platform/upstream/llvm.git
10 years agoOnly check os version for POSIX_SPAWN_CLOEXEC_DEFAULT on desktop builds of LLDB.
Greg Clayton [Wed, 8 Jan 2014 22:12:52 +0000 (22:12 +0000)]
Only check os version for POSIX_SPAWN_CLOEXEC_DEFAULT on desktop builds of LLDB.

llvm-svn: 198800

10 years agoUse getError and remove the error_code operator.
Rafael Espindola [Wed, 8 Jan 2014 22:03:39 +0000 (22:03 +0000)]
Use getError and remove the error_code operator.

llvm-svn: 198799

10 years ago[Mips] Remove unnecessary #include pragma.
Simon Atanasyan [Wed, 8 Jan 2014 22:00:35 +0000 (22:00 +0000)]
[Mips] Remove unnecessary #include pragma.

llvm-svn: 198798

10 years agoUse getError instead of the error_code operator.
Rafael Espindola [Wed, 8 Jan 2014 22:00:09 +0000 (22:00 +0000)]
Use getError instead of the error_code operator.

llvm-svn: 198797

10 years agoRemove vestigal bits of MC from the mangler. It no longer uses this, and
Chandler Carruth [Wed, 8 Jan 2014 21:59:22 +0000 (21:59 +0000)]
Remove vestigal bits of MC from the mangler. It no longer uses this, and
having the include could cause weird layering problems between the IR
and MC libraries.

llvm-svn: 198796

10 years agoConservatively handle multiple MMOs in MIsNeedChainEdge
Hal Finkel [Wed, 8 Jan 2014 21:52:02 +0000 (21:52 +0000)]
Conservatively handle multiple MMOs in MIsNeedChainEdge

MIsNeedChainEdge, which is used by -enable-aa-sched-mi (AA in misched), had an
llvm_unreachable when -enable-aa-sched-mi is enabled and we reach an
instruction with multiple MMOs. Instead, return a conservative answer. This
allows testing -enable-aa-sched-mi on x86.

Also, this moves the check above the isUnsafeMemoryObject checks.
isUnsafeMemoryObject is currently correct only for instructions with one MMO
(as noted in the comment in isUnsafeMemoryObject):

  // We purposefully do no check for hasOneMemOperand() here
  // in hope to trigger an assert downstream in order to
  // finish implementation.

The problem with this is that, had the candidate edge passed the
"!MIa->mayStore() && !MIb->mayStore()" check, the hoped-for assert would never
happen (which could, in theory, lead to incorrect behavior if one of these
secondary MMOs was volatile, for example).

llvm-svn: 198795

10 years agoMove declaration of variables down to first use.
Matt Arsenault [Wed, 8 Jan 2014 21:47:14 +0000 (21:47 +0000)]
Move declaration of variables down to first use.

llvm-svn: 198794

10 years agoAdd missing definitions of key_type and value_type to DenseSet.
Matt Arsenault [Wed, 8 Jan 2014 21:38:04 +0000 (21:38 +0000)]
Add missing definitions of key_type and value_type to DenseSet.

This matches std::set and allows using DenseSet with the functions
in SetOperations.h

llvm-svn: 198793

10 years agoAdd get and getError methods to ErrorOr.
Rafael Espindola [Wed, 8 Jan 2014 21:17:09 +0000 (21:17 +0000)]
Add get and getError methods to ErrorOr.

ErrorOr is modeled after boost::optional which has a get method.

llvm-svn: 198792

10 years ago[AArch64][NEON] Added UXTL and UXTL2 instruction aliases
Ana Pazos [Wed, 8 Jan 2014 21:02:13 +0000 (21:02 +0000)]
[AArch64][NEON] Added UXTL and UXTL2 instruction aliases

llvm-svn: 198791

10 years ago[Mips] GOT16 relocation against non-local symbol does not require a
Simon Atanasyan [Wed, 8 Jan 2014 20:43:03 +0000 (20:43 +0000)]
[Mips] GOT16 relocation against non-local symbol does not require a
paired LO16 relocation.

llvm-svn: 198790

10 years ago[Mips] Add assert to check that we handle all paired relocations.
Simon Atanasyan [Wed, 8 Jan 2014 20:42:52 +0000 (20:42 +0000)]
[Mips] Add assert to check that we handle all paired relocations.

llvm-svn: 198789

10 years ago[Mips] Factor out the code determines type of GOT entry (local/global)
Simon Atanasyan [Wed, 8 Jan 2014 20:42:45 +0000 (20:42 +0000)]
[Mips] Factor out the code determines type of GOT entry (local/global)
into the separate function.

