New test for PR1417

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37106 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid a "loss of precision" error in gcc 4.1.3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37105 91177308-0d34-0410-b5e6-96231b3b80d8

Output exception call-sites in address order, as required by the unwinding
runtime.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37104 91177308-0d34-0410-b5e6-96231b3b80d8

Conditional branch is not a barrier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37103 91177308-0d34-0410-b5e6-96231b3b80d8

Use a ptr set instead of a linear search to unique TokenFactor operands.
This fixes PR1423

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37102 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for maskmovq

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37101 91177308-0d34-0410-b5e6-96231b3b80d8

implement the missing maskmovq mmx intrinsic that akor hit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37100 91177308-0d34-0410-b5e6-96231b3b80d8

add missing mmx intrinsic

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37099 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37098 91177308-0d34-0410-b5e6-96231b3b80d8

Devang points out that we need an assertion here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37097 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37096 91177308-0d34-0410-b5e6-96231b3b80d8

Implement printing of instruction result values when debug info is turned
on. This helps to speed up the debugging time by showing computational
results as the program executes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37095 91177308-0d34-0410-b5e6-96231b3b80d8

Bug fix: should check ABI alignment, not pref. alignment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37094 91177308-0d34-0410-b5e6-96231b3b80d8

Hooks for predication support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37093 91177308-0d34-0410-b5e6-96231b3b80d8

Initial commit of (very basic) if converter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37092 91177308-0d34-0410-b5e6-96231b3b80d8

Add TargetInstrInfo predication hooks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37091 91177308-0d34-0410-b5e6-96231b3b80d8

New test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37090 91177308-0d34-0410-b5e6-96231b3b80d8

Remove extra CFG edges before doing these passes; it makes them happier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37089 91177308-0d34-0410-b5e6-96231b3b80d8

add PR number and testcase from the PR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37088 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37087 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an infinite recursion in GetNegatedExpression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37086 91177308-0d34-0410-b5e6-96231b3b80d8

Don't put test cases into regression test suite before the bug is fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37085 91177308-0d34-0410-b5e6-96231b3b80d8

The index into the actions table is a ULEB128 not a SLEB128.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37084 91177308-0d34-0410-b5e6-96231b3b80d8

Note on current implementation of ELF32/PPC ABI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37083 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for PR1420

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37082 91177308-0d34-0410-b5e6-96231b3b80d8

selects can also reach here

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37081 91177308-0d34-0410-b5e6-96231b3b80d8

implement the ModuleProvider::dematerializeFunction hook

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37080 91177308-0d34-0410-b5e6-96231b3b80d8

Un-brain-dead-ify the lowering of part set for the reverse case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37071 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CodeGen/PowerPC/2007-05-14-InlineAsmSelectCrash.ll, the other recent
patches are also needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37070 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some subtle issues handling immediate values.  This fixes
test/CodeGen/ARM/2007-05-14-InlineAsmCstCrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37069 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase that crashes llc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37068 91177308-0d34-0410-b5e6-96231b3b80d8

fix some subtle inline asm selection issues

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37067 91177308-0d34-0410-b5e6-96231b3b80d8

Add PredicateOperand to all ARM instructions that have the condition field.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37066 91177308-0d34-0410-b5e6-96231b3b80d8

fix subtle bugs in inline asm operand selection

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37065 91177308-0d34-0410-b5e6-96231b3b80d8

Added getNumExplicitOperands and findFirstPredOperand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37064 91177308-0d34-0410-b5e6-96231b3b80d8

Added \!con(a,b) syntax to concatnate two dag fragments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37063 91177308-0d34-0410-b5e6-96231b3b80d8

All operands that made up of the predicate operands are maked M_PREDICATE_OPERAND.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37062 91177308-0d34-0410-b5e6-96231b3b80d8

Mark all (not just the first) predicate operand M_PREDICATE_OPERAND.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37061 91177308-0d34-0410-b5e6-96231b3b80d8

PredicateOperand related bug fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37060 91177308-0d34-0410-b5e6-96231b3b80d8

testcase that crashes llc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37059 91177308-0d34-0410-b5e6-96231b3b80d8

Correct a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37058 91177308-0d34-0410-b5e6-96231b3b80d8

