Some improvements related to the computation of heights, depths of SUnits.

The basic idea is that all these algorithms are computing the longest paths from the root node or to the exit node. Therefore the existing implementation that uses and iterative and potentially
exponential algorithm was changed to a well-known graph algorithm based on dynamic programming. It has a linear run-time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47884 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor ExpandConstantFP so it can optimize load from constpool of types larger than f64 into extload from smaller types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47883 91177308-0d34-0410-b5e6-96231b3b80d8

Really fix the test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47882 91177308-0d34-0410-b5e6-96231b3b80d8

Fix broken test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47881 91177308-0d34-0410-b5e6-96231b3b80d8

80 column violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47878 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80 column violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47877 91177308-0d34-0410-b5e6-96231b3b80d8

Did I say 'e = getNumOperands()'? I meant --e, of course.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47875 91177308-0d34-0410-b5e6-96231b3b80d8

Add PR1501 test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47874 91177308-0d34-0410-b5e6-96231b3b80d8

Rename isOperand() to isOperandOf() (and other similar methods). It always confuses me.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47872 91177308-0d34-0410-b5e6-96231b3b80d8

Remove -always-fold-and-in-test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47871 91177308-0d34-0410-b5e6-96231b3b80d8

Miscellaneous clean-ups based on Evan's feedback:

- Cleaned up how the prologue-epilogue inserter loops over the instructions.
- Instead of restarting the processing of an instruction if we remove an
  implicit kill, just update the end iterator and make sure that the iterator
  isn't incremented.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47870 91177308-0d34-0410-b5e6-96231b3b80d8

Misc. APInt-ification in the DAGCombiner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47869 91177308-0d34-0410-b5e6-96231b3b80d8

More APInt-ification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47868 91177308-0d34-0410-b5e6-96231b3b80d8

Yet more APInt-ification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47867 91177308-0d34-0410-b5e6-96231b3b80d8

More APInt-ification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47866 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for lowering i64 SRA_PARTS and friends on x86-64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47865 91177308-0d34-0410-b5e6-96231b3b80d8

More APInt-ification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47864 91177308-0d34-0410-b5e6-96231b3b80d8

This is the initial check-in for adding register scavenging to PPC. (Currently,
PPC-64 doesn't work.) This also lowers the spilling of the CR registers so that
it uses a register other than the default R0 register (the scavenger scrounges
for one). A significant part of this patch fixes how kill information is
handled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47863 91177308-0d34-0410-b5e6-96231b3b80d8

Go through the machine instruction's operands to make sure that we're not
marking both a super- and sub-register as "killed". This removes implicit uses
that are marked as "killed".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47862 91177308-0d34-0410-b5e6-96231b3b80d8

Make the register scavenger update the bookkeeping values for sub/super
registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47861 91177308-0d34-0410-b5e6-96231b3b80d8

Multiple instructions can be inserted when eliminating frame indexes. We need
the register scavenger to process all of those new instructions instead of just
the last one inserted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47860 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47858 91177308-0d34-0410-b5e6-96231b3b80d8

s/isReturnStruct()/hasStructRetAttr()/g

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47857 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47849 91177308-0d34-0410-b5e6-96231b3b80d8

If a function uses multive values in ret instruction then it
is returning a struct value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47848 91177308-0d34-0410-b5e6-96231b3b80d8

Preserve paramater attributes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47847 91177308-0d34-0410-b5e6-96231b3b80d8

Renable lto2 build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47845 91177308-0d34-0410-b5e6-96231b3b80d8

Stub out a Path::GetMainExecutable call to find the path to the
main executable of a program.  This needs to be implemented on windows.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47835 91177308-0d34-0410-b5e6-96231b3b80d8

another random note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47831 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47830 91177308-0d34-0410-b5e6-96231b3b80d8

Evan implemented these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47828 91177308-0d34-0410-b5e6-96231b3b80d8

Evan implemented this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47827 91177308-0d34-0410-b5e6-96231b3b80d8

add a testcase for misc vector stuff

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47826 91177308-0d34-0410-b5e6-96231b3b80d8

add support for lvsl, fixing PR1481

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47825 91177308-0d34-0410-b5e6-96231b3b80d8

add support for the sse.cmp.* intrinsics, which fixes sse.isamax with the CBE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47824 91177308-0d34-0410-b5e6-96231b3b80d8

refactor intrinsic handling code out into its own method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47823 91177308-0d34-0410-b5e6-96231b3b80d8

fix printing of undef vectors, this fixes "simple" and "build" in UnitTests/Vector.
Now they all pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47820 91177308-0d34-0410-b5e6-96231b3b80d8

insertelement got the wrong operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47819 91177308-0d34-0410-b5e6-96231b3b80d8

