Simplify getIntPtrType, allowing it to work for arbitrary pointer sizes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42751 91177308-0d34-0410-b5e6-96231b3b80d8

Use correct parentheses with the '&& "..."' idiom in an assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42750 91177308-0d34-0410-b5e6-96231b3b80d8

In -debug mode, dump SelectionDAGs both before and after the
optimization passes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42749 91177308-0d34-0410-b5e6-96231b3b80d8

Fix grammar in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42748 91177308-0d34-0410-b5e6-96231b3b80d8

Add explicit keywords.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42747 91177308-0d34-0410-b5e6-96231b3b80d8

Use APInt::tcExtract.  It's cleaner, and works :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42746 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new function tcExtract for extracting a bignum from an
arbitrary range of bits embedded in the middle of another bignum.
This kind of operation is desirable in many cases of software
floating point, e.g. converting bignum integers to floating point
numbers of fixed precision (you want to extract the precision most
significant bits).

Elsewhere, add an assertion, and exit the shift functions early if
the shift count is zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42745 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed downrev Ocaml compatibility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42744 91177308-0d34-0410-b5e6-96231b3b80d8

Begin the process of allowing DomTree on MBB's.  Step One: template DomTreeNode by making it a typedef of a templated DomTreeNodeBase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42743 91177308-0d34-0410-b5e6-96231b3b80d8

Kill cycle of an live range is always the last use index + 1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42742 91177308-0d34-0410-b5e6-96231b3b80d8

Allow passes to just pass up "FunctionPass(&ID)" instead of "FunctionPass((intptr_t)&ID)"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42741 91177308-0d34-0410-b5e6-96231b3b80d8

C and Objective Caml bindings for getFunction and getNamedGlobal. Also enhanced
the Objective Caml 'declare_*' functions to behave more or less like
getOrInsertFunction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42740 91177308-0d34-0410-b5e6-96231b3b80d8

disable this entirely: it is causing use of invalidated iterators and infinite looping.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42739 91177308-0d34-0410-b5e6-96231b3b80d8

Fix many regressions on x86 by avoiding dereferencing the end iterator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42738 91177308-0d34-0410-b5e6-96231b3b80d8

Finish off PR1723, by working around some strange compiler bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42737 91177308-0d34-0410-b5e6-96231b3b80d8

C and Objective Caml bindings for GlobalVariable::isConstant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42736 91177308-0d34-0410-b5e6-96231b3b80d8

Add back convertFromSignExtendedInteger.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42735 91177308-0d34-0410-b5e6-96231b3b80d8

Now that convertFromUnsignedParts has a sane, constant interface,
convertFromZeroExtendedInteger can be simplified as it doesn't need
to make a copy of the source bignum.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42734 91177308-0d34-0410-b5e6-96231b3b80d8

Reimplement convertFromUnsignedInteger so it is passed a const bignum.
It used to modify its argument in-place.

This interface is saner and the implementation more efficient.  It will
be needed for decimal->binary conversion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42733 91177308-0d34-0410-b5e6-96231b3b80d8

convertFromInteger, as originally written, expected sign-extended
input.  APInt unfortunately zero-extends signed integers, so Dale
modified the function to expect zero-extended input.  Make this
assumption explicit in the function name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42732 91177308-0d34-0410-b5e6-96231b3b80d8

combineLostFractions does not need to be a member function

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42729 91177308-0d34-0410-b5e6-96231b3b80d8

Some more problems noticed by Curtis

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42717 91177308-0d34-0410-b5e6-96231b3b80d8

C and Objective Caml bindings for PATypeHolder.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42713 91177308-0d34-0410-b5e6-96231b3b80d8

Bindings for the verifier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42707 91177308-0d34-0410-b5e6-96231b3b80d8

Deleting some unnecessary glue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42702 91177308-0d34-0410-b5e6-96231b3b80d8

Adopting a uniform naming convention for type constructors in
bindings (part le deux).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42701 91177308-0d34-0410-b5e6-96231b3b80d8

Oops, I really wanted to commit this part also :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42700 91177308-0d34-0410-b5e6-96231b3b80d8

Move merge code into new helper function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42699 91177308-0d34-0410-b5e6-96231b3b80d8

Adopting a uniform naming convention for type constructors in bindings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42698 91177308-0d34-0410-b5e6-96231b3b80d8

