oota-llvm.git
10 years ago[mips] Add calling convention tests covering O32, N32, and N64.
Daniel Sanders [Wed, 16 Apr 2014 09:59:46 +0000 (09:59 +0000)]
[mips] Add calling convention tests covering O32, N32, and N64.

Summary:
I had difficulty finding tests for the N32 and N64 ABI so I've added a
collection of calling convention tests based on the document MIPS ABIs
Described (MD00305), the MIPSpro N32 Handbook, and the SYSV ABI. Where the
documents/implementations disagree, I've used GCC to resolve the conflict.

A few interesting details:
* For N32, LLVM uses 64-bit pointers when saving $ra despite pointers being
  32-bit. I've yet to find a supporting statement in the ABI documentation but
  the current behaviour matches GCC.

* For O32, the non-variable portion of a varargs argument list is also subject
  to the rule that floating-point is passed via GPR's (on N32/N64 only the
  variable portion is subject to this rule). This agrees with GCC's behaviour
  and the SYSV ABI but contradicts part of the MIPSpro N32 Handbook which talks about O32's behaviour.

* The N32 implementation has the wrong callee-saved register list.
  (I already have a fix for this but will commit it as a follow-up).

I've left RUN-TODO lines in for O32 on MIPS64. I don't plan to support this case
for now but we should revisit it.

Reviewers: matheusalmeida, vmedic

Reviewed By: matheusalmeida

Differential Revision: http://reviews.llvm.org/D3339

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206370 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Nuke to useless functions. The implicit ones are sufficient
Chandler Carruth [Wed, 16 Apr 2014 09:21:29 +0000 (09:21 +0000)]
[Allocator] Nuke to useless functions. The implicit ones are sufficient
here (obviously).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206369 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: explicitly ask for Apple NEON syntax so test passes on Linux
Tim Northover [Wed, 16 Apr 2014 09:13:44 +0000 (09:13 +0000)]
ARM64: explicitly ask for Apple NEON syntax so test passes on Linux

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206368 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: mark x7 as used when an i128 gets shunted onto the stack.
Tim Northover [Wed, 16 Apr 2014 09:03:25 +0000 (09:03 +0000)]
ARM64: mark x7 as used when an i128 gets shunted onto the stack.

The second half of a split i128 was ending up in x7, which is not a good thing.

This is another part of PR19432.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206366 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDAGCombiner: don't optimise non-existant litpool load
Tim Northover [Wed, 16 Apr 2014 09:03:09 +0000 (09:03 +0000)]
DAGCombiner: don't optimise non-existant litpool load

This particular DAG combine is designed to kick in when both ConstantFPs will
end up being loaded via a litpool, however those nodes have a semi-legal
status, dictated by isFPImmLegal so in some cases there wouldn't have been a
litpool in the first place. Don't try to be clever in those circumstances.

Picked up while merging some AArch64 tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206365 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSimplify a static_assert so VS2013 can build it
Timur Iskhodzhanov [Wed, 16 Apr 2014 08:30:32 +0000 (08:30 +0000)]
Simplify a static_assert so VS2013 can build it

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206363 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCOFF: fix an off by one error
Saleem Abdulrasool [Wed, 16 Apr 2014 06:22:53 +0000 (06:22 +0000)]
COFF: fix an off by one error

Adjust the tests to validate the number of auxiliary entries used to store the
filename.

Thanks to majnemer's sharp eye for catching the missing - 1 in the round up
calculation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206359 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConvert SelectionDAG::getVTList to use ArrayRef
Craig Topper [Wed, 16 Apr 2014 06:10:51 +0000 (06:10 +0000)]
Convert SelectionDAG::getVTList to use ArrayRef

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206357 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead...
Craig Topper [Wed, 16 Apr 2014 04:21:27 +0000 (04:21 +0000)]
[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead of comparing to nullptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206356 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCOFF: add support for .file symbols
Saleem Abdulrasool [Wed, 16 Apr 2014 04:15:32 +0000 (04:15 +0000)]
COFF: add support for .file symbols

Add support for emitting .file records.  This is mostly a quality of
implementation change (more complete support for COFF file emission) that was
noticed while working on COFF file emission for Windows on ARM.