Fix Transforms/InstCombine/2007-05-14-Crash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37057 91177308-0d34-0410-b5e6-96231b3b80d8

New testcase that crashes instcombine

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37056 91177308-0d34-0410-b5e6-96231b3b80d8

Fix previous patch. GOTOFF can be used only when the symbol has internal
linkage or hidden visibility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37055 91177308-0d34-0410-b5e6-96231b3b80d8

implement a simple fneg optimization/propagation thing.  This compiles:
CodeGen/PowerPC/fneg.ll into:

_t4:
        fmul f0, f3, f4
        fmadd f1, f1, f2, f0
        blr

instead of:

_t4:
        fneg f0, f3
        fmul f0, f0, f4
        fmsub f1, f1, f2, f0
        blr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37054 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37053 91177308-0d34-0410-b5e6-96231b3b80d8

Only worry about intervening kill if there are more than one live ranges in the interval.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37052 91177308-0d34-0410-b5e6-96231b3b80d8

Test for PR1406.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37051 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for PR1406:

v1 =
r2 = move v1
   = op r2<kill>
...
r2 = move v1
   = op r2<kill>

Clear the first r2 kill if v1 and r2 are joined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37050 91177308-0d34-0410-b5e6-96231b3b80d8

New test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37048 91177308-0d34-0410-b5e6-96231b3b80d8

Doh. .cpp -> .ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37047 91177308-0d34-0410-b5e6-96231b3b80d8

When marking a register as being implicitly defined, make sure to clear its partial use info as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37046 91177308-0d34-0410-b5e6-96231b3b80d8

Optimize PIC implementation. GOTOFF can be used when the symbol is defined
and used in the same module.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37044 91177308-0d34-0410-b5e6-96231b3b80d8

Enable aliases on arm-linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37042 91177308-0d34-0410-b5e6-96231b3b80d8

Remove duplicated line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37040 91177308-0d34-0410-b5e6-96231b3b80d8

Reverse last patch .. premature. Depends on uncommitted CBE patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37039 91177308-0d34-0410-b5e6-96231b3b80d8

Update this test to match the (corrected) output from the CBE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37038 91177308-0d34-0410-b5e6-96231b3b80d8

Give names to the final result values of the part_set computations. This
just aids in readability and debugability of the output. No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37037 91177308-0d34-0410-b5e6-96231b3b80d8

The wrapping behavior of part_set is not desirable. Adjust the
documentation to describe the desired behavior.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37036 91177308-0d34-0410-b5e6-96231b3b80d8

Revise definition of part_select. The wrapping behavior is not useful so
in the reverse case the only difference is that the bits are returned in
reverse order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37035 91177308-0d34-0410-b5e6-96231b3b80d8

Correct a few comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37034 91177308-0d34-0410-b5e6-96231b3b80d8

Add passes -view-cfg and -view-cfg-only that are like -print-cfg and
-print-cfg-only except they use the ViewCFG function, which displays the
CFG rendered with graphviz with gv.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37033 91177308-0d34-0410-b5e6-96231b3b80d8

Use templates for the GraphType for DefaultDOTGraphTraits' members instead
of just using void*. This allows it to be used with graph adapters like
Inverse.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37032 91177308-0d34-0410-b5e6-96231b3b80d8

Add a addRequiredTransitiveID member function, which is to
addRequiredTransitive as addRequiredID is to addRequired.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37031 91177308-0d34-0410-b5e6-96231b3b80d8

Add a skeleton for future contents

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37028 91177308-0d34-0410-b5e6-96231b3b80d8

update comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37027 91177308-0d34-0410-b5e6-96231b3b80d8

Make the results for the rotate functions correct when rotateAmt == 0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37026 91177308-0d34-0410-b5e6-96231b3b80d8

Add some things needed by the llvm-gcc version supporting bit accurate integer
types:
1. Functions to compute div/rem at the same time.
2. Further assurance that an APInt with 0 bitwidth cannot be constructed.
3. Left and right rotate operations.
4. An exactLogBase2 function which requires an exact power of two or it
   returns -1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37025 91177308-0d34-0410-b5e6-96231b3b80d8

add a link

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37024 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1413