llvm-svn: 198788

10 years ago[Mips] Do not use standard relocation identifier R_MIPS_NONE for
Simon Atanasyan [Wed, 8 Jan 2014 20:42:38 +0000 (20:42 +0000)]
[Mips] Do not use standard relocation identifier R_MIPS_NONE for
internal purpose. Use special LLD_R_MIPS_GLOBAL_GOT constant for that.

llvm-svn: 198787

10 years ago[Mips] Rename function to better reflect its purpose.
Simon Atanasyan [Wed, 8 Jan 2014 20:42:30 +0000 (20:42 +0000)]
[Mips] Rename function to better reflect its purpose.

llvm-svn: 198786

10 years ago[Mips] Do not save GOT headers atoms into the RelocationPass class
Simon Atanasyan [Wed, 8 Jan 2014 20:42:23 +0000 (20:42 +0000)]
[Mips] Do not save GOT headers atoms into the RelocationPass class
fields.

llvm-svn: 198785

10 years ago[Mips] Rename some classes to skip redundant mentioning of 'mips'.
Simon Atanasyan [Wed, 8 Jan 2014 20:42:17 +0000 (20:42 +0000)]
[Mips] Rename some classes to skip redundant mentioning of 'mips'.

llvm-svn: 198784

10 years ago[Mips] Move MipsGOTPass implementation to the separate file.
Simon Atanasyan [Wed, 8 Jan 2014 20:42:11 +0000 (20:42 +0000)]
[Mips] Move MipsGOTPass implementation to the separate file.

llvm-svn: 198783

10 years agoEnsure that ENABLE_SHARED is substituted for use in Windows builds
Alp Toker [Wed, 8 Jan 2014 20:06:24 +0000 (20:06 +0000)]
Ensure that ENABLE_SHARED is substituted for use in Windows builds

We (perhaps over-cautiously) disable the new plugin tests on static Windows
builds right now, matching what LLVM core does. This change was needed for the
lit check to work.

Thanks to Warren Hunt for spotting this.

llvm-svn: 198782

10 years ago[Serialization] In ASTReader::getInputFile record it when we didn't find the file...
Argyrios Kyrtzidis [Wed, 8 Jan 2014 19:13:34 +0000 (19:13 +0000)]
[Serialization] In ASTReader::getInputFile record it when we didn't find the file to avoid looking it up again.

Hopefully addresses rdar://14514222.

llvm-svn: 198781

10 years agoForce emit a relocation for @gnu_indirect_function symbols so that the indirect
Roman Divacky [Wed, 8 Jan 2014 18:50:32 +0000 (18:50 +0000)]
Force emit a relocation for @gnu_indirect_function symbols so that the indirect
resolution works.

llvm-svn: 198780

10 years ago[analyzer] Warn about double-delete in C++ at the second delete...
Jordan Rose [Wed, 8 Jan 2014 18:46:55 +0000 (18:46 +0000)]
[analyzer] Warn about double-delete in C++ at the second delete...

...rather somewhere in the destructor when we try to access something and
realize the object has already been deleted. This is necessary because
the destructor is processed before the 'delete' itself.

Patch by Karthik Bhat!

llvm-svn: 198779

10 years ago[x86] Remove OpSize16 flag from MOV32r0
David Woodhouse [Wed, 8 Jan 2014 18:38:26 +0000 (18:38 +0000)]
[x86] Remove OpSize16 flag from MOV32r0

It's not a real instruction any more and doesn't need encoding information.

llvm-svn: 198778

10 years agoTeach the DAGCombiner how to fold 'vselect' dag nodes according
Andrea Di Biagio [Wed, 8 Jan 2014 18:33:04 +0000 (18:33 +0000)]
Teach the DAGCombiner how to fold 'vselect' dag nodes according
to the following two rules:
  1) fold (vselect (build_vector AllOnes), A, B) -> A
  2) fold (vselect (build_vector AllZeros), A, B) -> B

llvm-svn: 198777

10 years agoFixed a kernel panic that would occur if you debug anything on MacOSX 10.7 or earlier...
Greg Clayton [Wed, 8 Jan 2014 18:02:51 +0000 (18:02 +0000)]
Fixed a kernel panic that would occur if you debug anything on MacOSX 10.7 or earlier due to the use of the POSIX_SPAWN_CLOEXEC_DEFAULT attribute flag that closes all file descriptors on exec. We now dyamically detect the OS version and do the right thing.

llvm-svn: 198776

10 years agoAdd missing rename from the previous commit.
Rafael Espindola [Wed, 8 Jan 2014 17:56:46 +0000 (17:56 +0000)]
Add missing rename from the previous commit.