A .file record is emitted as a symbol with storage class FILE (103) and the name
".file".  A series of auxiliary format 4 records follow which contain the file
name.  The filename is stored as an ANSI string and is padded with NULL if the
length is not a multiple of COFF::SymbolSize (18).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206355 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agotools: fix invalid printing, buffer overrun in llvm-readobj
Saleem Abdulrasool [Wed, 16 Apr 2014 04:15:29 +0000 (04:15 +0000)]
tools: fix invalid printing, buffer overrun in llvm-readobj

All auxiliary records are consumed when accessing a File record.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206354 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTarget: whitespace
Saleem Abdulrasool [Wed, 16 Apr 2014 04:15:25 +0000 (04:15 +0000)]
Target: whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206353 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Extend r600 sign_extend_inreg tests for EG
Matt Arsenault [Wed, 16 Apr 2014 01:41:34 +0000 (01:41 +0000)]
R600: Extend r600 sign_extend_inreg tests for EG

Patch by: Jan Vesely <jan.vesely@rutgers.edu>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206349 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Expand sign extension of vectors.
Matt Arsenault [Wed, 16 Apr 2014 01:41:30 +0000 (01:41 +0000)]
R600: Expand sign extension of vectors.

Setting vector types to expand will result in scalarization on pre SI hw,
as those gpus don't have vector shifts either.
Expand also i32 vectors, this helps llvm make the correct decision
about scalarizing the vector ops.

v2: move setOperation() calls to R600ISelLowering.cpp.
    cleanup the SI code to make it obvious that this patch does is nop for SI

Patch by: Jan Vesely <jan.vesely@rutgers.edu>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206348 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64,C++11] Tidy up branch relaxation a bit w/ c++11.
Jim Grosbach [Wed, 16 Apr 2014 00:42:46 +0000 (00:42 +0000)]
[ARM64,C++11] Tidy up branch relaxation a bit w/ c++11.

No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206344 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: Nuke some dead code.
Jim Grosbach [Wed, 16 Apr 2014 00:42:43 +0000 (00:42 +0000)]
ARM64: Nuke some dead code.

Missed in previous commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206343 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64,C++11] Clean up the ARM64 LOH collection pass.
Jim Grosbach [Tue, 15 Apr 2014 22:57:02 +0000 (22:57 +0000)]
[ARM64,C++11] Clean up the ARM64 LOH collection pass.

Range'ify a bunch of loops, mainly. As a result, we have a variety
of objects via reference rather than by pointer, so propogate that
through the various helper functions where it makes sense.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206337 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Print code size along with used registers
Matt Arsenault [Tue, 15 Apr 2014 22:40:47 +0000 (22:40 +0000)]
R600/SI: Print code size along with used registers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206336 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Print more immediates in hex format
Matt Arsenault [Tue, 15 Apr 2014 22:32:49 +0000 (22:32 +0000)]
R600/SI: Print more immediates in hex format

Print in decimal for inline immediates, and hex otherwise. Use hex
always for offsets in addressing offsets.

This approximately matches what the shader compiler does.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206335 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Cleanup parsing of register names.
Matt Arsenault [Tue, 15 Apr 2014 22:32:42 +0000 (22:32 +0000)]
R600/SI: Cleanup parsing of register names.

Try to figure out the class and number of subregisters.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206334 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake this test not match its own filename, when being run from a path that includes...
Nick Lewycky [Tue, 15 Apr 2014 22:29:32 +0000 (22:29 +0000)]
Make this test not match its own filename, when being run from a path that includes the string 'add'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206331 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix loads of i1
Matt Arsenault [Tue, 15 Apr 2014 22:28:39 +0000 (22:28 +0000)]
R600/SI: Fix loads of i1

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206330 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRegionInfo: Do not access a value that was just moved away
Tobias Grosser [Tue, 15 Apr 2014 22:09:36 +0000 (22:09 +0000)]
RegionInfo: Do not access a value that was just moved away

This fixes a regression introduced in r206310.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206328 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Fold the two templated overloads into a single one with
Chandler Carruth [Tue, 15 Apr 2014 21:51:14 +0000 (21:51 +0000)]
[Allocator] Fold the two templated overloads into a single one with
a default argument. The allocator interface we're modeling doesn't
distinguish between array and non-array allocation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206327 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Remove a really problematic overload. This is very confusing
Chandler Carruth [Tue, 15 Apr 2014 21:36:02 +0000 (21:36 +0000)]
[Allocator] Remove a really problematic overload. This is very confusing
because there is another (size_t, size_t) overload of Allocator, and the
only distinguishing factor is that one is a tempalte and the other
isn't. There was only one usage of this and that one was easily
converted to carry the alignment constraint in the type itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206325 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake FastISel::SelectInstruction return before target specific fast-isel code
Akira Hatanaka [Tue, 15 Apr 2014 21:30:06 +0000 (21:30 +0000)]
Make FastISel::SelectInstruction return before target specific fast-isel code
handles Intrinsic::trap if TargetOptions::TrapFuncName is set.

