Remove the CycleBound computation code from the ScheduleDAGRRList
schedulers. This doesn't have much immediate impact because
targets that use these schedulers by default don't yet provide
pipeline information.

This code also didn't have the benefit of register pressure
information. Also, removing it will avoid problems with list-burr
suddenly starting to do latency-oriented scheduling on x86 when we
start providing pipeline data, which would increase spilling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59775 91177308-0d34-0410-b5e6-96231b3b80d8

Update comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59766 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Do not rebuild the world when tblgen changes but the .inc files
it produces are not affected by the change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59763 91177308-0d34-0410-b5e6-96231b3b80d8

Implement ComputeLatency for MachineInstr ScheduleDAGs. Factor
some of the latency computation logic out of the SDNode
ScheduleDAG code into a TargetInstrItineraries helper method
to help with this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59761 91177308-0d34-0410-b5e6-96231b3b80d8

Add UADDO and SADDO nodes. These will be used for determining an overflow
condition in an addition operation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59760 91177308-0d34-0410-b5e6-96231b3b80d8

Change these schedulers to not emit no-ops. It turns out that
the RR scheduler actually does look at latency values, but it
doesn't use a hazard recognizer so it has no way to know when
a no-op is needed, as opposed to just stalling and incrementing
the cycle count.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59759 91177308-0d34-0410-b5e6-96231b3b80d8

Fix error where it wasn't getting the correct caller function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59758 91177308-0d34-0410-b5e6-96231b3b80d8

If the function being inlined has a higher stack protection level than the
inlining function, then increase the stack protection level on the inlining
function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59757 91177308-0d34-0410-b5e6-96231b3b80d8

Introduce two new "add" intrinsics. These return the sum plus a bit indicating
that an overflow/carry occured. These are converted into ISD::[SU]ADDO nodes,
which are lowered in a target-independent way into something sane. Eventually,
each target can implement their own method of checking the overflow/carry flags.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59756 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: More documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59755 91177308-0d34-0410-b5e6-96231b3b80d8

Delete redundant inline keywords.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59754 91177308-0d34-0410-b5e6-96231b3b80d8

Doxygenate comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59753 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Some documentation. Work in progress.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59752 91177308-0d34-0410-b5e6-96231b3b80d8

Treat mid-block labels the same as terminators when building the
MachineInstr scheduling DAG, meaning they implicitly depend on all
preceding defs. This fixes Benchmarks/Shootout-C++/except and
Regression/C++/EH/simple_rethrow in
  -relocation-model=pic -disable-post-RA-scheduler=false
mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59747 91177308-0d34-0410-b5e6-96231b3b80d8

Add another machine-code printing pass when post-pass scheduling is run.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59746 91177308-0d34-0410-b5e6-96231b3b80d8

Don't forget arguments!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59745 91177308-0d34-0410-b5e6-96231b3b80d8

Test -pre-RA-sched=fast too, for completeness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59741 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Option for building with -fPIC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59739 91177308-0d34-0410-b5e6-96231b3b80d8

Updating ignore lists.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59737 91177308-0d34-0410-b5e6-96231b3b80d8

CellSPU:
(a) Remove moved file (SPUAsmPrinter.cpp) to make svn happy.
(b) Remove truncated stores that will never be used.
(c) Add initial support for __muldi3 as a libcall.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59734 91177308-0d34-0410-b5e6-96231b3b80d8

Add some documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59727 91177308-0d34-0410-b5e6-96231b3b80d8

Allow XMM2 and XMM3 to be used for non ABI compliant code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59720 91177308-0d34-0410-b5e6-96231b3b80d8

80-column violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59718 91177308-0d34-0410-b5e6-96231b3b80d8

Reformatting. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59717 91177308-0d34-0410-b5e6-96231b3b80d8

add an operator= to assign to smallstring.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59715 91177308-0d34-0410-b5e6-96231b3b80d8

CellSPU: Custom lower truncating stores of i8 to i1 (should not have been
promote), fix signed conversion of indexed offsets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59707 91177308-0d34-0410-b5e6-96231b3b80d8

undef beats zero. Fix this missed optimization opportunity. Patch by Matt Elder!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59705 91177308-0d34-0410-b5e6-96231b3b80d8

Revert accidental last patch

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59704 91177308-0d34-0410-b5e6-96231b3b80d8

CellSPU: Adjust spacing/tabulation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59703 91177308-0d34-0410-b5e6-96231b3b80d8

Remove a remnant of list-burr's fast mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59702 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out the SethiUllman numbering logic from the list-burr and
list-tdrr schedulers into a common base class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59701 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the "fast" form of the list-burr scheduler, and use the
dedicated "fast" scheduler in -fast mode instead, which is
faster. This speeds up llc -fast by a few percent on some
testcases -- the speedup only happens for code not handled by
fast-isel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59700 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: when building shared libraries, use -fPIC for compiling
partially linked objects.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59699 91177308-0d34-0410-b5e6-96231b3b80d8

