platform/upstream/llvm.git
10 years agoDocs: remove extra {} around result types.
Tim Northover [Fri, 13 Jun 2014 14:24:23 +0000 (14:24 +0000)]
Docs: remove extra {} around result types.

It makes the types look like they're single-element structures. And
when we have instructions that *do* result in a struct, that can get
confusing rather quickly.

llvm-svn: 210905

10 years agoDocs: fix grammar error in description
Tim Northover [Fri, 13 Jun 2014 14:24:16 +0000 (14:24 +0000)]
Docs: fix grammar error in description

llvm-svn: 210904

10 years agoIR: add "cmpxchg weak" variant to support permitted failure.
Tim Northover [Fri, 13 Jun 2014 14:24:07 +0000 (14:24 +0000)]
IR: add "cmpxchg weak" variant to support permitted failure.

This commit adds a weak variant of the cmpxchg operation, as described
in C++11. A cmpxchg instruction with this modifier is permitted to
fail to store, even if the comparison indicated it should.

As a result, cmpxchg instructions must return a flag indicating
success in addition to their original iN value loaded. Thus, for
uniformity *all* cmpxchg instructions now return "{ iN, i1 }". The
second flag is 1 when the store succeeded.

At the DAG level, a new ATOMIC_CMP_SWAP_WITH_SUCCESS node has been
added as the natural representation for the new cmpxchg instructions.
It is a strong cmpxchg.

By default this gets Expanded to the existing ATOMIC_CMP_SWAP during
Legalization, so existing backends should see no change in behaviour.
If they wish to deal with the enhanced node instead, they can call
setOperationAction on it. Beware: as a node with 2 results, it cannot
be selected from TableGen.

Currently, no use is made of the extra information provided in this
patch. Test updates are almost entirely adapting the input IR to the
new scheme.

Summary for out of tree users:
------------------------------

+ Legacy Bitcode files are upgraded during read.
+ Legacy assembly IR files will be invalid.
+ Front-ends must adapt to different type for "cmpxchg".
+ Backends should be unaffected by default.

llvm-svn: 210903

10 years agoDo not store duplicate parents when memoization data is available.
Samuel Benzaquen [Fri, 13 Jun 2014 13:31:40 +0000 (13:31 +0000)]
Do not store duplicate parents when memoization data is available.

Summary:
Do not store duplicate parents when memoization data is available.
This does not solve the duplication problem, but ameliorates it.

Reviewers: klimek

Subscribers: klimek, cfe-commits

Differential Revision: http://reviews.llvm.org/D4124

llvm-svn: 210902

10 years agoFix bad copy-and-paste from r210652. AVX512 masked leading zero intrinsics.
Cameron McInally [Fri, 13 Jun 2014 13:20:01 +0000 (13:20 +0000)]
Fix bad copy-and-paste from r210652. AVX512 masked leading zero intrinsics.

llvm-svn: 210901

10 years ago[mips] Add cache and pref instructions
Daniel Sanders [Fri, 13 Jun 2014 13:15:59 +0000 (13:15 +0000)]
[mips] Add cache and pref instructions

Summary:
cache and pref were added in MIPS-III, and MIPS32 but were re-encoded in
MIPS32r6/MIPS64r6 to use a 9-bit offset rather than the 16-bit offset
available to earlier cores.

Resolved the decoding conflict between pref and lwc3.

Depends on D4115

Reviewers: zoran.jovanovic, jkolek, vmedic

Reviewed By: vmedic

Differential Revision: http://reviews.llvm.org/D4116

llvm-svn: 210900

10 years ago[mips][mips64r6] bc1any[24] are not available on MIPS32r6/MIPS64r6
Daniel Sanders [Fri, 13 Jun 2014 13:08:38 +0000 (13:08 +0000)]
[mips][mips64r6] bc1any[24] are not available on MIPS32r6/MIPS64r6

Summary:
These MIPS-3D instructions have never been implemented in LLVM so we only
add testcases.

Reviewers: zoran.jovanovic, jkolek, vmedic

Reviewed By: vmedic

Differential Revision: http://reviews.llvm.org/D4115

llvm-svn: 210899

10 years ago[mips][mips64r6] b(ge|lt)zal are not available on MIPS32r6/MIPS64r6 and bal is a...
Daniel Sanders [Fri, 13 Jun 2014 13:02:52 +0000 (13:02 +0000)]
[mips][mips64r6] b(ge|lt)zal are not available on MIPS32r6/MIPS64r6 and bal is a normal instruction

Summary:
b(ge|lt)zal have been removed in MIPS32r6/MIPS64r6. However, bal (an alias
for 'bgezal $zero, $offset') still remains with the same encoding it had
prior to MIPS32r6/MIPS64r6.

Updated the MipsNaCLELFStreamer, and MipsLongBranch to correctly handle the
MIPS32r6/MIPS64r6 BAL instruction in addition to the existing BAL_BR pseudo.

No changes were required to the CodeGen test that looks for BAL
(test/CodeGen/Mips/longbranch.ll) since the new instruction has the same
syntax.

