Darwin ABI issues: weak, linkonce, etc. dynamic-no-pic support is complete.
Also fixed a function stub bug. Added weak and linkonce support for
x86 Linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26038 91177308-0d34-0410-b5e6-96231b3b80d8

Remind myself to add PIC and static asm printer support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26037 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize MaskedValueIsZero into a ComputeMaskedNonZeroBits function, which
is just as efficient as MVIZ and is also more general.

Fix a few minor bugs introduced in recent patches

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26036 91177308-0d34-0410-b5e6-96231b3b80d8

Make MaskedValueIsZero take a uint64_t instead of a ConstantIntegral as a
mask.  This allows the code to be simpler and more efficient.

Also, generalize some of the cases in MVIZ a bit, making it slightly more aggressive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26035 91177308-0d34-0410-b5e6-96231b3b80d8

Use Type::getIntegralTypeMask() to simplify some code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26034 91177308-0d34-0410-b5e6-96231b3b80d8

Implement the beginnings of a facility for simplifying expressions based on
'demanded bits', inspired by Nate's work in the dag combiner.  This isn't
complete, but needs to unrelated instcombiner changes to continue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26033 91177308-0d34-0410-b5e6-96231b3b80d8

add a new Type::getIntegralTypeMask() method, which is useful for clients that
want to do bitwise inspection of integer types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26032 91177308-0d34-0410-b5e6-96231b3b80d8

The interpreter assumes that the caller of runFunction() must be lli, and
therefore the function being called must be a main() returning an int.  The
consequences when these assumptions are false are not good, so don't assume
them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26031 91177308-0d34-0410-b5e6-96231b3b80d8

Teach the interpreter to handle global variables that are added to a module after
interpretation has begun.  The JIT already handles this situation correctly, and
the interpreter can already handle new functions being added.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26030 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some truncation warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26029 91177308-0d34-0410-b5e6-96231b3b80d8

fix an error compiling with -pedantic

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26028 91177308-0d34-0410-b5e6-96231b3b80d8

Add the simple PPC integer constraints

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26027 91177308-0d34-0410-b5e6-96231b3b80d8

Hoist all SDOperand declarations within a Select_{opcode}() to the top level
to reduce stack memory usage. This is intended to work around the gcc bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26026 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate the printCallOperand method, using a 'call' modifier on
printOperand instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26025 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for modifier strings in machine instr descriptions.  This allows
us to avoid creating lots of "Operand" types with different printers, instead
we can fold several together and use modifiers.  For example, we can now use:

${target:call} to say that the operand should be printed like a 'call' operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26024 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify the variant handling code, no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26023 91177308-0d34-0410-b5e6-96231b3b80d8

Change prototype

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26022 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for modifier characters to operand printers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26021 91177308-0d34-0410-b5e6-96231b3b80d8

Change the prototype of PrintAsmOperand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26020 91177308-0d34-0410-b5e6-96231b3b80d8

Goodbye nasty macro.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26019 91177308-0d34-0410-b5e6-96231b3b80d8

Edit requests from Sabre.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26018 91177308-0d34-0410-b5e6-96231b3b80d8

see what this allignment thing will do

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26017 91177308-0d34-0410-b5e6-96231b3b80d8

Changing model for the construction of debug information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26016 91177308-0d34-0410-b5e6-96231b3b80d8

We seem to have settled to __DWARF for section name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26015 91177308-0d34-0410-b5e6-96231b3b80d8

At the end of isel, select a replacement node for each handle that does not
have one. This can happen if a load's real uses are dead (i.e. they do not
have uses themselves).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26014 91177308-0d34-0410-b5e6-96231b3b80d8

Name change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26013 91177308-0d34-0410-b5e6-96231b3b80d8

- Update load folding checks to match those auto-generated by tblgen.
- Manually select SDOperand's returned by TryFoldLoad which make up the
  load address.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26012 91177308-0d34-0410-b5e6-96231b3b80d8

Handle HANDLENODE: just return itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26011 91177308-0d34-0410-b5e6-96231b3b80d8

Complex pattern isel code shouldn't select nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26010 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the Sparc backend with Evan's recent tblgen changes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26009 91177308-0d34-0410-b5e6-96231b3b80d8

SparcV8 -> Sparc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26008 91177308-0d34-0410-b5e6-96231b3b80d8

This xform isn't safe

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26007 91177308-0d34-0410-b5e6-96231b3b80d8

Back out previous commit, it isn't safe.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26006 91177308-0d34-0410-b5e6-96231b3b80d8

fold c1 << (x + c2) into (c1 << c2) << x.  fix a warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26005 91177308-0d34-0410-b5e6-96231b3b80d8

some stuff is done

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26004 91177308-0d34-0410-b5e6-96231b3b80d8

Turn A % (C << N), where C is 2^k, into A & ((C << N)-1) [urem only].
Turn A / (C1 << N), where C1 is "1<<C2" into A >> (N+C2) [udiv only].

