oota-llvm.git
12 years agoRestrict InstCombine from converting varargs to or from fixed args.
Jim Grosbach [Fri, 3 Feb 2012 00:00:55 +0000 (00:00 +0000)]
Restrict InstCombine from converting varargs to or from fixed args.

More targetted fix replacing d0e277d272d517ca1cda368267d199f0da7cad95.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149648 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRevert "Disable InstCombine unsafe folding bitcasts of calls w/ varargs."
Jim Grosbach [Fri, 3 Feb 2012 00:00:50 +0000 (00:00 +0000)]
Revert "Disable InstCombine unsafe folding bitcasts of calls w/ varargs."

This reverts commit d0e277d272d517ca1cda368267d199f0da7cad95.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149647 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRequire non-NULL register masks.
Jakob Stoklund Olesen [Thu, 2 Feb 2012 23:52:57 +0000 (23:52 +0000)]
Require non-NULL register masks.

It doesn't seem worthwhile to give meaning to a NULL register mask
pointer. It complicates all the code using register mask operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149646 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agobuild/make: Ensure make clean removes the LLVMBuild makefile fragment.
Daniel Dunbar [Thu, 2 Feb 2012 23:27:34 +0000 (23:27 +0000)]
build/make: Ensure make clean removes the LLVMBuild makefile fragment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149643 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agobuild/Make: Add missing dependency, LLVMBuild makefile fragment implicitly depends...
Daniel Dunbar [Thu, 2 Feb 2012 23:27:32 +0000 (23:27 +0000)]
build/Make: Add missing dependency, LLVMBuild makefile fragment implicitly depends on Makefile.config.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149642 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd pseudo-registers for pairs, triples, and quads of D registers.
Jakob Stoklund Olesen [Thu, 2 Feb 2012 22:45:32 +0000 (22:45 +0000)]
Add pseudo-registers for pairs, triples, and quads of D registers.

NEON loads and stores accept single and double spaced pairs, triples,
and quads of D registers.  This patch adds new register classes to
accurately model those constraints:

  Dn, Dn+1    Dn, Dn+2
  ----------------------
  DPair       DPairSpc
  DTriple     DTripleSpc
  DQuad       DQuadSpc

Also extend the existing QQ and QQQQ register classes to contains all Q
pairs and quads instead of just the aligned ones.

These new register classes will make it possible to accurately model
constraints on NEON loads and stores, and we can get rid of all the NEON
pseudo-instructions.  The late scheduler will be able to accurately
model instruction dependencies from the explicit operands.

This more than doubles the number of ARM registers, but the backend
passes are quite good at handling this. The llc -O0 compile time only
regresses by 1.5%.  Future work on register mask operands will recover
this regression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149640 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoUnix line endings
Matt Beaumont-Gay [Thu, 2 Feb 2012 19:00:49 +0000 (19:00 +0000)]
Unix line endings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149615 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoBBVectorize: Simplify code, no functionality change.
Benjamin Kramer [Thu, 2 Feb 2012 18:52:15 +0000 (18:52 +0000)]
BBVectorize: Simplify code, no functionality change.

Also silences warnings about bodyless for loops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149612 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoMinor changes from review.
Hal Finkel [Thu, 2 Feb 2012 17:29:39 +0000 (17:29 +0000)]
Minor changes from review.

As suggested by Nick Lewycky, the tree traversal queues have been changed to SmallVectors and the associated loops have been rotated. Also, an 80-col violation was fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149607 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoMove test/CodeGen/Generic/2012-02-01-CoalescerBug.ll to CodeGen/ARM, for now. It...
NAKAMURA Takumi [Thu, 2 Feb 2012 11:44:58 +0000 (11:44 +0000)]
Move test/CodeGen/Generic/2012-02-01-CoalescerBug.ll to CodeGen/ARM, for now. It requires TARGETS=arm.