Depends on D4113

Reviewers: jkolek, zoran.jovanovic, vmedic

Reviewed By: vmedic

Differential Revision: http://reviews.llvm.org/D4114

llvm-svn: 210898

10 years ago[mips][mips64r6] daddi is not available on MIPS64r6
Daniel Sanders [Fri, 13 Jun 2014 12:49:06 +0000 (12:49 +0000)]
[mips][mips64r6] daddi is not available on MIPS64r6

Summary:
It's not emitted by the code generator so we only need assembler tests.

Also added missing daddi aliases from dsub mnemonics, and removed a couple
duplicate dsub tests.

Depends on D4112

Reviewers: zoran.jovanovic, jkolek, vmedic

Reviewed By: vmedic

Differential Revision: http://reviews.llvm.org/D4113

llvm-svn: 210897

10 years agoAdd files forgotten in the previous commit
Jeroen Ketema [Fri, 13 Jun 2014 12:33:40 +0000 (12:33 +0000)]
Add files forgotten in the previous commit

llvm-svn: 210896

10 years ago[Win32] Let utils/not aware of abort(), aka llvm_unreachable(), in msvcrt.
NAKAMURA Takumi [Fri, 13 Jun 2014 12:23:56 +0000 (12:23 +0000)]
[Win32] Let utils/not aware of abort(), aka llvm_unreachable(), in msvcrt.

It has exit code as 3. abort(), aka unreachable, may be handled as crash.

FIXME: Could we move this into Win32/Program.inc?
llvm-svn: 210895

10 years agollvm/test/CodeGen/X86/fast-isel-args-fail2.ll: Don't expect to fail with -Asserts...
NAKAMURA Takumi [Fri, 13 Jun 2014 12:05:06 +0000 (12:05 +0000)]
llvm/test/CodeGen/X86/fast-isel-args-fail2.ll: Don't expect to fail with -Asserts. It might or might not crash.

llvm-svn: 210894

10 years ago[Mips] Make exe-fileheader.test test case independent from external input files.
Simon Atanasyan [Fri, 13 Jun 2014 11:44:23 +0000 (11:44 +0000)]
[Mips] Make exe-fileheader.test test case independent from external input files.

llvm-svn: 210893

10 years agoAdd HasCDI predicate to AVX512 VPBROADCASTM*.
Cameron McInally [Fri, 13 Jun 2014 11:40:31 +0000 (11:40 +0000)]
Add HasCDI predicate to AVX512 VPBROADCASTM*.

llvm-svn: 210892

10 years agoImplementations for exp(float) and exp(double) v2
Jeroen Ketema [Fri, 13 Jun 2014 09:40:09 +0000 (09:40 +0000)]
Implementations for exp(float) and exp(double) v2

Use separate implementations instead of a macro
to ensure the constant multiplied with is of
higher precision.

v2: Use the correct formula, spotted by Dan Liew <daniel.liew@imperial.ac.uk>

Reviewed-by: Aaron Warty <awatry@gmail.com>
Reviewed-by: Tom Stellard <tom@stellard.net>
llvm-svn: 210891

10 years agoCPP backend: set volatile property on atomic instructions.
Tim Northover [Fri, 13 Jun 2014 09:14:50 +0000 (09:14 +0000)]
CPP backend: set volatile property on atomic instructions.

llvm-svn: 210890

10 years agoARM: Fix fastcc calling convention for Thumb1
Oliver Stannard [Fri, 13 Jun 2014 08:33:03 +0000 (08:33 +0000)]
ARM: Fix fastcc calling convention for Thumb1

When targetting Thumb1 on a processor which has a VFP unit (which
is not accessible from Thumb1), we were converting the fastcc calling
convention to AAPCS-VFP, which is not possible.

llvm-svn: 210889

10 years agoR600: Don't call setOperationAction with things that aren't opcodes.
Matt Arsenault [Fri, 13 Jun 2014 07:44:38 +0000 (07:44 +0000)]
R600: Don't call setOperationAction with things that aren't opcodes.

CondCode actions are set with setCondCodeAction.
This should have been a harmless bug since the values seem to only
collide only with nodes that don't need to be handled, and these are
already correctly setup elsewhere.

llvm-svn: 210888

10 years agoclang-format: [JS] Understand named function literals.
Daniel Jasper [Fri, 13 Jun 2014 07:02:04 +0000 (07:02 +0000)]
clang-format: [JS] Understand named function literals.

Before:
  return {a: function SomeFunction(){// ...
                                     return 1;
  }
  }
  ;

After:
  return {
    a: function SomeFunction() {
      // ...
      return 1;
    }
  };

llvm-svn: 210887

10 years agoMS ABI: Fix inheritance model calculation in CRTP
David Majnemer [Fri, 13 Jun 2014 06:43:46 +0000 (06:43 +0000)]
MS ABI: Fix inheritance model calculation in CRTP

CRTP-like patterns involve a class which inherits from another class
using itself as a template parameter.

However, the base class itself may try to create a pointer-to-member
which involves the derived class.  This is problematic because we
may not have finished parsing the most derived classes' base specifiers
yet.