Tested with: rem.ll:test5, div.ll:test10

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26003 91177308-0d34-0410-b5e6-96231b3b80d8

new testcases

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26002 91177308-0d34-0410-b5e6-96231b3b80d8

Handle urem by shifted powers of 2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26001 91177308-0d34-0410-b5e6-96231b3b80d8

handle combining A / (B << N) into A >>u (log2(B)+N) when B is a power of 2

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26000 91177308-0d34-0410-b5e6-96231b3b80d8

Use SelectRoot() as the entry to any tblgen based isel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25998 91177308-0d34-0410-b5e6-96231b3b80d8

Use SelectRoot() as entry of any tblgen based isel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25997 91177308-0d34-0410-b5e6-96231b3b80d8

Encourage use of the V8 ABI over the V9 ABI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25996 91177308-0d34-0410-b5e6-96231b3b80d8

Allow more loads to be folded which were previously prevented from happening
due to ordering issue. i.e. they were selected for chain use first.
Now at load select time, check if it is being selected for a chain use and if
it has only a single real use. If so, return a HANDLENODE (with the load as
its operand) in its place and record it.
When it is folded or the load is selected for a real use, the isel records it
as the replacement for the HANDLENODE. The replacement is done when all nodes
are selected.
This scheme exposed a couple of problems where cycles can happen. (See comments
in EmitMatchCode() for descriptions of the problems and their workaround /
solutions.) These problems have been resolved with a small compile time
penality.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25995 91177308-0d34-0410-b5e6-96231b3b80d8

This document is out of date.  :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25994 91177308-0d34-0410-b5e6-96231b3b80d8

V8 -> Sparc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25993 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the SparcV8 backend.  It has been renamed to be the Sparc backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25992 91177308-0d34-0410-b5e6-96231b3b80d8

remove V8 reference

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25991 91177308-0d34-0410-b5e6-96231b3b80d8

* Added SDNode::isOnlyUse().
* Fix hasNUsesOfValue(), it should be const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25990 91177308-0d34-0410-b5e6-96231b3b80d8

SparcV8 -> Sparc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25989 91177308-0d34-0410-b5e6-96231b3b80d8

SparcV8 -> Sparc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25988 91177308-0d34-0410-b5e6-96231b3b80d8

These were moved to ../SPARC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25987 91177308-0d34-0410-b5e6-96231b3b80d8

move V8 testcases here

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25986 91177308-0d34-0410-b5e6-96231b3b80d8

Rename SPARC V8 target to be the LLVM SPARC target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25985 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25984 91177308-0d34-0410-b5e6-96231b3b80d8

Re-commit the last bit of change that was backed out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25983 91177308-0d34-0410-b5e6-96231b3b80d8

Re-committing the last bit of change. It shouldn't break PPC this time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25982 91177308-0d34-0410-b5e6-96231b3b80d8

make sure that global doubles are aligned to 8 bytes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25981 91177308-0d34-0410-b5e6-96231b3b80d8

Use getPreferredAlignmentLog.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25980 91177308-0d34-0410-b5e6-96231b3b80d8

Use the asmprinter to find out what the preferred alignment of a global is.
This patch speeds up 172.mgrid from 31.81s to 11.39s on darwin/ppc.
Many many thanks to Nate for tracking down the root cause of the issue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25979 91177308-0d34-0410-b5e6-96231b3b80d8

Implement the AsmPrinter::getPreferredAlignmentLog method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25978 91177308-0d34-0410-b5e6-96231b3b80d8

add a new method, getPreferredAlignmentLog.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25977 91177308-0d34-0410-b5e6-96231b3b80d8

linkage fix for weak functions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25976 91177308-0d34-0410-b5e6-96231b3b80d8

Fix VC++ warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25975 91177308-0d34-0410-b5e6-96231b3b80d8

Use SCEVExpander::InsertCastOfTo instead of our own code.  This reduces
#LLVM LOC, and auto-cse's cast instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25974 91177308-0d34-0410-b5e6-96231b3b80d8

Pull the InsertCastOfTo out of the header, implement CSE'ing of arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25973 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor a bunch of code into a non-inlined method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25972 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert this patch, which probably breaks with the
tblgen patch reverted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25971 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert the last change, which breaks PPC and other targets that
DO select things.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25970 91177308-0d34-0410-b5e6-96231b3b80d8

Value# select instructions, allowing -gcse to remove duplicates

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25969 91177308-0d34-0410-b5e6-96231b3b80d8