Adopting a uniform naming convention for constant constructors in bindings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42697 91177308-0d34-0410-b5e6-96231b3b80d8

Adding C and Ocaml bindings for ConstantExpr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42696 91177308-0d34-0410-b5e6-96231b3b80d8

Include DataTypes.h to provide intptr_t type. This fixes OpenBSD compilation of this file.
Inspired by Laurence Tratt's patch

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42695 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply 42677.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42692 91177308-0d34-0410-b5e6-96231b3b80d8

Oops. Forgot this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42691 91177308-0d34-0410-b5e6-96231b3b80d8

Cleaner, more general exponent output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42690 91177308-0d34-0410-b5e6-96231b3b80d8

silence warnings in no-assert build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42687 91177308-0d34-0410-b5e6-96231b3b80d8

revert evan's patch until the header is committed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42686 91177308-0d34-0410-b5e6-96231b3b80d8

Added DAG xforms. e.g.
(vextract (v4f32 s2v (f32 load $addr)), 0) -> (f32 load $addr)
(vextract (v4i32 bc (v4f32 s2v (f32 load $addr))), 0) -> (i32 load $addr)
Remove x86 specific patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42677 91177308-0d34-0410-b5e6-96231b3b80d8

Next powerpc long double bits.  Comparisons work,
although not well, and shortening FP converts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42672 91177308-0d34-0410-b5e6-96231b3b80d8

Make test work on non-x86 hosts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42671 91177308-0d34-0410-b5e6-96231b3b80d8

Fix and clarify some comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42670 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize tcFullMultiply so that the operands can be of differing
part widths.  Also, return the number of parts actually required to
hold the result's value.
Remove an over-cautious condition from rounding of float->hex conversion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42669 91177308-0d34-0410-b5e6-96231b3b80d8

Wrapping Value::dump.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42668 91177308-0d34-0410-b5e6-96231b3b80d8

Removing the silly CHelpers header by rolling wrap and unwrap into the C
bindings headers themselves, hidden behind #ifdef __cplusplus.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42666 91177308-0d34-0410-b5e6-96231b3b80d8

Test case for 3-address conversion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42664 91177308-0d34-0410-b5e6-96231b3b80d8

Commute x86 cmove instructions by swapping the operands and change the condition
to its inverse.
Testing this as llcbeta

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42661 91177308-0d34-0410-b5e6-96231b3b80d8

This is done.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42656 91177308-0d34-0410-b5e6-96231b3b80d8

Enable convertToThreeAddress for X86 by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42655 91177308-0d34-0410-b5e6-96231b3b80d8

Fix bug in updating dominance frontier after loop
unswitch when frontier includes basic blocks that
are not inside loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42654 91177308-0d34-0410-b5e6-96231b3b80d8

INC64_32r -> LEA64_32r is better than INC64_32r -> LEA32r, but it still can
cause performance degradation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42653 91177308-0d34-0410-b5e6-96231b3b80d8

In 64-bit mode, avoid using leal with 32-bit 32-bit address size, e.g.
leal 1(%ecx), %edi, which requires 67H prefix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42647 91177308-0d34-0410-b5e6-96231b3b80d8

First round of ppc long double.  call/return and
basic arithmetic works.
Rename RTLIB long double functions to distinguish
different flavors of long double; the lib functions
have different names, alas.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42644 91177308-0d34-0410-b5e6-96231b3b80d8

Add support to convert more 64-bit instructions to 3-address instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42642 91177308-0d34-0410-b5e6-96231b3b80d8

ADC and SBB uses EFLAGS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42640 91177308-0d34-0410-b5e6-96231b3b80d8

Mark count method const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42639 91177308-0d34-0410-b5e6-96231b3b80d8

Change a few more spaces to tabs in assembly output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42638 91177308-0d34-0410-b5e6-96231b3b80d8

Change a space to a tab in the assembly output of a .globl directive
for consistency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42637 91177308-0d34-0410-b5e6-96231b3b80d8

Legalize support for MUL_LOHI and DIVREM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42636 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42635 91177308-0d34-0410-b5e6-96231b3b80d8

Provide names for MUL_LOHI and DIVREM operators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42634 91177308-0d34-0410-b5e6-96231b3b80d8

SMUL_LOHI and UMUL_LOHI are commutative.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42633 91177308-0d34-0410-b5e6-96231b3b80d8

