Shoot! Remove this debugging line again!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64617 91177308-0d34-0410-b5e6-96231b3b80d8

Don't discard definitions of common symbols. Not sure if this is the right fix.

Before this change, the program:
  int var;
  int main(void) { return 0; }
when run under 'nm -g' would show 'U var' with the gold plugin and
'B var' with gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64616 91177308-0d34-0410-b5e6-96231b3b80d8

Update the list of function annotations for nocapture. All of these came up
when I was looking at functions used by python.

Highlights include, better largefile support (64-bit file sizes on 32-bit
systems), fputs string is nocapture, popen/pclose added (popen being noalias
return), modf and frexp and friends. Also added some missing 'break' statements
and combined identical sections.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64615 91177308-0d34-0410-b5e6-96231b3b80d8

On 64bit we may have a personality function which requires 64 bits to
be encoded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64600 91177308-0d34-0410-b5e6-96231b3b80d8

Make this more useful for cleaning up after the
one-definition-rule llvm-gcc changes (coming soon
to a tree near you!).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64588 91177308-0d34-0410-b5e6-96231b3b80d8

80-column violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64586 91177308-0d34-0410-b5e6-96231b3b80d8

If the target of an alias has internal linkage, then the
alias can be morphed into the target.  Implement this
transform, and fix a crash in the existing transform at
the same time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64583 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR3522. It's not safe to sink into landing pad BB's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64582 91177308-0d34-0410-b5e6-96231b3b80d8

Fix pr3571: If stride is a value defined by an instruction, make sure it dominates the loop preheader. When IV users are strength reduced, the stride is inserted into the preheader. It could create a use before def situation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64579 91177308-0d34-0410-b5e6-96231b3b80d8

ifdef out unneeded if statement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64575 91177308-0d34-0410-b5e6-96231b3b80d8

Fix warning on gcc 4.3.

"system() declared with attribute warn_unused_result."

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64574 91177308-0d34-0410-b5e6-96231b3b80d8

Whitespace fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64573 91177308-0d34-0410-b5e6-96231b3b80d8

Add mode_t to the config.h generated by cmake on win32. Used by clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64555 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak the build on win32.
Cleanup some warning.

Remark: when struct/class are declared differently than they are defined, this make problem for VC++ since it seems to mangle class differently that struct. These error are very hard to understand and find. So please, try to keep your definition/declaration in sync.

Only tested with VS2008. hope it does not break anything. feel free to revert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64554 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize some alias analysis logic from atomic
intrinsics to any IntrWriteArgMem intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64551 91177308-0d34-0410-b5e6-96231b3b80d8

80-column violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64550 91177308-0d34-0410-b5e6-96231b3b80d8

The trampoline intrinsic can be marked IntrWriteArgMem,
now that this no longer implies nocapture.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64540 91177308-0d34-0410-b5e6-96231b3b80d8

IntrWriteArgMem implies nocapture, but this wasn't
taken advantage of anywhere.  Change the definition
of IntrWriteArgMem to no longer imply nocapture, and
explicitly add nocapture attributes everywhere (well,
not quite everywhere, because some of these intrinsics
did capture their arguments!).  Also, make clear that
the lack of other side-effects does not exclude doing
volatile loads or stores - the atomic intrinsics do
these, yet they are all marked IntrWriteArgMem (this
change is safe because nothing exploited it).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64539 91177308-0d34-0410-b5e6-96231b3b80d8

The x86-64 red zone is now being used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64535 91177308-0d34-0410-b5e6-96231b3b80d8

Extend the IndVarSimplify support for promoting induction variables:
 - Test for signed and unsigned wrapping conditions, instead of just
   testing for non-negative induction ranges.
 - Handle loops with GT comparisons, in addition to LT comparisons.
 - Support more cases of induction variables that don't start at 0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64532 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify debug output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64531 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify some code. hasComputableLoopEvolution is overkill in this case.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64530 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for llvm-gcc 64510.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64511 91177308-0d34-0410-b5e6-96231b3b80d8