It turns out that MSVC simply uses the unspecified inheritance model
instead of doing anything fancy.

This fixes PR19987.

llvm-svn: 210886

10 years agoR600/SI: Fix selection error on i64 rotl / rotr.
Matt Arsenault [Fri, 13 Jun 2014 04:00:30 +0000 (04:00 +0000)]
R600/SI: Fix selection error on i64 rotl / rotr.

Evergreen is still broken due to missing shl_parts.

llvm-svn: 210885

10 years agoMissed a space at the end of the line.
Brad Smith [Fri, 13 Jun 2014 03:53:07 +0000 (03:53 +0000)]
Missed a space at the end of the line.

llvm-svn: 210884

10 years agoUse dwarf-2 by default on OpenBSD and FreeBSD.
Brad Smith [Fri, 13 Jun 2014 03:35:37 +0000 (03:35 +0000)]
Use dwarf-2 by default on OpenBSD and FreeBSD.

The Tools.cpp part of the patch partially based on a patch from
FreeBSD's LLVM tree.

llvm-svn: 210883

10 years agoInterpreter: kill some dead code
Saleem Abdulrasool [Fri, 13 Jun 2014 03:30:47 +0000 (03:30 +0000)]
Interpreter: kill some dead code

Remove commented out code and an unnecessary associated scope.  No functional
change.

llvm-svn: 210882

10 years agoInterpreter: explicitly initialize base class
Saleem Abdulrasool [Fri, 13 Jun 2014 03:30:45 +0000 (03:30 +0000)]
Interpreter: explicitly initialize base class

Initialise base class std::enable_shared_from_this explicitly.  Identified by
GCC.

llvm-svn: 210881

10 years agoCore: address comparison of signed and unsigned types
Saleem Abdulrasool [Fri, 13 Jun 2014 03:30:42 +0000 (03:30 +0000)]
Core: address comparison of signed and unsigned types

Add a cast to ensure that the comparison is done with the same sign type.
Identified by GCC.

llvm-svn: 210880

10 years agoRemove unused variables
Saleem Abdulrasool [Fri, 13 Jun 2014 03:30:39 +0000 (03:30 +0000)]
Remove unused variables

Address the 'variable set but not used' warning from GCC.  In some cases a few
additional calls were removed where there should be no visible side effects of
the calls (i.e. should not effect any cached state).

llvm-svn: 210879

10 years agoRemove a 'using std::error_code' I missed in the previous pass.
Rafael Espindola [Fri, 13 Jun 2014 03:24:40 +0000 (03:24 +0000)]
Remove a 'using std::error_code' I missed in the previous pass.

llvm-svn: 210878

10 years agoRemove the last uses of 'using std::error_code'
Rafael Espindola [Fri, 13 Jun 2014 03:20:08 +0000 (03:20 +0000)]
Remove the last uses of 'using std::error_code'

This finishes the transition to std::error_code.

llvm-svn: 210877

10 years agoRemove 'using std::error_code' from tools.
Rafael Espindola [Fri, 13 Jun 2014 03:07:50 +0000 (03:07 +0000)]
Remove 'using std::error_code' from tools.

llvm-svn: 210876

10 years agowhitespace cleanup
Jason Molenda [Fri, 13 Jun 2014 02:44:21 +0000 (02:44 +0000)]
whitespace cleanup

llvm-svn: 210875

10 years agoInitial merge of some of the iOS 8 / Mac OS X Yosemite specific
Jason Molenda [Fri, 13 Jun 2014 02:37:02 +0000 (02:37 +0000)]
Initial merge of some of the iOS 8 / Mac OS X Yosemite specific
lldb support.  I'll be doing more testing & cleanup but I wanted to
get the initial checkin done.

This adds a new SBExpressionOptions::SetLanguage API for selecting a
language of an expression.

I added adds a new SBThread::GetInfoItemByPathString for retriving
information about a thread from that thread's StructuredData.

I added a new StructuredData class for representing
key-value/array/dictionary information (e.g. JSON formatted data).
Helper functions to read JSON and create a StructuredData object,
and to print a StructuredData object in JSON format are included.

A few Cocoa / Cocoa Touch data formatters were updated by Enrico
to track changes in iOS 8 / Yosemite.

Before we query a thread's extended information, the system runtime may
provide hints to the remote debug stub that it will use to retrieve values
out of runtime structures.  I added a new SystemRuntime method
AddThreadExtendedInfoPacketHints which allows the SystemRuntime to add
key-value type data to the initial request that we send to the remote stub.

The thread-format formatter string can now retrieve values out of a thread's
extended info structured data.  The default thread-format string picks up
two of these - thread.info.activity.name and thread.info.trace_messages.

I added a new "jThreadExtendedInfo" packet in debugserver; I will
add documentation to the lldb-gdb-remote.txt doc soon.  It accepts
JSON formatted arguments (most importantly, "thread":threadnum) and
it returns a variety of information regarding the thread to lldb
in JSON format.  This JSON return is scanned into a StructuredData
object that is associated with the thread; UI layers can query the
thread's StructuredData to see if key-values are present, and if
so, show them to the user.  These key-values are likely to be
specific to different targets with some commonality among many
targets.  For instance, many targets will be able to advertise the
pthread_t value for a thread.

