Add aditional patterns for vinsertf128 instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135659 91177308-0d34-0410-b5e6-96231b3b80d8

Add v16i16 type to VR256 class

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135658 91177308-0d34-0410-b5e6-96231b3b80d8

Move code around. No functionality changes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135657 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135656 91177308-0d34-0410-b5e6-96231b3b80d8

LSR, correct fix for rdar://9786536. Silly casting bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135654 91177308-0d34-0410-b5e6-96231b3b80d8

LSR must sometimes sign-extend before generating double constants.

rdar://9786536

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135650 91177308-0d34-0410-b5e6-96231b3b80d8

Mark instructions which are part of the frame setup with the MachineInstr::FrameSetup flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135645 91177308-0d34-0410-b5e6-96231b3b80d8

LSR crashes on an empty IVUsers list.

rdar://9786536

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135644 91177308-0d34-0410-b5e6-96231b3b80d8

X86 is the only target that uses coff format. This should fixes test failures running on Windows, Cygwin, or MingW hosts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135639 91177308-0d34-0410-b5e6-96231b3b80d8

docs/GettingStarted.html: Tweak style.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135637 91177308-0d34-0410-b5e6-96231b3b80d8

Sink ARMMCExpr and ARMAddressingModes into MC layer. First step to separate ARM MC code from target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135636 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135635 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the now defunct getCompactUnwindEncoding method from the frame lowering code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135634 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135633 91177308-0d34-0410-b5e6-96231b3b80d8

docs/GettingStarted.html: Fix a typo and tweak a command line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135632 91177308-0d34-0410-b5e6-96231b3b80d8

There are two ways to map a variable to its lexical scope. Lexical scope information is embedded in MDNode describing the variable. It is also available as a part of DebugLoc attached with DBG_VALUE instruction. DebugLoc attached with an instruction is less reliable in optimized code so use information embedded in the MDNode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135629 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up includes of llvm/Analysis/ConstantFolding.h so it's included where it's used and not included where it isn't.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135628 91177308-0d34-0410-b5e6-96231b3b80d8

While emitting constant value, look through derived type and use underlying basic type to determine size and signness of the constant value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135627 91177308-0d34-0410-b5e6-96231b3b80d8

ARM PKH shift ammount operand printing tweaks.

Move the shift operator and special value (32 encoded as 0 for PKHTB) handling
into the instruction printer. This cleans up a bit of the disassembler
special casing for these instructions, more easily handles not printing the
operand at all for "lsl #0" and prepares for correct asm parsing of these
operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135626 91177308-0d34-0410-b5e6-96231b3b80d8

Bring LICM into compliance with the new "Memory Model for Concurrent Operations" in LangRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135625 91177308-0d34-0410-b5e6-96231b3b80d8

Commit LangRef changes for LLVM concurrency model.  Start of supporting C++0x memory model and atomics.  See thread on llvmdev titled "Reviving the new LLVM concurrency model".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135624 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak the MSVC build. Since the "next" function already exists in the MSVC headers, we need the explicit llvm:: qualifier to prevent a conflict.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135623 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up a bit.

Move common definitions for ARM and Thumb2 into ARMInstrFormats.td and rename
them to be a bit more descriptive that they're for the PKH instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135617 91177308-0d34-0410-b5e6-96231b3b80d8

ARM: Tidy up representation of PKH instruction.

The shift type is implied by the instruction (PKHBT vs. PKHTB) and so shouldn't
be also encoded as part of the shift value immediate. Otherwise we're able to
represent invalid instructions, plus it needlessly complicates the
representation. Preparatory work for asm parsing of these instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135616 91177308-0d34-0410-b5e6-96231b3b80d8

Fix cmake again :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135613 91177308-0d34-0410-b5e6-96231b3b80d8

Goodbye TargetAsmInfo. This eliminate last bit of CodeGen and Target in llvm-mc.

There is still a bit more refactoring left to do in Targets. But we are now very
close to fixing all the layering issues in MC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135611 91177308-0d34-0410-b5e6-96231b3b80d8

Fix cmake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135609 91177308-0d34-0410-b5e6-96231b3b80d8

Sketch out an CFG reconstruction mode for llvm-objdump.

- Not great yet, but it's a start.
- Requires an object file with a symbol table. (I really want to fix this, but it'll need a whole new algorithm)
- ELF and COFF won't work at the moment due to libObject shortcomings.

To try it out run
$ llvm-objdump -d --cfg foo.o

This will create a graphviz file for every symbol in the object file's text section containing a CFG.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135608 91177308-0d34-0410-b5e6-96231b3b80d8

Extend the hack for _GLOBAL_OFFSET_TABLE_ slightly; PR10389.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135607 91177308-0d34-0410-b5e6-96231b3b80d8

Add parsing/encoding tests for ARM ORR instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135602 91177308-0d34-0410-b5e6-96231b3b80d8

Consolidate ARM NOP encoding test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135600 91177308-0d34-0410-b5e6-96231b3b80d8

ARM parsing and encoding tests for MVN

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135599 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembly parsing of MUL instruction.