I cannot reproduce a fixed issue with other targets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149604 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoMinor change in signature of the getZeroVector()
Elena Demikhovsky [Thu, 2 Feb 2012 09:20:18 +0000 (09:20 +0000)]
Minor change in signature of the getZeroVector()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149601 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoOptimization for SIGN_EXTEND operation on AVX.
Elena Demikhovsky [Thu, 2 Feb 2012 09:10:43 +0000 (09:10 +0000)]
Optimization for SIGN_EXTEND operation on AVX.
Special handling was added for v4i32 -> v4i64 and v8i16 -> v8i32
extensions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149600 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoUnbreak the MSVC build.
Francois Pichet [Thu, 2 Feb 2012 08:36:09 +0000 (08:36 +0000)]
Unbreak the MSVC build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149599 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRe-apply the coalescer fix from r149147. Commit r149597 should have fixed the llvm...
Lang Hames [Thu, 2 Feb 2012 08:01:53 +0000 (08:01 +0000)]
Re-apply the coalescer fix from r149147. Commit r149597 should have fixed the llvm-gcc and clang self-host issues.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149598 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoSet EFLAGS correctly in EmitLoweredSelect on X86.
Lang Hames [Thu, 2 Feb 2012 07:48:37 +0000 (07:48 +0000)]
Set EFLAGS correctly in EmitLoweredSelect on X86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149597 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoBreak as soon as the MustMapCurValNos flag is set - no need to reiterate.
Lang Hames [Thu, 2 Feb 2012 06:55:45 +0000 (06:55 +0000)]
Break as soon as the MustMapCurValNos flag is set - no need to reiterate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149596 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoVectorize long blocks in groups.
Hal Finkel [Thu, 2 Feb 2012 06:14:56 +0000 (06:14 +0000)]
Vectorize long blocks in groups.

Long basic blocks with many candidate pairs (such as in the SHA implementation in Perl 5.14; thanks to Roman Divacky for the example) used to take an unacceptably-long time to compile. Instead, break long blocks into groups so that no group has too many candidate pairs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149595 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoPR11868. The previous loop in LiveIntervals::join would sometimes fall over if
Lang Hames [Thu, 2 Feb 2012 05:37:34 +0000 (05:37 +0000)]
PR11868. The previous loop in LiveIntervals::join would sometimes fall over if
more than two adjacent ranges needed to be merged. The new version should be
able to handle an arbitrary sequence of adjancent ranges.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149588 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoSet the correct stack pointer register.
Akira Hatanaka [Thu, 2 Feb 2012 03:17:04 +0000 (03:17 +0000)]
Set the correct stack pointer register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149585 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoExpand EHSELECTION and EHSELECTION nodes. Set the correct exception pointer and
Akira Hatanaka [Thu, 2 Feb 2012 03:13:40 +0000 (03:13 +0000)]
Expand EHSELECTION and EHSELECTION nodes. Set the correct exception pointer and
selector registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149584 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd DWARF numbers of 64-bit registers.
Akira Hatanaka [Thu, 2 Feb 2012 02:56:14 +0000 (02:56 +0000)]
Add DWARF numbers of 64-bit registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149583 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRegen one last time.
Eric Christopher [Thu, 2 Feb 2012 01:11:30 +0000 (01:11 +0000)]
Regen one last time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149576 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoautoconf: fix build/src dir confusion
Dylan Noblesmith [Thu, 2 Feb 2012 00:54:18 +0000 (00:54 +0000)]
autoconf: fix build/src dir confusion

This was the cause of the silent failure to generate
clang's config.h. My bad.

Fix on r149563 / r149568.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149574 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRegenerate again.
Eric Christopher [Thu, 2 Feb 2012 00:19:05 +0000 (00:19 +0000)]
Regenerate again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149569 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoautoconf: restore old clang-srcdir behavior
Dylan Noblesmith [Thu, 2 Feb 2012 00:17:33 +0000 (00:17 +0000)]
autoconf: restore old clang-srcdir behavior

