platform/upstream/llvm.git
10 years agoUpdate MachOObjectFile::getSymbolAddress so it returns UnknownAddressOrSize
Kevin Enderby [Tue, 20 May 2014 23:04:47 +0000 (23:04 +0000)]
Update MachOObjectFile::getSymbolAddress so it returns UnknownAddressOrSize
for undefined symbols, so it matches what COFFObjectFile::getSymbolAddress
does.  This allows llvm-nm to print spaces instead of 0’s for the value
of undefined symbols in Mach-O files.

To make this change other uses of MachOObjectFile::getSymbolAddress
are updated to handle when the Value is returned as UnknownAddressOrSize.
Which is needed to keep two of the ExecutionEngine tests working for example.

llvm-svn: 209253

10 years agoFix test added in r209242: llc shouldn't create files in source tree
Alexey Samsonov [Tue, 20 May 2014 22:40:31 +0000 (22:40 +0000)]
Fix test added in r209242: llc shouldn't create files in source tree

llvm-svn: 209252

10 years agoRevert "DebugInfo: Assume all subprogram DIEs have been created before any abstract...
David Blaikie [Tue, 20 May 2014 22:33:09 +0000 (22:33 +0000)]
Revert "DebugInfo: Assume all subprogram DIEs have been created before any abstract subprograms are constructed."

This reverts commit r209178.

This seems to be asserting in an LTO build on some internal Apple
buildbots. No upstream reproduction (and I don't have an LLVM-aware gold
built right now to reproduce it personally) but it's a small patch & the
failure's semi-plausible so I'm going to revert first while I try to
reproduce this.

llvm-svn: 209251

10 years agoVirtualFileSystem: Fix false positives in YAMLVFSWriter::containedIn
Justin Bogner [Tue, 20 May 2014 22:12:58 +0000 (22:12 +0000)]
VirtualFileSystem: Fix false positives in YAMLVFSWriter::containedIn

Checking if a path starts with another path isn't sufficient for
determining if one is contained within the heirarchy of the other.
We need to ensure that the substring ends at a directory boundary.

llvm-svn: 209250

10 years ago[ubsan] fix vptr test on ARM
Greg Fitzgerald [Tue, 20 May 2014 22:07:58 +0000 (22:07 +0000)]
[ubsan] fix vptr test on ARM

Differential Revision: http://reviews.llvm.org/D3751

llvm-svn: 209249

10 years agoRun common and profile tests in cross-compiled builds.
Greg Fitzgerald [Tue, 20 May 2014 22:04:27 +0000 (22:04 +0000)]
Run common and profile tests in cross-compiled builds.

Route target_cflags to common and profile tests

llvm-svn: 209248

10 years agoReduce string duplication
Alp Toker [Tue, 20 May 2014 22:03:47 +0000 (22:03 +0000)]
Reduce string duplication

If we're so keen on saving a dynamic allocation to add the trailing space, we
might as well do it in style.

llvm-svn: 209247

10 years agoRAV reunification: merge Lambda body visitation to DRAV
Alp Toker [Tue, 20 May 2014 22:03:39 +0000 (22:03 +0000)]
RAV reunification: merge Lambda body visitation to DRAV

llvm-svn: 209246

10 years agoRAV reunification: merge DISPATCH_STMT() macro back to standard RAV
Alp Toker [Tue, 20 May 2014 22:03:27 +0000 (22:03 +0000)]
RAV reunification: merge DISPATCH_STMT() macro back to standard RAV

Also add the missing undef in both files.

llvm-svn: 209245

10 years agoRAV reunification: merge r190728
Alp Toker [Tue, 20 May 2014 22:03:18 +0000 (22:03 +0000)]
RAV reunification: merge r190728

llvm-svn: 209244

10 years ago[PECOFF] Do not use anonymous namespace in a header.
Rui Ueyama [Tue, 20 May 2014 21:58:34 +0000 (21:58 +0000)]
[PECOFF] Do not use anonymous namespace in a header.

llvm-svn: 209243

10 years ago[ARM64] PR19792: Fix cycle in DAG after performPostLD1Combine
Adam Nemet [Tue, 20 May 2014 21:47:07 +0000 (21:47 +0000)]
[ARM64] PR19792: Fix cycle in DAG after performPostLD1Combine

Povray and dealII currently assert with "Overran sorted position" in
AssignTopologicalOrder.  The problem is that performPostLD1Combine can
introduce cycles.