No idea how this was compiling locally. Found by the bots.

llvm-svn: 198775

10 years agoRename get to getStorage and getError to getErrorStorage.
Rafael Espindola [Wed, 8 Jan 2014 17:43:26 +0000 (17:43 +0000)]
Rename get to getStorage and getError to getErrorStorage.

These private functions return pointers to the internal storage.

llvm-svn: 198774

10 years agoAdd missing test case for r198737.
Lang Hames [Wed, 8 Jan 2014 16:31:16 +0000 (16:31 +0000)]
Add missing test case for r198737.

llvm-svn: 198772

10 years agoRemove mention of old deleted test scripts from testing guide
Nico Rieck [Wed, 8 Jan 2014 16:30:03 +0000 (16:30 +0000)]
Remove mention of old deleted test scripts from testing guide

llvm-svn: 198771

10 years agoclang-format: Don't hang forever when encountering a stray "}" in an @implementation...
Benjamin Kramer [Wed, 8 Jan 2014 15:59:42 +0000 (15:59 +0000)]
clang-format: Don't hang forever when encountering a stray "}" in an @implementation block.

PR18406.

llvm-svn: 198770

10 years agoclang-format: Fix spacing in Cpp11 braced lists:
Daniel Jasper [Wed, 8 Jan 2014 15:41:13 +0000 (15:41 +0000)]
clang-format: Fix spacing in Cpp11 braced lists:

Before:
  vector<int> foo{ ::SomeFunction()};

After:
  vector<int> foo{::SomeFunction()};

llvm-svn: 198769

10 years ago[DAGCombiner] Factor duplicated rotate code into a separate function
Richard Sandiford [Wed, 8 Jan 2014 15:40:47 +0000 (15:40 +0000)]
[DAGCombiner] Factor duplicated rotate code into a separate function

No functional change intended.

llvm-svn: 198768

10 years agoDon't emit diagnostics for system headers.
Alexander Kornienko [Wed, 8 Jan 2014 15:21:08 +0000 (15:21 +0000)]
Don't emit diagnostics for system headers.

Reviewers: djasper

Reviewed By: djasper

CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D2519

llvm-svn: 198767

10 years agolit: Provide file location in cfg error messages
Alp Toker [Wed, 8 Jan 2014 14:20:59 +0000 (14:20 +0000)]
lit: Provide file location in cfg error messages

Python doesn't do a good job at diagnosing string exec() so use execfile()
where available.

This should be a timesaver when trying to get to the bottom of build bot
failures.

Before:

    File "llvm/utils/lit/lit/TestingConfig.py", line 93, in load_from_path
      exec("exec data in cfg_globals")
    File "<string>", line 1, in <module>
    File "<string>", line 194, in <module>
  NameError: name 'typo' is not defined

After:

    File "llvm/utils/lit/lit/TestingConfig.py", line 95, in load_from_path
    execfile(path, cfg_globals)
    File "clang/test/lit.cfg", line 194, in <module>
      typo
      ^~~~
  NameError: name 'typo' is not defined

llvm-svn: 198766

10 years agoTreating the RegParmAttr as a TypeAttr because that is what it is.
Aaron Ballman [Wed, 8 Jan 2014 13:23:01 +0000 (13:23 +0000)]
Treating the RegParmAttr as a TypeAttr because that is what it is.

Patch reviewed by Rafael Espindola.

llvm-svn: 198765

10 years agoDefine ENABLE_CLANG_EXAMPLES instead of relying on BUILD_EXAMPLES
Alp Toker [Wed, 8 Jan 2014 13:00:32 +0000 (13:00 +0000)]
Define ENABLE_CLANG_EXAMPLES instead of relying on BUILD_EXAMPLES

This is a further build fix attempt for r198747 on some Makefile builders where
the value wasn't set at all.

llvm-svn: 198764

10 years ago[x86] Support R_386_PC8, R_386_PC16 and R_X86_64_PC8
David Woodhouse [Wed, 8 Jan 2014 12:58:40 +0000 (12:58 +0000)]
[x86] Support R_386_PC8, R_386_PC16 and R_X86_64_PC8

llvm-svn: 198763

10 years ago[x86] Add JMP_2 and other 16-bit PC-relative branch instructions
David Woodhouse [Wed, 8 Jan 2014 12:58:36 +0000 (12:58 +0000)]
[x86] Add JMP_2 and other 16-bit PC-relative branch instructions

Mark them as requiring 16-bit mode for now, since we don't yet have
relaxation support for FK_Data_2.

llvm-svn: 198762

10 years ago[x86] Do not relax PUSHi16 to PUSHi32 (PR18414)
David Woodhouse [Wed, 8 Jan 2014 12:58:32 +0000 (12:58 +0000)]
[x86] Do not relax PUSHi16 to PUSHi32 (PR18414)

They do *different* things to %esp, so they are not equivalent.

