oota-llvm.git
11 years ago[Object][ELF] Refactor ELFRelocationIterator into ELFEntityIterator. No functionality...
Michael J. Spencer [Sun, 6 Jan 2013 03:56:14 +0000 (03:56 +0000)]
[Object][ELF] Refactor ELFRelocationIterator into ELFEntityIterator. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171647 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd reference to dcommit on GettingStarted
Renato Golin [Sun, 6 Jan 2013 00:14:27 +0000 (00:14 +0000)]
Add reference to dcommit on GettingStarted

Signed-off-by: Renato Golin <renato.golin@linaro.org>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171642 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoInclude access modifiers in subprogram metadata IR comment.
David Blaikie [Sat, 5 Jan 2013 21:39:33 +0000 (21:39 +0000)]
Include access modifiers in subprogram metadata IR comment.

Based on code review feedback in r171604 from Chandler Carruth &
Eric Christopher.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171636 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGettingStarted: improve formatting and document that configure checks for
Dmitri Gribenko [Sat, 5 Jan 2013 18:10:06 +0000 (18:10 +0000)]
GettingStarted: improve formatting and document that configure checks for
'clang' to use it as the compiler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171630 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded debug message to ObjCARC when we transform objc_retainAutorelasedReturnValue...
Michael Gottesman [Sat, 5 Jan 2013 17:55:42 +0000 (17:55 +0000)]
Added debug message to ObjCARC when we transform objc_retainAutorelasedReturnValue => objc_retain since the operand to said function is not a return value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171629 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded debug message for ObjCARC when we zap an objc_autoreleaseReturnValue/objc_retai...
Michael Gottesman [Sat, 5 Jan 2013 17:55:35 +0000 (17:55 +0000)]
Added debug message for ObjCARC when we zap an objc_autoreleaseReturnValue/objc_retainAutoreleasedValue pair.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171628 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoswitch from pointer equality comparison to MDNode::getMostGenericTBAA
Chris Lattner [Sat, 5 Jan 2013 16:44:07 +0000 (16:44 +0000)]
switch from pointer equality comparison to MDNode::getMostGenericTBAA
when merging two TBAA tags, pointed out by Nuno.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171627 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFunnel the actual TargetTransformInfo pass from the SelectionDAGISel
Chandler Carruth [Sat, 5 Jan 2013 12:32:17 +0000 (12:32 +0000)]
Funnel the actual TargetTransformInfo pass from the SelectionDAGISel
pass into the SelectionDAG itself rather than snooping on the
implementation of that pass as exposed by the TargetMachine. This
removes the last direct client of the ScalarTargetTransformInfo class
outside of the TTI pass implementation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171625 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAttribute: Make hashes match when looking up AttributeImpls.
Benjamin Kramer [Sat, 5 Jan 2013 12:08:00 +0000 (12:08 +0000)]
Attribute: Make hashes match when looking up AttributeImpls.

This isn't optimal either but fixes a massive compile time regression from the
attribute uniquing work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171624 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix another place where we build the TTI pass to the new interface.
Chandler Carruth [Sat, 5 Jan 2013 11:54:35 +0000 (11:54 +0000)]
Fix another place where we build the TTI pass to the new interface.

Sorry for the noise here, 'make check' doesn't build this code. =/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171623 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoConvert the TargetTransformInfo from an immutable pass with dynamic
Chandler Carruth [Sat, 5 Jan 2013 11:43:11 +0000 (11:43 +0000)]
Convert the TargetTransformInfo from an immutable pass with dynamic
interfaces which could be extracted from it, and must be provided on
construction, to a chained analysis group.

The end goal here is that TTI works much like AA -- there is a baseline
"no-op" and target independent pass which is in the group, and each
target can expose a target-specific pass in the group. These passes will
naturally chain allowing each target-specific pass to delegate to the
generic pass as needed.

In particular, this will allow a much simpler interface for passes that
would like to use TTI -- they can have a hard dependency on TTI and it
will just be satisfied by the stub implementation when that is all that
is available.