Teach x86 target -soft-float.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64496 91177308-0d34-0410-b5e6-96231b3b80d8

In CodeGenPrepare's debug output, use WriteAsOperand instead of
printing getName(), so that unnamed values are printed correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64468 91177308-0d34-0410-b5e6-96231b3b80d8

Complete the sentance in this comment. I have reservations
about the code it describes, but at least now the comment
is right.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64465 91177308-0d34-0410-b5e6-96231b3b80d8

Correct a comment: since AccessArguments is now
being used for atomic intrinsics, it seems the
access may be volatile.  No code was exploiting
the original non-volatile definition, so only
the comment needs changing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64464 91177308-0d34-0410-b5e6-96231b3b80d8

If a function only reads memory, then we know that...
it only reads memory!  The other change has no
functional effect, it just seems more logical to
go in order of decreasing knowledge.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64463 91177308-0d34-0410-b5e6-96231b3b80d8

Mark strto* as readonly when the endptr is null.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64460 91177308-0d34-0410-b5e6-96231b3b80d8

On strtod and friends, mark 'endptr' nocapture in the function prototype, and
mark the first argument nocapture if endptr=NULL for each particular call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64453 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r64301.  These uses of "inline" can cause strange
link-time failures when building with optimization.  Just
get rid of them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64449 91177308-0d34-0410-b5e6-96231b3b80d8

Switch from new[] + delete[] to malloc + free since llvm does not catch C++ exceptions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64448 91177308-0d34-0410-b5e6-96231b3b80d8

If new[] fails, return 0 rather then trying to dereference a null pointer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64444 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r64300:

Make sure the SCC pass manager initializes any contained
function pass managers.  Without this, simplify-libcalls
would add nocapture attributes when run on its own, but
not when run as part of -std-compile-opts or similar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64443 91177308-0d34-0410-b5e6-96231b3b80d8

BasicAA was making the assumption that a local allocation which hadn't escaped
couldn't ever be the return of call instruction. However, it's quite possible
that said local allocation is itself the return of a function call. That's
what malloc and calloc are for, actually.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64442 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the code that checked if a SCEVAddRecExpr Start contains an
addrec in a different loop to check the value being added to
the accumulated Start value, not the Start value before it has
the new value added to it. This prevents LSR from going crazy
on the included testcase. Dale, please review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64440 91177308-0d34-0410-b5e6-96231b3b80d8

Remove non-DebugLoc versions of BuildMI.
"I got blisters on my fingers."

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64439 91177308-0d34-0410-b5e6-96231b3b80d8

Remove non-DebugLoc versions of BuildMI from IA64, Mips.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64438 91177308-0d34-0410-b5e6-96231b3b80d8

Remove non-DebugLoc versions of BuildMI from X86.
There were some that might even matter in X86FastISel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64437 91177308-0d34-0410-b5e6-96231b3b80d8

missed file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64436 91177308-0d34-0410-b5e6-96231b3b80d8

Remove non-DebugLoc versions of buildMI from Sparc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64435 91177308-0d34-0410-b5e6-96231b3b80d8

Remove non-DebugLoc versions of BuildMI from Alpha and Cell.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64433 91177308-0d34-0410-b5e6-96231b3b80d8

Remove refs to non-DebugLoc version of BuildMI from XCore, PIC16.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64432 91177308-0d34-0410-b5e6-96231b3b80d8

Remove refs to non-DebugLoc version of BuildMI from PowerPC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64431 91177308-0d34-0410-b5e6-96231b3b80d8

and one more file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64430 91177308-0d34-0410-b5e6-96231b3b80d8

Remove refs to non-DebugLoc versions of BuildMI from ARM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64429 91177308-0d34-0410-b5e6-96231b3b80d8

Revert this. It was breaking stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64428 91177308-0d34-0410-b5e6-96231b3b80d8

Turn off the old way of handling debug information in the code generator. Use
the new way, where all of the information is passed on SDNodes and machine
instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64427 91177308-0d34-0410-b5e6-96231b3b80d8