Rename PUSHi8 to PUSH32i8 and add the missing PUSH16i8.

llvm-svn: 198761

10 years ago[x86] Make AsmParser validate registers for memory operands a bit better
David Woodhouse [Wed, 8 Jan 2014 12:58:28 +0000 (12:58 +0000)]
[x86] Make AsmParser validate registers for memory operands a bit better

We can't do a perfect job here. We *have* to allow (%dx) even in 64-bit
mode, for example, because it might be used for an unofficial form of
the in/out instructions. We actually want to do a better job of validation
*later*. Perhaps *instead* of doing it where we are at the moment.

But for now, doing what validation we *can* do in the place that the code
already has its validation, is an improvement.

llvm-svn: 198760

10 years ago[x86] Fix MOV8ao8 et al for 16-bit mode, fix up disassembler to understand
David Woodhouse [Wed, 8 Jan 2014 12:58:24 +0000 (12:58 +0000)]
[x86] Fix MOV8ao8 et al for 16-bit mode, fix up disassembler to understand

It seems there is no separate instruction class for having AdSize *and*
OpSize bits set, which is required in order to disambiguate between all
these instructions. So add that to the disassembler.

Hm, perhaps we do need an AdSize16 bit after all?

llvm-svn: 198759

10 years ago[x86] Use 16-bit addressing where possible in 16-bit mode
David Woodhouse [Wed, 8 Jan 2014 12:58:18 +0000 (12:58 +0000)]
[x86] Use 16-bit addressing where possible in 16-bit mode

Where "where possible" means that it's an immediate value and it's below
0x10000. In fact GAS will either truncate or error with larger values,
and will insist on using the addr32 prefix to get 32-bit addressing. So
perhaps we should do that, in a later patch.

llvm-svn: 198758

10 years ago[x86] Fix JCXZ,JECXZ_32 for 16-bit mode
David Woodhouse [Wed, 8 Jan 2014 12:58:12 +0000 (12:58 +0000)]
[x86] Fix JCXZ,JECXZ_32 for 16-bit mode

JCXZ should have the 0x67 prefix only if we're in 32-bit mode, so make that
appropriately conditional. And JECXZ needs the prefix instead.

llvm-svn: 198757

10 years ago[x86] Disambiguate RET[QL] and fix aliases for 16-bit mode
David Woodhouse [Wed, 8 Jan 2014 12:58:07 +0000 (12:58 +0000)]
[x86] Disambiguate RET[QL] and fix aliases for 16-bit mode

I couldn't see how to do this sanely without splitting RETQ from RETL.

Eric says: "sad about the inability to roundtrip them now, but...".
I have no idea what that means, but perhaps it wants preserving in the
commit comment.

llvm-svn: 198756

10 years ago[x86] Disambiguate [LS][IG]DT{32,64}m and add 16-bit versions, fix aliases
David Woodhouse [Wed, 8 Jan 2014 12:57:55 +0000 (12:57 +0000)]
[x86] Disambiguate [LS][IG]DT{32,64}m and add 16-bit versions, fix aliases

llvm-svn: 198755

10 years ago[x86] Add JMP16[rm],CALL16[rm] instructions, and fix up aliases
David Woodhouse [Wed, 8 Jan 2014 12:57:49 +0000 (12:57 +0000)]
[x86] Add JMP16[rm],CALL16[rm] instructions, and fix up aliases

llvm-svn: 198754

10 years ago[x86] Add PUSHA16,POPA16 instructions, and fix aliases for 16-bit mode
David Woodhouse [Wed, 8 Jan 2014 12:57:45 +0000 (12:57 +0000)]
[x86] Add PUSHA16,POPA16 instructions, and fix aliases for 16-bit mode

llvm-svn: 198753

10 years ago[x86] Add OpSize16 to instructions that need it
David Woodhouse [Wed, 8 Jan 2014 12:57:40 +0000 (12:57 +0000)]
[x86] Add OpSize16 to instructions that need it

This fixes the bulk of 16-bit output, and the corresponding test case
x86-16.s now looks mostly like the x86-32.s test case that it was
originally based on. A few irrelevant instructions have been dropped,
and there are still some corner cases to be fixed in subsequent patches.

llvm-svn: 198752

10 years agoProspective build fix for the Makefile system
Alp Toker [Wed, 8 Jan 2014 12:03:17 +0000 (12:03 +0000)]
Prospective build fix for the Makefile system

llvm-svn: 198751

10 years agoBuild fix following r198747
Alp Toker [Wed, 8 Jan 2014 11:55:49 +0000 (11:55 +0000)]
Build fix following r198747

Convert CMake CLANG_BUILD_EXAMPLES to a boolean value for consumption by
Python, otherwise the raw config string gets passed through which may not be
0/1.

