]> Gitweb @ Texas Instruments - Open Source Git Repositories - git.TI.com/gitweb - opencl/llvm.git/log
opencl/llvm.git
10 years agoSwitch the type field in DIVariable and DIGlobalVariable over to DITypeRefs.
Adrian Prantl [Tue, 18 Mar 2014 02:34:58 +0000 (02:34 +0000)]
Switch the type field in DIVariable and DIGlobalVariable over to DITypeRefs.
This allows us to catch more opportunities for ODR-based type uniquing
during LTO.
Paired commit with CFE which updates some testcases to verify the new
DIBuilder behavior.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204106 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReplace unnecessary #include directive with forward declarations.
Adrian Prantl [Tue, 18 Mar 2014 02:34:52 +0000 (02:34 +0000)]
Replace unnecessary #include directive with forward declarations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204104 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd explanatory comment.
Eric Christopher [Tue, 18 Mar 2014 02:18:27 +0000 (02:18 +0000)]
Add explanatory comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204103 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoShorten DotDebugLocEntry to just DebugLocEntry and reformat.
Eric Christopher [Tue, 18 Mar 2014 02:18:24 +0000 (02:18 +0000)]
Shorten DotDebugLocEntry to just DebugLocEntry and reformat.
No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204102 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Avoid emitting standard opcode lengths in debug_line.dwo headers where...
David Blaikie [Tue, 18 Mar 2014 02:13:23 +0000 (02:13 +0000)]
DebugInfo: Avoid emitting standard opcode lengths in debug_line.dwo headers where opcodes are never used anyway

Introduce a slightly tighter wrapper around the header structure that
handles this use case. (MCDwarfDwoLineTable)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204101 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Implement debug_line.dwo for file names used in type units during -gsplit...
David Blaikie [Tue, 18 Mar 2014 01:17:26 +0000 (01:17 +0000)]
DebugInfo: Implement debug_line.dwo for file names used in type units during -gsplit-dwarf

This removes an attribute (and more importantly, a relocation) from
skeleton type units and removes some unnecessary file names from the
debug_line section that remains in the .o (and linked executable) file.

There's still a few places we could shave off some more space here:

* use compilation dir of the underlying compilation unit (since all the
  type units share that compilation dir - though this would be more
  complicated in LTO cases where they don't (keep a map of compilation
  dir->line table header?))

* Remove some of the unnecessary header fields from the line table since
  they're not needed in this situation (about 12 bytes per table).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204099 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Flag test as requiring object emission support
David Blaikie [Tue, 18 Mar 2014 00:12:25 +0000 (00:12 +0000)]
DebugInfo: Flag test as requiring object emission support

Cleans up buildbot failures on R600 and similar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204095 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Do not rely on the compilation dir (index 0) for files in line tables...
David Blaikie [Tue, 18 Mar 2014 00:11:48 +0000 (00:11 +0000)]
DebugInfo: Do not rely on the compilation dir (index 0) for files in line tables shared between compilation units

When emitting assembly there's no support for emitting separate line
tables for each compilation unit - so LLVM emits .loc directives
producing a single line table.

Line tables have an implicit directory (index 0) equal to the
compilation directory (DW_AT_comp_dir) of the compilation unit that
references them.

If multiple compilation units (with possibly disparate compilation
directories) reference the same line table, we must avoid relying on
this ambiguous directory.

Achieve this my simply not setting the compilation directory on the line
table when we're in this situation (multiple units while emitting
assembly).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204094 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Move getOrCreateSourceID from DwarfDebug to DwarfCompileUnit
David Blaikie [Mon, 17 Mar 2014 23:53:25 +0000 (23:53 +0000)]
DebugInfo: Move getOrCreateSourceID from DwarfDebug to DwarfCompileUnit

We still do a few lookups into the line table mapping in MCContext that
could be factored out into a single lookup (rather than looking it up
once for the table label, once to set the compilation unit, once for
each time we need a file ID, etc... ) but assembly output complicates
that somewhat as we still need a virtual dispatch back to the
MCAsmStreamer in that case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204092 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Move line table zero-directory-index (compilation dir) handling into MCDwarf
David Blaikie [Mon, 17 Mar 2014 23:29:40 +0000 (23:29 +0000)]
DebugInfo: Move line table zero-directory-index (compilation dir) handling into MCDwarf

Our handling of compilation directory in DwarfDebug was broken
(incorrectly using the 'last' compilation directory (that of the last
CU in the metadata list) for all function emission in any CU). By moving
this handling down into MCDwarf the issue is fixed as the compilation
dir is tracked correctly per line table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204089 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake methods static
Matt Arsenault [Mon, 17 Mar 2014 22:23:09 +0000 (22:23 +0000)]
Make methods static

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204085 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse range metadata instead of introducing selects.
Dan Gohman [Mon, 17 Mar 2014 19:57:04 +0000 (19:57 +0000)]
Use range metadata instead of introducing selects.