This fixes a bug in which the trap function was not taken into consideration
when a program was compiled without optimization (at -O0).

<rdar://problem/16291933>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206323 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[X86] Improve the lowering of packed shifts by constant build_vector.
Andrea Di Biagio [Tue, 15 Apr 2014 19:30:48 +0000 (19:30 +0000)]
[X86] Improve the lowering of packed shifts by constant build_vector.

This patch teaches the backend how to efficiently lower logical and
arithmetic packed shifts on both SSE and AVX/AVX2 machines.

When possible, instead of scalarizing a vector shift, the backend should try
to expand the shift into a sequence of two packed shifts by immedate count
followed by a MOVSS/MOVSD.

Example
  (v4i32 (srl A, (build_vector < X, Y, Y, Y>)))

Can be rewritten as:
  (v4i32 (MOVSS (srl A, <Y,Y,Y,Y>), (srl A, <X,X,X,X>)))

[with X and Y ConstantInt]

The advantage is that the two new shifts from the example would be lowered into
X86ISD::VSRLI nodes. This is always cheaper than scalarizing the vector into
four scalar shifts plus four pairs of vector insert/extract.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206316 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64] Set default CPU to generic instead of cyclone.
Quentin Colombet [Tue, 15 Apr 2014 19:08:46 +0000 (19:08 +0000)]
[ARM64] Set default CPU to generic instead of cyclone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206313 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert r191049/r191059 as it can produce wrong code (see PR17975).
Robert Lougher [Tue, 15 Apr 2014 18:34:24 +0000 (18:34 +0000)]
Revert r191049/r191059 as it can produce wrong code (see PR17975).
It has already been reverted on the 3.4 branch in r196521.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206311 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to manage ownership of child Regions within llvm::Region
David Blaikie [Tue, 15 Apr 2014 18:32:43 +0000 (18:32 +0000)]
Use unique_ptr to manage ownership of child Regions within llvm::Region

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206310 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd lifetime markers for allocas created to hold byval arguments, make them
Julien Lerouge [Tue, 15 Apr 2014 18:06:46 +0000 (18:06 +0000)]
Add lifetime markers for allocas created to hold byval arguments, make them
appear in the InlineFunctionInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206308 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSplit byval argument initialization so the memcpy(s) are injected at the
Julien Lerouge [Tue, 15 Apr 2014 18:01:54 +0000 (18:01 +0000)]
Split byval argument initialization so the memcpy(s) are injected at the
beginning of the first new block after inlining.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206307 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLTO: Add more loop simplification passes to LTO
Duncan P. N. Exon Smith [Tue, 15 Apr 2014 17:48:15 +0000 (17:48 +0000)]
LTO: Add more loop simplification passes to LTO

Similar to r202051, add missing loop simplification passes to the LTO
optimization pipeline.

Patch by Rafael Espindola.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206306 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoverify-di: Add back braces for MSVC compatability
Duncan P. N. Exon Smith [Tue, 15 Apr 2014 17:28:26 +0000 (17:28 +0000)]
verify-di: Add back braces for MSVC compatability

Fixup after r206300.

<rdar://problem/15500563>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206305 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoverify-di: Implement DebugInfoVerifier
Duncan P. N. Exon Smith [Tue, 15 Apr 2014 16:27:38 +0000 (16:27 +0000)]
verify-di: Implement DebugInfoVerifier

Implement DebugInfoVerifier, which steals verification relying on
DebugInfoFinder from Verifier.

  - Adds LegacyDebugInfoVerifierPassPass, a ModulePass which wraps
    DebugInfoVerifier.  Uses -verify-di command-line flag.

  - Change verifyModule() to invoke DebugInfoVerifier as well as
    Verifier.

  - Add a call to createDebugInfoVerifierPass() wherever there was a
    call to createVerifierPass().

This implementation as a module pass should sidestep efficiency issues,
allowing us to turn debug info verification back on.

<rdar://problem/15500563>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206300 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoverify-di: split out VerifierSupport
Duncan P. N. Exon Smith [Tue, 15 Apr 2014 16:27:32 +0000 (16:27 +0000)]
verify-di: split out VerifierSupport

Split out assertion and output helpers from Verifier in preparation for
writing the DebugInfoVerifier.