This patch is a WIP however. In particular, the "stub" pass is actually
the one and only pass, and everything there is implemented by delegating
to the target-provided interfaces. As a consequence the tools still have
to explicitly construct the pass. Switching targets to provide custom
passes and sinking the stub behavior into the NoTTI pass is the next
step.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171621 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSwitch the loop vectorizer from VTTI to just use TTI directly.
Chandler Carruth [Sat, 5 Jan 2013 10:16:02 +0000 (10:16 +0000)]
Switch the loop vectorizer from VTTI to just use TTI directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171620 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSwitch the cost model analysis over to just the TTI interface.
Chandler Carruth [Sat, 5 Jan 2013 10:09:33 +0000 (10:09 +0000)]
Switch the cost model analysis over to just the TTI interface.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171619 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSwitch the BB vectorizer from the VTTI interface to the simple TTI
Chandler Carruth [Sat, 5 Jan 2013 10:05:28 +0000 (10:05 +0000)]
Switch the BB vectorizer from the VTTI interface to the simple TTI
interface.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171618 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSwitch SimplifyCFG over to the TargetTransformInfo interface rather than
Chandler Carruth [Sat, 5 Jan 2013 10:05:26 +0000 (10:05 +0000)]
Switch SimplifyCFG over to the TargetTransformInfo interface rather than
the ScalarTargetTransformInfo interface.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171617 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSwitch LoopIdiomRecognize to directly use the TargetTransformInfo
Chandler Carruth [Sat, 5 Jan 2013 10:00:09 +0000 (10:00 +0000)]
Switch LoopIdiomRecognize to directly use the TargetTransformInfo
interface rather than the ScalarTargetTransformInterface.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171616 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReplicate the APIs of ScalarTargetTransformInfo and
Chandler Carruth [Sat, 5 Jan 2013 09:56:20 +0000 (09:56 +0000)]
Replicate the APIs of ScalarTargetTransformInfo and
VectorTargetTransformInfo into the TargetTransformInfo pass,
implementing them be delegating back out to the two subobjects.

This is the first step to folding the interfaces together and making
TargetTransformInfo a normal analysis pass (specifically an analysis
group which targets can provide target-specific analysis pass
implementations of).

No callers are migrated here, this just stubs out the interface. Next
step will be to migrate all the callers to directly operate on TTI
instead of STTI or VTTI respectively. That will allow replacing the
machinery for delivering TTI without changing every caller at once.

WIP, I promise all the duplicated interfaces will be removed in the end,
this just decouples the steps of the process.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171615 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSwitch the empty and tombstone key enumerators to not have explicit
Chandler Carruth [Sat, 5 Jan 2013 08:47:26 +0000 (08:47 +0000)]
Switch the empty and tombstone key enumerators to not have explicit
values -- that's not required to fix the bug that was cropping up, and
the values selected made the enumeration's underlying type signed and
introduced some warnings. This fixes the -Werror build.

The underlying issue here was that the DenseMapInfo was casting values
completely outside the range of the underlying storage of the
enumeration to the enumeration's type. GCC went and "optimized" that
into infloops and other misbehavior. By providing designated special
values for these keys in the dense map, we ensure they are indeed
representable and that they won't be used for anything else.

It might be better to reuse None for the empty key and have the
tombstone share the value of the sentinel enumerator, but honestly
having 2 extra enumerators seemed not to matter and this seems a bit
simpler. I'll let Bill shuffle this around (or ask me to shuffle it
around) if he prefers it to look a different way.

I also made the switch a bit more clear (and produce a better assert)
that the enumerators are *never* going to show up and are errors if they
do.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171614 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoWhile the struct being defined in the AddressingMode.h header was
Chandler Carruth [Sat, 5 Jan 2013 08:19:20 +0000 (08:19 +0000)]
While the struct being defined in the AddressingMode.h header was
unused, there were transitive includes needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171613 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove unnecessary include.
Chandler Carruth [Sat, 5 Jan 2013 08:12:59 +0000 (08:12 +0000)]
Remove unnecessary include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171612 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoIR/Attributes: Provide EmptyKey and TombstoneKey in part of enum, as workaround for...
NAKAMURA Takumi [Sat, 5 Jan 2013 07:55:47 +0000 (07:55 +0000)]
IR/Attributes: Provide EmptyKey and TombstoneKey in part of enum, as workaround for gcc-4.4 take #2.

I will investigate, later, what was wrong. I am too tired for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171611 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEmit DW_TAG_formal_parameter for unnamed parameters.
David Blaikie [Sat, 5 Jan 2013 07:43:02 +0000 (07:43 +0000)]
Emit DW_TAG_formal_parameter for unnamed parameters.

This change essentially reverts r87069 which came without a test case. It
causes no regressions in the GDB 7.5 test suite & fixes 25 xfails (commit
to the test suite to follow). If anyone can present a test case that
demonstrates why this check is necessary I'd be happy to account for it in one
way or another.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171609 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRecommit r171461 which was incorrectly reverted. Mark DIV/IDIV instructions hasSideEf...
Craig Topper [Sat, 5 Jan 2013 07:39:25 +0000 (07:39 +0000)]
Recommit r171461 which was incorrectly reverted. Mark DIV/IDIV instructions hasSideEffects=1 because they can trap when dividing by 0. This is needed to keep early if conversion from moving them across basic blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171608 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert revision 171524. Original message:
Nadav Rotem [Sat, 5 Jan 2013 05:42:48 +0000 (05:42 +0000)]
Revert revision 171524. Original message:

URL: http://llvm.org/viewvc/llvm-project?rev=171524&view=rev
Log:
The current Intel Atom microarchitecture has a feature whereby when a function
returns early then it is slightly faster to execute a sequence of NOP
instructions to wait until the return address is ready,
as opposed to simply stalling on the ret instruction
until the return address is ready.