When GlobalOpt has determined that a GlobalVariable only ever has two values,
it would convert the GlobalVariable to a boolean, and introduce SelectInsts
at every load, to choose between the two possible values. These SelectInsts
introduce overhead and other unpleasantness.

This patch makes GlobalOpt just add range metadata to loads from such
GlobalVariables instead. This enables the same main optimization (as seen in
test/Transforms/GlobalOpt/integer-bool.ll), without introducing selects.

The main downside is that it doesn't get the memory savings of shrinking such
GlobalVariables, but this is expected to be negligible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204076 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSwitch a number of loops in lib/CodeGen over to range-based for-loops, now that
Owen Anderson [Mon, 17 Mar 2014 19:36:09 +0000 (19:36 +0000)]
Switch a number of loops in lib/CodeGen over to range-based for-loops, now that
the MachineRegisterInfo iterators are compatible with it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204075 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReally REALLY finish adding const to the MachineRegisterInfo iterator range methods.
Owen Anderson [Mon, 17 Mar 2014 19:34:51 +0000 (19:34 +0000)]
Really REALLY finish adding const to the MachineRegisterInfo iterator range methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204074 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMaking a guess to fix the test case with r204056 to get the build bot working.
Kevin Enderby [Mon, 17 Mar 2014 19:00:03 +0000 (19:00 +0000)]
Making a guess to fix the test case with r204056 to get the build bot working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204073 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Match sign_extend_inreg to BFE instructions
Matt Arsenault [Mon, 17 Mar 2014 18:58:11 +0000 (18:58 +0000)]
R600: Match sign_extend_inreg to BFE instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204072 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake DAGCombiner work on vector bitshifts with constant splat vectors.
Matt Arsenault [Mon, 17 Mar 2014 18:58:01 +0000 (18:58 +0000)]
Make DAGCombiner work on vector bitshifts with constant splat vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204071 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd one more const marker that I missed.
Owen Anderson [Mon, 17 Mar 2014 18:16:10 +0000 (18:16 +0000)]
Add one more const marker that I missed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204070 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Use MC line table file entry uniquing for non-asm input as well.
David Blaikie [Mon, 17 Mar 2014 18:13:58 +0000 (18:13 +0000)]
DebugInfo: Use MC line table file entry uniquing for non-asm input as well.

See r204027 for the precursor to this that applied to asm debug info.

This required some non-obvious API changes to handle the case of asm
output (we never go asm->asm so this didn't come up in r204027): the
modification of the file/directory name by MCDwarfLineTableHeader needed
to be reflected in the MCAsmStreamer caller so it could print the
appropriate .file directive, so those StringRef parameters are now
non-const ref (in/out) parameters rather than just const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204069 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMark MachineRegisterInfo's iterator range methods as const.
Owen Anderson [Mon, 17 Mar 2014 18:05:05 +0000 (18:05 +0000)]
Mark MachineRegisterInfo's iterator range methods as const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204067 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd iterator range definitions for the MachineRegisterInfo iterators.
Owen Anderson [Mon, 17 Mar 2014 18:01:41 +0000 (18:01 +0000)]
Add iterator range definitions for the MachineRegisterInfo iterators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204066 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[X86] Fix unused variable warning with NDEBUG from r204058
Adam Nemet [Mon, 17 Mar 2014 17:32:53 +0000 (17:32 +0000)]
[X86] Fix unused variable warning with NDEBUG from r204058

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204063 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMC: change runtime check to an assertion
Saleem Abdulrasool [Mon, 17 Mar 2014 17:13:57 +0000 (17:13 +0000)]
MC: change runtime check to an assertion

It is unclear how it would be possible to get M to be NULL in normal scenarios.
Change this to an assert rather than a runtime check as per dblakie's
suggestion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204060 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM IAS: support .thumb_set
Saleem Abdulrasool [Mon, 17 Mar 2014 17:13:54 +0000 (17:13 +0000)]
ARM IAS: support .thumb_set

This performs the equivalent of a .set directive in that it creates a symbol
which is an alias for another symbol or value which may possibly be yet
undefined.  This directive also has the added property in that it marks the
aliased symbol as being a thumb function entry point, in the same way that the
.thumb_func directive does.