Keep the string empty when unspecified. Undoes
part of r149563.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149568 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRegenerate configure.
Eric Christopher [Thu, 2 Feb 2012 00:16:55 +0000 (00:16 +0000)]
Regenerate configure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149567 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoautoconf: honor --with-clang-srcdir
Dylan Noblesmith [Thu, 2 Feb 2012 00:11:14 +0000 (00:11 +0000)]
autoconf: honor --with-clang-srcdir

configure was silently failing to produce anything in the case
where clang wasn't at tools/clang/, resulting in compilation
errors much later in the build when config.h didn't exist.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149563 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoTypo
Pete Cooper [Wed, 1 Feb 2012 23:43:12 +0000 (23:43 +0000)]
Typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149562 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoFix the cmake build
Rafael Espindola [Wed, 1 Feb 2012 23:40:51 +0000 (23:40 +0000)]
Fix the cmake build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149561 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoInstruction scheduling itinerary for Intel Atom.
Andrew Trick [Wed, 1 Feb 2012 23:20:51 +0000 (23:20 +0000)]
Instruction scheduling itinerary for Intel Atom.

Adds an instruction itinerary to all x86 instructions, giving each a default latency of 1, using the InstrItinClass IIC_DEFAULT.

Sets specific latencies for Atom for the instructions in files X86InstrCMovSetCC.td, X86InstrArithmetic.td, X86InstrControl.td, and X86InstrShiftRotate.td. The Atom latencies for the remainder of the x86 instructions will be set in subsequent patches.

Adds a test to verify that the scheduler is working.

Also changes the scheduling preference to "Hybrid" for i386 Atom, while leaving x86_64 as ILP.

Patch by Preston Gurd!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149558 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoMove ARM subreg index compositions to the SubRegIndex itself.
Jakob Stoklund Olesen [Wed, 1 Feb 2012 23:16:43 +0000 (23:16 +0000)]
Move ARM subreg index compositions to the SubRegIndex itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149557 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoSpecify SubRegIndex components on the index itself.
Jakob Stoklund Olesen [Wed, 1 Feb 2012 23:16:41 +0000 (23:16 +0000)]
Specify SubRegIndex components on the index itself.

It is simpler to define a composite index directly:

  def ssub_2 : SubRegIndex<[dsub_1, ssub_0]>;
  def ssub_3 : SubRegIndex<[dsub_1, ssub_1]>;

Than specifying the composite indices on each register:

  CompositeIndices = [(ssub_2 dsub_1, ssub_0),
                      (ssub_3 dsub_1, ssub_1)] in ...

This also makes it clear that SubRegIndex composition is supposed to be
unique.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149556 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agofix cmake
Andrew Trick [Wed, 1 Feb 2012 22:28:29 +0000 (22:28 +0000)]
fix cmake

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149553 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoFix a bug in the TopoOrderRC comparison function.
Jakob Stoklund Olesen [Wed, 1 Feb 2012 22:19:26 +0000 (22:19 +0000)]
Fix a bug in the TopoOrderRC comparison function.

The final tie breaker comparison also needs to return +/-1, or 0.
This is not a less() function.

This could cause otherwise identical super-classes to be ordered
unstably, depending on what the system qsort routine does with a bad
compare function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149549 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAvoid creating an extract element to an illegal type after LegalizeTypes has run.
Mon P Wang [Wed, 1 Feb 2012 22:15:20 +0000 (22:15 +0000)]
Avoid creating an extract element to an illegal type after LegalizeTypes has run.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149548 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoVLIW specific scheduler framework that utilizes deterministic finite automaton (DFA).
Andrew Trick [Wed, 1 Feb 2012 22:13:57 +0000 (22:13 +0000)]
VLIW specific scheduler framework that utilizes deterministic finite automaton (DFA).

This new scheduler plugs into the existing selection DAG scheduling framework. It is a top-down critical path scheduler that tracks register pressure and uses a DFA for pipeline modeling.

Patch by Sergei Larin!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149547 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAvoid emitting empty arrays, they're not standard C++.
Jakob Stoklund Olesen [Wed, 1 Feb 2012 22:12:51 +0000 (22:12 +0000)]
Avoid emitting empty arrays, they're not standard C++.