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37023 91177308-0d34-0410-b5e6-96231b3b80d8

this crashes globalopt

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37021 91177308-0d34-0410-b5e6-96231b3b80d8

Fix Transforms/GlobalOpt/2007-05-13-Crash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37020 91177308-0d34-0410-b5e6-96231b3b80d8

Emit function debug frames in one atom. This will prevent us from generating incorrect assembler in case of both
debug information & exception information presented.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37019 91177308-0d34-0410-b5e6-96231b3b80d8

Emit multiple common EH frames for multiple (including blank) personality
functions. This partly fixes PR1414: now we're restricted only to one
personality function per eh frame, not per module. Further work on
"multiple personalities" topic needs representative example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37018 91177308-0d34-0410-b5e6-96231b3b80d8

add initial description of llvm top-level stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37017 91177308-0d34-0410-b5e6-96231b3b80d8

finish the description of the bitstream format.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37016 91177308-0d34-0410-b5e6-96231b3b80d8

More DWARF-related things cleanup:
1. Fix PR1380
2. Apply Duncan's patch from PR1410
3. Insert workaround for "one personality function per module" as noted in PR1414
4. Emit correct debug frames for x86/linux. This partly fixes DebugInfo/2006-11-06-StackTrace.cpp: stack trace is
shown correctly, but arguments for function on top of stack are displayed incorrectly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37015 91177308-0d34-0410-b5e6-96231b3b80d8

Fix shl to produce the correct result when the bitwidth is > 64 and the
shift amount is 0. Previously this code would do a lshr by the bit width
which can lead to incorrect results.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37010 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for shl of APInt integers > 64 bits by 0 shift amount.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37009 91177308-0d34-0410-b5e6-96231b3b80d8

Get the size of auto arrays right, regardless of its changing size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37006 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a grammaro.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37005 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37004 91177308-0d34-0410-b5e6-96231b3b80d8

continued description

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37003 91177308-0d34-0410-b5e6-96231b3b80d8

add a bunch of content.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37002 91177308-0d34-0410-b5e6-96231b3b80d8

first step

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37001 91177308-0d34-0410-b5e6-96231b3b80d8

Add a known QEMU problem.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37000 91177308-0d34-0410-b5e6-96231b3b80d8

allow partially materialized modules to be written out, which just strips out
the functions which haven't been read.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36999 91177308-0d34-0410-b5e6-96231b3b80d8

Fix lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070507/049516.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36998 91177308-0d34-0410-b5e6-96231b3b80d8

New test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36997 91177308-0d34-0410-b5e6-96231b3b80d8

significantly improve debug output of lsr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36996 91177308-0d34-0410-b5e6-96231b3b80d8

Update comments to say "vector" instead of "packed".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36995 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36994 91177308-0d34-0410-b5e6-96231b3b80d8

Remove forward-declarations for classes that don't exist.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36993 91177308-0d34-0410-b5e6-96231b3b80d8

Add explicit keywords to several constructors that now have one argument.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36992 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify BranchInst::getSuccessor, avoiding a conditional operator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36991 91177308-0d34-0410-b5e6-96231b3b80d8

This patch extends the LoopUnroll pass to be able to unroll loops
with unknown trip counts. This is left off by default, and a
command-line option enables it. It also begins to separate loop
unrolling into a utility routine; eventually it might be made usable
from other passes.

It currently works by inserting conditional branches between each
unrolled iteration, unless it proves that the trip count is a
multiple of a constant integer > 1, which it currently only does in
the rare case that the trip count expression is a Mul operator with
a ConstantInt operand. Eventually this information might be provided
by other sources, for example by a pass that peels/splits the loop
for this purpose.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36990 91177308-0d34-0410-b5e6-96231b3b80d8

fix regressions from my previous checking, including
Transforms/InstCombine/2006-12-08-ICmp-Combining.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36989 91177308-0d34-0410-b5e6-96231b3b80d8

Perform correct actions numbers/sizes computation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36988 91177308-0d34-0410-b5e6-96231b3b80d8

Fix action No calculation in multiple-invoke-one-LP mode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36987 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1411:
Don't try to use {} bracketing when setting a variable in site.exp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36985 91177308-0d34-0410-b5e6-96231b3b80d8