The current implementation fails one test due to an unrelated issue.  Functions
within .thumb sections are not marked as thumb_func.  The result is that
the aliasee function is not valued correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204059 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[VectorLegalizer/X86] Don't unvectorize fp_to_uint for v8f32->v8i16
Adam Nemet [Mon, 17 Mar 2014 17:06:14 +0000 (17:06 +0000)]
[VectorLegalizer/X86] Don't unvectorize fp_to_uint for v8f32->v8i16

Rather than LegalizeAction::Expand, this needs LegalizeAction::Promote to get
promoted to fp_to_sint v8f32->v8i32.  This is a legal operation on AVX.

For that to work properly, we also need to teach the legalizer about the
specific promotion required here.  The default vector promotion uses
bitcasting to a vector type of the same total size.  We want to promote the
vector element type, effectively widening the operation and then truncating
the result.  This is analogous to the current logic of how int_to_fp is
promoted.

The change also factors out some code from the int_to_fp promotion code to
ValueType::widenIntegerVectorElementType.  This is now shared between
int_to_fp and fp_to_int.

There is no longer need for the custom lowering of fp_to_sint f32->v8i16 in
X86.  It can now go through the new target-independent fp_to_*int promotion
logic.

I also checked that no other target uses Promote for these ops yet, so there
shouldn't be any unexpected change in behavior.

Fixes <rdar://problem/16202247>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204058 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix implementation of isInlineConstant() used by the verifier
Tom Stellard [Mon, 17 Mar 2014 17:03:52 +0000 (17:03 +0000)]
R600/SI: Fix implementation of isInlineConstant() used by the verifier

The type of the immediates should not matter as long as the encoding is
equivalent to the encoding of one of the legal inline constants.

Tested-by: Michel Dänzer <michel.daenzer@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204056 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Use correct dest register class for V_READFIRSTLANE_B32
Tom Stellard [Mon, 17 Mar 2014 17:03:51 +0000 (17:03 +0000)]
R600/SI: Use correct dest register class for V_READFIRSTLANE_B32

This instructions writes to an 32-bit SGPR.  This change required adding
the 32-bit VCC_LO and VCC_HI registers, because the full VCC register
is 64 bits.

This fixes verifier errors on several of the indirect addressing piglit
tests.

Tested-by: Michel Dänzer <michel.daenzer@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204055 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Add generic checks to SIInstrInfo::verifyInstruction()
Tom Stellard [Mon, 17 Mar 2014 17:03:49 +0000 (17:03 +0000)]
R600/SI: Add generic checks to SIInstrInfo::verifyInstruction()

Added checks for number of operands and operand register classes.

Tested-by: Michel Dänzer <michel.daenzer@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204054 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConsistent use of the noduplicate attribute.
Eli Bendersky [Mon, 17 Mar 2014 16:19:07 +0000 (16:19 +0000)]
Consistent use of the noduplicate attribute.

The "noduplicate" attribute of call instructions is sometimes queried directly
and sometimes through the cannotDuplicate() predicate. This patch streamlines
all queries to use the cannotDuplicate() predicate. It also adds this predicate
to InvokeInst, to mirror what CallInst has.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204049 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Introduce ObjectFile::symbols() to use range-based loops.
Alexey Samsonov [Mon, 17 Mar 2014 07:28:19 +0000 (07:28 +0000)]
[C++11] Introduce ObjectFile::symbols() to use range-based loops.

Reviewers: rafael

Reviewed By: rafael

CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D3081

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204031 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm/test/MC/MachO/gen-dwarf-cpp.s: Relax an expression to match DOS pat.
NAKAMURA Takumi [Mon, 17 Mar 2014 05:31:54 +0000 (05:31 +0000)]
llvm/test/MC/MachO/gen-dwarf-cpp.s: Relax an expression to match DOS pat.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204030 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConsider the base pointer for setting the symbol type.
Rafael Espindola [Mon, 17 Mar 2014 04:29:51 +0000 (04:29 +0000)]
Consider the base pointer for setting the symbol type.

This is really a consistency fix. Since given

a = b

we propagate the information, we should propagate it too given

a = b + (1 - 1)

Fixes pr19145.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204028 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Improve reuse of file table entries in asm debug info
David Blaikie [Mon, 17 Mar 2014 01:52:11 +0000 (01:52 +0000)]
DebugInfo: Improve reuse of file table entries in asm debug info

The previous deduping strategy was woefully inadequate - it only
considered the most recent file used and avoided emitting a duplicate in
that case - never considering the a/b/a scenario.

It was also lacking when it came to directory paths as the previous
filename would never match the current if the filename had been split
into file and directory components.