It's only by luck that we haven't produced any yet, and clang refuses to
compile them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149546 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoTidy up.
Chad Rosier [Wed, 1 Feb 2012 18:45:51 +0000 (18:45 +0000)]
Tidy up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149521 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agocmake: make BUILD_SHARED_LIBS a visible option
Dylan Noblesmith [Wed, 1 Feb 2012 14:49:39 +0000 (14:49 +0000)]
cmake: make BUILD_SHARED_LIBS a visible option

It could only be specified on the commandline, and wouldn't show
up as an option in the GUI or when invoked via `cmake -i` at all.

This also tells CMake that it's a BOOL, rather than "UNINITIALIZED".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149506 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agotest/CodeGen/X86/avx-minmax.ll: Relax expressions for Win32 targets. YMM arguments...
NAKAMURA Takumi [Wed, 1 Feb 2012 14:35:29 +0000 (14:35 +0000)]
test/CodeGen/X86/avx-minmax.ll: Relax expressions for Win32 targets. YMM arguments are passed as indirect on Win32 x64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149505 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoautoconf: generate clang's private config.h header
Dylan Noblesmith [Wed, 1 Feb 2012 14:06:21 +0000 (14:06 +0000)]
autoconf: generate clang's private config.h header

The CMake build already generated one. Follows clang r149497.

This brings us one step closer to compiling and configuring clang
separately from LLVM using the autoconf build, too.

(I lack the right version of autoconf et al. to regen, but it
was a simple change, so I just updated configure manually.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149498 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoPassing AVX 256-bit structures in Win64 was wrong.
Elena Demikhovsky [Wed, 1 Feb 2012 10:46:14 +0000 (10:46 +0000)]
Passing AVX 256-bit structures in Win64 was wrong.
Fixed Win64 calling conventions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149494 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoShortened code in shuffle masks
Elena Demikhovsky [Wed, 1 Feb 2012 10:33:05 +0000 (10:33 +0000)]
Shortened code in shuffle masks

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149493 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoOptimization for "truncate" operation on AVX.
Elena Demikhovsky [Wed, 1 Feb 2012 07:56:44 +0000 (07:56 +0000)]
Optimization for "truncate" operation on AVX.
Truncating v4i64 -> v4i32 and v8i32 -> v8i16 may be done with set of shuffles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149485 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoSwitchInst refactoring.
Stepan Dyatkovskiy [Wed, 1 Feb 2012 07:49:51 +0000 (07:49 +0000)]
SwitchInst refactoring.
The purpose of refactoring is to hide operand roles from SwitchInst user (programmer). If you want to play with operands directly, probably you will need lower level methods than SwitchInst ones (TerminatorInst or may be User). After this patch we can reorganize SwitchInst operands and successors as we want.

What was done:

1. Changed semantics of index inside the getCaseValue method:
getCaseValue(0) means "get first case", not a condition. Use getCondition() if you want to resolve the condition. I propose don't mix SwitchInst case indexing with low level indexing (TI successors indexing, User's operands indexing), since it may be dangerous.
2. By the same reason findCaseValue(ConstantInt*) returns actual number of case value. 0 means first case, not default. If there is no case with given value, ErrorIndex will returned.
3. Added getCaseSuccessor method. I propose to avoid usage of TerminatorInst::getSuccessor if you want to resolve case successor BB. Use getCaseSuccessor instead, since internal SwitchInst organization of operands/successors is hidden and may be changed in any moment.
4. Added resolveSuccessorIndex and resolveCaseIndex. The main purpose of these methods is to see how case successors are really mapped in TerminatorInst.
4.1 "resolveSuccessorIndex" was created if you need to level down from SwitchInst to TerminatorInst. It returns TerminatorInst's successor index for given case successor.
4.2 "resolveCaseIndex" converts low level successors index to case index that curresponds to the given successor.

Note: There are also related compatability fix patches for dragonegg, klee, llvm-gcc-4.0, llvm-gcc-4.2, safecode, clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149481 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd pass printer passes in the right place.
Andrew Trick [Wed, 1 Feb 2012 07:16:20 +0000 (07:16 +0000)]
Add pass printer passes in the right place.