When compiling for X86 Atom only, this patch will run a pass, called
"X86PadShortFunction" which will add NOP instructions where less than four
cycles elapse between function entry and return.

It includes tests.

Patch by Andy Zhang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171603 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoWhitespace.
NAKAMURA Takumi [Sat, 5 Jan 2013 05:16:53 +0000 (05:16 +0000)]
Whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171601 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDenseMap: Appease -fstrict-aliasing on g++-4.4.
NAKAMURA Takumi [Sat, 5 Jan 2013 05:14:23 +0000 (05:14 +0000)]
DenseMap: Appease -fstrict-aliasing on g++-4.4.

With DenseMapInfo<Enum>, it is miscompiled on g++-4.4.

    static inline Enum getEmptyKey() { return Enum(<arbitrary int/unsigned value>); }

    isEauql(getEmptyKey(), ...)

The compiler mis-assumes the return value is not aliased to Enum.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171600 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDon't call destructors on MachineInstr and MachineOperand.
Jakob Stoklund Olesen [Sat, 5 Jan 2013 05:05:51 +0000 (05:05 +0000)]
Don't call destructors on MachineInstr and MachineOperand.

The series of patches leading up to this one makes llc -O0 run 8% faster.

When deallocating a MachineFunction, there is no need to visit all
MachineInstr and MachineOperand objects to deallocate them. All their
memory come from a BumpPtrAllocator that is about to be purged, and they
have empty destructors anyway.

This only applies when deallocating the MachineFunction.
DeleteMachineInstr() should still be used to recycle MI memory during
the codegen passes.

Remove the LeakDetector support for MachineInstr. I've never seen it
used before, and now it definitely doesn't work. With this patch, leaked
MachineInstrs would be much less of a problem since all of their memory
will be reclaimed by ~MachineFunction().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171599 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse ArrayRecycler for MachineInstr operand lists.
Jakob Stoklund Olesen [Sat, 5 Jan 2013 05:00:09 +0000 (05:00 +0000)]
Use ArrayRecycler for MachineInstr operand lists.

Instead of an std::vector<MachineOperand>, use MachineOperand arrays
from an ArrayRecycler living in MachineFunction.

This has several advantages:

- MachineInstr now has a trivial destructor, making it possible to
  delete them in batches when destroying MachineFunction. This will be
  enabled in a later patch.

- Bypassing malloc() and free() can be faster, depending on the system
  library.

- MachineInstr objects and their operands are allocated from the same
  BumpPtrAllocator, so they will usually be next to each other in
  memory, providing better locality of reference.

- Reduce MachineInstr footprint. A std::vector is 24 bytes, the new
  operand array representation only uses 8+4+1 bytes in MachineInstr.

- Better control over operand array reallocations. In the old
  representation, the use-def chains would be reordered whenever a
  std::vector reached its capacity. The new implementation never changes
  the use-def chain order.

Note that some decisions in the code generator depend on the use-def
chain orders, so this patch may cause different assembly to be produced
in a few cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171598 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd MachineRegisterInfo::moveOperands().
Jakob Stoklund Olesen [Sat, 5 Jan 2013 04:38:12 +0000 (04:38 +0000)]
Add MachineRegisterInfo::moveOperands().

This function works like memmove() for MachineOperands, except it also
updates any use-def chains containing the moved operands.

The use-def chains are updated without affecting the order of operands
in the list. That isn't possible when using the
removeRegOperandFromUseList() and addRegOperandToUseList() functions.

Callers to follow soon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171597 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRefactor the ScalarTargetTransformInfo API for querying about the
Chandler Carruth [Sat, 5 Jan 2013 03:36:17 +0000 (03:36 +0000)]
Refactor the ScalarTargetTransformInfo API for querying about the
legality of an address mode to not use a struct of four values and
instead to accept them as parameters. I'd love to have named parameters
here as most callers only care about one or two of these, but the
defaults aren't terribly scary to write out.

That said, there is no real impact of this as the passes aren't yet
using STTI for this and are still relying upon TargetLowering.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171595 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSink the AddressingModeMatcher helper class into an anonymous namespace
Chandler Carruth [Sat, 5 Jan 2013 02:09:22 +0000 (02:09 +0000)]
Sink the AddressingModeMatcher helper class into an anonymous namespace
next to its only user. This helper relies on TargetLowering information
that shouldn't be generally used throughout the Transfoms library, and
so it made little sense as a generic utility.

This also consolidates the file where we need to remove the remaining
uses of TargetLowering in favor of the IR-layer abstract interface in
TargetTransformInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171590 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRename the unittest from ArrayRecylerTest.cpp to ArrayRecyclerTest.cpp.
Chandler Carruth [Sat, 5 Jan 2013 02:08:43 +0000 (02:08 +0000)]
Rename the unittest from ArrayRecylerTest.cpp to ArrayRecyclerTest.cpp.

