The result of getSetCCResultType (eg: i32) may be larger
than the type an i1 is promoted to (eg: i8).  Account
for this.  Noticed by Tilmann Scheller on CellSPU; he
will hopefully take care of fixing this in LegalizeDAG
and adding a testcase!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56997 91177308-0d34-0410-b5e6-96231b3b80d8

Teach internalize to preserve the callgraph.
Why?  Because it was there!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56996 91177308-0d34-0410-b5e6-96231b3b80d8

Acquire the lock only when necessary. More precisely, do not acquire
the lock when calling a method which may materialize the llvm::Function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56995 91177308-0d34-0410-b5e6-96231b3b80d8

SplitBlock should only attempt to update LoopInfo if it is actually being used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56994 91177308-0d34-0410-b5e6-96231b3b80d8

Implement fast-isel support for zero-extending from i1.
It turns out that this is a fairly common operation,
and it's easy enough to handle.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56990 91177308-0d34-0410-b5e6-96231b3b80d8

Fix X86FastISel to handle dynamic allocas that have avoided
getting inserted into the ValueMap. This avoids infinite
recursion in some rare cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56989 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak the build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56988 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56986 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for Canadian Cross builds where the host executables are not
runnable on the build machine.

There are a few bits that need built for the build environment (TableGen).
This patch builds those bits, and the associated libraries, for the build
environment as well as the (usual) host environment.

Thanks to Eric C. and Devang P. for pre-commit review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56975 91177308-0d34-0410-b5e6-96231b3b80d8

Fix build breakage.

Forgot to include Makefile.plugins.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56970 91177308-0d34-0410-b5e6-96231b3b80d8

Optimize conditional branches in X86FastISel. This replaces
sequences like this:
       sete    %al
       testb   %al, %al
       jne     LBB11_1
with this:
       je      LBB11_1

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56969 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new MachineBasicBlock utility function, isLayoutSuccessor, that
can be used when deciding if a block can transfer control to another
via a fall-through instead of a branch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56968 91177308-0d34-0410-b5e6-96231b3b80d8

Build system tweaks to make it more convenient for the plugin authors.

Plugins can be now compiled in with a slight Makefile change.
For example, to compile the new Clang driver, use:

cd $LLVMC2_DIR
make TOOLNAME=ccc2 BUILTIN_PLUGINS=Clang

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56967 91177308-0d34-0410-b5e6-96231b3b80d8

fix build gcc 4.3

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56965 91177308-0d34-0410-b5e6-96231b3b80d8

Use a multimap rather than a map for holding the list of copies to insert, so we don't lose copies when two of them have
the same source.  I don't know what I was thinking when I wrote this originally.
Note: There's probably a more efficient way to do this, but I need to think about it some more, and about what determinism
guarantees need to be present.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56964 91177308-0d34-0410-b5e6-96231b3b80d8

Handle some 64-bit atomics on x86-32, some of the time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56963 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid name shadowing with E variable defined in for(). This was giving VC++
grief.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56961 91177308-0d34-0410-b5e6-96231b3b80d8

Remove redundant check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56960 91177308-0d34-0410-b5e6-96231b3b80d8

A Partitioned Boolean Quadratic Programming (PBQP) based register allocator.

Contributed by Lang Hames.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56959 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Added Host.cpp to lib/System/CMakeLists.txt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56957 91177308-0d34-0410-b5e6-96231b3b80d8

Rename IRBuilder::IsNonNull -> IsNotNull in response to feedback.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56953 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a think-o in isSafeToMove. This fixes it from thinking that
volatile memory references are safe to move.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56948 91177308-0d34-0410-b5e6-96231b3b80d8

Work around an interaction between fast-isel and regalloc=local. The
local register allocator's physreg liveness doesn't recognize subregs,
so it doesn't know that defs of %ecx that are immediately followed by
uses of %cl aren't dead. This comes up due to the way fast-isel emits
shift instructions.

This is a temporary workaround. Arguably, local regalloc should
handle subreg references correctly. On the other hand, perhaps
fast-isel should use INSERT_SUBREG instead of just assigning to the
most convenient super-register of %cl when lowering shifts.