<rdar://problem/15500563>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206299 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to manage PassInfo instances in the PassRegistry
David Blaikie [Tue, 15 Apr 2014 15:17:14 +0000 (15:17 +0000)]
Use unique_ptr to manage PassInfo instances in the PassRegistry

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206297 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMipsAsmParser.cpp: Fix vg_leak in MipsOperand::CreateMem(). Mem.Base is managed by...
NAKAMURA Takumi [Tue, 15 Apr 2014 14:13:21 +0000 (14:13 +0000)]
MipsAsmParser.cpp: Fix vg_leak in MipsOperand::CreateMem(). Mem.Base is managed by k_Memory itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206293 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMipsAsmParser::ParseRegister(): Be responsible to delete an Operand on a temporary...
NAKAMURA Takumi [Tue, 15 Apr 2014 14:06:27 +0000 (14:06 +0000)]
MipsAsmParser::ParseRegister(): Be responsible to delete an Operand on a temporary Operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206292 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: enable more AArch64 tests on ARM64.
Tim Northover [Tue, 15 Apr 2014 14:00:29 +0000 (14:00 +0000)]
AArch64/ARM64: enable more AArch64 tests on ARM64.

No code changes for this bunch, just some test rejigs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206291 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: add missing pattern for extending load.
Tim Northover [Tue, 15 Apr 2014 14:00:19 +0000 (14:00 +0000)]
AArch64/ARM64: add missing pattern for extending load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206290 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: only mangle MOVZ/MOVN during encoding when needed
Tim Northover [Tue, 15 Apr 2014 14:00:15 +0000 (14:00 +0000)]
AArch64/ARM64: only mangle MOVZ/MOVN during encoding when needed

Sometimes we need emit the bits that would actually be a MOVN when producing a
relocated MOVZ instruction (don't ask). But not always, a check which ARM64 got
wrong until now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206289 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: add support for large code-model jump tables.
Tim Northover [Tue, 15 Apr 2014 14:00:11 +0000 (14:00 +0000)]
AArch64/ARM64: add support for large code-model jump tables.

I've left the MachO CodeGen as it is, there's a reasonable chance it should use
the GOT like ConstPools, but I'm not certain.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206288 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: add patterns for various commutations of FNMADD.
Tim Northover [Tue, 15 Apr 2014 14:00:06 +0000 (14:00 +0000)]
AArch64/ARM64: add patterns for various commutations of FNMADD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206287 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: add half as a storage type on ARM64.
Tim Northover [Tue, 15 Apr 2014 14:00:03 +0000 (14:00 +0000)]
AArch64/ARM64: add half as a storage type on ARM64.

This brings it into line with the AArch64 behaviour and should open the way for
certain OpenCL features.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206286 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: copy patterns for fixed-point conversions
Tim Northover [Tue, 15 Apr 2014 13:59:57 +0000 (13:59 +0000)]
AArch64/ARM64: copy patterns for fixed-point conversions

Code is mostly copied directly across, with a slight extension of the
ISelDAGToDAG function so that it can cope with the floating-point constants
being behind a litpool.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206285 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: add constraints to various FastISel operations
Tim Northover [Tue, 15 Apr 2014 13:59:53 +0000 (13:59 +0000)]
ARM64: add constraints to various FastISel operations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206284 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFastISel: constrain the RegClass of operands when emitting instructions.
Tim Northover [Tue, 15 Apr 2014 13:59:49 +0000 (13:59 +0000)]
FastISel: constrain the RegClass of operands when emitting instructions.

ARM64 suffered multiple -verify-machineinstr failures (principally over the
xsp/xzr issue) because FastISel was completely ignoring which subset of the
general-purpose registers each instruction required.