Fixes the CMake build. It took me cutting and pasting this before
I managed to see the missing character. =]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171589 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Fix data layout string. Add 64 to the list of native integer widths
Akira Hatanaka [Sat, 5 Jan 2013 02:00:56 +0000 (02:00 +0000)]
[mips] Fix data layout string. Add 64 to the list of native integer widths
and add stack alignment information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171587 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a method to create an AttributeSet from an AttrBuilder.
Bill Wendling [Sat, 5 Jan 2013 01:36:54 +0000 (01:36 +0000)]
Add a method to create an AttributeSet from an AttrBuilder.

The Attribute class is eventually going to represent one attribute. So we need
this class to create the set of attributes. Add some iterator methods to the
builder to access its internal bits in a nice way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171586 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a typo. Remove the duplicated test.
Nadav Rotem [Sat, 5 Jan 2013 01:17:46 +0000 (01:17 +0000)]
Fix a typo. Remove the duplicated test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171584 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoiLoopVectorize: Non commutative operators can be used as reduction variables as long...
Nadav Rotem [Sat, 5 Jan 2013 01:15:47 +0000 (01:15 +0000)]
iLoopVectorize: Non commutative operators can be used as reduction variables as long as the reduction chain is used in the LHS.

PR14803.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171583 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoForce a fixed unroll count on the target independent tests.
Nadav Rotem [Sat, 5 Jan 2013 00:58:48 +0000 (00:58 +0000)]
Force a fixed unroll count on the target independent tests.
This should fix clang-native-arm-cortex-a9. Thanks Renato.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171582 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd an ArrayRecycler class.
Jakob Stoklund Olesen [Sat, 5 Jan 2013 00:57:11 +0000 (00:57 +0000)]
Add an ArrayRecycler class.

This is similar to the existing Recycler allocator, but instead of
recycling individual objects from a BumpPtrAllocator, arrays of
different sizes can be allocated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171581 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEric thought that Darwin was right to use -1 consistently rather than
Chandler Carruth [Sat, 5 Jan 2013 00:42:50 +0000 (00:42 +0000)]
Eric thought that Darwin was right to use -1 consistently rather than
leaving this undefined, and despite the sentence in the standard that
seems to require it, I'll cede the point and assume its a bug in the
wording. Other parts of POSIX regularly allow for things to be -1
instead of undefined, this should too. Makes things more consistent too.

This should have to real impact for folks though.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171574 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a stray 'dnl' that my editor line-wrapped into this comment. Thanks
Chandler Carruth [Sat, 5 Jan 2013 00:34:40 +0000 (00:34 +0000)]
Fix a stray 'dnl' that my editor line-wrapped into this comment. Thanks
to filcab on IRC for spotting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171573 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake this an integer so we have enumeral types in the conditional
Eric Christopher [Sat, 5 Jan 2013 00:32:04 +0000 (00:32 +0000)]
Make this an integer so we have enumeral types in the conditional
expression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171571 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFinally, fix the autoconf setup to allow for a missing clock_gettime;
Chandler Carruth [Sat, 5 Jan 2013 00:29:06 +0000 (00:29 +0000)]
Finally, fix the autoconf setup to allow for a missing clock_gettime;
the source code should now be set up to handle this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171570 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoProvide a default constructor for TimeValue. This was used, but only in
Chandler Carruth [Sat, 5 Jan 2013 00:23:09 +0000 (00:23 +0000)]
Provide a default constructor for TimeValue. This was used, but only in
if-ed out code paths and on Windows. Hopefully restores the Windows
build. Thanks to Reid Kleckner for helping triage this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171568 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix warnings from llvm-gcc as seen on darwin10 (10.6).
Alex Rosenberg [Sat, 5 Jan 2013 00:21:12 +0000 (00:21 +0000)]
Fix warnings from llvm-gcc as seen on darwin10 (10.6).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171567 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTry to suppress the use of clock_gettime on Darwin which apparantly
Chandler Carruth [Sat, 5 Jan 2013 00:11:21 +0000 (00:11 +0000)]
Try to suppress the use of clock_gettime on Darwin which apparantly
defines _POSIX_CPUTIME but doesn't support the clock_* functions.

I don't test the value of _POSIX_CPUTIME because the spec merely says
that if it is defined, the CPU-specific timers are available, whereas it
says that _POSIX_TIMERS must be defined and defined to a value greater
than zero. However, this may not work, as the POSIX spec clearly states:

  "If the symbolic constant _POSIX_CPUTIME is defined, then the symbolic
  constant _POSIX_TIMERS shall also be defined by the implementation to
  have the value 200112L."