This fixes MultiSource/Benchmarks/MallocBench/espresso,
MultiSource/Applications/hexxagon, and others, under -fast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56947 91177308-0d34-0410-b5e6-96231b3b80d8

"The original bug was a complaint that _mm_srli_si128 mis-compiled when passed
a constant vector ("{0x123, 0x456}" syntax).  The fix is to simplify the
_mm_srli_si128 macro, and  move the "* 8" from the macro into the compiler
back-end.  I can't change the existing __builtins because so many people are
using them :-(."
Patch by Stuart Hastings!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56944 91177308-0d34-0410-b5e6-96231b3b80d8

Add llvm::sys::{osName,osVersion} for retrieving operating system name
& version as strings.
 - Win32 code is untested.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56942 91177308-0d34-0410-b5e6-96231b3b80d8

Disable fast-isel for this test, as it doesn't emit the same
number of instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56940 91177308-0d34-0410-b5e6-96231b3b80d8

Attributes noinline alwaysinline are incompatible

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56939 91177308-0d34-0410-b5e6-96231b3b80d8

Remove OptimizeForSize global. Use function attribute optsize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56937 91177308-0d34-0410-b5e6-96231b3b80d8

Enable FastISel by default (on x86 and x86-64) with the -fast option.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56930 91177308-0d34-0410-b5e6-96231b3b80d8

add a new form of Type::dump that takes a module for type names,
patch provided by Tomas Lindquist Olsen!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56929 91177308-0d34-0410-b5e6-96231b3b80d8

Make some implicit conversions explicit, to avoid compiler warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56927 91177308-0d34-0410-b5e6-96231b3b80d8

Split this test and move it into target-specific directories.
This fixes failures on configurations that don't have one or the
other targets enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56926 91177308-0d34-0410-b5e6-96231b3b80d8

Split x86's ADJCALLSTACK instructions into 32-bit and 64-bit forms.
This allows the 64-bit forms to use+def RSP instead of ESP. This
doesn't fix any real bugs today, but it is more precise and it
makes the debug dumps on x86-64 look more consistent.

Also, add some comments describing the CALL instructions' physreg
operand uses and defs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56925 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo s/ther/there/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56924 91177308-0d34-0410-b5e6-96231b3b80d8

Factorize code: remove variants of "strip off
pointer bitcasts and GEP's", and centralize the
logic in Value::getUnderlyingObject.  The
difference with stripPointerCasts is that
stripPointerCasts only strips GEPs if all
indices are zero, while getUnderlyingObject
strips GEPs no matter what the indices are.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56922 91177308-0d34-0410-b5e6-96231b3b80d8

Fold trivial two-operand tokenfactors where the operands are equal
immediately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56921 91177308-0d34-0410-b5e6-96231b3b80d8

Don't prepend a space character for constants in Value::print.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56920 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typos in comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56919 91177308-0d34-0410-b5e6-96231b3b80d8

nounwind-ify this test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56918 91177308-0d34-0410-b5e6-96231b3b80d8

revert the addition of Preverves(CallGraph), per Duncan's comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56917 91177308-0d34-0410-b5e6-96231b3b80d8

Mark CALL instructions as having a Use of ESP/RSP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56911 91177308-0d34-0410-b5e6-96231b3b80d8

Don't leave an output file in the test directory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56910 91177308-0d34-0410-b5e6-96231b3b80d8

Call ScalarEvolution's deleteValueFromRecords before deleting an
instruction, not after. This fixes some uses of free'd memory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56908 91177308-0d34-0410-b5e6-96231b3b80d8

Moved this option to the front-end.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56901 91177308-0d34-0410-b5e6-96231b3b80d8

Implement the -fno-builtin option in the front-end, not in the back-end.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56900 91177308-0d34-0410-b5e6-96231b3b80d8

Use explicit target-triples to unbreak this test on non-darwin systems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56896 91177308-0d34-0410-b5e6-96231b3b80d8

Mark merged-in VNInfo's as being PHIKilled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56893 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a simple error in renumbering kill markaers, that took an inordinant amount of time to track down.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56889 91177308-0d34-0410-b5e6-96231b3b80d8

Just don't transform this memset into "bzero" if no-builtin is specified.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56888 91177308-0d34-0410-b5e6-96231b3b80d8

add preserversCFG() + preservers(CallGraph)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56887 91177308-0d34-0410-b5e6-96231b3b80d8

- Initialize "--no-builtin" to "false".
- Testcase for r56885.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56886 91177308-0d34-0410-b5e6-96231b3b80d8

Add the new `-no-builtin' flag. This flag is meant to mimic the GCC
`-fno-builtin' flag. Currently, it's used to replace "memset" with "_bzero"
instead of "__bzero" on Darwin10+. This arguably violates the meaning of this
flag, but is currently sufficient. The meaning of this flag should become more
specific over time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56885 91177308-0d34-0410-b5e6-96231b3b80d8

Move the primary fast-isel top-level comments to FastISel.cpp, where
they'll be a little more visible. Also, update and reword them a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56877 91177308-0d34-0410-b5e6-96231b3b80d8

add AU.setPreservesCFG() since this pass only adds and removes function attributes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56868 91177308-0d34-0410-b5e6-96231b3b80d8

Optimize SelectionDAG's AssignTopologicalOrder even further.