This change builds caching functionality into the line table at the
lowest level in an optional form (a file number of 0 indicates that one
should be chosen and returned) and will eventually be reused by the
normal source level debugging DWARF emission.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204027 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTest case
David Blaikie [Mon, 17 Mar 2014 01:52:04 +0000 (01:52 +0000)]
Test case

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204026 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDeleted copy-constructor/copy-assignment broke a buildbot. Removing while I
Lang Hames [Mon, 17 Mar 2014 01:51:51 +0000 (01:51 +0000)]
Deleted copy-constructor/copy-assignment broke a buildbot. Removing while I
investigate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204025 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm-readobj: Print referred symbol name for CLR token definition
Nico Rieck [Mon, 17 Mar 2014 01:46:52 +0000 (01:46 +0000)]
llvm-readobj: Print referred symbol name for CLR token definition

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204024 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm-readobj: Add test for COFF auxiliary symbols as used by C++/CLI
Nico Rieck [Mon, 17 Mar 2014 01:46:28 +0000 (01:46 +0000)]
llvm-readobj: Add test for COFF auxiliary symbols as used by C++/CLI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204023 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRename PBQP RegAllocSolver local variables to conform to LLVM coding style.
Lang Hames [Mon, 17 Mar 2014 01:33:21 +0000 (01:33 +0000)]
Rename PBQP RegAllocSolver local variables to conform to LLVM coding style.

No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204022 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[X86] New and improved VZeroUpperInserter optimization.
Lang Hames [Mon, 17 Mar 2014 01:22:54 +0000 (01:22 +0000)]
[X86] New and improved VZeroUpperInserter optimization.

- Adds support for inserting vzerouppers before tail-calls.
  This is enabled implicitly by having MachineInstr::copyImplicitOps preserve
  regmask operands, which allows VZeroUpperInserter to see where tail-calls use
  vector registers.

- Fixes a bug that caused the previous version of this optimization to miss some
  vzeroupper insertion points in loops. (Loops-with-vector-code that followed
  loops-without-vector-code were mistakenly overlooked by the previous version).

- New algorithm never revisits instructions.

Fixes <rdar://problem/16228798>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204021 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUntabify.
NAKAMURA Takumi [Sun, 16 Mar 2014 23:58:43 +0000 (23:58 +0000)]
Untabify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204020 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd FIXMEs to use DIScopeRef instead of DIScope for LTO type uniqueing.
Manman Ren [Sun, 16 Mar 2014 18:44:23 +0000 (18:44 +0000)]
Add FIXMEs to use DIScopeRef instead of DIScope for LTO type uniqueing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204019 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CMake][cygming] Disable --out-implib from executables.
NAKAMURA Takumi [Sun, 16 Mar 2014 13:51:24 +0000 (13:51 +0000)]
[CMake][cygming] Disable --out-implib from executables.

It doesn't make sense even with --export-all-symbols.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204017 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove named Twine.
David Blaikie [Sun, 16 Mar 2014 01:36:18 +0000 (01:36 +0000)]
Remove named Twine.

While technically correct, we generally disallow any instance of named
Twines due to their subtlety.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204016 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse a fixed subtarget for test so atom scheduling can't change the addresses this...
Benjamin Kramer [Sat, 15 Mar 2014 23:01:29 +0000 (23:01 +0000)]
Use a fixed subtarget for test so atom scheduling can't change the addresses this test relies on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204014 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove some dead assignements found by scan-build
Arnaud A. de Grandmaison [Sat, 15 Mar 2014 22:13:15 +0000 (22:13 +0000)]
Remove some dead assignements found by scan-build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204013 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake some assertions on constant expressions static.
Benjamin Kramer [Sat, 15 Mar 2014 18:47:07 +0000 (18:47 +0000)]
Make some assertions on constant expressions static.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204011 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoPointerIntPair: Avoid an (academic) case of undefined behavior in the DenseMapInfo...
Benjamin Kramer [Sat, 15 Mar 2014 18:10:49 +0000 (18:10 +0000)]
PointerIntPair: Avoid an (academic) case of undefined behavior in the DenseMapInfo specialization.

If we use a pair with an enum type this could create values outside
of the enum range. Avoid it by creating the bit pattern directly.
While there turn a dynamic assert into a static one. No functionality
change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204010 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLSR: Compress a pair (and get rid of the DenseMapInfo for it).
Benjamin Kramer [Sat, 15 Mar 2014 17:17:48 +0000 (17:17 +0000)]
LSR: Compress a pair (and get rid of the DenseMapInfo for it).

Also convert a horrible hash function to use our hashing infrastructure.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204008 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReplace ValueTypes.h with MachineValueType.h if possible.
Patrik Hagglund [Sat, 15 Mar 2014 09:11:41 +0000 (09:11 +0000)]
Replace ValueTypes.h with MachineValueType.h if possible.

Utilize the previous move of MVT to a separate header for all trivial
cases (that don't need any further restructuring).