The if/else matches the technique used to solve the problem in LLVM's own
CMakeLists.txt.

llvm-svn: 198750

10 years agoUse -std=gnu99 in tools/llvm-c-test/CMakeLists.txt
Rafael Espindola [Wed, 8 Jan 2014 11:48:19 +0000 (11:48 +0000)]
Use -std=gnu99 in tools/llvm-c-test/CMakeLists.txt

With a current mingw (gcc 4.8.1) it looks like we hit some variation of

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40278

The end result is that off_t is not defined and the build fails without this
patch.

llvm-svn: 198749

10 years agoUse -std=gnu89 in tools/c-index-test/CMakeLists.txt
Rafael Espindola [Wed, 8 Jan 2014 11:44:42 +0000 (11:44 +0000)]
Use -std=gnu89 in tools/c-index-test/CMakeLists.txt

With the old use of -std=c89 off_t is not defined and the build fails.

This seems to be another variation of
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40278.

llvm-svn: 198748

10 years agoAdd tests for clang plugins
Alp Toker [Wed, 8 Jan 2014 11:38:47 +0000 (11:38 +0000)]
Add tests for clang plugins

Somehow the entire plugin infrastructure went wholly untested until now.

The only plugins available for use in testing are the examples, so plugin tests
will only be run if CLANG_BUILD_EXAMPLES is enabled in the build.

(The examples should really be enabled by default, not just to aid testing but
also to prevent bitrot in some key user-facing code. I'll propose that
shortly.)

Requires supporting changes in LLVM r198746.

llvm-svn: 198747

10 years agoCMake: Provide LLVM_PLUGIN_EXT definition
Alp Toker [Wed, 8 Jan 2014 11:10:24 +0000 (11:10 +0000)]
CMake: Provide LLVM_PLUGIN_EXT definition

This is needed to support the addition of tests for clang loadable plugins.

In clang, plugins are built as modules (bundles on OS X) rather than dynamic
libraries (dylib) so the build system needs to inform lit of the actual
file extension in use, typically '.so' on Unix and '.dll' on Windows.

(LLVM itself should probably switch to this scheme to fix PR14903 once and for
all.)

No change in build output or functionality intended.

llvm-svn: 198746

10 years agoAVX-512: Added more intrinsics for pmin/pmax, pabs, blend, pmuldq.
Elena Demikhovsky [Wed, 8 Jan 2014 10:54:22 +0000 (10:54 +0000)]
AVX-512: Added more intrinsics for pmin/pmax, pabs, blend, pmuldq.

llvm-svn: 198745

10 years ago[patch] Adjust behavior of FDE cross-section relocs for targets that don't support...
Iain Sandoe [Wed, 8 Jan 2014 10:22:54 +0000 (10:22 +0000)]
[patch] Adjust behavior of FDE cross-section relocs for targets that don't support abs-differences.

Modern versions of OSX/Darwin's ld (ld64 > 97.17) have an optimisation present that allows the back end to omit relocations (and replace them with an absolute difference) for FDE some text section refs.

This patch allows a backend to opt-in to this behaviour by setting "DwarfFDESymbolsUseAbsDiff".  At present, this is only enabled for modern x86 OSX ports.

test changes by David Fang.

llvm-svn: 198744

10 years ago[AArch64 NEON] Fix generating incorrect value type of NEON_VDUPLANE
Kevin Qin [Wed, 8 Jan 2014 08:06:14 +0000 (08:06 +0000)]
[AArch64 NEON] Fix generating incorrect value type of NEON_VDUPLANE
when lower build_vector if result value type mismatch with operand
value type.

llvm-svn: 198743

10 years agoFixes a bug preventing reading of the python register file.
Todd Fiala [Wed, 8 Jan 2014 07:52:40 +0000 (07:52 +0000)]
Fixes a bug preventing reading of the python register file.

This change fixes a bug recently introduced in ProcessGDBRemote that
prevented the Python register definition file from getting loaded when
the qRegisterInfo0 response returned $00#.