Several changes:
  * Simplify handling of byval, making it easier to understand and more
    consistent.  This fixes PR2065.
  * Clean up and simplify handling of GEPs.  I can actually understand it now!
  * Implement support for GEP'ing into vectors, this fixes
    SingleSource/UnitTests/Vector/build2 among others.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47818 91177308-0d34-0410-b5e6-96231b3b80d8

suppress extra -install_name on Tiger.  The 10.4 tools don't support it.  Fix whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47817 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug I introduced in constant array and constant vector handling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47816 91177308-0d34-0410-b5e6-96231b3b80d8

implement shufflevector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47815 91177308-0d34-0410-b5e6-96231b3b80d8

make operand accessors const-correct

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47814 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new ShuffleVectorInst::getMaskValue method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47813 91177308-0d34-0410-b5e6-96231b3b80d8

implement extractelement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47812 91177308-0d34-0410-b5e6-96231b3b80d8

implement insertelement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47811 91177308-0d34-0410-b5e6-96231b3b80d8

respect isSigned for vector types, fixing sdiv of vectors etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47810 91177308-0d34-0410-b5e6-96231b3b80d8

print the attribute in the right place, this fixes function returning vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47809 91177308-0d34-0410-b5e6-96231b3b80d8

Oops, can't unwind to the entry block (entry block may have no preds).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47808 91177308-0d34-0410-b5e6-96231b3b80d8

vector types are simple types.  This fixes div/rem of vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47807 91177308-0d34-0410-b5e6-96231b3b80d8

Print vector types appropriately.  This gets basic vector code working
(PR1126)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47806 91177308-0d34-0410-b5e6-96231b3b80d8

rename PT -> VT for VectorTypes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47805 91177308-0d34-0410-b5e6-96231b3b80d8

Print i32/i64 integer constants as 1u instead of ((unsigned int)1).
Use dyn_cast better.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47804 91177308-0d34-0410-b5e6-96231b3b80d8

Move pr717 to here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47803 91177308-0d34-0410-b5e6-96231b3b80d8

Add an unwind_to field to basic blocks, making them Users instead of Values.
This is the first checkin for PR1269, the new EH infrastructure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47802 91177308-0d34-0410-b5e6-96231b3b80d8

good catch anton

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47800 91177308-0d34-0410-b5e6-96231b3b80d8

make CAS work

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47799 91177308-0d34-0410-b5e6-96231b3b80d8

all but CAS working on x86

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47798 91177308-0d34-0410-b5e6-96231b3b80d8

Also handle "getresult".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47797 91177308-0d34-0410-b5e6-96231b3b80d8

Print the name, not a pointer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47796 91177308-0d34-0410-b5e6-96231b3b80d8

Add lock prefix support to x86.  Also add the instructions necessary for the atomic ops.  They are still marked pseudo, since I cannot figure out what format to use, but they are the correct opcode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47795 91177308-0d34-0410-b5e6-96231b3b80d8

Remove llvm-upgrade and update test cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47793 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR2113 by verifying allocations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47792 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47791 91177308-0d34-0410-b5e6-96231b3b80d8

allow specified inline threshold to be negative, as the value is
itself sometimes negative.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47786 91177308-0d34-0410-b5e6-96231b3b80d8

Remove llvm-upgrade and update tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47784 91177308-0d34-0410-b5e6-96231b3b80d8

Disable lto because its broken on tiger.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47783 91177308-0d34-0410-b5e6-96231b3b80d8

Add MVT::is128BitVector and is64BitVector.  Shrink
unaligned load/store code using them.  Per review
of unaligned load/store vector patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47782 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor / clean up code; remove td list scheduler special tie breaker (no real benefit).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47779 91177308-0d34-0410-b5e6-96231b3b80d8

Fix cut-n-pasto.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47777 91177308-0d34-0410-b5e6-96231b3b80d8

Add pass to promote sret.
This pass transforms

  %struct._Point = type { i32, i32, i32, i32, i32, i32 }
  define internal void @foo(%struct._Point* sret  %agg.result)

into

  %struct._Point = type { i32, i32, i32, i32, i32, i32 }
  define internal %struct._Point @foo()

This pass updates foo() clients appropriately to use
getresult instruction to extract return values.