Reviewed By: Tim Northover

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204003 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSupport: Make error_category's constructor public
Justin Bogner [Sat, 15 Mar 2014 04:05:59 +0000 (04:05 +0000)]
Support: Make error_category's constructor public

Since our error_category is based on the std one, we should have the
same visibility for the constructor.  This also allows us to avoid
using the _do_message implementation detail in our own categories.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203998 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm/test/Transforms/SampleProfile/syntax.ll: Suppress checking the message catalog...
NAKAMURA Takumi [Sat, 15 Mar 2014 02:32:21 +0000 (02:32 +0000)]
llvm/test/Transforms/SampleProfile/syntax.ll: Suppress checking the message catalog in ENOENT. It is locale-dependent on Windows.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203997 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSampleProfile.cpp: Fix take #2. The issue was abuse of StringRef here.
NAKAMURA Takumi [Sat, 15 Mar 2014 01:56:17 +0000 (01:56 +0000)]
SampleProfile.cpp: Fix take #2. The issue was abuse of StringRef here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203996 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSampleProfile.cpp: Quick fix to r203976 about abuse of Twine. The life of Twine was...
NAKAMURA Takumi [Sat, 15 Mar 2014 00:10:12 +0000 (00:10 +0000)]
SampleProfile.cpp: Quick fix to r203976 about abuse of Twine. The life of Twine was too short.

FIXME: DiagnosticInfoSampleProfile should not hold Twine&.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203990 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Remove unnecessary attempt to zext a pointer.
Matt Arsenault [Sat, 15 Mar 2014 00:08:26 +0000 (00:08 +0000)]
R600: Remove unnecessary attempt to zext a pointer.

Private pointers are now always 32-bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203989 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Code cleanup.
Matt Arsenault [Sat, 15 Mar 2014 00:08:22 +0000 (00:08 +0000)]
R600: Code cleanup.

Use sign_extend_inreg and getZeroExtendInReg instead of
using the bit operations they expand into.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203988 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoObject/COFF: change data type of SymbolNumber from int16 to uint16.
Rui Ueyama [Sat, 15 Mar 2014 00:04:08 +0000 (00:04 +0000)]
Object/COFF: change data type of SymbolNumber from int16 to uint16.

Microsoft PE/COFF Spec clearly states that the field is of signed interger
type. However, in reality, it's unsigned. If cl.exe needs to create a large
number of sections for COMDAT sections, it will just create more than 32768
sections. Handling large section number as negative number is not correct.
I think this is a spec bug.

Differential Revision: http://llvm-reviews.chandlerc.com/D3088

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203986 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug info: Unique types before emitting them to DWARF, where applicable.
Adrian Prantl [Fri, 14 Mar 2014 23:08:29 +0000 (23:08 +0000)]
Debug info: Unique types before emitting them to DWARF, where applicable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203983 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug Info: Fix LTO type uniquing for C++ member declarations
Adrian Prantl [Fri, 14 Mar 2014 23:08:25 +0000 (23:08 +0000)]
Debug Info: Fix LTO type uniquing for C++ member declarations
based on the ODR.

This adds an OdrMemberMap to DwarfDebug which is used to unique C++
member function declarations based on the unique identifier of their
containing class and their mangled name.
We can't use the usual DIRef mechanism here because DIScopes are indexed
using their entire MDNode, including decl_file and decl_line, which need
not be unique (see testcase).

Prior to this change multiple redundant member function declarations would
end up in the same uniqued DW_TAG_class_type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203982 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRe-add checks that were in this testcase before it was converted to dwarfdump.
Adrian Prantl [Fri, 14 Mar 2014 23:08:21 +0000 (23:08 +0000)]
Re-add checks that were in this testcase before it was converted to dwarfdump.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203981 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agotypo
Adrian Prantl [Fri, 14 Mar 2014 23:08:17 +0000 (23:08 +0000)]
typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203980 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMC: don't create a backtrace for diagnostics.
Jim Grosbach [Fri, 14 Mar 2014 22:41:58 +0000 (22:41 +0000)]
MC: don't create a backtrace for diagnostics.

For better or worse, this is currently the normal error reporting path
when dealing with backend errors from inline assembly. It's not just
internal compiler issues that come through here, so we shouldn't be
creating a backtrace on this path.

rdar://16329947

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203979 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRe-format SampleProfile.cpp with clang-format. No functional changes.
Diego Novillo [Fri, 14 Mar 2014 22:07:18 +0000 (22:07 +0000)]
Re-format SampleProfile.cpp with clang-format. No functional changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203977 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse DiagnosticInfo facility.
Diego Novillo [Fri, 14 Mar 2014 21:58:59 +0000 (21:58 +0000)]
Use DiagnosticInfo facility.