Fix LSR's IV sorting function to explicitly sort by bitwidth
after sorting by stride value. This prevents it from missing
IV reuse opportunities in a host-sensitive manner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64415 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid order files for now, as they aren't supported in 3.79.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64412 91177308-0d34-0410-b5e6-96231b3b80d8

Check in missing file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64410 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate a couple of non-DebugLoc BuildMI variants.
Modify callers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64409 91177308-0d34-0410-b5e6-96231b3b80d8

Teach IndVarSimplify to optimize code using the C "int" type for
loop induction on LP64 targets. When the induction variable is
used in addressing, IndVars now is usually able to inserst a
64-bit induction variable and eliminates the sign-extending cast.
This is also useful for code using C "short" types for
induction variables on targets with 32-bit addressing.

Inserting a wider induction variable is easy; the tricky part is
determining when trunc(sext(i)) expressions are no-ops. This
requires range analysis of the loop trip count. A common case is
when the original loop iteration starts at 0 and exits when the
induction variable is signed-less-than a fixed value; this case
is now handled.

This replaces IndVarSimplify's OptimizeCanonicalIVType. It was
doing the same optimization, but it was limited to loops with
constant trip counts, because it was running after the loop
rewrite, and the information about the original induction
variable is lost by that point.

Rename ScalarEvolution's executesAtLeastOnce to
isLoopGuardedByCond, generalize it to be able to test for
ICMP_NE conditions, and move it to be a public function so that
IndVars can use it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64407 91177308-0d34-0410-b5e6-96231b3b80d8

Add suppport for ConstantExprs of shufflevectors whose result type is not equal to the
type of the vectors being shuffled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64401 91177308-0d34-0410-b5e6-96231b3b80d8

Arrange to print constants that match "n" and "i" constraints
in inline asm as signed (what gcc does).  Add partial support
for x86-specific "e" and "Z" constraints, with appropriate
signedness for printing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64400 91177308-0d34-0410-b5e6-96231b3b80d8

Rename bugpoint's error message file so that if it somehow
gets left behind, it's less cryptic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64399 91177308-0d34-0410-b5e6-96231b3b80d8

Add a utility function to LoopInfo to return the exit block
when the loop has exactly one exit, and make use of it in
LoopIndexSplit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64388 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR3538 for ARM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64384 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR3538 for PPC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64383 91177308-0d34-0410-b5e6-96231b3b80d8

fix the X86 backend to just drop llvm.declare nodes for VLAs instead of
leaving them in the DAG and then getting selection errors.  This is a
fix for PR3538.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64382 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust the sizes for a few SmallVectors to reflect their usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64381 91177308-0d34-0410-b5e6-96231b3b80d8

make fast isel fall back to selectiondags for VLA llvm.declare intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64379 91177308-0d34-0410-b5e6-96231b3b80d8

Add operator->, patch by Ben Laurie!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64378 91177308-0d34-0410-b5e6-96231b3b80d8

add PR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64377 91177308-0d34-0410-b5e6-96231b3b80d8

This code doesn't actually use the ExitingBlocks list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64376 91177308-0d34-0410-b5e6-96231b3b80d8

It's (currently) not safe to keep certain physical registers live across basic blocks, e.g. x86 fp stack registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64374 91177308-0d34-0410-b5e6-96231b3b80d8

Oops. Last second clean up messed things up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64373 91177308-0d34-0410-b5e6-96231b3b80d8

If availability info is kept when fallthrough into a bb, add the available registers to live-in set.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64372 91177308-0d34-0410-b5e6-96231b3b80d8

Replace one of burr scheduling heuristic with something more sensible. Now calcMaxScratches simply compute the number of true data dependencies. This actually improve a couple of tests in dejagnu suite as many tests in llvm nightly test suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64369 91177308-0d34-0410-b5e6-96231b3b80d8

Fix MingW build: define GTEST_OS_WINDOWS if OS is MingW, but disable
exceptions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64367 91177308-0d34-0410-b5e6-96231b3b80d8