More fixes are coming in ARM64 specific FastISel, but this should cover the
generic problems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206283 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: add more arm64 lines to AArch64 regression tests
Tim Northover [Tue, 15 Apr 2014 13:59:44 +0000 (13:59 +0000)]
AArch64/ARM64: add more arm64 lines to AArch64 regression tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206282 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAArch64/ARM64: add dp tests from AArch64
Tim Northover [Tue, 15 Apr 2014 13:59:40 +0000 (13:59 +0000)]
AArch64/ARM64: add dp tests from AArch64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206281 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64AsmParser.cpp: Fix vg_leak in MC/ARM64/fp-encoding.s.
NAKAMURA Takumi [Tue, 15 Apr 2014 13:22:11 +0000 (13:22 +0000)]
ARM64AsmParser.cpp: Fix vg_leak in MC/ARM64/fp-encoding.s.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206279 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOptional hash symbol feature support for ARM64
Stepan Dyatkovskiy [Tue, 15 Apr 2014 11:43:09 +0000 (11:43 +0000)]
Optional hash symbol feature support for ARM64
http://reviews.llvm.org/D3328

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206276 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCurrent definition of subtract with immediate instruction aliases uses CodeGenOnly...
Vladimir Medic [Tue, 15 Apr 2014 10:14:49 +0000 (10:14 +0000)]
Current definition of subtract with immediate instruction aliases uses CodeGenOnly defined instructions and post matcher expansion methods to emit real instructions add with immediate. However, they can directly alias add with immediate instruction and remove unnecessary definitions and code in MipsAsmParser.cpp. This patch makes no change in functionality, just removes unnecessary definitions and code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206272 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agovect.omp.persistence.ll REQUIRES asserts due to -debug-only.
NAKAMURA Takumi [Tue, 15 Apr 2014 10:12:47 +0000 (10:12 +0000)]
vect.omp.persistence.ll REQUIRES asserts due to -debug-only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206271 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Finally, finish nuking the redundant code that led me here
Chandler Carruth [Tue, 15 Apr 2014 09:44:09 +0000 (09:44 +0000)]
[Allocator] Finally, finish nuking the redundant code that led me here
by removing the MallocSlabAllocator entirely and just using
MallocAllocator directly. This makes all off these allocators expose and
utilize the same core interface.

The only ugly part of this is that it exposes the fact that the JIT
allocator has no real handling of alignment, any more than the malloc
allocator does. =/ It would be nice to fix both of these to support
alignments, and then to leverage that in the BumpPtrAllocator to do less
over allocation in order to manually align pointers. But, that's another
patch for another day. This patch has no functional impact, it just
removes the somewhat meaningless wrapper around MallocAllocator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206267 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoD3348 - [BUG] "Rotate Loop" pass kills "llvm.vectorizer.enable" metadata
Alexey Bataev [Tue, 15 Apr 2014 09:37:30 +0000 (09:37 +0000)]
D3348 - [BUG] "Rotate Loop" pass kills "llvm.vectorizer.enable" metadata

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206266 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Pass the size to the deallocation function. This, on some
Chandler Carruth [Tue, 15 Apr 2014 08:59:52 +0000 (08:59 +0000)]
[Allocator] Pass the size to the deallocation function. This, on some
allocation libraries, may allow more efficient allocation and
deallocation. It at least makes the interface implementable by the JIT
memory manager.

However, this highlights problematic overloading between the void* and
the T* deallocation functions. I'm looking into a better way to do this,
but as it happens, it comes up rarely in the codebase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206265 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoX86JITInfo: [x86] Rework r206240, X86CompilationCallback_SSE() should be called for...
NAKAMURA Takumi [Tue, 15 Apr 2014 08:28:23 +0000 (08:28 +0000)]
X86JITInfo: [x86] Rework r206240, X86CompilationCallback_SSE() should be called for SSE-enabled code generator, even if LLVM is not built with -msse.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206261 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Fix r206256 which got the enabling case backwards on these
Chandler Carruth [Tue, 15 Apr 2014 08:14:48 +0000 (08:14 +0000)]
[Allocator] Fix r206256 which got the enabling case backwards on these
overloads. This doesn't matter *that* much yet, but it will in
a subsequent patch. I had tested the original pattern, but not my
attempt to pacify MSVC. This at least appears to work. Still fixing the
rest of the fallout in the final patch that uses these overloads, but it
will follow shortly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206259 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix broken build of llvm using clang.
Nick Lewycky [Tue, 15 Apr 2014 08:10:46 +0000 (08:10 +0000)]
Fix broken build of llvm using clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206257 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] MSVC apparantly has broken SFINAE context handling of
Chandler Carruth [Tue, 15 Apr 2014 08:02:29 +0000 (08:02 +0000)]
[Allocator] MSVC apparantly has broken SFINAE context handling of
'sizeof(T)' for T == void and produces a hard error. I cannot fathom why
this is OK. Oh well. switch to an explicit test for being the
(potentially qualified) void type, which is the only specific case I was
worried about. Hopefully this survives the libstdc++ build bots which
have limited type traits implementations...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206256 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoBreak PseudoSourceValue out of the Value hierarchy. It is now the root of its own...
Nick Lewycky [Tue, 15 Apr 2014 07:22:52 +0000 (07:22 +0000)]
Break PseudoSourceValue out of the Value hierarchy. It is now the root of its own tree containing FixedStackPseudoSourceValue (which you can use isa/dyn_cast on) and MipsCallEntry (which you can't). Anything that needs to use either a PseudoSourceValue* and Value* is strongly encouraged to use a MachinePointerInfo instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206255 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead...
Craig Topper [Tue, 15 Apr 2014 07:20:03 +0000 (07:20 +0000)]
[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead of comparing to nullptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206254 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a DenseMapInfo specialization for PointerUnion. In tree user to land shortly.
Nick Lewycky [Tue, 15 Apr 2014 07:08:40 +0000 (07:08 +0000)]
Add a DenseMapInfo specialization for PointerUnion. In tree user to land shortly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206253 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead...
Craig Topper [Tue, 15 Apr 2014 06:32:26 +0000 (06:32 +0000)]
[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead of comparing to nullptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206252 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Constrain the Deallocate templated overloads to only apply
Chandler Carruth [Tue, 15 Apr 2014 06:29:04 +0000 (06:29 +0000)]
[Allocator] Constrain the Deallocate templated overloads to only apply
to types which we can compute the size of. The comparison with zero
isn't actually interesting here, it's mostly about putting sizeof into
a sfinae context.