Summary:
The sample profiler pass emits several error messages. Instead of
just aborting the compiler with report_fatal_error, we can emit
better messages using DiagnosticInfo.

This adds a new sub-class of DiagnosticInfo to handle the sample
profiler.

Reviewers: chandlerc, qcolombet

CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D3086

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203976 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove command line option for CU hashing. This is on by default now.
Eric Christopher [Fri, 14 Mar 2014 21:20:07 +0000 (21:20 +0000)]
Remove command line option for CU hashing. This is on by default now.
Fix up testcases and use of flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203973 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake the arbitrary section name be something mach-o compatible.
Eric Christopher [Fri, 14 Mar 2014 21:16:54 +0000 (21:16 +0000)]
Make the arbitrary section name be something mach-o compatible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203972 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoIf we see that we're emitting code for a function that doesn't have
Eric Christopher [Fri, 14 Mar 2014 20:53:49 +0000 (20:53 +0000)]
If we see that we're emitting code for a function that doesn't have
any lexical scopes then go ahead and turn on DW_AT_ranges for the
compile unit since we would be claiming to describe in the CU
a range for which we don't have information in the CU otherwise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203969 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove the -generate-dwarf-cu-ranges flag.
Eric Christopher [Fri, 14 Mar 2014 20:53:43 +0000 (20:53 +0000)]
Remove the -generate-dwarf-cu-ranges flag.
Rewrite a couple of testcases to cover areas that would be normally
by turning it on into testcases that will follow the logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203968 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMCDwarf: Initialize MCLineTableHeader::Label
David Blaikie [Fri, 14 Mar 2014 20:36:44 +0000 (20:36 +0000)]
MCDwarf: Initialize MCLineTableHeader::Label

This sometimes remains null into MCLineTableHeader::Emit where we
conditionally construct a label if one isn't provided for us. We need it
to remain null (rather than just always constructing the label) so we
can identify unused line tables... which is a bit weird and maybe we can
do away with that logic one day (& on that day we can always construct
the label up-front and just have compilation units query the line table
for its label, etc)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203967 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMCContext: Remove redundant assignment
David Blaikie [Fri, 14 Mar 2014 20:09:26 +0000 (20:09 +0000)]
MCContext: Remove redundant assignment

The member variable is not initialized in the ctor so it's already
empty. No need to empty it again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203963 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCorrectly handle an ELF symbol defined with "a = b + expr".
Rafael Espindola [Fri, 14 Mar 2014 20:09:04 +0000 (20:09 +0000)]
Correctly handle an ELF symbol defined with "a = b + expr".

We were marking the symbol as absolute instead of computing b's offset + the
expression value.

This fixes pr19126.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203962 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unnecessary StringRef::str() call where an implicit conversion works just...
David Blaikie [Fri, 14 Mar 2014 19:53:39 +0000 (19:53 +0000)]
Remove unnecessary StringRef::str() call where an implicit conversion works just fine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203960 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfDebug: Remove some needless recursion.
David Blaikie [Fri, 14 Mar 2014 16:33:32 +0000 (16:33 +0000)]
DwarfDebug: Remove some needless recursion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203946 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agox86: Add missing break to getCallPreservedMask()
Duncan P. N. Exon Smith [Fri, 14 Mar 2014 16:29:21 +0000 (16:29 +0000)]
x86: Add missing break to getCallPreservedMask()

This change brings getCallPreservedMask()'s logic in line with
getCalleeSavedRegs().

While this changes the control flow slightly, the change is not
currently observable.  is64Bit must be false to get to the accidental
fallthrough, but the case that we fall into (coldcc) does nothing unless
is64Bit is true.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203943 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agox86: NFC: Make getCallPreservedMask() more similar to getCalleeSavedRegs()
Duncan P. N. Exon Smith [Fri, 14 Mar 2014 16:09:13 +0000 (16:09 +0000)]
x86: NFC: Make getCallPreservedMask() more similar to getCalleeSavedRegs()

Changing order of checks in getCallPreservedMask() to match
getCalleeSavedRegs() so that the logic is easier to compare.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203939 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Add links to XMOS XCore documentation.
Richard Osborne [Fri, 14 Mar 2014 15:53:50 +0000 (15:53 +0000)]
[docs] Add links to XMOS XCore documentation.

Summary: Add links to XCore ISA and ABI documents.