If this doesn't work, I'll add more hacks for Darwin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171565 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix an obvious typo spotted by Reid Kleckner, and breaking windows builds.
Chandler Carruth [Fri, 4 Jan 2013 23:46:04 +0000 (23:46 +0000)]
Fix an obvious typo spotted by Reid Kleckner, and breaking windows builds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171559 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGet rid of the 'Bits' mask in the attribute builder.
Bill Wendling [Fri, 4 Jan 2013 23:27:34 +0000 (23:27 +0000)]
Get rid of the 'Bits' mask in the attribute builder.

The bit mask thing will be a thing of the past. It's not extensible enough. Get
rid of its use here. Opt instead for using a vector to hold the attributes.

Note: Some of this code will become obsolete once the rewrite is further along.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171553 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd time getters to the process interface for requesting the elapsed
Chandler Carruth [Fri, 4 Jan 2013 23:19:55 +0000 (23:19 +0000)]
Add time getters to the process interface for requesting the elapsed
wall time, user time, and system time since a process started.

For walltime, we currently use TimeValue's interface and a global
initializer to compute a close approximation of total process runtime.

For user time, this adds support for an somewhat more precise timing
mechanism -- clock_gettime with the CLOCK_PROCESS_CPUTIME_ID clock
selected.

For system time, we have to do a full getrusage call to extract the
system time from the OS. This is expensive but unavoidable.

In passing, clean up the implementation of the old APIs and fix some
latent bugs in the Windows code. This might have manifested on Windows
ARM systems or other systems with strange 64-bit integer behavior.

The old API for this both user time and system time simultaneously from
a single getrusage call. While this results in fewer system calls, it
also results in a lower precision user time and if only user time is
desired, it introduces a higher overhead. It may be worthwhile to switch
some of the pass timers to not track system time and directly track user
and wall time. The old API also tracked walltime in a confusing way --
it just set it to the current walltime rather than providing any measure
of wall time since the process started the way buth user and system time
are tracked. The new API is more consistent here.

The plan is to eventually implement these methods for a *child* process
by using the wait3(2) system call to populate an rusage struct
representing the whole subprocess execution. That way, after waiting on
a child process its stats will become accurate and cheap to query.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171551 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agotabs-to-spaces
Andrew Trick [Fri, 4 Jan 2013 23:11:35 +0000 (23:11 +0000)]
tabs-to-spaces

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171550 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove 'break' to the right place to prevent fallthru. There is no test-case
Jakub Staszak [Fri, 4 Jan 2013 23:01:26 +0000 (23:01 +0000)]
Move 'break' to the right place to prevent fallthru. There is no test-case
because conditions in the next case prevented from doing anything nasty.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171549 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSpecial case Recycler::clear(BumpPtrAllocator).
Jakob Stoklund Olesen [Fri, 4 Jan 2013 22:35:45 +0000 (22:35 +0000)]
Special case Recycler::clear(BumpPtrAllocator).

A BumpPtrAllocator has an empty Deallocate() method, but
Recycler::clear() would still call it for every single object ever
allocated, bringing all those objects into cache. As a bonus,
iplist::remove() will also write to the Prev/Next pointers on all the
objects, so all those cache lines have to be written back to RAM before
the pages are given back to the OS.

Stop wasting time and memory bandwith by using the new
clearAndLeakUnsafely() function to jettison all the recycled objects.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171541 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd an iplist::clearAndLeakNodesUnsafely() function.
Jakob Stoklund Olesen [Fri, 4 Jan 2013 22:35:42 +0000 (22:35 +0000)]
Add an iplist::clearAndLeakNodesUnsafely() function.

The iplist::clear() function can be quite expensive because it traverses
the entire list, calling deleteNode() and removeNodeFromList() on each
element. If node destruction and deallocation can be handled some other
way, clearAndLeakNodesUnsafely() can be used to jettison all nodes
without bringing them into cache.

The function name is meant to be ominous.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171540 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove an assertion so it doesn't dereference end().
Jakob Stoklund Olesen [Fri, 4 Jan 2013 22:17:31 +0000 (22:17 +0000)]
Move an assertion so it doesn't dereference end().

The R600 target has test cases that exercises this code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171538 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDo not vectorize loops with subtraction reductions
Paul Redmond [Fri, 4 Jan 2013 22:10:16 +0000 (22:10 +0000)]
Do not vectorize loops with subtraction reductions

Since subtraction does not commute the loop vectorizer incorrectly vectorizes
reductions such as x = A[i] - x.

Disabling for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171537 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a name for the anonymous type we're creating for subrange
Eric Christopher [Fri, 4 Jan 2013 21:51:53 +0000 (21:51 +0000)]
Add a name for the anonymous type we're creating for subrange
types and a FIXME for what we should be doing. Should solve the
immediacy of PR12069 where our debug info is crashing another
tool.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171536 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded DEBUG message to ObjCARC when we optimize objc_retain => objc_retainAutorelased...
Michael Gottesman [Fri, 4 Jan 2013 21:30:38 +0000 (21:30 +0000)]
Added DEBUG message to ObjCARC when we optimize objc_retain => objc_retainAutorelasedReturnValue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171535 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFixed up some DEBUG messages where I was putting in the text of a message the method...
Michael Gottesman [Fri, 4 Jan 2013 21:29:57 +0000 (21:29 +0000)]
Fixed up some DEBUG messages where I was putting in the text of a message the method where it was being called when I should have just prefixed the actual message with Pass::Method.