Correctly handle 's' bit and predication suffices. Add parsing and encoding
tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135596 91177308-0d34-0410-b5e6-96231b3b80d8

PR10421: Fix a straightforward bug in the widening logic for CONCAT_VECTORS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135595 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize the EHFrameSection pointer to zero.

This should fix the spurious buildbot errors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135594 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate configure and friends for Chad.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135592 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a GCC warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135581 91177308-0d34-0410-b5e6-96231b3b80d8

- Move CodeModel from a TargetMachine global option to MCCodeGenInfo.
- Introduce JITDefault code model. This tells targets to set different default
  code model for JIT. This eliminates the ugly hack in TargetMachine where
  code model is changed after construction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135580 91177308-0d34-0410-b5e6-96231b3b80d8

Include MCRegisterInfo to eliminate a compilation warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135575 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the CMake build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135573 91177308-0d34-0410-b5e6-96231b3b80d8

Add MCObjectFileInfo and sink the MCSections initialization code from
TargetLoweringObjectFileImpl down to MCObjectFileInfo.

TargetAsmInfo is done to one last method. It's *almost* gone!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135569 91177308-0d34-0410-b5e6-96231b3b80d8

indvars: Added getInsertPointForUses to find a valid place to truncate the IV.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135568 91177308-0d34-0410-b5e6-96231b3b80d8

indvars -disable-iv-rewrite: Add NarrowIVDefUse to cache def-use
info. Holding Use* pointers is bad form even though it happened to
work in this case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135566 91177308-0d34-0410-b5e6-96231b3b80d8

X86Subtarget.h: Assume "x86_64-cygwin", though it has not been released yet, to appease test/CodeGen/X86 on cygwin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135564 91177308-0d34-0410-b5e6-96231b3b80d8

New pointer rotate test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135562 91177308-0d34-0410-b5e6-96231b3b80d8

Extra semi-colon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135561 91177308-0d34-0410-b5e6-96231b3b80d8

indvars test case for r135558.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135559 91177308-0d34-0410-b5e6-96231b3b80d8

indvars -disable-iv-rewrite fix: derived GEP IVs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135558 91177308-0d34-0410-b5e6-96231b3b80d8

Don't leak CodeGenInfos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135555 91177308-0d34-0410-b5e6-96231b3b80d8

Change name of class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135550 91177308-0d34-0410-b5e6-96231b3b80d8

Define classes for definitions of atomic instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135546 91177308-0d34-0410-b5e6-96231b3b80d8

Build and install the archive when building the Apple way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135538 91177308-0d34-0410-b5e6-96231b3b80d8

Lower memory barriers to sync instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135537 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an obvious typo that's preventing x86 (32-bit) from using .literal16.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135535 91177308-0d34-0410-b5e6-96231b3b80d8

PR10386: Don't try to split an edge from an indirectbr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135534 91177308-0d34-0410-b5e6-96231b3b80d8

Fix off-by-one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135533 91177308-0d34-0410-b5e6-96231b3b80d8

Tweak ARM assembly parsing and printing of MSR instruction.

The system register spec should be case insensitive. The preferred form for
output with mask values of 4, 8, and 12 references APSR rather than CPSR.
Update and tidy up tests accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135532 91177308-0d34-0410-b5e6-96231b3b80d8

Distinguish between two copies of one inlined variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135528 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembly parsing of MRS instruction.

Teach the parser to recognize the APSR and SPSR system register names. Add
and update tests accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135527 91177308-0d34-0410-b5e6-96231b3b80d8

Enhance the FixedLengthDecoder to be able to generate plausible-looking decoders for ARM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135524 91177308-0d34-0410-b5e6-96231b3b80d8

Change variable name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135522 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembly parsing for MRC/MRC2/MRRC/MRRC2.

Add range checking to the immediate operands. Update tests accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135521 91177308-0d34-0410-b5e6-96231b3b80d8

Use the correct opcodes: SLLV/SRLV or AND must be used instead of SLL/SRL or
ANDi, when the instruction does not have any immediate operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135520 91177308-0d34-0410-b5e6-96231b3b80d8

Move mr[r]c[2] ARM tests and tidy up a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135517 91177308-0d34-0410-b5e6-96231b3b80d8

ARM testcases for MOVT.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135516 91177308-0d34-0410-b5e6-96231b3b80d8

Use descriptive variable names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135514 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembly parsing for MOV (register).

Correct the handling of the 's' suffix when parsing ARM mode. It's only a
truly separate opcode in Thumb. Add test cases to make sure we handle
the s and condition suffices correctly, including diagnostics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135513 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135507 91177308-0d34-0410-b5e6-96231b3b80d8

Tighten conditional for 'mov' cc_out.