The pass pointer should never be referenced after sending it to
schedulePass(), which may delete the pass. To fix this bug I had to
clean up the design leading to more goodness.

You may notice now that any non-analysis pass is printed. So things like loop-simplify and lcssa show up, while target lib, target data, alias analysis do not show up. Normally, analysis don't mutate the IR, but you can now check this by using both -print-after and -print-before. The effects of analysis will now show up in between the two.

The llc path is still in bad shape. But I'll be improving it in my next checkin. Meanwhile, print-machineinstrs still works the same way. With print-before/after, many llc passes that were not printed before now are, some of these should be converted to analysis. A few very important passes, isel and scheduler, are not properly initialized, so not printed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149480 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agowhitespace
Andrew Trick [Wed, 1 Feb 2012 07:16:17 +0000 (07:16 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149479 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoDon't create VBROADCAST nodes if any nodes use the chain result from the load. Fixes...
Craig Topper [Wed, 1 Feb 2012 06:51:58 +0000 (06:51 +0000)]
Don't create VBROADCAST nodes if any nodes use the chain result from the load. Fixes PR11900.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149478 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoBBVectorize.cpp: Try to fix MSVC build. map::iterator and multimap::iterator are...
NAKAMURA Takumi [Wed, 1 Feb 2012 06:11:58 +0000 (06:11 +0000)]
BBVectorize.cpp: Try to fix MSVC build. map::iterator and multimap::iterator are incompatible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149475 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoA few of the changes suggested in code review (by Nick Lewycky)
Hal Finkel [Wed, 1 Feb 2012 05:51:45 +0000 (05:51 +0000)]
A few of the changes suggested in code review (by Nick Lewycky)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149472 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRevert Chris' commits up to r149348 that started causing VMCoreTests unit test to...
Argyrios Kyrtzidis [Wed, 1 Feb 2012 04:51:17 +0000 (04:51 +0000)]
Revert Chris' commits up to r149348 that started causing VMCoreTests unit test to fail.

These are:

r149348
r149351
r149352
r149354
r149356
r149357
r149361
r149362
r149364
r149365

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149470 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd a basic-block autovectorization pass.
Hal Finkel [Wed, 1 Feb 2012 03:51:43 +0000 (03:51 +0000)]
Add a basic-block autovectorization pass.

This is the initial checkin of the basic-block autovectorization pass along with some supporting vectorization infrastructure.
Special thanks to everyone who helped review this code over the last several months (especially Tobias Grosser).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149468 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoDisable InstCombine unsafe folding bitcasts of calls w/ varargs.
Jim Grosbach [Wed, 1 Feb 2012 00:08:17 +0000 (00:08 +0000)]
Disable InstCombine unsafe folding bitcasts of calls w/ varargs.

Changing arguments from being passed as fixed to varargs is unsafe, as
the ABI may require they be handled differently (stack vs. register, for
example).

Remove two tests which rely on the bitcast being folded into the direct
call, which is exactly the transformation that's unsafe.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149457 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoTidy up. One more return type mismatch fix.
Jim Grosbach [Tue, 31 Jan 2012 23:51:09 +0000 (23:51 +0000)]
Tidy up. One more return type mismatch fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149452 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRevert test commit
Brendon Cahoon [Tue, 31 Jan 2012 23:18:33 +0000 (23:18 +0000)]
Revert test commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149446 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agobz11794 : EarlyCSE stack overflow on long functions.
Lenny Maiorani [Tue, 31 Jan 2012 23:14:41 +0000 (23:14 +0000)]
bz11794 : EarlyCSE stack overflow on long functions.
Make the EarlyCSE optimizer not use recursion to do a depth first iteration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149445 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agotest commit, adding a blank space
Brendon Cahoon [Tue, 31 Jan 2012 23:13:42 +0000 (23:13 +0000)]
test commit, adding a blank space