Define target-indepenent SDNode types for multiply and divide that
produce two results.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42632 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo in comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42631 91177308-0d34-0410-b5e6-96231b3b80d8

Testing convertToThreeeAddress as X86 llcbeta.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42630 91177308-0d34-0410-b5e6-96231b3b80d8

Fix intersectWithComplement bug noticed by Curtis Dunham
Optimize |= case for empty RHS

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42629 91177308-0d34-0410-b5e6-96231b3b80d8

New test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42628 91177308-0d34-0410-b5e6-96231b3b80d8

Chain producing nodes cannot be moved, not chain reading nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42627 91177308-0d34-0410-b5e6-96231b3b80d8

Oops. Didn't mean to leave this in.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42626 91177308-0d34-0410-b5e6-96231b3b80d8

If a node that defines a physical register that is expensive to copy. The
scheduler will try a number of tricks in order to avoid generating the
copies. This may not be possible in case the node produces a chain value
that prevent movement. Try unfolding the load from the node before to allow
it to be moved / cloned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42625 91177308-0d34-0410-b5e6-96231b3b80d8

Added storeRegToAddr, loadRegFromAddr, and unfoldMemoryOperand's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42624 91177308-0d34-0410-b5e6-96231b3b80d8

Not needed any more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42623 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42622 91177308-0d34-0410-b5e6-96231b3b80d8

- Added a few target hooks to generate load / store instructions from / to any
address (not just from / to frameindexes).
- Added target hooks to unfold load / store instructions / SDNodes into separate
load, data processing, store instructions / SDNodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42621 91177308-0d34-0410-b5e6-96231b3b80d8

Add a variant of getTargetNode() that takes a vector of MVT::ValueType.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42620 91177308-0d34-0410-b5e6-96231b3b80d8

Silence a warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42619 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the previous bug a slightly different way (by modifying how find_next works)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42613 91177308-0d34-0410-b5e6-96231b3b80d8

Fix off by one error in iterator

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42612 91177308-0d34-0410-b5e6-96231b3b80d8

Support Objective C++ too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42610 91177308-0d34-0410-b5e6-96231b3b80d8

Added testcase for creation of metadata with null pointers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42609 91177308-0d34-0410-b5e6-96231b3b80d8

add a note.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42607 91177308-0d34-0410-b5e6-96231b3b80d8

un-XFAIL this test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42606 91177308-0d34-0410-b5e6-96231b3b80d8

It helps a lot when you check for the end of your list before randomly
dereferencing things.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42605 91177308-0d34-0410-b5e6-96231b3b80d8

Make SparseBitVector::set actually work properly when sets are not in ascending or descending order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42604 91177308-0d34-0410-b5e6-96231b3b80d8

Do use the actual ocaml stdlib (not the install dir) to find the
caml/*.h headers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42599 91177308-0d34-0410-b5e6-96231b3b80d8

Add APFloat -> hexadecimal string conversion, as per %a and %A in C99.
Useful for diagnostics and debugging.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42598 91177308-0d34-0410-b5e6-96231b3b80d8

Completely merge the implementation details of DomTree and PostDomTree.

Also, add a FIXME for a bug in PostDomTree calculation I noticed while writing this,

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42593 91177308-0d34-0410-b5e6-96231b3b80d8

Add a GraphTraits partial specialization to make the inverse of an inverse be the same as the underlying graph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42592 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80 col violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42591 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor code in a separate method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42590 91177308-0d34-0410-b5e6-96231b3b80d8

Add initial iterator support for folding set.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42589 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify that shifts that are too large are undefined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42588 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify implementation of the FoldingSet circular list, a necessary step
to giving it iterators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42586 91177308-0d34-0410-b5e6-96231b3b80d8

Use empty() member functions when that's what's being tested for instead
of comparing begin() and end().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42585 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a  using namespace llvm;  in a header file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42584 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily XFAIL this test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42583 91177308-0d34-0410-b5e6-96231b3b80d8

Tone down an overzealous optimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42582 91177308-0d34-0410-b5e6-96231b3b80d8

remove extraneous tag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42581 91177308-0d34-0410-b5e6-96231b3b80d8

remove misleading documentation: LLVM has no atomic support yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42580 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42579 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1719, by not marking llvm.global.annotations internal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42578 91177308-0d34-0410-b5e6-96231b3b80d8