I added an initial rough cut of "thread info" command which will print
the information about a thread from the jThreadExtendedInfo result.
I need to do more work to make this format reasonably.

Han Ming added calls into the pmenergy and pmsample libraries if
debugserver is run on Mac OS X Yosemite to get information about the
inferior's power use.

I added support to debugserver for gathering the Genealogy information
about threads, if it exists, and returning it in the jThreadExtendedInfo
JSON result.

llvm-svn: 210874

10 years agoFix build on windows.
Rafael Espindola [Fri, 13 Jun 2014 02:36:09 +0000 (02:36 +0000)]
Fix build on windows.

llvm-svn: 210873

10 years agoExtend AST dump to include 'used' and 'referenced' flags, and put 'invalid' flag...
Richard Smith [Fri, 13 Jun 2014 02:24:47 +0000 (02:24 +0000)]
Extend AST dump to include 'used' and 'referenced' flags, and put 'invalid' flag in the right place.

llvm-svn: 210872

10 years agoRemove 'using std::errro_code' from lib.
Rafael Espindola [Fri, 13 Jun 2014 02:24:39 +0000 (02:24 +0000)]
Remove 'using std::errro_code' from lib.

llvm-svn: 210871

10 years ago[FastISel][X86] Add support for cvttss2si/cvttsd2si intrinsics.
Juergen Ributzka [Fri, 13 Jun 2014 02:21:58 +0000 (02:21 +0000)]
[FastISel][X86] Add support for cvttss2si/cvttsd2si intrinsics.

This adds support for the cvttss2si/cvttsd2si intrinsics. Preceding
insertelement instructions are folded into the conversion instruction (if
possible).

llvm-svn: 210870

10 years agoR600: Move AMDGPUInstrInfo from AMDGPUTargetMachine into AMDGPUSubtarget
Tom Stellard [Fri, 13 Jun 2014 01:32:00 +0000 (01:32 +0000)]
R600: Move AMDGPUInstrInfo from AMDGPUTargetMachine into AMDGPUSubtarget

llvm-svn: 210869

10 years agoR600: Drop use of cached TargetMachine in R600InstrInfo.cpp
Tom Stellard [Fri, 13 Jun 2014 01:31:56 +0000 (01:31 +0000)]
R600: Drop use of cached TargetMachine in R600InstrInfo.cpp

llvm-svn: 210868

10 years agoprepare-builtins: Use std:: prefix for error_code
Tom Stellard [Fri, 13 Jun 2014 01:30:14 +0000 (01:30 +0000)]
prepare-builtins: Use std:: prefix for error_code

This fixes the build with with newer LLVM.

llvm-svn: 210867

10 years agoRemove all uses of 'using std::error_code' from headers.
Rafael Espindola [Fri, 13 Jun 2014 01:25:41 +0000 (01:25 +0000)]
Remove all uses of 'using std::error_code' from headers.

llvm-svn: 210866

10 years agoR600: Drop use of cached TargetMachine in AMDGPUInstrInfo.cpp
Tom Stellard [Fri, 13 Jun 2014 01:02:57 +0000 (01:02 +0000)]
R600: Drop use of cached TargetMachine in AMDGPUInstrInfo.cpp

llvm-svn: 210865

10 years agoAdded the ability to save core files:
Greg Clayton [Fri, 13 Jun 2014 00:54:12 +0000 (00:54 +0000)]
Added the ability to save core files:

(lldb) file /bin/ls
(lldb) b malloc
(lldb) run
(lldb) process save-core /tmp/ls.core

Each ObjectFile plug-in now has the option to save core files by registering a new static callback.

llvm-svn: 210864

10 years ago[FastISel][X86] - Add branch weights
Juergen Ributzka [Fri, 13 Jun 2014 00:45:11 +0000 (00:45 +0000)]
[FastISel][X86] - Add branch weights

Add branch weights to branch instructions, so that the following passes can
optimize based on it (i.e. basic block ordering).

llvm-svn: 210863

10 years agoMove ARMSelectionDAGInfo from the TargetMachine to the subtarget.
Eric Christopher [Fri, 13 Jun 2014 00:20:39 +0000 (00:20 +0000)]
Move ARMSelectionDAGInfo from the TargetMachine to the subtarget.

llvm-svn: 210862

10 years agoMove to a private function to initialize subtarget dependencies
Eric Christopher [Fri, 13 Jun 2014 00:20:35 +0000 (00:20 +0000)]
Move to a private function to initialize subtarget dependencies
so we can use initializer lists for the ARMSubtarget and then
use this to initialize a moved DataLayout on the subtarget from
the TargetMachine.

llvm-svn: 210861

10 years ago[DWARF parser] Fix broken address ranges construction.
Alexey Samsonov [Thu, 12 Jun 2014 23:58:49 +0000 (23:58 +0000)]
[DWARF parser] Fix broken address ranges construction.