This is particular important for Deallocate as otherwise the void*
overload can quickly become ambiguous.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206251 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to manage TypePromotionActions owned by TypePromotionTransaction.
David Blaikie [Tue, 15 Apr 2014 06:17:44 +0000 (06:17 +0000)]
Use unique_ptr to manage TypePromotionActions owned by TypePromotionTransaction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206250 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to manage ownership of GCFunctionInfos in GCStrategy
David Blaikie [Tue, 15 Apr 2014 06:07:26 +0000 (06:07 +0000)]
Use unique_ptr to manage ownership of GCFunctionInfos in GCStrategy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206249 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr for the result of Registry entries.
David Blaikie [Tue, 15 Apr 2014 05:53:26 +0000 (05:53 +0000)]
Use unique_ptr for the result of Registry entries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206248 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to manage ownership of GCStrategy objects in GCMetadata
David Blaikie [Tue, 15 Apr 2014 05:34:49 +0000 (05:34 +0000)]
Use unique_ptr to manage ownership of GCStrategy objects in GCMetadata

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206246 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr for section/segment ownership in WinCOFFObjectWriter
David Blaikie [Tue, 15 Apr 2014 05:25:03 +0000 (05:25 +0000)]
Use unique_ptr for section/segment ownership in WinCOFFObjectWriter

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206245 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to own MCFunctions within MCModule.
David Blaikie [Tue, 15 Apr 2014 05:15:19 +0000 (05:15 +0000)]
Use unique_ptr to own MCFunctions within MCModule.

MCModule's ctor had to be moved out of line so the definition of
MCFunction was available. (ctor requires the dtor of members (in case
the ctor throws) which required access to the dtor of MCFunction)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206244 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead...
Craig Topper [Tue, 15 Apr 2014 04:59:12 +0000 (04:59 +0000)]
[C++11] More 'nullptr' conversion. In some cases just using a boolean check instead of comparing to nullptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206243 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse std::unique_ptr to manage MCBasicBlocks in MCFunction.
David Blaikie [Tue, 15 Apr 2014 04:56:29 +0000 (04:56 +0000)]
Use std::unique_ptr to manage MCBasicBlocks in MCFunction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206242 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[MC] Require an MCContext when constructing an MCDisassembler.
Lang Hames [Tue, 15 Apr 2014 04:40:56 +0000 (04:40 +0000)]
[MC] Require an MCContext when constructing an MCDisassembler.

This patch re-introduces the MCContext member that was removed from
MCDisassembler in r206063, and requires that an MCContext be passed in at
MCDisassembler construction time. (Previously the MCContext member had been
initialized in an ad-hoc fashion after construction). The MCCContext member
can be used by MCDisassembler sub-classes to construct constant or
target-specific MCExprs.