Patch by Steve Pucci.

llvm-svn: 198742

10 years agoFor AArch64, support builtin neon vector type with 'long' as base element type.
Jiangning Liu [Wed, 8 Jan 2014 07:51:48 +0000 (07:51 +0000)]
For AArch64, support builtin neon vector type with 'long' as base element type.

llvm-svn: 198741

10 years ago[SparcV9] Rename operands in some sparc64 instructions so that TableGen can encode...
Venkatraman Govindaraju [Wed, 8 Jan 2014 07:47:57 +0000 (07:47 +0000)]
[SparcV9] Rename operands in some sparc64 instructions so that TableGen can encode them correctly.

llvm-svn: 198740

10 years ago[Sparc] Correct the mask for fixup_sparc_br19.
Venkatraman Govindaraju [Wed, 8 Jan 2014 06:46:51 +0000 (06:46 +0000)]
[Sparc] Correct the mask for fixup_sparc_br19.

llvm-svn: 198739

10 years ago[Sparc] Add support for parsing branch instructions and conditional moves.
Venkatraman Govindaraju [Wed, 8 Jan 2014 06:14:52 +0000 (06:14 +0000)]
[Sparc] Add support for parsing branch instructions and conditional moves.

llvm-svn: 198738

10 years agoRe-apply r196639: Add support for archives and object file caching under MCJIT.
Lang Hames [Wed, 8 Jan 2014 04:09:09 +0000 (04:09 +0000)]
Re-apply r196639: Add support for archives and object file caching under MCJIT.

I believe the bot failures on linux systems were due to overestimating the
alignment of object-files within archives, which are only guaranteed to be
two-byte aligned. I have reduced the alignment in
RuntimeDyldELF::createObjectImageFromFile accordingly.

llvm-svn: 198737

10 years agotests: disable ARM unwinding tests if ARM is unavailable
Saleem Abdulrasool [Wed, 8 Jan 2014 03:44:01 +0000 (03:44 +0000)]
tests: disable ARM unwinding tests if ARM is unavailable

Appease the buildbots for targets which do not build the ARM support by moving
the ARM specific test into a subdirectory and use the lit configuration to
disable them appropriately.

Thanks to chapuni and thakis for explaining how to do this!

llvm-svn: 198736

10 years agoARM IAS: properly handle expression operands
Saleem Abdulrasool [Wed, 8 Jan 2014 03:28:14 +0000 (03:28 +0000)]
ARM IAS: properly handle expression operands

Operands which involved label arithemetic would previously fail to parse.  This
corrects that by adding the additional case for the shift operand validation.

llvm-svn: 198735

10 years agollvm-readobj: add support for ARM EHABI unwind info
Saleem Abdulrasool [Wed, 8 Jan 2014 03:28:09 +0000 (03:28 +0000)]
llvm-readobj: add support for ARM EHABI unwind info

This adds some preliminary support for decoding ARM EHABI unwinding information.
The major functionality that remains from complete support is bytecode
translation.

Each Unwind Index Table is printed out as a separate entity along with its
section index, name, offset, and entries.

Each entry lists the function address, and if possible, the name, of the
function to which it corresponds.  The encoding model, personality routine or
index, and byte code is also listed.

llvm-svn: 198734

10 years agoUndo test
Enrico Granata [Wed, 8 Jan 2014 03:20:06 +0000 (03:20 +0000)]
Undo test

llvm-svn: 198733

10 years agoTest
Enrico Granata [Wed, 8 Jan 2014 03:19:56 +0000 (03:19 +0000)]
Test

llvm-svn: 198732

10 years agoAlignment fix
Enrico Granata [Wed, 8 Jan 2014 03:14:18 +0000 (03:14 +0000)]
Alignment fix

llvm-svn: 198731

10 years agoFix a bug about generating undef operand when optimising shuffle vector and insert...
Hao Liu [Wed, 8 Jan 2014 03:06:15 +0000 (03:06 +0000)]
Fix a bug about generating undef operand when optimising shuffle vector and insert element in instruction combine.

llvm-svn: 198730

10 years ago<rdar://problem/15453076>
Enrico Granata [Wed, 8 Jan 2014 03:02:33 +0000 (03:02 +0000)]
<rdar://problem/15453076>

When determining the type of array members, do not see-through typedefs
For instance, in BOOL arr[4], we want the elements to be typed as BOOL, not signed char

llvm-svn: 198729

10 years ago[mach-o] properly extract atom content from subrange of section content
Nick Kledzik [Wed, 8 Jan 2014 02:52:58 +0000 (02:52 +0000)]
[mach-o] properly extract atom content from subrange of section content

llvm-svn: 198728

10 years ago__CFString should also format as an NSString
Enrico Granata [Wed, 8 Jan 2014 02:34:42 +0000 (02:34 +0000)]
__CFString should also format as an NSString

llvm-svn: 198727

10 years agoRP18408: If a member template is used as a template template argument, it does
Richard Smith [Wed, 8 Jan 2014 01:51:59 +0000 (01:51 +0000)]
RP18408: If a member template is used as a template template argument, it does
not cause the template specialization to have no linkage.