Completely eliminate the TopOrder std::vector. Instead, sort
the AllNodes list in place. This also eliminates the need to
call AllNodes.size(), a linear-time operation, before
performing the sort.

Also, eliminate the Sources temporary std::vector, since it
essentially duplicates the sorted result as it is being
built.

This also changes the direction of the topological sort
from bottom-up to top-down. The AllNodes list starts out in
roughly top-down order, so this reduces the amount of
reordering needed. Top-down is also more convenient for
Legalize, and ISel needed only minor adjustments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56867 91177308-0d34-0410-b5e6-96231b3b80d8

ignore generated files

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56866 91177308-0d34-0410-b5e6-96231b3b80d8

Add runStaticConstructorsDestructors which runs ctors / dtors of a single module. Patch by David Chisnall.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56849 91177308-0d34-0410-b5e6-96231b3b80d8

Re-apply 56835 along with header file changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56848 91177308-0d34-0410-b5e6-96231b3b80d8

Stacker is dead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56844 91177308-0d34-0410-b5e6-96231b3b80d8

Revert commit 56835 since it breaks the build.
"If a re-materializable instruction has a register
operand, the spiller will change the register operand's
spill weight to HUGE_VAL to avoid it being spilled.
However, if the operand is already in the queue ready
to be spilled, avoid re-materializing it".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56837 91177308-0d34-0410-b5e6-96231b3b80d8

If a re-materializable instruction has a register operand, the spiller will change the register operand's spill weight to HUGE_VAL to avoid it being spilled. However, if the operand is already in the queue ready to be spilled, avoid re-materializing it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56835 91177308-0d34-0410-b5e6-96231b3b80d8

Fix misoptimization of:  xor i1 (icmp eq (X, C1), icmp s[lg]t (X, C2))

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56834 91177308-0d34-0410-b5e6-96231b3b80d8

Fix X86FastISel's output for x86-32 PIC constant pool addresses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56829 91177308-0d34-0410-b5e6-96231b3b80d8

Move the GlobalBaseReg field out of X86ISelDAGToDAG.cpp
and X86FastISel.cpp into X86MachineFunction.h, so that it
can be shared, instead of having each selector keep track
of its own.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56825 91177308-0d34-0410-b5e6-96231b3b80d8

Disable all x87 usage, including f32 and f64 when the subtarget
doesn't have SSE(2), with X86FastISel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56823 91177308-0d34-0410-b5e6-96231b3b80d8

Remove misuse of ReplaceNodeResults for atomics with
valid types.  No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56808 91177308-0d34-0410-b5e6-96231b3b80d8

Fix FastISel to not initialize the PIC-base register multiple times
in functions with PIC references from more than one basic block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56807 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an over-pessimization about GlobalVariable addresses in X86FastISel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56802 91177308-0d34-0410-b5e6-96231b3b80d8

Support inreg, zext and sext as return value attributes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56801 91177308-0d34-0410-b5e6-96231b3b80d8

Fix grammar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56796 91177308-0d34-0410-b5e6-96231b3b80d8

Fix grammer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56795 91177308-0d34-0410-b5e6-96231b3b80d8

Fix sentence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56794 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR2835. Do not change the width of a volatile load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56792 91177308-0d34-0410-b5e6-96231b3b80d8

Speed up these passes when the callgraph has
huge simply connected components.  Suggested
by Chris.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56787 91177308-0d34-0410-b5e6-96231b3b80d8

remove redundant test (mayBeOverriden() includes hasLinkOnceLinkage)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56786 91177308-0d34-0410-b5e6-96231b3b80d8

Tweak some comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56784 91177308-0d34-0410-b5e6-96231b3b80d8

Add some hands-on documentation about which files to create and edit when
adding a backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56783 91177308-0d34-0410-b5e6-96231b3b80d8

Rename isWeakForLinker to mayBeOverridden.  Use it
instead of hasWeakLinkage in a bunch of optimization
passes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56782 91177308-0d34-0410-b5e6-96231b3b80d8

Add a testcase showing that scalarrepl supports first class structs.