This patch updates disassemblers for in-tree targets, and provides the
MCRegisterInfo instance that some disassemblers were using through the
MCContext (previously those backends were constructing their own
MCRegisterInfo instances).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206241 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoX86JITInfo: [x86] Use X86CompilationCallback_SSE() along;
NAKAMURA Takumi [Tue, 15 Apr 2014 04:12:21 +0000 (04:12 +0000)]
X86JITInfo: [x86] Use X86CompilationCallback_SSE() along;

  *not* Subtarget->hasSSE1()
  *but* __SSE__, the flag that LLVM libraries are compiled

The callback calls internal LLVM JIT libraries. It may be built with -msse (or above).

FIXME: JIT may use "host" instead of "generic" by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206240 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64,C++11]: Range'ify the dead-register-definition pass.
Jim Grosbach [Tue, 15 Apr 2014 02:14:09 +0000 (02:14 +0000)]
[ARM64,C++11]: Range'ify the dead-register-definition pass.

Range-based for loops. No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206239 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd iterator_range for MachineInstr defs.
Jim Grosbach [Tue, 15 Apr 2014 02:14:06 +0000 (02:14 +0000)]
Add iterator_range for MachineInstr defs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206238 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[MC] Emit an error if cfi_startproc is used before a symbol is defined.
Quentin Colombet [Tue, 15 Apr 2014 01:17:45 +0000 (01:17 +0000)]
[MC] Emit an error if cfi_startproc is used before a symbol is defined.
Currently, we bind those directives with the last symbol, so if none
has been defined, this would lead to a crash of the compiler.

<rdar://problem/15939159>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206236 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Register Coalescer] Add a test case for 206060.
Quentin Colombet [Tue, 15 Apr 2014 01:15:32 +0000 (01:15 +0000)]
[Register Coalescer] Add a test case for 206060.

<rdar://problem/16582185>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206235 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Add Deallocate support to the AllocatorBase CRTP class,
Chandler Carruth [Tue, 15 Apr 2014 00:47:47 +0000 (00:47 +0000)]
[Allocator] Add Deallocate support to the AllocatorBase CRTP class,
along with templated overloads much like we have for Allocate. These
will facilitate switching the Deallocate interface of all the Allocator
classes to accept the size by pre-filling it from the type size where we
can do so. I plan to convert several uses to the template variants in
subsequent patches prior to adding the Size parameter.

No functionality changed, WIP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206230 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64][MC] Set the default CPU string to generic.
Quentin Colombet [Tue, 15 Apr 2014 00:28:39 +0000 (00:28 +0000)]
[ARM64][MC] Set the default CPU string to generic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206228 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Hack around the fact that GCC can't compile the
Chandler Carruth [Tue, 15 Apr 2014 00:22:53 +0000 (00:22 +0000)]
[Allocator] Hack around the fact that GCC can't compile the
static_assert added in r206225. I'm looking into a proper fix, but
wanted the bots back.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206226 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Allocator] Factor the Allocate template overloads into a base class
Chandler Carruth [Tue, 15 Apr 2014 00:19:41 +0000 (00:19 +0000)]
[Allocator] Factor the Allocate template overloads into a base class
rather than defining them (differently!) in both allocators. This also
serves as a basis for documenting and even enforcing some of the
LLVM-style "allocator" concept methods which must exist with various
signatures.

I plan on extending and changing the signatures of these to further
simplify our allocator model in subsequent commits, so I wanted to
factor things as best as I could first. Notably, I'm working to add the
'Size' to the deallocation method of all allocators. This has several
implications not the least of which are faster deallocation times on
certain allocation libraries (tcmalloc). It also will allow the JIT
allocator to fully model the existing allocation interfaces and allow
sanitizer poisoning of deallocated regions. The list of advantages goes
on. =] But by factoring things first I'll be able to make this easier by
first introducing template helpers for the deallocation path.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206225 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse std::unique_ptr for DIE children
David Blaikie [Mon, 14 Apr 2014 22:45:02 +0000 (22:45 +0000)]
Use std::unique_ptr for DIE children

Got bored, removed some manual memory management.

Pushed references (rather than pointers) through a few APIs rather than
replacing *x with x.get().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206222 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoX86: Nuke one more CPU autodetect blurb.
Jim Grosbach [Mon, 14 Apr 2014 22:23:30 +0000 (22:23 +0000)]
X86: Nuke one more CPU autodetect blurb.