CC: llvm-commits, rafael
Differential Revision: http://llvm-reviews.chandlerc.com/D2981

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203936 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agox86: getCalleeSavedRegs() would crash on 0 (so don't default to it)
Duncan P. N. Exon Smith [Fri, 14 Mar 2014 15:38:12 +0000 (15:38 +0000)]
x86: getCalleeSavedRegs() would crash on 0 (so don't default to it)

The current logic assumes that MF is not 0.  Assert that it isn't, and
remove the default of 0 from the header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203934 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix a bug introduced during the transition to PathV2.
Rafael Espindola [Fri, 14 Mar 2014 15:13:35 +0000 (15:13 +0000)]
Fix a bug introduced during the transition to PathV2.

sys::fs::createUniqueFile returns an absolute path, so MakeSharedObject does
too and we don't need to add a './' prefix.

Patch by Jon McLachlan.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203931 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDelete unused ObjectFile::{begin,end}_symbols()
Alexey Samsonov [Fri, 14 Mar 2014 14:52:03 +0000 (14:52 +0000)]
Delete unused ObjectFile::{begin,end}_symbols()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203928 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Introduce SectionRef::relocations() to use range-based loops
Alexey Samsonov [Fri, 14 Mar 2014 14:22:49 +0000 (14:22 +0000)]
[C++11] Introduce SectionRef::relocations() to use range-based loops

Reviewers: rafael

Reviewed By: rafael

CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D3077

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203927 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSeparate out MVT in a separate header file: MachineValueType.h
Patrik Hagglund [Fri, 14 Mar 2014 13:41:09 +0000 (13:41 +0000)]
Separate out MVT in a separate header file: MachineValueType.h

The idea behind this split of ValueTypes.h, is to make it easier to
ensure that stuff after type legalization only use MVT (rather than
EVT), by watching include dependencies.

Reviewed By: Tim Northover

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203926 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ppc64] Avoid copy relocs in named rodata sections
Ulrich Weigand [Fri, 14 Mar 2014 12:45:22 +0000 (12:45 +0000)]
[ppc64] Avoid copy relocs in named rodata sections

Commit r181723 introduced code to avoid placing initialized variables
needing relocations into the .rodata section, which avoid copy relocs
that do not work as expected on ppc64 function references.

The same treatment is also needed for *named* .rodata.XXX sections.
This patch changes PPC64LinuxTargetObjectFile::SelectSectionForGlobal
to modify "Kind" *before* calling the default SelectSectionForGlobal
routine, instead of first calling the default routine and then just
checking for the (main) .rodata section afterwards.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203921 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ASan] Fix https://code.google.com/p/address-sanitizer/issues/detail?id=274
Alexander Potapenko [Fri, 14 Mar 2014 10:41:49 +0000 (10:41 +0000)]
[ASan] Fix https://code.google.com/p/address-sanitizer/issues/detail?id=274
by ignoring globals from __TEXT,__cstring,cstring_literals during instrumenation.
Add a regression test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203916 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse ctor instead of initializer list to appease Windows buildbots
Alexey Samsonov [Fri, 14 Mar 2014 10:37:36 +0000 (10:37 +0000)]
Use ctor instead of initializer list to appease Windows buildbots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203915 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse temporary instead of a local variable here
Alexey Samsonov [Fri, 14 Mar 2014 10:20:10 +0000 (10:20 +0000)]
Use temporary instead of a local variable here

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203914 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoGeneralise assembly tests to not rely on anonymous symbol names
Oliver Stannard [Fri, 14 Mar 2014 09:10:26 +0000 (09:10 +0000)]
Generalise assembly tests to not rely on anonymous symbol names

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203909 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAddressSanitizer instrumentation for MOV and MOVAPS.
Evgeniy Stepanov [Fri, 14 Mar 2014 08:58:04 +0000 (08:58 +0000)]
AddressSanitizer instrumentation for MOV and MOVAPS.

This is an initial version of *Sanitizer instrumentation of assembly code.