Additionally I fixed some whitespace issues.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171534 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a warning
Nadav Rotem [Fri, 4 Jan 2013 21:08:44 +0000 (21:08 +0000)]
Fix a warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171525 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoThe current Intel Atom microarchitecture has a feature whereby when a function
Preston Gurd [Fri, 4 Jan 2013 20:54:54 +0000 (20:54 +0000)]
The current Intel Atom microarchitecture has a feature whereby when a function
returns early then it is slightly faster to execute a sequence of NOP
instructions to wait until the return address is ready,
as opposed to simply stalling on the ret instruction
until the return address is ready.

When compiling for X86 Atom only, this patch will run a pass, called
"X86PadShortFunction" which will add NOP instructions where less than four
cycles elapse between function entry and return.

It includes tests.

Patch by Andy Zhang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171524 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGeneral cleanups.
Bill Wendling [Fri, 4 Jan 2013 20:54:35 +0000 (20:54 +0000)]
General cleanups.

* Remove dead methods.
* Use the 'operator==' method instead of 'contains', which isn't needed.
* Fix some comments.

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171523 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[Object][ELF] Add a maximum alignment. This is used by createELFObjectFile to create...
Michael J. Spencer [Fri, 4 Jan 2013 20:36:28 +0000 (20:36 +0000)]
[Object][ELF] Add a maximum alignment. This is used by createELFObjectFile to create a properly aligned reader.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171520 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] MipsTargetLowering::getSetCCResultType should return a vector type if
Akira Hatanaka [Fri, 4 Jan 2013 20:06:01 +0000 (20:06 +0000)]
[mips] MipsTargetLowering::getSetCCResultType should return a vector type if
vectors are being compared.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171517 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] 80 columns.
Akira Hatanaka [Fri, 4 Jan 2013 19:38:05 +0000 (19:38 +0000)]
[mips] 80 columns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171515 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix how YAML I/O detects flow sequences.
Nick Kledzik [Fri, 4 Jan 2013 19:32:00 +0000 (19:32 +0000)]
Fix how YAML I/O detects flow sequences.
Update test case to verify flow sequence is
written as a flow sequence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171514 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Reorder template parameters. Remove class shift_rotate_imm32 and
Akira Hatanaka [Fri, 4 Jan 2013 19:25:46 +0000 (19:25 +0000)]
[mips] Reorder template parameters. Remove class shift_rotate_imm32 and
shift_rotate_imm64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171513 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMemory Dependence Analysis: fix a miscompile that uses DT to approxmiate the
Manman Ren [Fri, 4 Jan 2013 19:19:47 +0000 (19:19 +0000)]
Memory Dependence Analysis: fix a miscompile that uses DT to approxmiate the
reachablity.

We conservatively approximate the reachability analysis by saying it is not
reachable if there is a single path starting from "From" and the path does not
reach "To".

rdar://12801584

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171512 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Refactor conditional move instructions.
Akira Hatanaka [Fri, 4 Jan 2013 19:16:38 +0000 (19:16 +0000)]
[mips] Refactor conditional move instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171511 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Refactor instructions which move data from or to coprocessors.
Akira Hatanaka [Fri, 4 Jan 2013 19:13:49 +0000 (19:13 +0000)]
[mips] Refactor instructions which move data from or to coprocessors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171510 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate the gcc-loops benchmarks results with the new automatic unrolling feature.
Nadav Rotem [Fri, 4 Jan 2013 19:10:34 +0000 (19:10 +0000)]
Update the gcc-loops benchmarks results with the new automatic unrolling feature.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171509 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agofix a couple of typos
Eli Bendersky [Fri, 4 Jan 2013 19:09:15 +0000 (19:09 +0000)]
fix a couple of typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171508 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove unused #include
Eli Bendersky [Fri, 4 Jan 2013 19:08:43 +0000 (19:08 +0000)]
Remove unused #include

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171507 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPowerPC: Fix eh_frame relocation for PIC
Adhemerval Zanella [Fri, 4 Jan 2013 19:08:13 +0000 (19:08 +0000)]
PowerPC: Fix eh_frame relocation for PIC

This patch fixes the PPC eh_frame definitions for the personality and
frame unwinding for PIC objects. It makes PIC build correctly creates
relative relocations in the '.rela.eh_frame' segments and thus avoiding
a text relocation that generates a DT_TEXTREL segments in link phase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171506 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd linpack-pc bench
Nadav Rotem [Fri, 4 Jan 2013 19:00:42 +0000 (19:00 +0000)]
Add linpack-pc bench