Fix MingW build, patch by Kenneth Boyd!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64366 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a nasty bug (PR3550) where the inline pass could incorrectly mark
calls with the tail marker when inlining them through an invoke.  Patch,
testcase, and perfect analysis by Jay Foad!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64364 91177308-0d34-0410-b5e6-96231b3b80d8

improve naming of values in GVN, patch by Jay Foad!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64363 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR3537: if resetting bbi back to the start of a block, we need to
forget about already inserted expressions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64362 91177308-0d34-0410-b5e6-96231b3b80d8

rename test to avoid messing with tab completion of dates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64361 91177308-0d34-0410-b5e6-96231b3b80d8

Add method 'isSingleton()' to ImmutableSet.  This returns true if the set contains exactly one element.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64359 91177308-0d34-0410-b5e6-96231b3b80d8

Don't mark all args to strtod and friends as nocapture.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64352 91177308-0d34-0410-b5e6-96231b3b80d8

Move debug loc info along when the spiller creates new instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64342 91177308-0d34-0410-b5e6-96231b3b80d8

Apparently some MachineBasicBlock's don't have corresponding llvm basic blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64340 91177308-0d34-0410-b5e6-96231b3b80d8

Remove a bogus assertion. It's possible a live-in available value is used by a previous instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64339 91177308-0d34-0410-b5e6-96231b3b80d8

the two non-mask arguments to a shufflevector must be the same width, but they do not have to be the same
width as the result value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64335 91177308-0d34-0410-b5e6-96231b3b80d8

Edit description of floating point constants to
reflect reality.  Acknowledgements to John Clements
for prodding me into this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64332 91177308-0d34-0410-b5e6-96231b3b80d8

Fill in a glaring omission in derived User classes, namely
add efficient versions of op_begin and op_end. Up to now always those from User have been
called, which in most cases follow an indirection (OperandList) even if the exact Instruction
type is known.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64331 91177308-0d34-0410-b5e6-96231b3b80d8

Propagate DebugLoc info for spiller call-backs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64329 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64328 91177308-0d34-0410-b5e6-96231b3b80d8

Don't use special heuristics for nodes with no data predecessors
unless they actually have data successors, and likewise for nodes
with no data successors unless they actually have data precessors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64327 91177308-0d34-0410-b5e6-96231b3b80d8

Add FoldingSet profile method for bools.  Patch by Ben Laurie!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64324 91177308-0d34-0410-b5e6-96231b3b80d8

Update to match space changes in .ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64322 91177308-0d34-0410-b5e6-96231b3b80d8

Delete the heuristic for non-livein CopyFromReg nodes. Non-liveinness
is determined by whether the node has a Flag operand. However, if the
node does have a Flag operand, it will be glued to its register's
def, so the heuristic would end up spuriously applying to whatever
node is the def.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64319 91177308-0d34-0410-b5e6-96231b3b80d8

Don't try to set an EFLAGS operand to dead if no instruction was created.
This fixes a bug introduced by r61215.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64316 91177308-0d34-0410-b5e6-96231b3b80d8

Make a transformation added in 63266 a bit less aggressive.
It was transforming (x&y)==y to (x&y)!=0 in the case where
y is variable and known to have at most one bit set (e.g. z&1).
This is not correct; the expressions are not equivalent when y==0.
I believe this patch salvages what can be salvaged, including
all the cases in bt.ll.  Dan, please review.
Fixes gcc.c-torture/execute/20040709-[12].c

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64314 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r64300 and r64301. These were causing the following errors respectively:

During llvm-gcc bootstrap:

Undefined symbols:
  "llvm::FPPassManager::doFinalization(llvm::Module&)", referenced from:
      (anonymous namespace)::CGPassManager::doFinalization(llvm::CallGraph&, llvm::Module&)
        in libLLVMipa.a(CallGraphSCCPass.o)
  "llvm::FPPassManager::doInitialization(llvm::Module&)", referenced from:
      (anonymous namespace)::CGPassManager::doInitialization(llvm::CallGraph&, llvm::Module&)
        in libLLVMipa.a(CallGraphSCCPass.o)
ld: symbol(s) not found
collect2: ld returned 1 exit status
make[3]: *** [/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore.roots/llvmCore~obj/obj-llvm/Release/bin/opt] Error 1