Previous algorithm for constructing [Address ranges]->[Compile Units]
mapping was wrong. It somewhat relied on the assumption that address ranges
for different compile units may not overlap. It is not so.
For example, two compile units may contain the definition of the same
linkonce_odr function. These definitions will be merged at link-time,
resulting in equivalent .debug_ranges entries for both these units

Instead of sorting and merging original address ranges (from .debug_ranges
and .debug_aranges), implement a different approach: save endpoints
of all ranges, and then use a sweep-line approach to construct
the desired mapping. If we find that certain address maps to
several compilation units, we just pick any of them.

llvm-svn: 210860

10 years agoHave ARMSelectionDAGInfo take a DataLayout as it's argument as the
Eric Christopher [Thu, 12 Jun 2014 23:39:49 +0000 (23:39 +0000)]
Have ARMSelectionDAGInfo take a DataLayout as it's argument as the
DAG has access to the subtarget and TargetSelectionDAGInfo only
needs a DataLayout.

llvm-svn: 210859

10 years ago[FastISel][X86] Add MachineMemOperand to load/store instructions.
Juergen Ributzka [Thu, 12 Jun 2014 23:27:57 +0000 (23:27 +0000)]
[FastISel][X86] Add MachineMemOperand to load/store instructions.

This commit adds MachineMemOperands to load and store instructions. This allows
the peephole optimizer to fold load instructions. Unfortunatelly the peephole
optimizer currently doesn't run at -O0.

llvm-svn: 210858

10 years agoRevert "fixes duplicate header installation"
David Fang [Thu, 12 Jun 2014 23:19:55 +0000 (23:19 +0000)]
Revert "fixes duplicate header installation"

This reverts commit 0bd40d6c3da6719fecf77038673d453ff1eab25b.

llvm-svn: 210857

10 years agoRecover from missing 'typename' in sizeof(T::InnerType)
Reid Kleckner [Thu, 12 Jun 2014 23:03:48 +0000 (23:03 +0000)]
Recover from missing 'typename' in sizeof(T::InnerType)

Summary:
'sizeof' is a UnaryExprOrTypeTrait, and it can contain either a type or
an expression.  This change threads a RecoveryTSI parameter through the
layers between TransformUnaryExprOrTypeTrait the point at which we look
up the type.  If lookup finds a single type result after instantiation,
we now build TypeSourceInfo for it just like a normal transformation
would.

This fixes the last error in the hello world ATL app that I've been
working with, and it now links and runs with clang.  Please try it and
file bugs!

Reviewers: rsmith

Subscribers: cfe-commits

Differential Revision: http://reviews.llvm.org/D4108

llvm-svn: 210855

10 years agoMove the PPCSelectionDAGInfo off the TargetMachine and onto the
Eric Christopher [Thu, 12 Jun 2014 23:02:32 +0000 (23:02 +0000)]
Move the PPCSelectionDAGInfo off the TargetMachine and onto the
subtarget.

llvm-svn: 210854

10 years agoMake PPCSelectionDAGInfo take a DataLayout instead of a TargetMachine
Eric Christopher [Thu, 12 Jun 2014 22:56:48 +0000 (22:56 +0000)]
Make PPCSelectionDAGInfo take a DataLayout instead of a TargetMachine
since that's all it needs.

llvm-svn: 210853

10 years agoMove PPCTargetLowering off of the TargetMachine and onto the subtarget.
Eric Christopher [Thu, 12 Jun 2014 22:50:10 +0000 (22:50 +0000)]
Move PPCTargetLowering off of the TargetMachine and onto the subtarget.

llvm-svn: 210852

10 years agoRemove stale part of comment.
Eli Bendersky [Thu, 12 Jun 2014 22:47:08 +0000 (22:47 +0000)]
Remove stale part of comment.

llvm-svn: 210851

10 years agoCheck the access of operator delete from the destructor context
Reid Kleckner [Thu, 12 Jun 2014 22:39:12 +0000 (22:39 +0000)]
Check the access of operator delete from the destructor context

Previously we would do the access check from the context of
MarkVTableUsed.

Also update this test to C++11, since that is typically used with the MS
C++ ABI.

Fixes PR20005.

llvm-svn: 210850

10 years agoRemove an extraneous this-> to access the subtarget.
Eric Christopher [Thu, 12 Jun 2014 22:38:20 +0000 (22:38 +0000)]
Remove an extraneous this-> to access the subtarget.

llvm-svn: 210849

10 years agoRename PPCSubTarget to Subtarget in PPCTargetLowering for consistency.
Eric Christopher [Thu, 12 Jun 2014 22:38:18 +0000 (22:38 +0000)]
Rename PPCSubTarget to Subtarget in PPCTargetLowering for consistency.
Also remove an extra local subtarget in the initialization functions.

llvm-svn: 210848

10 years agoFix the polly build.
Rafael Espindola [Thu, 12 Jun 2014 22:37:15 +0000 (22:37 +0000)]
Fix the polly build.

llvm-svn: 210847

10 years agoFix the scheduler's MaxObservedStall computation.
Andrew Trick [Thu, 12 Jun 2014 22:36:28 +0000 (22:36 +0000)]
Fix the scheduler's MaxObservedStall computation.