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171499 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd Textmate bundle for Tablegen syntax-highlighting
Joel Jones [Fri, 4 Jan 2013 18:55:24 +0000 (18:55 +0000)]
Add Textmate bundle for Tablegen syntax-highlighting

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171496 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoChange the default number of registers to prevent unrolling on targets that dont...
Nadav Rotem [Fri, 4 Jan 2013 18:40:39 +0000 (18:40 +0000)]
Change the default number of registers to prevent unrolling on targets that dont have this hook.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171489 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a space to the end of the line so we don't get "itsbounds" in output.
Eric Christopher [Fri, 4 Jan 2013 18:30:36 +0000 (18:30 +0000)]
Add a space to the end of the line so we don't get "itsbounds" in output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171487 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agosmall fixes to enable the reuse of the pass manager across multiple modules
Pedro Artigas [Fri, 4 Jan 2013 18:04:42 +0000 (18:04 +0000)]
small fixes to enable the reuse of the pass manager across multiple modules

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171475 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd section information for the DWARF5 split debug proposal
Eric Christopher [Fri, 4 Jan 2013 17:59:22 +0000 (17:59 +0000)]
Add section information for the DWARF5 split debug proposal
string offset section.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171474 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake comment a bit more clear.
Eric Christopher [Fri, 4 Jan 2013 17:59:17 +0000 (17:59 +0000)]
Make comment a bit more clear.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171473 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove the loop vectorizer from O2 to O3. It looks like the increase in code size actua...
Nadav Rotem [Fri, 4 Jan 2013 17:57:44 +0000 (17:57 +0000)]
Move the loop vectorizer from O2 to O3. It looks like the increase in code size actually hurts the performance on many programs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171471 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate the docs about the new unroll features.
Nadav Rotem [Fri, 4 Jan 2013 17:49:45 +0000 (17:49 +0000)]
Update the docs about the new unroll features.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171470 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorizer:
Nadav Rotem [Fri, 4 Jan 2013 17:48:25 +0000 (17:48 +0000)]
LoopVectorizer:

1. Add code to estimate register pressure.
2. Add code to select the unroll factor based on register pressure.
3. Add bits to TargetTransformInfo to provide the number of registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171469 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert revision: 171467. This transformation is incorrect and makes some tests fail...
Nadav Rotem [Fri, 4 Jan 2013 17:35:21 +0000 (17:35 +0000)]
Revert revision: 171467. This transformation is incorrect and makes some tests fail. Original message:

Simplified TRUNCATE operation that comes after SETCC. It is possible since SETCC result is 0 or -1.
Added a test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171468 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSimplified TRUNCATE operation that comes after SETCC. It is possible since SETCC...
Elena Demikhovsky [Thu, 3 Jan 2013 08:48:33 +0000 (08:48 +0000)]
Simplified TRUNCATE operation that comes after SETCC. It is possible since SETCC result is 0 or -1.
Added a test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171467 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert "Mark DIV/IDIV instructions hasSideEffects=1 because they can trap when dividi...
Michael Gottesman [Thu, 3 Jan 2013 08:18:30 +0000 (08:18 +0000)]
Revert "Mark DIV/IDIV instructions hasSideEffects=1 because they can trap when dividing by 0. This is needed to keep early if conversion from moving them across basic blocks."

This reverts commit r171461 since it breaks the following tests:

Clang :: Analysis/outofbound-notwork.c
Clang :: Analysis/string-fail.c
Clang :: CXX/basic/basic.lookup/basic.lookup.qual/p6-0x.cpp
Clang :: CXX/basic/basic.lookup/basic.lookup.unqual/p15.cpp
Clang :: CXX/dcl.dcl/dcl.spec/dcl.fct.spec/p4.cpp
Clang :: CXX/dcl.dcl/dcl.spec/dcl.stc/p10.cpp
Clang :: CXX/temp/temp.param/p14.cpp
Clang :: CXX/temp/temp.res/temp.dep.res/temp.point/p1.cpp
Clang :: CodeGen/2009-02-13-zerosize-union-field-ppc.c
Clang :: CodeGen/blocks-2.c
Clang :: CodeGen/libcalls-d.c
Clang :: CodeGen/libcalls-ld.c
Clang :: CodeGenCXX/conversion-function.cpp
Clang :: CodeGenCXX/debug-info-limit-type.cpp
Clang :: CodeGenCXX/inheriting-constructor.cpp
Clang :: FixIt/fixit-errors.c
Clang :: FixIt/fixit-pmem.cpp
Clang :: Modules/namespaces.cpp
Clang :: PCH/changed-files.c
Clang :: PCH/pr4489.c
Clang :: PCH/source-manager-stack.c
Clang :: Parser/cxx-ambig-decl-expr-xfail.cpp
Clang :: SemaCXX/switch-implicit-fallthrough-cxx98.cpp
Clang :: SemaTemplate/instantiate-function-1.mm

