Fix bug 9905: Failure in code selection for llvm intrinsics sqrt/exp (fix for FSQRT, FSIN, FCOS, FPOWI, FPOW, FLOG, FLOG2, FLOG10, FEXP, FEXP2).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146143 91177308-0d34-0410-b5e6-96231b3b80d8

MTCTR needs to be glued to BCTR so that CTR is not marked dead in MTCTR (another find by -verify-machineinstrs)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146137 91177308-0d34-0410-b5e6-96231b3b80d8

Reverting r145899 as it breaks clang self-hosting

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146136 91177308-0d34-0410-b5e6-96231b3b80d8

ARM NEON two-operand aliases for VSHL(immediate).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146125 91177308-0d34-0410-b5e6-96231b3b80d8

Drop the HasInlineAsm flag.

It is not used any more. We are tracking inline assembly misalignments
directly through the BBInfo.Unalign and KnownBits fields.

A simple conservative size estimate is not good enough since it can
cause alignment padding to be underestimated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146124 91177308-0d34-0410-b5e6-96231b3b80d8

ARM NEON two-operand aliases for VSHL(register).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146123 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify offset verification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146121 91177308-0d34-0410-b5e6-96231b3b80d8

Fix copy/past-o.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146120 91177308-0d34-0410-b5e6-96231b3b80d8

ARM NEON two-operand aliases for VMUL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146119 91177308-0d34-0410-b5e6-96231b3b80d8

Don't include alignment padding in BBInfo.Size.

Compute alignment padding before and after basic blocks dynamically.

Heed basic block alignment.

This simplifies bookkeeping because we don't have to constantly add and
remove padding from BBInfo.Size.  It also makes it possible to track the
extra known alignment bits we get after a tBR_JTr terminator and when
entering an aligned basic block.

This makes the ARMConstantIslandPass aware of aligned basic blocks.

It is tricky to model block alignment correctly when dealing with inline
assembly and tBR_JTr instructions that have variable size.  If inline
assembly turns out to be smaller than expected, that may cause following
alignment padding to be larger than expected.  This could cause constant
pool entries to move out of range.

To avoid that problem, we use the worst case alignment padding following
inline assembly. This may cause slightly suboptimal constant island
placement in aligned basic blocks following inline assembly.  Normal
functions should be unaffected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146118 91177308-0d34-0410-b5e6-96231b3b80d8

ARM VFP support 'fmrs/fmsr' aliases for 'vldr'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146116 91177308-0d34-0410-b5e6-96231b3b80d8

ARM VFP support 'flds/fldd' aliases for 'vldr'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146115 91177308-0d34-0410-b5e6-96231b3b80d8

ARM optional destination operand variants for VEXT instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146114 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146113 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80-column.
Simplify code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146112 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembler aliases for "add Rd, #-imm" to "sub Rd, #imm".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146111 91177308-0d34-0410-b5e6-96231b3b80d8

Add Tsan annotations to the pass system.

Perhaps once(&func) should be hoisted into lib/Support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146110 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146109 91177308-0d34-0410-b5e6-96231b3b80d8

EngineBuilder: support for custom TargetOptions.  Fixes the
ExceptionDemo example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146108 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146107 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembly, allow 'asl' as a synonym for 'lsl' in shifted-register operands.

For 'gas' compatibility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146106 91177308-0d34-0410-b5e6-96231b3b80d8

Modify class ReadHardware and add definition of 64-bit version of instruction
RDHWR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146101 91177308-0d34-0410-b5e6-96231b3b80d8

Add newline.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146100 91177308-0d34-0410-b5e6-96231b3b80d8

Add 64-bit HWR29 register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146099 91177308-0d34-0410-b5e6-96231b3b80d8

32 to 64-bit anyext pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146097 91177308-0d34-0410-b5e6-96231b3b80d8

32 to 64-bit zext pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146096 91177308-0d34-0410-b5e6-96231b3b80d8

ARM two-operand aliases for VAND/VEOR/VORR instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146095 91177308-0d34-0410-b5e6-96231b3b80d8

ARM two-operand aliases for VADDW instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146093 91177308-0d34-0410-b5e6-96231b3b80d8

ARM two-operand aliases for VADD instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146091 91177308-0d34-0410-b5e6-96231b3b80d8

Flesh out a bit more of the bitcode use-list ordering preservation code.
Nothing too interesting at this point, but comments are welcome.
Part of rdar://9860654 and PR5680.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146090 91177308-0d34-0410-b5e6-96231b3b80d8

Variable cleanup. Based on past patch submittals variable names have
been normalized and more descriptive comments added. Patch by Reed
Kotler and Jack Carter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146088 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure we correctly set LiveRegGens when a call is unscheduled.  <rdar://problem/10460321>.  No testcase because this is very sensitive to scheduling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146087 91177308-0d34-0410-b5e6-96231b3b80d8

64-bit WrapperPICPat patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146086 91177308-0d34-0410-b5e6-96231b3b80d8

configure: Don't warn about compiler-rt checked out into projects/.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146084 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an assertion in the scheduler.  PR11386.  No testcase included because it's rather delicate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146083 91177308-0d34-0410-b5e6-96231b3b80d8