This pass is not yet ready for prime time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47776 91177308-0d34-0410-b5e6-96231b3b80d8

Use enumeration for preffered EH dwarf encoding reason

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47770 91177308-0d34-0410-b5e6-96231b3b80d8

Don't fill eh frames even though these are text sections.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47765 91177308-0d34-0410-b5e6-96231b3b80d8

update to document new lto API

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47764 91177308-0d34-0410-b5e6-96231b3b80d8

install new lto dylib.  pass down LLVM_SUBMIT_VERSION

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47763 91177308-0d34-0410-b5e6-96231b3b80d8

stop building lto on all platforms.  Start building lto2 on Darwin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47762 91177308-0d34-0410-b5e6-96231b3b80d8

fix default RC_ProjectSourceSubversion to be zero

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47761 91177308-0d34-0410-b5e6-96231b3b80d8

DCE'ed this testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47760 91177308-0d34-0410-b5e6-96231b3b80d8

If we reload a virtual register that's already been assigned, we want to mark
that instruction as its "last use". This fixes PR1925.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47758 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR2112: don't run loop aligner if target doesn't have a TargetLowering object.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47755 91177308-0d34-0410-b5e6-96231b3b80d8

fix a bug Anders ran into where scalarrepl would crash when promoting
a union containing a vector and an array whose elements were smaller than
the vector elements.  this means we need to compile the load of the
array elements into an extract element plus a truncate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47752 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor some code out of ConvertUsesToScalar into their own methods, no
functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47751 91177308-0d34-0410-b5e6-96231b3b80d8

Folding or(fcmp,fcmp) only works if the operands of the fcmps are the same fp type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47750 91177308-0d34-0410-b5e6-96231b3b80d8

No need for coalescer to update kills. Only copies are coalesced and those instructions will be deleted. Doh.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47749 91177308-0d34-0410-b5e6-96231b3b80d8

Remove redundant #include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47748 91177308-0d34-0410-b5e6-96231b3b80d8

also pass -compatibility_version when linking shared object for Darwin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47747 91177308-0d34-0410-b5e6-96231b3b80d8

More APInt-ification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47746 91177308-0d34-0410-b5e6-96231b3b80d8

Use the new convertFromAPInt instead of convertFromZeroExtendedInteger,
which allows more of the surrounding arithmetic to be done with APInt
instead of uint64_t.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47745 91177308-0d34-0410-b5e6-96231b3b80d8

Use the new convertFromAPInt instead of convertFromZeroExtendedInteger.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47744 91177308-0d34-0410-b5e6-96231b3b80d8

Use the new APInt-enabled form of getConstant instead of converting
an APInt into a uint64_t to call getConstant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47742 91177308-0d34-0410-b5e6-96231b3b80d8

Add support to APInt for shift and rotate operations with APInt
instead of uint32_t for the shift/rotate count operand type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47741 91177308-0d34-0410-b5e6-96231b3b80d8

Use the correct instruction encodings for the 64-bit MMX movd.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47740 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify code using convertFromZeroExtendedInteger with an APInt
by using the new convertFromAPInt directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47739 91177308-0d34-0410-b5e6-96231b3b80d8

Add a method to APFloat to convert directly from APInt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47738 91177308-0d34-0410-b5e6-96231b3b80d8

Added option -align-loops=<true/false> to disable loop aligner pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47736 91177308-0d34-0410-b5e6-96231b3b80d8

Update testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47735 91177308-0d34-0410-b5e6-96231b3b80d8

Interface of getByValTypeAlignment differed between
generic & x86 versions; change generic to follow x86
and improve comments.  Add PPC version (not right
for non-Darwin.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47734 91177308-0d34-0410-b5e6-96231b3b80d8

Make llvm-ar behave like ar, if you create an empty archive, ar creates an empty archive.  llvm-ar would not generate an output file in this case

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47733 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize TextAlignFillValue. Fix revision 47703.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47732 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid collisions in value names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47725 91177308-0d34-0410-b5e6-96231b3b80d8