Complex pattern's custom matcher should not call Select() on any operands.
Select them afterwards if it returns true.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25968 91177308-0d34-0410-b5e6-96231b3b80d8

Custom lower VAARG for the case when we are doing vaarg(double).  In this
case, the double being loaded may not be 8-byte aligned, so we have to use
our standard bit_convert game.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25967 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a nasty typo that broke functions with big stack frames.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25966 91177308-0d34-0410-b5e6-96231b3b80d8

fix a bug in my last checkin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25965 91177308-0d34-0410-b5e6-96231b3b80d8

Fix two significant bugs in LSR:

1. When rewriting code in outer loops, sometimes we would insert code into
   inner loops that is invariant in that loop.
2. Notice that 4*(2+x) is 8+4*x and use that to simplify expressions.

This is a performance neutral change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25964 91177308-0d34-0410-b5e6-96231b3b80d8

Remove some stuff that now works

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25963 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25962 91177308-0d34-0410-b5e6-96231b3b80d8

Two changes:

1. Treat FMOVD as a copy instruction, to help with coallescing in V9 mode
2. When in V9 mode, insert FMOVD instead of FpMOVD instructions, as we don't
   ever rewrite FpMOVD instructions into FMOVS instructions, thus we just end
   up with commented out copies!
This should fix a bunch of failures in V9 mode on sparc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25961 91177308-0d34-0410-b5e6-96231b3b80d8

Get rid of some memory leaks identified by Valgrind

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25960 91177308-0d34-0410-b5e6-96231b3b80d8

add a method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25959 91177308-0d34-0410-b5e6-96231b3b80d8

Let bugpoint work on sparc with v9 instructions enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25958 91177308-0d34-0410-b5e6-96231b3b80d8

Fix VC++ warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25957 91177308-0d34-0410-b5e6-96231b3b80d8

Keep Visual Studio informed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25956 91177308-0d34-0410-b5e6-96231b3b80d8

Add initial support for immediates.  This allows us to compile this:

int %rlwnm(int %A, int %B) {
  %C = call int asm "rlwnm $0, $1, $2, $3, $4", "=r,r,r,n,n"(int %A, int %B, int 4, int 17)
  ret int %C
}

into:

_rlwnm:
        or r2, r3, r3
        or r3, r4, r4
        rlwnm r2, r2, r3, 4, 17    ;; note the immediates :)
        or r3, r2, r2
        blr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25955 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an unnecessary predicate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25954 91177308-0d34-0410-b5e6-96231b3b80d8

Separate FILD and FILD_FLAG, the later is only used for SSE2. It produces a
flag so it can be flagged to a FST.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25953 91177308-0d34-0410-b5e6-96231b3b80d8

Initial early support for non-register operands, like immediates

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25952 91177308-0d34-0410-b5e6-96231b3b80d8

implementation of some methods for inlineasm

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25951 91177308-0d34-0410-b5e6-96231b3b80d8

Add some methods for inline asm support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25950 91177308-0d34-0410-b5e6-96231b3b80d8

Handle another case exposed on X86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25949 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a nasty problem on two-address machines in the following situation:

store EAX -> [ss#0]
[ss#0] += 1
...
use(EAX)

In this case, it is not valid to rewrite this as:

store EAX -> [ss#0]
EAX += 1
store EAX -> [ss#0]  ;;; this would also delete the store above
...
use(EAX)

... because EAX is not a dead at that point.  Keep track of which registers
we are allowed to clobber, and which ones we aren't, and don't clobber the
ones we're not supposed to.  :)

This should resolve the issues on X86 last night.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25948 91177308-0d34-0410-b5e6-96231b3b80d8

significantly simplify the VirtRegMap code by pulling the SpillSlotsAvailable
and PhysRegsAvailable maps out into a new AvailableSpills struct.  No
functionality change.

This paves the way for a bugfix, coming up next.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25947 91177308-0d34-0410-b5e6-96231b3b80d8

Implement some feedback from sabre

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25946 91177308-0d34-0410-b5e6-96231b3b80d8

Add a framework for eliminating instructions that produces undemanded bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25945 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25944 91177308-0d34-0410-b5e6-96231b3b80d8

another case Nate came up with

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25943 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25942 91177308-0d34-0410-b5e6-96231b3b80d8

remove some #ifdef'd out code, which should properly be in the dag combiner anyway.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25941 91177308-0d34-0410-b5e6-96231b3b80d8

remove an old comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25940 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the X86PeepholeOptimizerPass, a truly horrible old hack that is now
obsolete.  yaay :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25939 91177308-0d34-0410-b5e6-96231b3b80d8

When rewriting frame instructions, emit the appropriate small-immediate
instruction when possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25938 91177308-0d34-0410-b5e6-96231b3b80d8