Define base class for WrapperPICPat.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146081 91177308-0d34-0410-b5e6-96231b3b80d8

Modify LowerFCOPYSIGN to handle Mips64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146080 91177308-0d34-0410-b5e6-96231b3b80d8

Update bcanalyzer to handle new USELIST_BLOCK/USELIST_CODE_ENTRY.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146079 91177308-0d34-0410-b5e6-96231b3b80d8

Begin adding experimental support for preserving use-list ordering of bitcode
files.  First, add a new block USELIST_BLOCK to the bitcode format.  This is
where USELIST_CODE_ENTRYs will be stored.  The format of the USELIST_CODE_ENTRYs
have not yet been defined.  Add support in the BitcodeReader for parsing the
USELIST_BLOCK.
Part of rdar://9860654 and PR5680.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146078 91177308-0d34-0410-b5e6-96231b3b80d8

These global variables aren't thread-safe, STATISTIC is. Andy Trick tells me
that he isn't using these any more, so just delete them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146076 91177308-0d34-0410-b5e6-96231b3b80d8

Have cmake build llvm-cov.  Patch by arrowdodger.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146071 91177308-0d34-0410-b5e6-96231b3b80d8

ValueEnumerator - debug dump().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146070 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146063 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146062 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 64-bit immediate patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146059 91177308-0d34-0410-b5e6-96231b3b80d8

Nuke inadvertant debugging commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146057 91177308-0d34-0410-b5e6-96231b3b80d8

Darwin assembler improved relocs when w/o subsections_via_symbols.

When the file isn't being built with subsections-via-symbols, symbol
differences involving non-local symbols can be resolved more aggressively.
Needed for gas compatibility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146054 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unneeded semicolon.
Skip two looking up at BlockChain.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146053 91177308-0d34-0410-b5e6-96231b3b80d8

Thumb2 alias for long-form pop and friends.
rdar://10542474

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146046 91177308-0d34-0410-b5e6-96231b3b80d8

Also pass in correct initializer here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146044 91177308-0d34-0410-b5e6-96231b3b80d8

ARM support the .arm and .thumb directives for assembly mode switching.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146042 91177308-0d34-0410-b5e6-96231b3b80d8

Correct initializer in example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146041 91177308-0d34-0410-b5e6-96231b3b80d8

ARM NEON VCLT(register) is a pseudo aliasing VCGT(register).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146039 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up. Move MachO tests to MachO directory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146038 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146037 91177308-0d34-0410-b5e6-96231b3b80d8

When doing "opt -O2" verify the bitcode like is done for
"opt -std-compile-opts".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146036 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bunch of SSE/AVX patterns to use proper memop types. In particular, not using integer loads other than v2i64/v4i64 since the others are all promoted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146031 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust the stack by one pointer size for all frameless stacks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146030 91177308-0d34-0410-b5e6-96231b3b80d8

Fix off-by-one error when encoding the stack size for a frameless stack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146029 91177308-0d34-0410-b5e6-96231b3b80d8

Add bundle aware API for querying instruction properties and switch the code
generator to it. For non-bundle instructions, these behave exactly the same
as the MC layer API.

For properties like mayLoad / mayStore, look into the bundle and if any of the
bundled instructions has the property it would return true.
For properties like isPredicable, only return true if *all* of the bundled
instructions have the property.
For properties like canFoldAsLoad, isCompare, conservatively return false for
bundles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146026 91177308-0d34-0410-b5e6-96231b3b80d8

Adding missing anchor to DATDeltaAlgorithm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146025 91177308-0d34-0410-b5e6-96231b3b80d8

make CR spill and restore 64-bit clean (no functional change), and fix some other problems found with -verify-machineinstrs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146024 91177308-0d34-0410-b5e6-96231b3b80d8

make base register selection used in eliminateFrameIndex 64-bit clean

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146023 91177308-0d34-0410-b5e6-96231b3b80d8

set mayStore and mayLoad on CR pseudos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146022 91177308-0d34-0410-b5e6-96231b3b80d8

64-bit LR8 load should use X11 not R11

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146021 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate delta argument from AdjustBBOffsetsAfter.

The block offset can be computed from the previous block. That is more
robust than keeping track of a delta.

Eliminate one redundant AdjustBBOffsetsAfter call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146018 91177308-0d34-0410-b5e6-96231b3b80d8

Compute some alignment information for each basic block.

These fields are not used for anything yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146017 91177308-0d34-0410-b5e6-96231b3b80d8

Zap unnecessary isIntDivCheap() check.  PR11485.  No testcase because this doesn't affect any in-tree target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146015 91177308-0d34-0410-b5e6-96231b3b80d8

ARM tidy up and remove no longer needed InstAlias definitions.

The TokenAlias handling of data type suffices renders these unnecessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146010 91177308-0d34-0410-b5e6-96231b3b80d8

Move common expression into a method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146008 91177308-0d34-0410-b5e6-96231b3b80d8

ARM Implement ARM ARM Table A7-3 via TokenAlias.