Patch by Yuri Gorshenin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203908 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMergeFunctions, cmpType: fixed variable names from XXTy1 and XXTy2 to XXTyL and XXTyR.
Stepan Dyatkovskiy [Fri, 14 Mar 2014 08:48:52 +0000 (08:48 +0000)]
MergeFunctions, cmpType: fixed variable names from XXTy1 and XXTy2 to XXTyL and XXTyR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203907 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMergeFunctions, cmpType: Fixed comments wrapping.
Stepan Dyatkovskiy [Fri, 14 Mar 2014 08:17:19 +0000 (08:17 +0000)]
MergeFunctions, cmpType: Fixed comments wrapping.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203905 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[yaml2obj][ELF] Use range-based for loops.
Simon Atanasyan [Fri, 14 Mar 2014 06:53:30 +0000 (06:53 +0000)]
[yaml2obj][ELF] Use range-based for loops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203900 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[yaml2obj][ELF] Refer to a section in the error message by its name not
Simon Atanasyan [Fri, 14 Mar 2014 06:53:25 +0000 (06:53 +0000)]
[yaml2obj][ELF] Refer to a section in the error message by its name not
index.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203899 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[yaml2obj][ELF] Remove unused ELFState class field.
Simon Atanasyan [Fri, 14 Mar 2014 06:53:21 +0000 (06:53 +0000)]
[yaml2obj][ELF] Remove unused ELFState class field.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203898 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[yaml2obj][ELF] Assign name (.shstrtab) to the section holds sections names.
Simon Atanasyan [Fri, 14 Mar 2014 06:53:16 +0000 (06:53 +0000)]
[yaml2obj][ELF] Assign name (.shstrtab) to the section holds sections names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203897 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Mips] Add one more MIPS relocation type constant.
Simon Atanasyan [Fri, 14 Mar 2014 06:53:10 +0000 (06:53 +0000)]
[Mips] Add one more MIPS relocation type constant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203896 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert r203883 (which was more of a bandaid) and fix the real underlying
Owen Anderson [Fri, 14 Mar 2014 05:02:18 +0000 (05:02 +0000)]
Revert r203883 (which was more of a bandaid) and fix the real underlying
issue in that the new MachineRegisterInfo bundle iterators didn't
dereference to the START of the bundle, while the old skipBundle()
method did.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203890 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agostatic link polly into tools
Sebastian Pop [Fri, 14 Mar 2014 04:04:14 +0000 (04:04 +0000)]
static link polly into tools

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203886 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix issue with r203865. The old behaviour would get a MachineOperand then find the...
Pete Cooper [Fri, 14 Mar 2014 02:28:05 +0000 (02:28 +0000)]
Fix issue with r203865.  The old behaviour would get a MachineOperand then find the MI for the bundle the MI was in.  The new behaviour was failing to get the parent bundle and instead just used the MI from the MachineOperand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203883 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReverted r203879.
Galina Kistanova [Fri, 14 Mar 2014 01:56:55 +0000 (01:56 +0000)]
Reverted r203879.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203880 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFixed misuse of isascii. Also fixes mingw32 build, see http://msdn.microsoft.com...
Galina Kistanova [Fri, 14 Mar 2014 01:43:46 +0000 (01:43 +0000)]
Fixed misuse of isascii. Also fixes mingw32 build, see http://msdn.microsoft.com/en-us/library/ms235417.aspx

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203879 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse DW_AT_linkage_name when we're emitting DWARF4 or above.
Eric Christopher [Thu, 13 Mar 2014 23:26:25 +0000 (23:26 +0000)]
Use DW_AT_linkage_name when we're emitting DWARF4 or above.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203867 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove the linker_private and linker_private_weak linkages.
Rafael Espindola [Thu, 13 Mar 2014 23:18:37 +0000 (23:18 +0000)]
Remove the linker_private and linker_private_weak linkages.

These linkages were introduced some time ago, but it was never very
clear what exactly their semantics were or what they should be used
for. Some investigation found these uses:

* utf-16 strings in clang.
* non-unnamed_addr strings produced by the sanitizers.

It turns out they were just working around a more fundamental problem.
For some sections a MachO linker needs a symbol in order to split the
section into atoms, and llvm had no idea that was the case. I fixed
that in r201700 and it is now safe to use the private linkage. When
the object ends up in a section that requires symbols, llvm will use a
'l' prefix instead of a 'L' prefix and things just work.

With that, these linkages were already dead, but there was a potential
future user in the objc metadata information. I am still looking at
CGObjcMac.cpp, but at this point I am convinced that linker_private
and linker_private_weak are not what they need.

The objc uses are currently split in

* Regular symbols (no '\01' prefix). LLVM already directly provides
whatever semantics they need.
* Uses of a private name (start with "\01L" or "\01l") and private
linkage. We can drop the "\01L" and "\01l" prefixes as soon as llvm
agrees with clang on L being ok or not for a given section. I have two
patches in code review for this.
* Uses of private name and weak linkage.

The last case is the one that one could think would fit one of these
linkages. That is not the case. The semantics are

* the linker will merge these symbol by *name*.
* the linker will hide them in the final DSO.

Given that the merging is done by name, any of the private (or
internal) linkages would be a bad match. They allow llvm to rename the
symbols, and that is really not what we want. From the llvm point of
view, these objects should really be (linkonce|weak)(_odr)?.

For now, just keeping the "\01l" prefix is probably the best for these
symbols. If we one day want to have a more direct support in llvm,
IMHO what we should add is not a linkage, it is just a hidden_symbol
attribute. It would be applicable to multiple linkages. For example,
on weak it would produce the current behavior we have for objc
metadata. On internal, it would be equivalent to private (and we
should then remove private).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203866 91177308-0d34-0410-b5e6-96231b3b80d8