WenHan Gu pointed out this bug that results in an assert
not being effective in some cases.

llvm-svn: 210846

10 years agoMove PPCJITInfo off of the TargetMachine and onto the subtarget.
Eric Christopher [Thu, 12 Jun 2014 22:28:06 +0000 (22:28 +0000)]
Move PPCJITInfo off of the TargetMachine and onto the subtarget.
Needed to migrate a few functions around to avoid circular header
dependencies.

llvm-svn: 210845

10 years agoRemove the use of TargetMachine from PPCJITInfo and replace with
Eric Christopher [Thu, 12 Jun 2014 22:19:51 +0000 (22:19 +0000)]
Remove the use of TargetMachine from PPCJITInfo and replace with
the subtarget. Also remove unnecessary argument to the constructor
at the same time, we already have access via the subtarget.

llvm-svn: 210844

10 years agoFix the build of KillTheDoctor.
Rafael Espindola [Thu, 12 Jun 2014 22:16:55 +0000 (22:16 +0000)]
Fix the build of KillTheDoctor.

llvm-svn: 210843

10 years ago[modules] The LLVM C API does not require C++!
Richard Smith [Thu, 12 Jun 2014 22:09:39 +0000 (22:09 +0000)]
[modules] The LLVM C API does not require C++!

llvm-svn: 210842

10 years agoThe Clang C API does not require C++. Thanks to Jordan Rose for pointing out this...
Richard Smith [Thu, 12 Jun 2014 22:09:11 +0000 (22:09 +0000)]
The Clang C API does not require C++. Thanks to Jordan Rose for pointing out this error!

llvm-svn: 210841

10 years agoPrefix error_code with std.
Rafael Espindola [Thu, 12 Jun 2014 22:08:48 +0000 (22:08 +0000)]
Prefix error_code with std.

llvm-svn: 210840

10 years agoMove PPCInstrInfo off of the target machine and onto the subtarget.
Eric Christopher [Thu, 12 Jun 2014 22:05:46 +0000 (22:05 +0000)]
Move PPCInstrInfo off of the target machine and onto the subtarget.

llvm-svn: 210839

10 years agoQuick build fix.
Rafael Espindola [Thu, 12 Jun 2014 22:01:48 +0000 (22:01 +0000)]
Quick build fix.

llvm-svn: 210838

10 years agoTry to fix the windows build.
Rafael Espindola [Thu, 12 Jun 2014 21:53:57 +0000 (21:53 +0000)]
Try to fix the windows build.

llvm-svn: 210837

10 years agoRemove TargetMachine from PPCInstrInfo and all dependencies and
Eric Christopher [Thu, 12 Jun 2014 21:48:52 +0000 (21:48 +0000)]
Remove TargetMachine from PPCInstrInfo and all dependencies and
replace with the current subtarget.

llvm-svn: 210836

10 years agoDon't use 'using std::error_code' in include/llvm.
Rafael Espindola [Thu, 12 Jun 2014 21:46:39 +0000 (21:46 +0000)]
Don't use 'using std::error_code' in include/llvm.

This should make sure that most new uses use the std prefix.

llvm-svn: 210835

10 years agos/pr2007/20007/ in a test
Reid Kleckner [Thu, 12 Jun 2014 21:46:38 +0000 (21:46 +0000)]
s/pr2007/20007/ in a test

llvm-svn: 210834

10 years agoMove test for r210734 to Feature/aliases.ll.
Bob Wilson [Thu, 12 Jun 2014 21:37:30 +0000 (21:37 +0000)]
Move test for r210734 to Feature/aliases.ll.

llvm-svn: 210833

10 years agoPrefix another use of error_code.
Rafael Espindola [Thu, 12 Jun 2014 21:36:35 +0000 (21:36 +0000)]
Prefix another use of error_code.

llvm-svn: 210832

10 years agoMore prefixing of error_code.
Rafael Espindola [Thu, 12 Jun 2014 21:35:33 +0000 (21:35 +0000)]
More prefixing of error_code.

llvm-svn: 210831

10 years agoDelete trailing whitespace.
Matt Arsenault [Thu, 12 Jun 2014 21:27:03 +0000 (21:27 +0000)]
Delete trailing whitespace.

Hopefully this forces cmake to re-run.

llvm-svn: 210830

10 years agoUpdate test case to use "not" instead of "XFAIL".
Juergen Ributzka [Thu, 12 Jun 2014 21:17:40 +0000 (21:17 +0000)]
Update test case to use "not" instead of "XFAIL".

llvm-svn: 210829

10 years agoGVN: Enable value forwarding for calloc
Duncan P. N. Exon Smith [Thu, 12 Jun 2014 21:16:19 +0000 (21:16 +0000)]
GVN: Enable value forwarding for calloc

Enable value forwarding for loads from `calloc()` without an intervening
store.