Missed one in r206094. This brings MC and TargetMachine back into sync.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206220 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoChange argument order and add explanatory comment to r206130
David Blaikie [Mon, 14 Apr 2014 22:23:06 +0000 (22:23 +0000)]
Change argument order and add explanatory comment to r206130

Changes requested in code review by Eric Christopher of r206130.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206219 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse FrameSetup on frame instructions for the Mips port.
Eric Christopher [Mon, 14 Apr 2014 22:21:22 +0000 (22:21 +0000)]
Use FrameSetup on frame instructions for the Mips port.

I can't seem to get a testcase to show a difference here, but it's
part of the unconditional-br.ll line table weirdness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206218 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CMake] Reformat, if(MSVC)...else()...endif()
NAKAMURA Takumi [Mon, 14 Apr 2014 21:58:19 +0000 (21:58 +0000)]
[CMake] Reformat, if(MSVC)...else()...endif()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206215 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "Revert r206045, "Fix shift by constants for vector.""
Matt Arsenault [Mon, 14 Apr 2014 21:50:37 +0000 (21:50 +0000)]
Revert "Revert r206045, "Fix shift by constants for vector.""

Fix cases where the Value itself is used, and not the constant value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206214 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64][MC] Set the default CPU to cyclone when initilizating the MC layer.
Quentin Colombet [Mon, 14 Apr 2014 21:25:53 +0000 (21:25 +0000)]
[ARM64][MC] Set the default CPU to cyclone when initilizating the MC layer.
This matches that ARM64Subtarget does for now.

This is related to <rdar://problem/16573920>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206211 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRe-apply r206096 after investigating the gdb buildbot failure.
Adrian Prantl [Mon, 14 Apr 2014 21:16:04 +0000 (21:16 +0000)]
Re-apply r206096 after investigating the gdb buildbot failure.
Thanks to dblaikie for updating the testcase!

Debug info: (bugfix) C++ C/Dtors can be compiled to multiple functions,
therefore, their declaration cannot have one DW_AT_linkage_name.
The specific instances however can and should have that attribute.

This patch reorders the code in DwarfUnit::getOrCreateSubprogramDIE()
to emit linkage names for C/Dtors.

rdar://problem/16362674.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206210 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[cleanup] Run clang-format over most of YAMLParser.h to fix a bunch of
Chandler Carruth [Mon, 14 Apr 2014 21:12:15 +0000 (21:12 +0000)]
[cleanup] Run clang-format over most of YAMLParser.h to fix a bunch of
small formatting inconsistencies with the rest of LLVM and even this
file. I looked at all the changes and they seemed like just better
formatting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206209 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix for codegen bug that could cause illegal cmn instruction generation
Louis Gerbarg [Mon, 14 Apr 2014 21:05:05 +0000 (21:05 +0000)]
Fix for codegen bug that could cause illegal cmn instruction generation

In rare cases the dead definition elimination pass code can cause illegal cmn
instructions when it replaces dead registers on instructions that use
unmaterialized frame indexes. This patch disables the dead definition
optimization for instructions which include frame index operands.

rdar://16438284

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206208 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a flag to disable the ARM64DeadRegisterDefinitionsPass
Louis Gerbarg [Mon, 14 Apr 2014 21:05:02 +0000 (21:05 +0000)]
Add a flag to disable the ARM64DeadRegisterDefinitionsPass

This patch adds a -arm64-dead-def-elimination flag so that it is possible to
disable dead definition elimination. Includes test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206207 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64] Port over missing subtarget features, and CPU definitions from AArch64.
James Molloy [Mon, 14 Apr 2014 17:38:00 +0000 (17:38 +0000)]
[ARM64] Port over missing subtarget features, and CPU definitions from AArch64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206198 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM64] Add big endian target arm64_be.
James Molloy [Mon, 14 Apr 2014 17:37:53 +0000 (17:37 +0000)]
[ARM64] Add big endian target arm64_be.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206197 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReplace two calls to object::symbol_iterator::increment(), which had
Kaelyn Takata [Mon, 14 Apr 2014 17:26:50 +0000 (17:26 +0000)]
Replace two calls to object::symbol_iterator::increment(), which had
been removed in r200442.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206196 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove a variable from r206192 that is only used in an assert.
Kaelyn Takata [Mon, 14 Apr 2014 17:21:50 +0000 (17:21 +0000)]
Remove a variable from r206192 that is only used in an assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206195 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic block...
Akira Hatanaka [Mon, 14 Apr 2014 16:56:19 +0000 (16:56 +0000)]
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly.

Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights.

This fixes PR18705 and <rdar://problem/15991090>.

Differential Revision: http://reviews.llvm.org/D3363

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8