I originally made this script to show that scalarrepl didn't support them, but
it turned out it does. Better to still add the testcase then.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56781 91177308-0d34-0410-b5e6-96231b3b80d8

Move the code that handles DAGISel error conditions into
helper functions instead of duplicating it inline each time
it is needed. This eliminates a few hundred or so copies
of this code in each target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56759 91177308-0d34-0410-b5e6-96231b3b80d8

Add IRBuilder::{CreateIsNull, CreateIsNonNull} helper methods.
 - I'm open to the idea that these could have better names. I think
   these read better than CreateEQNull and CreateNENull.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56757 91177308-0d34-0410-b5e6-96231b3b80d8

Re-apply 56683 with fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56748 91177308-0d34-0410-b5e6-96231b3b80d8

Update tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56730 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56727 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing #include for asserts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56718 91177308-0d34-0410-b5e6-96231b3b80d8

Implement function notes as function attributes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56716 91177308-0d34-0410-b5e6-96231b3b80d8

Generated files for warning fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56715 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a compiler warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56714 91177308-0d34-0410-b5e6-96231b3b80d8

Implement "punpckldq  %xmm0, $xmm0" as "pshufd  $0x50, %xmm0, %xmm" unless optimizing for code size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56711 91177308-0d34-0410-b5e6-96231b3b80d8

Now Attributes are divided in three groups
- return attributes - inreg, zext and sext
- parameter attributes
- function attributes - nounwind, readonly, readnone, noreturn

Return attributes use 0 as the index.
Function attributes use ~0U as the index.

This patch requires corresponding changes in llvm-gcc and clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56704 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily reverting r56683. This is causing a failure during the build of llvm-gcc:

/Volumes/Gir/devel/llvm/clean/llvm-gcc.obj/./gcc/xgcc -B/Volumes/Gir/devel/llvm/clean/llvm-gcc.obj/./gcc/ -B/Volumes/Gir/devel/llvm/clean/llvm-gcc.install/i386-apple-darwin9.5.0/bin/ -B/Volumes/Gir/devel/llvm/clean/llvm-gcc.install/i386-apple-darwin9.5.0/lib/ -isystem /Volumes/Gir/devel/llvm/clean/llvm-gcc.install/i386-apple-darwin9.5.0/include -isystem /Volumes/Gir/devel/llvm/clean/llvm-gcc.install/i386-apple-darwin9.5.0/sys-include -mmacosx-version-min=10.4 -O2  -O2 -g -O2  -DIN_GCC    -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition  -isystem ./include  -fPIC -pipe -g -DHAVE_GTHR_DEFAULT -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED  -I. -I. -I../../llvm-gcc.src/gcc -I../../llvm-gcc.src/gcc/. -I../../llvm-gcc.src/gcc/../include -I./../intl -I../../llvm-gcc.src/gcc/../libcpp/include  -I../../llvm-gcc.src/gcc/../libdecnumber -I../libdecnumber -I/Volumes/Gir/devel/llvm/clean/llvm.obj/include -I/Volumes/Gir/devel/llvm/clean/llvm.src/include -fexceptions -fvisibility=hidden -DHIDE_EXPORTS -c ../../llvm-gcc.src/gcc/unwind-dw2-fde-darwin.c -o libgcc/./unwind-dw2-fde-darwin.o
Assertion failed: (TargetRegisterInfo::isVirtualRegister(regA) && TargetRegisterInfo::isVirtualRegister(regB) && "cannot update physical register live information"), function runOnMachineFunction, file /Volumes/Gir/devel/llvm/clean/llvm.src/lib/CodeGen/TwoAddressInstructionPass.cpp, line 311.
../../llvm-gcc.src/gcc/unwind-dw2.c:1527: internal compiler error: Abort trap
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
{standard input}:3521:non-relocatable subtraction expression, "_dwarf_reg_size_table" minus "L20$pb"
{standard input}:3521:symbol: "_dwarf_reg_size_table" can't be undefined in a subtraction expression
{standard input}:3520:non-relocatable subtraction expression, "_dwarf_reg_size_table" minus "L20$pb"
...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56703 91177308-0d34-0410-b5e6-96231b3b80d8

Fix missing whitespace in the printing of function notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56702 91177308-0d34-0410-b5e6-96231b3b80d8

Rename ConstantSDNode's getSignExtended to getSExtValue, for
consistancy with ConstantInt, and re-implement it in terms
of ConstantInt's getSExtValue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56700 91177308-0d34-0410-b5e6-96231b3b80d8

Make some minor reorganizations and comment changes to eliminate trivial
differences between the CallInst and InvokeInst classes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56699 91177308-0d34-0410-b5e6-96231b3b80d8

Mark lea fi# as being really rematerializable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56698 91177308-0d34-0410-b5e6-96231b3b80d8

unpckhps requires sse1, punpckhdq requires sse2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56697 91177308-0d34-0410-b5e6-96231b3b80d8

Update LICENSE link.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56689 91177308-0d34-0410-b5e6-96231b3b80d8

Fix X86FastISel's address folding to check displacement
values for overflow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56686 91177308-0d34-0410-b5e6-96231b3b80d8

Fix @llvm.frameaddress codegen. FP elimination optimization should be disabled when frame address is desired. Also add support for depth > 0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56683 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Builds all examples. Corrected name of CBackend target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56682 91177308-0d34-0410-b5e6-96231b3b80d8