This change extends GVN to handle the following case:

    %1 = tail call noalias i8* @calloc(i64 1, i64 4)
    %2 = bitcast i8* %1 to i32*
    ; This load is trivially constant zero
    %3 = load i32* %2, align 4

This is analogous to the handling for `malloc()` in the same places.
`malloc()` returns `undef`; `calloc()` returns a zero value.  Note that
it is correct to return zero even for out of bounds GEPs since the
result of such a GEP would be undefined.

Patch by Philip Reames!

llvm-svn: 210828

10 years agoR600: Mostly remove remaining AMDIL intrinsics.
Matt Arsenault [Thu, 12 Jun 2014 21:15:44 +0000 (21:15 +0000)]
R600: Mostly remove remaining AMDIL intrinsics.

Delete all unused ones, and add new AMDGPU named intrinsics for
the ones that are. Handle the old AMDIL names for comptability (although
remove their GCCBuiltin names) and add tests since there weren't any
for these before.

llvm-svn: 210827

10 years agoTweak documentation.
Nico Weber [Thu, 12 Jun 2014 21:15:10 +0000 (21:15 +0000)]
Tweak documentation.

1. Having "get started", "get involved", and "hacking" makes it hard to find
   how to send patches, so add a link from "get involved" to "hacking".
2. Remove an almost 5 year old note on the test running meachanism changing
   soon.
3. Let "hacking" link to the LLVM developer policy.

llvm-svn: 210826

10 years agoRemove unused include which breaks build after r210803
Jeroen Ketema [Thu, 12 Jun 2014 21:10:17 +0000 (21:10 +0000)]
Remove unused include which breaks build after r210803

Tested with llvm 3.4 and trunk.

llvm-svn: 210825

10 years agoMove DataLayout from the PPCTargetMachine to the subtarget.
Eric Christopher [Thu, 12 Jun 2014 21:08:06 +0000 (21:08 +0000)]
Move DataLayout from the PPCTargetMachine to the subtarget.

llvm-svn: 210824

10 years agoObjective-C ARC. Blocks that strongly capture themselves
Fariborz Jahanian [Thu, 12 Jun 2014 20:57:14 +0000 (20:57 +0000)]
Objective-C ARC. Blocks that strongly capture themselves
to call themselves will get the warning:
"Capturing <itself> strongly in this block is likely to
lead to a retain cycle". Cut down on the amount of noise
by noticing that user at some point sets the captured variable
to null in order to release it (and break the cycle).
// rdar://16944538

llvm-svn: 210823

10 years agoMove PPCFrameLowering into PPCSubtarget from PPCTargetMachine. Use
Eric Christopher [Thu, 12 Jun 2014 20:54:11 +0000 (20:54 +0000)]
Move PPCFrameLowering into PPCSubtarget from PPCTargetMachine. Use
the initializeSubtargetDependencies code to obtain an initialized
subtarget and migrate a couple of subtarget using functions to the
.cpp file to avoid circular includes.

llvm-svn: 210822

10 years agoFix up extra whitespace from previous commit.
Todd Fiala [Thu, 12 Jun 2014 20:50:30 +0000 (20:50 +0000)]
Fix up extra whitespace from previous commit.

llvm-svn: 210821

10 years agoRefer to error_code with an std prefix.
Rafael Espindola [Thu, 12 Jun 2014 20:42:12 +0000 (20:42 +0000)]
Refer to error_code with an std prefix.

llvm-svn: 210820

10 years agoLangRef: clarify that global declarations can have section and alignment info.
Bob Wilson [Thu, 12 Jun 2014 20:40:33 +0000 (20:40 +0000)]
LangRef: clarify that global declarations can have section and alignment info.

I'm not sure what it means to set a section for a declaration in another
translation unit, but there are some tests in the tree that do it so it seems
to be legal now regardless.

llvm-svn: 210819

10 years agoAdded gdb-remote P/p thread suffix test.
Todd Fiala [Thu, 12 Jun 2014 20:38:56 +0000 (20:38 +0000)]
Added gdb-remote P/p thread suffix test.

Improved the P writes all GPR register values test.  It now
limits itself to GPR registers that are not containers for
other registers.  Pulled in improvements from the llgs branch.

Note on Linux llgs I'm able to write a much wider range
of registers successfully with $P using the bitflip test than I am able
to write with debugserver.  Might be worth drilling into.

llvm-svn: 210818

10 years agoRefer to error_code with the std prefix.
Rafael Espindola [Thu, 12 Jun 2014 20:37:59 +0000 (20:37 +0000)]
Refer to error_code with the std prefix.

llvm-svn: 210817

10 years ago[FastISel][X86] Argument lowering test case
Juergen Ributzka [Thu, 12 Jun 2014 20:34:09 +0000 (20:34 +0000)]
[FastISel][X86] Argument lowering test case

This test case is supposed to xfail, because we do not handle structs or byval
arguments.

llvm-svn: 210816

10 years ago[FastIsel][X86] Add support for lowering the first 8 floating-point arguments.
Juergen Ributzka [Thu, 12 Jun 2014 20:12:34 +0000 (20:12 +0000)]
[FastIsel][X86] Add support for lowering the first 8 floating-point arguments.