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149444 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoFixed a crash in llvm-mc for Mach-O when a symbol difference expression uses a
Kevin Enderby [Tue, 31 Jan 2012 23:02:57 +0000 (23:02 +0000)]
Fixed a crash in llvm-mc for Mach-O when a symbol difference expression uses a
symbol from an assignment.  In this case the symbol did not have a fragment so
MCObjectWriter::IsSymbolRefDifferenceFullyResolved() should not have been
calling IsSymbolRefDifferenceFullyResolvedImpl() with a NULL fragment and should
just have returned false in that case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149442 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd Triple::getMacOSXVersion to replace crufty code in the clang driver.
Bob Wilson [Tue, 31 Jan 2012 22:32:29 +0000 (22:32 +0000)]
Add Triple::getMacOSXVersion to replace crufty code in the clang driver.

This new function provides a way to get the Mac OS X version number from
either generic "darwin" triples of macosx triples.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149438 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoDon't assign a value to NUM_TARGET_NAMED_SUBREGS.
Jakob Stoklund Olesen [Tue, 31 Jan 2012 21:51:53 +0000 (21:51 +0000)]
Don't assign a value to NUM_TARGET_NAMED_SUBREGS.

It was wrong and completely unused.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149433 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoMove the composite map into CodeGenSubRegIndex.
Jakob Stoklund Olesen [Tue, 31 Jan 2012 21:44:11 +0000 (21:44 +0000)]
Move the composite map into CodeGenSubRegIndex.

Each SubRegIndex keeps track of how it composes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149423 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd a TableGen CodeGenSubRegIndex class.
Jakob Stoklund Olesen [Tue, 31 Jan 2012 20:57:55 +0000 (20:57 +0000)]
Add a TableGen CodeGenSubRegIndex class.

This class is used to represent SubRegIndex instances instead of the raw
Record pointers that were used before.

No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149418 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRefactor loop for better readability.
Jim Grosbach [Tue, 31 Jan 2012 20:56:55 +0000 (20:56 +0000)]
Refactor loop for better readability.

Excellent suggestion from Ben Kramer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149417 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd explanatory comment.
Jim Grosbach [Tue, 31 Jan 2012 20:34:53 +0000 (20:34 +0000)]
Add explanatory comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149416 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRefCountedBaseVPTR needs the IntrusiveRefCntPtrInfo as friend,
Manuel Klimek [Tue, 31 Jan 2012 19:58:34 +0000 (19:58 +0000)]
RefCountedBaseVPTR needs the IntrusiveRefCntPtrInfo as friend,
now that this handles the release / retain calls.

Adds a regression test for that bug (which is a compile-time
regression) and for the last two changes to the IntrusiveRefCntPtr,
especially tests for the memory leak due to copy construction of the
ref-counted object and ensuring that the traits are used for release /
retain calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149411 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoTidy up. Trailing whitespace.
Jim Grosbach [Tue, 31 Jan 2012 19:47:32 +0000 (19:47 +0000)]
Tidy up. Trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149408 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoObvious unnecessary loop removal. Follow through from previous checkin.
Andrew Trick [Tue, 31 Jan 2012 18:54:19 +0000 (18:54 +0000)]
Obvious unnecessary loop removal. Follow through from previous checkin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149398 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd assembler dialect attribute in asm parser which lets target specific asm parser...
Devang Patel [Tue, 31 Jan 2012 18:14:05 +0000 (18:14 +0000)]
Add assembler dialect attribute in asm parser which lets target specific asm parser change dialect on the fly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149396 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRegenerate configure.
Rafael Espindola [Tue, 31 Jan 2012 17:18:47 +0000 (17:18 +0000)]
Regenerate configure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149391 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoFix BrainF compilation.
Francois Pichet [Tue, 31 Jan 2012 09:35:01 +0000 (09:35 +0000)]
Fix BrainF compilation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149375 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoIncrease the initial vector size to be equivalent to the size of the Deps
Bill Wendling [Tue, 31 Jan 2012 07:04:52 +0000 (07:04 +0000)]
Increase the initial vector size to be equivalent to the size of the Deps
vector. This potentially saves a resizing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149369 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoCache the size of the vector instead of calling .size() all over the place.
Bill Wendling [Tue, 31 Jan 2012 06:57:53 +0000 (06:57 +0000)]
Cache the size of the vector instead of calling .size() all over the place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149368 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove pcmpgt/pcmpeq intrinsics as clang is not using them.
Craig Topper [Tue, 31 Jan 2012 06:52:44 +0000 (06:52 +0000)]
Remove pcmpgt/pcmpeq intrinsics as clang is not using them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149367 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoeliminate the "string" form of ConstantArray::get, using
Chris Lattner [Tue, 31 Jan 2012 06:18:43 +0000 (06:18 +0000)]
eliminate the "string" form of ConstantArray::get, using
ConstantDataArray::getString instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149365 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoremove commented-out code.
Chris Lattner [Tue, 31 Jan 2012 06:17:26 +0000 (06:17 +0000)]
remove commented-out code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149364 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agowith recent changes, ConstantArray is never a "string". Remove the associated
Chris Lattner [Tue, 31 Jan 2012 06:05:00 +0000 (06:05 +0000)]
with recent changes, ConstantArray is never a "string".  Remove the associated
methods and constant fold the clients to false.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149362 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoupdate this to ConstantDataArray. There are no tests and this isn't using the prefer...
Chris Lattner [Tue, 31 Jan 2012 06:03:46 +0000 (06:03 +0000)]
update this to ConstantDataArray.  There are no tests and this isn't using the preferred functionality for ripping apart strings, so I have no way to test this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149361 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRAFast: Generalize the logic for return operands.
Andrew Trick [Tue, 31 Jan 2012 05:55:32 +0000 (05:55 +0000)]
RAFast: Generalize the logic for return operands.