llvm-svn: 198726

10 years agoUse the mach-o MH_* name for file types in yaml
Nick Kledzik [Wed, 8 Jan 2014 01:38:07 +0000 (01:38 +0000)]
Use the mach-o MH_* name for file types in yaml

llvm-svn: 198725

10 years agoNew and improved data formatter for std::shared_ptr<> and std::weak_ptr<>
Enrico Granata [Wed, 8 Jan 2014 01:36:59 +0000 (01:36 +0000)]
New and improved data formatter for std::shared_ptr<> and std::weak_ptr<>

llvm-svn: 198724

10 years agoPR18400: ignore cv-qualifiers on the underlying type of an enumeration.
Richard Smith [Wed, 8 Jan 2014 01:16:19 +0000 (01:16 +0000)]
PR18400: ignore cv-qualifiers on the underlying type of an enumeration.

llvm-svn: 198723

10 years ago[Driver] Fix a typo in the setting of the arch name when -arch x86_64h is used.
Quentin Colombet [Wed, 8 Jan 2014 01:02:06 +0000 (01:02 +0000)]
[Driver] Fix a typo in the setting of the arch name when -arch x86_64h is used.
<rdar://problem/15711488>

llvm-svn: 198722

10 years agoPR18234: Mark a tag definition as invalid early if it appears in a
Richard Smith [Wed, 8 Jan 2014 00:56:48 +0000 (00:56 +0000)]
PR18234: Mark a tag definition as invalid early if it appears in a
type-specifier in C++. Some checks will assert in this case otherwise (in
particular, the access specifier may be missing if this happens inside a class
definition, due to a violation of an AST invariant).

llvm-svn: 198721

10 years ago[x86] Kill gratuitous X86_{32,64}TargetMachine subclasses, use X86TargetMachine
David Woodhouse [Wed, 8 Jan 2014 00:08:50 +0000 (00:08 +0000)]
[x86] Kill gratuitous X86_{32,64}TargetMachine subclasses, use X86TargetMachine

llvm-svn: 198720

10 years agoFixed a bug where the expression parser doesn't
Sean Callanan [Tue, 7 Jan 2014 23:15:26 +0000 (23:15 +0000)]
Fixed a bug where the expression parser doesn't
materialize a variable in a register correctly
if the variable is a pointer.  This fixes a
regression introduced by my commit of Oct. 22nd
(r193191).

llvm-svn: 198718

10 years agoFixed an issue when attaching to a process without specifying a file where we wouldn...
Greg Clayton [Tue, 7 Jan 2014 23:15:02 +0000 (23:15 +0000)]
Fixed an issue when attaching to a process without specifying a file where we wouldn't set the dyld gdb image notifier breakpoint correctly.

<rdar://problem/15720040>

llvm-svn: 198717

10 years agoFix the top header that corresponds to this source file. Thanks for
Chandler Carruth [Tue, 7 Jan 2014 22:15:39 +0000 (22:15 +0000)]
Fix the top header that corresponds to this source file. Thanks for
spotting this Cody Maloney!

llvm-svn: 198716

10 years agoRe-applying r198699 after reverting r198461.
Adrian Prantl [Tue, 7 Jan 2014 22:05:55 +0000 (22:05 +0000)]
Re-applying r198699 after reverting r198461.
Debug info: Implement a cleaner version of r198461. For symmetry with
C and C++ don't emit an extra lexical scope for the compound statement
that is the body of an Objective-C method.

llvm-svn: 198715

10 years agoRevert "Debug info: Ensure that the last stop point in a function is still within"
Adrian Prantl [Tue, 7 Jan 2014 22:05:52 +0000 (22:05 +0000)]
Revert "Debug info: Ensure that the last stop point in a function is still within"

This reverts commit r198461.

llvm-svn: 198714

10 years agoRevert "Debug info: Implement a cleaner version of r198461. For symmetry with"
Adrian Prantl [Tue, 7 Jan 2014 22:05:45 +0000 (22:05 +0000)]
Revert "Debug info: Implement a cleaner version of r198461. For symmetry with"

This reverts commit 198699 so we can get a cleaner patch.

llvm-svn: 198713

10 years agoFixed issues with ptr_refs:
Greg Clayton [Tue, 7 Jan 2014 21:55:00 +0000 (21:55 +0000)]
Fixed issues with ptr_refs:
- If there is only 1 frame ptr_refs now works (fixed issue with stack detection)
- Fixed test for result now that it isn't a pointer anymore

llvm-svn: 198712

10 years ago[analyzer] Files with .c extensions are still C++ files if the compiler is CXX.
Jordan Rose [Tue, 7 Jan 2014 21:39:51 +0000 (21:39 +0000)]
[analyzer] Files with .c extensions are still C++ files if the compiler is CXX.