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171466 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoChanged two debug statements that state that a queue had finished being processed...
Michael Gottesman [Thu, 3 Jan 2013 08:09:27 +0000 (08:09 +0000)]
Changed two debug statements that state that a queue had finished being processed when said queue was really a list to state a list had finished being processed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171465 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded DEBUG message for ObjCARC when we zap a push/pop pair in ObjCARCAPElim::OptimizeBB.
Michael Gottesman [Thu, 3 Jan 2013 08:09:17 +0000 (08:09 +0000)]
Added DEBUG message for ObjCARC when we zap a push/pop pair in ObjCARCAPElim::OptimizeBB.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171464 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded DEBUG message to ObjCARC when we transform objc_initWeak(p, null) => *p = null.
Michael Gottesman [Thu, 3 Jan 2013 07:32:53 +0000 (07:32 +0000)]
Added DEBUG message to ObjCARC when we transform objc_initWeak(p, null) => *p = null.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171463 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded DEBUG message for ObjCARC when an inline asm marker is inserted for architectur...
Michael Gottesman [Thu, 3 Jan 2013 07:32:41 +0000 (07:32 +0000)]
Added DEBUG message for ObjCARC when an inline asm marker is inserted for architectures where this is required to perform a retainAutoreleasedReturnValue optimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171462 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMark DIV/IDIV instructions hasSideEffects=1 because they can trap when dividing by...
Craig Topper [Thu, 3 Jan 2013 06:40:20 +0000 (06:40 +0000)]
Mark DIV/IDIV instructions hasSideEffects=1 because they can trap when dividing by 0. This is needed to keep early if conversion from moving them across basic blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171461 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoBetter comment on VTTI::getShuffleCost
Hal Finkel [Thu, 3 Jan 2013 05:02:41 +0000 (05:02 +0000)]
Better comment on VTTI::getShuffleCost

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171459 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCompiler.h: Leave LLVM_BUILTIN_UNREACHABLE undefined if it is unavailable in host...
NAKAMURA Takumi [Thu, 3 Jan 2013 03:30:22 +0000 (03:30 +0000)]
Compiler.h: Leave LLVM_BUILTIN_UNREACHABLE undefined if it is unavailable in host compiler.

Users of LLVM_BUILTIN_UNREACHABLE should be responsible in the case when LLVM_BUILTIN_UNREACHABLE is undefined.

Actually, (0, (p)) in LLVM_ASSUME_ALIGNED(p, a) caused thousands of warnings on g++-4.4. It was a motivation in this commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171455 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a subtype parameter to VTTI::getShuffleCost
Hal Finkel [Thu, 3 Jan 2013 02:34:09 +0000 (02:34 +0000)]
Add a subtype parameter to VTTI::getShuffleCost

In order to cost subvector insertion and extraction, we need to know
the type of the subvector being extracted.

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171453 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert everything to r171366 to try to fix the build.
Bill Wendling [Thu, 3 Jan 2013 02:01:50 +0000 (02:01 +0000)]
Revert everything to r171366 to try to fix the build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171450 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReformat the rst text.
Nadav Rotem [Thu, 3 Jan 2013 01:56:33 +0000 (01:56 +0000)]
Reformat the rst text.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171449 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTry again to revert the bad patch. The tree was reverted for some unknown reason
Bill Wendling [Thu, 3 Jan 2013 01:54:39 +0000 (01:54 +0000)]
Try again to revert the bad patch. The tree was reverted for some unknown reason
before the last time.

--- Reverse-merging r171442 into '.':
U    include/llvm/IR/Attributes.h
U    lib/IR/Attributes.cpp
U    lib/IR/AttributeImpl.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171448 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a default Index for VTTI::getShuffleCost
Hal Finkel [Thu, 3 Jan 2013 01:50:51 +0000 (01:50 +0000)]
Add a default Index for VTTI::getShuffleCost

When Kind == (Broadcast or Reverse) then Index is not used; make it an optional parameter.

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171447 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorizer: Test the unrolling flag.
Nadav Rotem [Thu, 3 Jan 2013 01:47:31 +0000 (01:47 +0000)]
LoopVectorizer: Test the unrolling flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171446 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorizer: Document the unrolling feature.
Nadav Rotem [Thu, 3 Jan 2013 01:47:02 +0000 (01:47 +0000)]
LoopVectorizer: Document the unrolling feature.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171445 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert patch. Something snuck in there that shouldn't be.
Bill Wendling [Thu, 3 Jan 2013 01:46:27 +0000 (01:46 +0000)]
Revert patch. Something snuck in there that shouldn't be.

--- Reverse-merging r171441 into '.':
U    include/llvm/IR/Attributes.h
U    lib/IR/Attributes.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171444 91177308-0d34-0410-b5e6-96231b3b80d8