This removes implicit assumption about the form of MI coming into regalloc. In particular, it should be independent of ProcessImplicitDefs which will eventually become a standard part of coming out of SSA--unless we simply can eliminate IMPLICIT_DEF completely. Current unit tests expose this once I remove incidental pass ordering restrictions.

This is not a final fix. Just a temporary workaround until I figure out the right way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149360 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agofix a small oversight that broke the fhourstones app.
Chris Lattner [Tue, 31 Jan 2012 05:18:56 +0000 (05:18 +0000)]
fix a small oversight that broke the fhourstones app.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149357 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoremove the last vestiges of llvm::GetConstantStringInfo, in CodeGen.
Chris Lattner [Tue, 31 Jan 2012 05:09:17 +0000 (05:09 +0000)]
remove the last vestiges of llvm::GetConstantStringInfo, in CodeGen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149356 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoeliminate the last uses of GetConstantStringInfo from this file, I didn't realize...
Chris Lattner [Tue, 31 Jan 2012 04:54:27 +0000 (04:54 +0000)]
eliminate the last uses of GetConstantStringInfo from this file, I didn't realize I was that close...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149354 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoAdd various coarse bit-width architecture predicates to llvm::Triple.
Chandler Carruth [Tue, 31 Jan 2012 04:52:32 +0000 (04:52 +0000)]
Add various coarse bit-width architecture predicates to llvm::Triple.
These are very useful for frontends and other utilities reasoning about
or selecting between triples.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149353 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agostart moving SimplifyLibcalls over to getConstantStringInfo, which is
Chris Lattner [Tue, 31 Jan 2012 04:43:11 +0000 (04:43 +0000)]
start moving SimplifyLibcalls over to getConstantStringInfo, which is
dramatically more efficient than GetConstantStringInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149352 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoChange ConstantArray::get to form a ConstantDataArray when possible,
Chris Lattner [Tue, 31 Jan 2012 04:42:22 +0000 (04:42 +0000)]
Change ConstantArray::get to form a ConstantDataArray when possible,
kicking in the big win of ConstantDataArray.  As part of this, change
the implementation of GetConstantStringInfo in ValueTracking to work
with ConstantDataArray (and not ConstantArray) making it dramatically,
amazingly, more efficient in the process and renaming it to
getConstantStringInfo.