During an LLVM release build:

/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/Release/bin/tblgen -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86 -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target -gen-register-desc -o  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Target/X86/Release/X86GenRegisterInfo.inc.tmp /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86/X86.td
llvm[3]: Building X86.td instruction names with tblgen
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/Release/bin/tblgen -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86 -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target -gen-instr-enums -o  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Target/X86/Release/X86GenInstrNames.inc.tmp /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86/X86.td
llvm[3]: Building X86.td instruction information with tblgen
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/Release/bin/tblgen -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86 -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target -gen-instr-desc -o  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Target/X86/Release/X86GenInstrInfo.inc.tmp /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86/X86.td
llvm[3]: Building X86.td assembly writer with tblgen
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/Release/bin/tblgen -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86 -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target -gen-asm-writer -o  /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Target/X86/Release/X86GenAsmWriter.inc.tmp /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Target/X86/X86.td
llvm[3]: Compiling InstructionCombining.cpp for Release build
if /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmgcc42.roots/llvmgcc42~dst/Developer/usr/bin/llvm-g++-4.2 -I/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/include -I/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Transforms/Scalar -I/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/include -I/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Transforms/Scalar  -D_DEBUG  -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -D__STDC_CONSTANT_MACROS -O3  -fno-exceptions  -Woverloaded-virtual -pedantic -Wall -W -Wwrite-strings -Wno-long-long -Wunused -Wno-unused-parameter  -fstrict-aliasing -Wstrict-aliasing -c -MMD -MP -MF "/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Transforms/Scalar/Release/InstructionCombining.d.tmp" -MT "/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Transforms/Scalar/Release/InstructionCombining.lo" -MT "/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Transforms/Scalar/Release/InstructionCombining.o" -MT "/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Transforms/Scalar/Release/InstructionCombining.d" /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/lib/Transforms/Scalar/InstructionCombining.cpp -o /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Transforms/Scalar/Release/InstructionCombining.o ; \
           then /bin/mv -f "/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Transforms/Scalar/Release/InstructionCombining.d.tmp" "/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.sh.build/lib/Trans

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64311 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r64299: it breaks the build when configured
without --enable-pic, like my nightly tester.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64302 91177308-0d34-0410-b5e6-96231b3b80d8

These uses of "inline" can cause strange link-time
failures when building with optimization.  Just get
rid of them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64301 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure the SCC pass manager initializes any contained
function pass managers.  Without this, simplify-libcalls
would add nocapture attributes when run on its own, but
not when run as part of -std-compile-opts or similar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64300 91177308-0d34-0410-b5e6-96231b3b80d8

Try this. Darwin -> LTO, PIC -> LTO + possibly gold too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64299 91177308-0d34-0410-b5e6-96231b3b80d8

Implement PR3495: local spiller optimization. The local spiller can now keep availability information over BB boundaries. It visits BB's in depth first order. After visiting a BB if it find a successor which has a single predecessor it visits the successor next without clearing the availability information. This allows the successor to omit reloads or change them into copies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64298 91177308-0d34-0410-b5e6-96231b3b80d8

When scheduling a block in parts, keep track of the overall
instruction index across each part. Instruction indices are used
to make live range queries, and live ranges can extend beyond
scheduling region boundaries.

Refactor the ScheduleDAGSDNodes class some more so that it
doesn't have to worry about this additional information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64288 91177308-0d34-0410-b5e6-96231b3b80d8

62987 disables LTO build on darwin.
Revert 62987 for now. Nicolas please investigate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64285 91177308-0d34-0410-b5e6-96231b3b80d8

If llvm.dbg.region.end is disappearing then remove corresponding llvm.dbg.func.start also.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64278 91177308-0d34-0410-b5e6-96231b3b80d8