Recommit with fixed argument attribute checking code, which is required to bail
out of all the cases we don't handle yet.

llvm-svn: 210815

10 years agoCodeGen: enable mov.w/mov.t pairs with minsize for WoA
Saleem Abdulrasool [Thu, 12 Jun 2014 20:06:33 +0000 (20:06 +0000)]
CodeGen: enable mov.w/mov.t pairs with minsize for WoA

Windows on ARM uses COFF/PE which is intrinsically position independent.  For
the case of 32-bit immediates, use a pair-wise relocation as otherwise we may
exceed the range of operators.  This fixes a code generation crash when using
-Oz when targeting Windows on ARM.

llvm-svn: 210814

10 years agoMS ABI: Fix forming pointers to members of a base class
Reid Kleckner [Thu, 12 Jun 2014 19:49:17 +0000 (19:49 +0000)]
MS ABI: Fix forming pointers to members of a base class

Previously we would calculate the inheritance model of a class when
requiring a pointer to member type of that class to be complete.  The
inheritance model is used to figure out how many fields are used by the
member pointer.

However, once we require a pointer to member of a derived class type to
be complete, we can form pointers to members of bases without
calculating the inheritance model for those bases.  This was causing
crashes on this simple test case:

  struct A {
    void f();
    void f(int);
  };
  struct B : public A {};
  void g() { void (B::*a)() = &B::f; }

Now we calculate the inheritance models of all base classes when
completing a member pointer type.

Fixes PR2007.

llvm-svn: 210813

10 years agoPermit -Wswitch coverage for enumerations.
Saleem Abdulrasool [Thu, 12 Jun 2014 19:33:26 +0000 (19:33 +0000)]
Permit -Wswitch coverage for enumerations.

Thanks to David Blakie and Richard Smith for pointing out that we can retain the
-Wswitch coverage while avoiding the warning from GCC by pushing the unreachable
outside of the switch!

llvm-svn: 210812

10 years agoTry to fix the polly build.
Rafael Espindola [Thu, 12 Jun 2014 19:22:17 +0000 (19:22 +0000)]
Try to fix the polly build.

llvm-svn: 210811

10 years agoRevert "[FastIsel][X86] Add support for lowering the first 8 floating-point arguments."
Juergen Ributzka [Thu, 12 Jun 2014 19:21:43 +0000 (19:21 +0000)]
Revert "[FastIsel][X86] Add support for lowering the first 8 floating-point arguments."

Reverting it because it breaks several tests.

llvm-svn: 210810

10 years ago[llvm-symbolizer] Fix parsing DW_AT_ranges in Fission skeleton compile unit DIEs.
Alexey Samsonov [Thu, 12 Jun 2014 18:52:35 +0000 (18:52 +0000)]
[llvm-symbolizer] Fix parsing DW_AT_ranges in Fission skeleton compile unit DIEs.

Turns out that DW_AT_ranges_base attribute sets the offset for
DW_AT_ranges values specified in the .dwo file, but not for DW_AT_ranges specified
in the skeleton compile unit DIE in the main executable. This is extremely confusing,
and would hopefully be fixed in DWARF-5 when it's finalized. For now this
behavior makes sense, as otherwise Fission would break DWARF consumers who
doesn't know anything about DW_AT_ranges_base.

llvm-svn: 210809

10 years agoAdd missing "InitializerConstant" to global variable syntax in LangRef.
Bob Wilson [Thu, 12 Jun 2014 18:42:55 +0000 (18:42 +0000)]
Add missing "InitializerConstant" to global variable syntax in LangRef.

The syntax for Global Variables in LangRef is missing the initializer.
This syntax section was added in r199218 along with changes to the
dllexport/dllimport handling, and I guess it was just an oversight to omit the
initializer values. I’ve marked the initializer as optional because this syntax
is used for both declarations and definitions.

llvm-svn: 210808

10 years agoRevert r210721 as it causes breakage in internal builds (and possibly GDB).
Eli Bendersky [Thu, 12 Jun 2014 18:05:39 +0000 (18:05 +0000)]
Revert r210721 as it causes breakage in internal builds (and possibly GDB).

llvm-svn: 210807

10 years agoX86: stifle GCC warning
Saleem Abdulrasool [Thu, 12 Jun 2014 17:56:18 +0000 (17:56 +0000)]
X86: stifle GCC warning

lib/Target/X86/X86TargetTransformInfo.cpp: In member function ‘virtual unsigned int {anonymous}::X86TTI::getIntImmCost(unsigned int, unsigned int, const llvm::APInt&, llvm::Type*) const’:
lib/Target/X86/X86TargetTransformInfo.cpp:920:60: warning: enumeral and non-enumeral type in conditional expression [enabled by default]

This seems like an unhelpful warning, but there doesnt seem to be a controlling
flag, so add an explicit cast to silence the warning.

llvm-svn: 210806

10 years agoTrying to fix the windows build.
Rafael Espindola [Thu, 12 Jun 2014 17:49:35 +0000 (17:49 +0000)]
Trying to fix the windows build.

llvm-svn: 210805