Facter AddPseudoTwoAddrDeps and associated infrasructure out of
the list-burr scheduler so that it can be used by the list-tdrr
scheduler too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59698 91177308-0d34-0410-b5e6-96231b3b80d8

- Register scavenger should use MachineRegisterInfo and internal map to find the first use of a register after a given machine instruction.
- When scavenging a register, in addition to the spill, insert a restore before the first use.
- Abort if client is looking to scavenge a register even when a previously scavenged register is still live.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59697 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a thinko. MO is getOperand(i-1) so we don't have to adjust e.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59696 91177308-0d34-0410-b5e6-96231b3b80d8

Add #include <climits> to get the definition of INT_MAX.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59692 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out the code for verifying the work of the scheduler,
extend it a bit, and make use of it in all schedulers, to
ensure consistent checking.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59689 91177308-0d34-0410-b5e6-96231b3b80d8

Do not forget llvm.dbg.declare's first argument while removing debugging information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59688 91177308-0d34-0410-b5e6-96231b3b80d8

Copy the tblgen utility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59681 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify this code a little. In the fast scheduler, CreateNewSUnit
and CreateClone don't add any extra value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59679 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate a compile time warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59678 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate a compile time warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59677 91177308-0d34-0410-b5e6-96231b3b80d8

Experimental post-pass scheduling support. Post-pass scheduling
is currently off by default, and can be enabled with
-disable-post-RA-scheduler=false.

This doesn't have a significant impact on most code yet because it doesn't
yet do anything to address anti-dependencies and it doesn't attempt to
disambiguate memory references. Also, several popular targets
don't have pipeline descriptions yet.

The majority of the changes here are splitting the SelectionDAG-specific
code out of ScheduleDAG, so that ScheduleDAG can be moved to
libLLVMCodeGen.a. The interface between ScheduleDAG-using code and
the rest of the scheduling code is somewhat rough and will evolve.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59676 91177308-0d34-0410-b5e6-96231b3b80d8

Move the code for printing a graph node label for an SUnit into
a virtual method of SelectionDAG.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59667 91177308-0d34-0410-b5e6-96231b3b80d8

Convert SUnit's dump method into a print method and implement
dump in terms of it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59665 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Removed source file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59662 91177308-0d34-0410-b5e6-96231b3b80d8

Do not use separate utility to walk all instructions and remove dead dbg intrinsics. Let instcombiner do this job.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59659 91177308-0d34-0410-b5e6-96231b3b80d8

Let instcombiner remove redundant dbg intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59658 91177308-0d34-0410-b5e6-96231b3b80d8

If there are two consecutive llvm.dbg.stoppoint calls then
it is likely that the optimizer deleted code in between these
two intrinsics. Keep only the last llvm.dbg.stoppoint in this case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59657 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Removed source file from lib/Target/PIC16/CMakeLists.txt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59655 91177308-0d34-0410-b5e6-96231b3b80d8

CellSPU: Do not custom lower i1 stores, rely on type legalization to do the
right thing and promote the store to i8.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59648 91177308-0d34-0410-b5e6-96231b3b80d8

<rdar://problem/6351057>
Discourage (allocate last) use of x86_64 R12 and R13 due to their
longer instruction encodings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59644 91177308-0d34-0410-b5e6-96231b3b80d8

ignore the -m elf_i386 directive used in the linux kernel

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59642 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r59640. It broke this test for builds that aren't
configured with llvm-gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59641 91177308-0d34-0410-b5e6-96231b3b80d8

Use %llvmgcc -xassembler instead of invoking as directly. This avoids
problems for example when LLVM is built with --with-extra-options=-m64
and as defaults to x86-32 mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59640 91177308-0d34-0410-b5e6-96231b3b80d8

Temporary check-in for Duncan to demonstrate CellSPU store problem.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59637 91177308-0d34-0410-b5e6-96231b3b80d8

Fix compilation error on MSVC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59629 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot to add this in the previous commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59623 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed build warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59621 91177308-0d34-0410-b5e6-96231b3b80d8

Added a more function PIC16 backend. However to get this working a patch in
ExpandIntegerOperand (LegalizeIntegerTypes.cpp) is needed which is yet to be reworked and submitted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59617 91177308-0d34-0410-b5e6-96231b3b80d8

Use stripPointerCasts when checking for AllocaInsts for the stackprotector intrinsic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59614 91177308-0d34-0410-b5e6-96231b3b80d8

Int type for PIC16 is i16. Added i16 intrinsics for memmove, memcpy and memset.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59612 91177308-0d34-0410-b5e6-96231b3b80d8

add a write method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59595 91177308-0d34-0410-b5e6-96231b3b80d8