This keeps around a GetConstantStringInfo entrypoint that (grossly)
forwards to getConstantStringInfo and constructs the std::string
required, but existing clients should move over to
getConstantStringInfo instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149351 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agorework this logic to not depend on the last argument to GetConstantStringInfo,
Chris Lattner [Tue, 31 Jan 2012 04:39:22 +0000 (04:39 +0000)]
rework this logic to not depend on the last argument to GetConstantStringInfo,
which is going away.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149348 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agodon't emit a 1-byte object as a .fill. This is silly and causes
Chris Lattner [Tue, 31 Jan 2012 03:39:24 +0000 (03:39 +0000)]
don't emit a 1-byte object as a .fill.  This is silly and causes
CodeGen/X86/global-sections.ll to fail with CDArray

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149343 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agouse the right accessor for ConstantDataArray.
Chris Lattner [Tue, 31 Jan 2012 03:16:39 +0000 (03:16 +0000)]
use the right accessor for ConstantDataArray.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149342 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agofix asmwriting of ConstantDataArray to use the right element count,
Chris Lattner [Tue, 31 Jan 2012 03:15:40 +0000 (03:15 +0000)]
fix asmwriting of ConstantDataArray to use the right element count,
simplify ConstantArray handling, since they can never be empty.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149341 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoenhance logic to support ConstantDataArray.
Chris Lattner [Tue, 31 Jan 2012 02:55:06 +0000 (02:55 +0000)]
enhance logic to support ConstantDataArray.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149340 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove all references to the old EH.
Bill Wendling [Tue, 31 Jan 2012 02:09:07 +0000 (02:09 +0000)]
Remove all references to the old EH.

There was always the current EH. -- Ministry of Truth

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149335 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoUpdate test to new EH model.
Bill Wendling [Tue, 31 Jan 2012 02:05:13 +0000 (02:05 +0000)]
Update test to new EH model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149333 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoUpdate test to new EH model.
Bill Wendling [Tue, 31 Jan 2012 02:04:20 +0000 (02:04 +0000)]
Update test to new EH model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149332 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove the now-dead llvm.eh.exception and llvm.eh.selector intrinsics.
Bill Wendling [Tue, 31 Jan 2012 01:58:48 +0000 (01:58 +0000)]
Remove the now-dead llvm.eh.exception and llvm.eh.selector intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149331 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agos/getInnerUnwindDest/getInnerResumeDest/g
Bill Wendling [Tue, 31 Jan 2012 01:48:40 +0000 (01:48 +0000)]
s/getInnerUnwindDest/getInnerResumeDest/g

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149328 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove the eh.exception and eh.selector intrinsics. Also remove a hack to copy
Bill Wendling [Tue, 31 Jan 2012 01:46:13 +0000 (01:46 +0000)]
Remove the eh.exception and eh.selector intrinsics. Also remove a hack to copy
over the catch information. The catch information is now tacked to the invoke
instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149326 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove ivar which is identical to another ivar.
Bill Wendling [Tue, 31 Jan 2012 01:25:54 +0000 (01:25 +0000)]
Remove ivar which is identical to another ivar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149323 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove unused ivars and s/getOuterUnwindDest/getOuterResumeDest/g.
Bill Wendling [Tue, 31 Jan 2012 01:22:03 +0000 (01:22 +0000)]
Remove unused ivars and s/getOuterUnwindDest/getOuterResumeDest/g.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149322 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove more dead functions.
Bill Wendling [Tue, 31 Jan 2012 01:18:21 +0000 (01:18 +0000)]
Remove more dead functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149318 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agos/getInnerUnwindDestNewEH/getInnerUnwindDest/g
Bill Wendling [Tue, 31 Jan 2012 01:15:59 +0000 (01:15 +0000)]
s/getInnerUnwindDestNewEH/getInnerUnwindDest/g

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149317 91177308-0d34-0410-b5e6-96231b3b80d8

12 years agoRemove some unused, old-EH methods.
Bill Wendling [Tue, 31 Jan 2012 01:14:49 +0000 (01:14 +0000)]
Remove some unused, old-EH methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149316 91177308-0d34-0410-b5e6-96231b3b80d8