Make sure we only clobber the cc_out operand if it is indeed a default
non-setting operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135506 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r135457. This needs llvm-gcc change, that I forgot to check-in yesterday.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135504 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembly parsing for MOV (immediate).

Add range checking for the immediate operand and handle the "mov" mnemonic
choosing between encodings based on the value of the immediate. Add tests
for fixups, encoding choice and values, and diagnostic for out of range values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135500 91177308-0d34-0410-b5e6-96231b3b80d8

Whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135499 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused code.

cc_out and pred operands are added during parsing via custom C++ now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135497 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135496 91177308-0d34-0410-b5e6-96231b3b80d8

Remove redundant instructions.
- In EmitAtomicBinaryPartword, mask incr in loopMBB only if atomic.swap is the
  instruction being expanded, instead of masking it in thisMBB.
- Remove redundant Or in EmitAtomicCmpSwap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135495 91177308-0d34-0410-b5e6-96231b3b80d8

Separate code that modifies control flow from code that adds instruction to
basic blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135490 91177308-0d34-0410-b5e6-96231b3b80d8

ARM range checking for so_imm operands in assembly parsing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135489 91177308-0d34-0410-b5e6-96231b3b80d8

Revert "Make a provision to encode inline location in a variable. This will enable dwarf writer to easily distinguish between two instances of a inlined variable in one basic block."

This reverts commit 9fec5e346efdf744b151ae6604f912908315fa7a.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135486 91177308-0d34-0410-b5e6-96231b3b80d8

Revert "Update docs to reflect r135457."

This reverts commit ba034c0a2e71303c7cf3f43ca8e69dc8436b32e2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135485 91177308-0d34-0410-b5e6-96231b3b80d8

Convert ConstantFoldGetElementPtr to use ArrayRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135483 91177308-0d34-0410-b5e6-96231b3b80d8

Convert SimplifyGEPInst to use ArrayRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135482 91177308-0d34-0410-b5e6-96231b3b80d8

Convert gep_type_begin and gep_type_end to use ArrayRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135481 91177308-0d34-0410-b5e6-96231b3b80d8

Convert TargetData::getIndexedOffset to use ArrayRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135478 91177308-0d34-0410-b5e6-96231b3b80d8

Use ArrayRef in ConstantFoldInstOperands and ConstantFoldCall.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135477 91177308-0d34-0410-b5e6-96231b3b80d8

Add intrinsics for the zext / sext instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135476 91177308-0d34-0410-b5e6-96231b3b80d8

Add intrinsics for the testct, testwct instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135475 91177308-0d34-0410-b5e6-96231b3b80d8

Add intrinsics for the peek and endin instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135474 91177308-0d34-0410-b5e6-96231b3b80d8

Remove bogus test: for all possible inputs of %X, the 'sub nsw' is guaranteed
to perform a signed wrap. Don't rely on any particular handling of that case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135471 91177308-0d34-0410-b5e6-96231b3b80d8

Introduce MCCodeGenInfo, which keeps information that can affect codegen
(including compilation, assembly). Move relocation model Reloc::Model from
TargetMachine to MCCodeGenInfo so it's accessible even without TargetMachine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135468 91177308-0d34-0410-b5e6-96231b3b80d8

Make EmitAtomic functions return the correct MachineBasicBlocks so that
ExpandISelPseudos::runOnMachineFunction does not visit instructions that have
just been added.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135465 91177308-0d34-0410-b5e6-96231b3b80d8

Do not insert instructions in reverse order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135464 91177308-0d34-0410-b5e6-96231b3b80d8

Make isLoadExtLegal and isTruncStoreLegal check what the name says. :)  This might have some minor effect on CellSPU, but all other targets should be unaffected.  Fixing per report from Damien Vincent on llvmdev.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135462 91177308-0d34-0410-b5e6-96231b3b80d8

Update docs to reflect r135457.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135458 91177308-0d34-0410-b5e6-96231b3b80d8

Make a provision to encode inline location in a variable. This will enable dwarf writer to easily distinguish between two instances of a inlined variable in one basic block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135457 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r135423.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135454 91177308-0d34-0410-b5e6-96231b3b80d8

Document how to maintain a git-svn clone of the LLVM git repositories.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135453 91177308-0d34-0410-b5e6-96231b3b80d8

Micro-opt: Only emit compact unwind if there is a compact unwind encoding to emit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135452 91177308-0d34-0410-b5e6-96231b3b80d8

Use the CompactUnwindEncoding from the Frame, if it's defined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135451 91177308-0d34-0410-b5e6-96231b3b80d8

Add a frame with the compact unwind encoding if it exists.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135450 91177308-0d34-0410-b5e6-96231b3b80d8

Add a method to set compact unwind encoding information in a frame.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@135449 91177308-0d34-0410-b5e6-96231b3b80d8