- Move the stackprotector intrinsic to the general section.
- Rewrite the sentence to make it look as if English is my first language.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59592 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for rematerialization in pre-alloc-splitting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59587 91177308-0d34-0410-b5e6-96231b3b80d8

LLVMC2: -emit-llvm stops compilation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59586 91177308-0d34-0410-b5e6-96231b3b80d8

LLVMC2: Teach llvm_gcc_c tool about -include and -fsyntax-only.
 - Only focusing on llvm_gcc_c for now, eventually this needs to be
   refactored so it can be shared via all the gcc-like tools.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59582 91177308-0d34-0410-b5e6-96231b3b80d8

Grammar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59581 91177308-0d34-0410-b5e6-96231b3b80d8

Rearrange code to reduce the nesting level. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59580 91177308-0d34-0410-b5e6-96231b3b80d8

Use dyn_cast instead of cast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59577 91177308-0d34-0410-b5e6-96231b3b80d8

Match an element of the return type if it returns a structure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59576 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59570 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59569 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Support for building 32 bits shared libraries on 64 bits GNU
systems. BUILD_32_BITS option renamed to LLVM_BUILD_32_BITS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59568 91177308-0d34-0410-b5e6-96231b3b80d8

Fix debug printing of flagged SDNodes in SUnits so that they
print in the correct order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59567 91177308-0d34-0410-b5e6-96231b3b80d8

Make the same change to RegScavenger::backward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59566 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Option for enabling/disabling threads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59565 91177308-0d34-0410-b5e6-96231b3b80d8

Verify that the second parameter of the stacprotector intrinsic is an alloca
instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59563 91177308-0d34-0410-b5e6-96231b3b80d8

We also need to keep the operand index for two address check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59562 91177308-0d34-0410-b5e6-96231b3b80d8

Register scavenger should process early clobber defs first. A dead early clobber def should not interfere with a normal def which happens one slot later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59559 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59558 91177308-0d34-0410-b5e6-96231b3b80d8

Documentation for the llvm.stackprotector intrinsic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59557 91177308-0d34-0410-b5e6-96231b3b80d8

Add new helper pass that strips all symbol names except debugging information.
This pass makes it easier to test wheter debugging info. influences optimization passes or not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59552 91177308-0d34-0410-b5e6-96231b3b80d8

Don't set neverHasSideEffects on x86's divide instructions, since
they trap on divide-by-zero, and this side effect is otherwise
unmodeled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59551 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up ScheduleNodeBottomUp methods, and make them more
consistent with ScheduleNodeTopDown methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59550 91177308-0d34-0410-b5e6-96231b3b80d8

Update a comment to reflect the current code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59549 91177308-0d34-0410-b5e6-96231b3b80d8

Remove integer promotion support for FP_EXTEND
and FP_ROUND.  Not sure what these were doing
here - probably they were sometimes (wrongly)
created with integer operands somewhere that
has since been fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59548 91177308-0d34-0410-b5e6-96231b3b80d8

Remove even more llvm.dbg variables.
Remove all dead globals from llvm.metadata.
Ignore linkonce linkage for selected llvm.dbg values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59547 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Remove HAVE_LT_DLOPEN from config.h.cmake because it was
removed from config.h.in.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59546 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify code using helper routines.  There is not
supposed to be any functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59545 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug introduced by my previous patch.  With this change, SPEC is now clean with prealloc splitting enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59544 91177308-0d34-0410-b5e6-96231b3b80d8

Add more const qualifiers. This fixes build breakage from r59540.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59542 91177308-0d34-0410-b5e6-96231b3b80d8

Make some methods const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59540 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize MallocFunc and FreeFunc properly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59538 91177308-0d34-0410-b5e6-96231b3b80d8

Whitespace cleanups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59532 91177308-0d34-0410-b5e6-96231b3b80d8

Add svn:ignore for build directories.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59531 91177308-0d34-0410-b5e6-96231b3b80d8

LegalizeTypes support for splitting and scalarizing
SCALAR_TO_VECTOR.  I didn't add the testcase, because
once llc gets past scalar-to-vector it hits a SPU target
lowering bug and explodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59530 91177308-0d34-0410-b5e6-96231b3b80d8

Add a utility function that detects whether a loop is guaranteed to be finite.

Use it to safely handle less-than-or-equals-to exit conditions in loops. These
also occur when the loop exit branch is exit on true because SCEV inverses the
icmp predicate.

Use it again to handle non-zero strides, but only with an unsigned comparison
in the exit condition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59528 91177308-0d34-0410-b5e6-96231b3b80d8

Rename stackprotector_create intrinsic to stackprotector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59519 91177308-0d34-0410-b5e6-96231b3b80d8