PR18339

llvm-svn: 198711

10 years ago[analyzer] Pointers escape into +[NSValue valueWithPointer:]...
Jordan Rose [Tue, 7 Jan 2014 21:39:48 +0000 (21:39 +0000)]
[analyzer] Pointers escape into +[NSValue valueWithPointer:]...

...even though the argument is declared "const void *", because this is
just a way to pass pointers around as objects. (Though NSData is often
a better one.)

PR18262

llvm-svn: 198710

10 years ago[analyzer] Remove unused ARCNotOwnedSymbol retain count return effect.
Jordan Rose [Tue, 7 Jan 2014 21:39:41 +0000 (21:39 +0000)]
[analyzer] Remove unused ARCNotOwnedSymbol retain count return effect.

RetainCountChecker has to track returned object values to know if they are
retained or not. Under ARC, even methods that return +1 are tracked by the
system and should be treated as +0. However, this effect behaves exactly
like NotOwned(ObjC), i.e. a generic Objective-C method that actually returns
+0, so we don't need a special case for it.

No functionality change.

llvm-svn: 198709

10 years agoMove the llvm mangler to lib/IR.
Rafael Espindola [Tue, 7 Jan 2014 21:19:40 +0000 (21:19 +0000)]
Move the llvm mangler to lib/IR.

This makes it available to tools that don't link with target (like llvm-ar).

llvm-svn: 198708

10 years agoIn the ELFWriter when writing aliased (.set) symbols dont blindly
Roman Divacky [Tue, 7 Jan 2014 20:17:03 +0000 (20:17 +0000)]
In the ELFWriter when writing aliased (.set) symbols dont blindly
take type from the new symbol but merge them so that the type
is never "downgraded".

This is probably quite rare, except for IFUNC symbols which
we used to misassemble, losing the IFUNC type.

Fixes #18372.

llvm-svn: 198706

10 years agoUpdating the documentation about how to add attributes based on the rather extensive...
Aaron Ballman [Tue, 7 Jan 2014 20:12:20 +0000 (20:12 +0000)]
Updating the documentation about how to add attributes based on the rather extensive refactorings that have happened over the past several months.

llvm-svn: 198705

10 years ago[cmake] Write Version.inc at cmake time, not at build time.
Nico Weber [Tue, 7 Jan 2014 20:10:39 +0000 (20:10 +0000)]
[cmake] Write Version.inc at cmake time, not at build time.

In SVN checkouts, clang_revision_tag is rerun on every build, even if nothing
else is dirty.  After this change, Version.inc is only written at cmake time,
so that empty builds run 0 build steps (like r191784 apparently did for git).

llvm-svn: 198704

10 years agoRun llvm/utils/sort_includes.py over the Clang tools code. This doesn't
Chandler Carruth [Tue, 7 Jan 2014 20:05:01 +0000 (20:05 +0000)]
Run llvm/utils/sort_includes.py over the Clang tools code. This doesn't
always produce as pretty of results as it does in LLVM and Clang, but
I don't mind and the value of having a single canonical ordering is very
high IMO.

Let me know if you spot really serious problems here.

llvm-svn: 198703

10 years agoFix uninitialized variable warning in DataLayout.
Cameron McInally [Tue, 7 Jan 2014 19:51:38 +0000 (19:51 +0000)]
Fix uninitialized variable warning in DataLayout.

llvm-svn: 198702

10 years agoDon't assert with private type info variables.
Rafael Espindola [Tue, 7 Jan 2014 19:38:47 +0000 (19:38 +0000)]
Don't assert with private type info variables.

With the gnu objc runtime private strings are used. Since we only need to
produce a unique label, the fix is to just drop the asserts.

llvm-svn: 198701

10 years agoEmit arange padding with a single directive.
Benjamin Kramer [Tue, 7 Jan 2014 19:28:14 +0000 (19:28 +0000)]
Emit arange padding with a single directive.

llvm-svn: 198700

10 years agoDebug info: Implement a cleaner version of r198461. For symmetry with
Adrian Prantl [Tue, 7 Jan 2014 19:24:24 +0000 (19:24 +0000)]
Debug info: Implement a cleaner version of r198461. For symmetry with
C and C++ don't emit an extra lexical scope for the compound statement
that is the body of an Objective-C method.

rdar://problem/15010825

llvm-svn: 198699

10 years agoReplace casts of __impl_ with the correct reinterpret_cast of the
Joerg Sonnenberger [Tue, 7 Jan 2014 19:21:13 +0000 (19:21 +0000)]
Replace casts of __impl_ with the correct reinterpret_cast of the
address. Restores the assembly of before r198504.

llvm-svn: 198698