but
conservative. We should check the uses of the register and select the
largest register class that can be used in all of them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108122 91177308-0d34-0410-b5e6-96231b3b80d8

Fix up the ocaml kaleidoscope example as well.

Note that the example currently doesn't appear to work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108121 91177308-0d34-0410-b5e6-96231b3b80d8

A basic block that only uses RFP registers still needs the FP_REG_KILL marker.

This fixes PR7375.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108120 91177308-0d34-0410-b5e6-96231b3b80d8

Don't use getPhysicalRegisterRegClass in PBQP. The existing checks that the
physical register can be allocated in the class of the virtual are sufficient.

I think that the test for virtual registers is more strict than it needs to be,
it should be possible to coalesce two virtual registers the class of one
is a subclass of the other.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108118 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR7311 by avoiding breaking casts when a bitcast from scalar->vector
is involved.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108117 91177308-0d34-0410-b5e6-96231b3b80d8

make the prototypes for CreateMalloc and CreateFree more consistent.  Patch
by Hans Vandierendonck from PR7605

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108116 91177308-0d34-0410-b5e6-96231b3b80d8

Convert the last getPhysicalRegisterRegClass in VirtRegRewriter.cpp to
getMinimalPhysRegClass. It was used to produce spills, and it is better to
use the most specific class if possible.

Update getLoadStoreRegOpcode to handle GR32_AD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108115 91177308-0d34-0410-b5e6-96231b3b80d8

if jump threading is able to infer interesting values on both
the LHS and RHS of an and/or instruction, don't multiply add
known predecessor values.  This fixes the crash on testcase
from PR7498

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108114 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR7429, a crash turning a load from a string into a float.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108113 91177308-0d34-0410-b5e6-96231b3b80d8

convert to filechecconvert to filecheckk

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108112 91177308-0d34-0410-b5e6-96231b3b80d8

merge two tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108111 91177308-0d34-0410-b5e6-96231b3b80d8

improve Path::makeUnique when mkstemp/mktemp are not available
patch by Lasse Kärkkäinen in PR7404.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108110 91177308-0d34-0410-b5e6-96231b3b80d8

change machinelicm to use MachineInstr::isSafeToMove.  No
intended functionality change.

The avoidance of hoistiing implicitdef seems wrong though.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108109 91177308-0d34-0410-b5e6-96231b3b80d8

first part of JIT support for address of labels, part of PR7264,
patch by Yuri!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108107 91177308-0d34-0410-b5e6-96231b3b80d8

introduce WinCOFFObjectWriter, patch by Michael Spencer!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108103 91177308-0d34-0410-b5e6-96231b3b80d8

introduce WinCOFFStreamer.cpp, patch by Michael Spencer!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108102 91177308-0d34-0410-b5e6-96231b3b80d8

If it's safe to speculatively execute load(alloca) the it's safe to execute
load(gep(alloca)) where the gep is all-zeros. There's more we could do here but
this is a common case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108101 91177308-0d34-0410-b5e6-96231b3b80d8