Data type suffix aliasing. Previously handled via lots of instruction
aliases. Cleanup of those forthcoming.

rdar://10435076

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146007 91177308-0d34-0410-b5e6-96231b3b80d8

Group BBSizes and BBOffsets into a single vector<BasicBlockInfo>.

No functional change is intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146005 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146004 91177308-0d34-0410-b5e6-96231b3b80d8

ARM: NEON SHLL instruction immediate operand range checking.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146003 91177308-0d34-0410-b5e6-96231b3b80d8

Support vector bitcasts in the AsmPrinter.  PR11495.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146001 91177308-0d34-0410-b5e6-96231b3b80d8

Add a few moreLocal/Global R_MIPS_GOT related fixups and
make the addend fixup code a bit more generic

Patch by Jack Carter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145998 91177308-0d34-0410-b5e6-96231b3b80d8

Add MachineOperand IsInternalRead flag.

This flag is used when bundling machine instructions.  It indicates
whether the operand reads a value defined inside or outside its bundle.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145997 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an optimization involving EXTRACT_SUBVECTOR in DAGCombine so it behaves correctly.  PR11494.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145996 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unneeded type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145995 91177308-0d34-0410-b5e6-96231b3b80d8

ARM: Parameterize the immediate operand type for NEON VSHLL.

No functional change yet. Will be implementing range-checked immediates
for better diagnostics and disambiguation of instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145994 91177308-0d34-0410-b5e6-96231b3b80d8

- Remove unneeded #includes.
- Remove unused types/fields.
- Add some constantness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145993 91177308-0d34-0410-b5e6-96231b3b80d8

Extend AsmMatcher token literal matching to allow aliasing.

For example, ARM allows:
    vmov.u32 s4, #0  -> vmov.i32, #0
'u32' is a more specific designator for the 32-bit integer type specifier
and is legal for any instruction which accepts 'i32' as a datatype suffix.

We want to say,
    def : TokenAlias<".u32", ".i32">;

This works by marking the match class of 'From' as a subclass of the
match class of 'To'.

rdar://10435076

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145992 91177308-0d34-0410-b5e6-96231b3b80d8

Make Release Notes HTML 4.01 Strict.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145991 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-build: Don't generate duplicate dependencies when LLVMBuild files define
multiple components.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145989 91177308-0d34-0410-b5e6-96231b3b80d8

Add documentation for machine-independent DFA packetizer

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145988 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r145971: "Use conservative size estimate for tBR_JTr."

This caused more offset errors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145980 91177308-0d34-0410-b5e6-96231b3b80d8

Add link to builtin_expect in Release Notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145979 91177308-0d34-0410-b5e6-96231b3b80d8

Re-enable compact unwind. It seems to work now. <rdar://problem/10441838>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145977 91177308-0d34-0410-b5e6-96231b3b80d8

Explicitly check for the different SUB instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145976 91177308-0d34-0410-b5e6-96231b3b80d8

First chunk of MachineInstr bundle support.
1. Added opcode BUNDLE
2. Taught MachineInstr class to deal with bundled MIs
3. Changed MachineBasicBlock iterator to skip over bundled MIs; added an iterator to walk all the MIs
4. Taught MachineBasicBlock methods about bundled MIs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145975 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up. Fix naming convention stuff for some internal functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145974 91177308-0d34-0410-b5e6-96231b3b80d8

Use conservative size estimate for tBR_JTr.

This pseudo-instruction contains a .align directive in its expansion, so
the total size may vary by 2 bytes.

It is too difficult to accurately keep track of this alignment
directive, just use the worst-case size instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145971 91177308-0d34-0410-b5e6-96231b3b80d8

Remove alignment from deserted constant islands.

ARMConstantIslandPass may sometimes leave empty constant islands behind
(it really shouldn't). Remove the alignment from the empty islands so
the size calculations are still correct.

This should fix the many Thumb1 assembler errors in the nightly test
suite.

The reduced test case for this problem is way too big. That is to be
expected for ARMConstantIslandPass bugs.

<rdar://problem/10534709>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145970 91177308-0d34-0410-b5e6-96231b3b80d8

Encode the total stack if there isn't a frame.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145969 91177308-0d34-0410-b5e6-96231b3b80d8

* Add a macro to remove a magic number.
* Rename variables to reflect what they're actually used for.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145968 91177308-0d34-0410-b5e6-96231b3b80d8

Pretty-print basic block alignment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145965 91177308-0d34-0410-b5e6-96231b3b80d8

Add link to llvm.expect in Release Notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145964 91177308-0d34-0410-b5e6-96231b3b80d8

delaying restore-cr changed assigned registers in some tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145963 91177308-0d34-0410-b5e6-96231b3b80d8

add a test case that uses RESTORE_CR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145962 91177308-0d34-0410-b5e6-96231b3b80d8

add RESTORE_CR and support CR unspills

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145961 91177308-0d34-0410-b5e6-96231b3b80d8