Consider:

(insert_vector_elt (INSERT_SUBREG undef,
                                  (load (add %vreg0, Constant<8>), undef),  <= A
                                  TargetConstant<2>),
                   (load %vreg0, undef),                                    <= B
                   Constant<1>)

This is turned into a LD1LANEpost node.  However the address in A is not a
valid user of the post-incremented address of B in LD1LANEpost.

llvm-svn: 209242

10 years agoVirtualFileSystem: Add YAMLVFSWriter to generate VFS mapping files
Justin Bogner [Tue, 20 May 2014 21:43:27 +0000 (21:43 +0000)]
VirtualFileSystem: Add YAMLVFSWriter to generate VFS mapping files

This moves the logic to write a JSON VFS mapping from the C api into
VirtualFileSystem, so that we can use it internally.

No functional change.

llvm-svn: 209241

10 years agoUnbreak the sanitizer buildbots after r209226 due to SROA issue described in http...
David Blaikie [Tue, 20 May 2014 21:40:13 +0000 (21:40 +0000)]
Unbreak the sanitizer buildbots after r209226 due to SROA issue described in reviews.llvm.org/D3714

Undecided whether this should include a test case - SROA produces bad
dbg.value metadata describing a value for a reference that is actually
the value of the thing the reference refers to. For now, loosening the
assert lets this not assert, but it's still bogus/wrong output...

If someone wants to tell me to add a test, I'm willing/able, just
undecided. Hopefully we'll get SROA fixed soon & we can tighten up this
assertion again.

llvm-svn: 209240

10 years agoUpdate for llvm change to avoid having global flag setting in TargetMachine.
Eric Christopher [Tue, 20 May 2014 21:25:41 +0000 (21:25 +0000)]
Update for llvm change to avoid having global flag setting in TargetMachine.

llvm-svn: 209239

10 years agoMove the function and data section flags into the options struct and
Eric Christopher [Tue, 20 May 2014 21:25:34 +0000 (21:25 +0000)]
Move the function and data section flags into the options struct and
make the functions to set them non-static.
Move and rename the llvm specific backend options to avoid conflicting
with the clang option.

Paired with a backend commit to update.

llvm-svn: 209238

10 years agoUpdate .arcconfig to point to reviews.llvm.org
Reid Kleckner [Tue, 20 May 2014 21:24:59 +0000 (21:24 +0000)]
Update .arcconfig to point to reviews.llvm.org

Mostly a test review and commit.

Reviewers: tfiala

Differential Revision: http://reviews.llvm.org/D3834

llvm-svn: 209237

10 years agoRevert r209235 as it broke two tests:
Kevin Enderby [Tue, 20 May 2014 21:10:15 +0000 (21:10 +0000)]
Revert r209235 as it broke two tests:
Failing Tests (2):
    LLVM :: ExecutionEngine/MCJIT/stubs-sm-pic.ll
    LLVM :: ExecutionEngine/MCJIT/stubs.ll

llvm-svn: 209236

10 years agoUpdate MachOObjectFile::getSymbolAddress so it returns UnknownAddressOrSize
Kevin Enderby [Tue, 20 May 2014 20:32:18 +0000 (20:32 +0000)]
Update MachOObjectFile::getSymbolAddress so it returns UnknownAddressOrSize
for undefined symbols.  Allowing llvm-nm to print spaces instead of 0’s for
the value of undefined symbols in Mach-O files.

llvm-svn: 209235

10 years agoChange gdb remote test support multi-request format.
Todd Fiala [Tue, 20 May 2014 20:02:08 +0000 (20:02 +0000)]
Change gdb remote test support multi-request format.

The multi request-response test infrastructure support was adding
the optional request suffix iteration index as decimal but it needed
to be hex per spec.  This change fixes that.

llvm-svn: 209234

10 years agoCleaning up some range-based for loops so that the automatic type deduction is more...
Aaron Ballman [Tue, 20 May 2014 19:47:14 +0000 (19:47 +0000)]
Cleaning up some range-based for loops so that the automatic type deduction is more explicit about pointers and const. Did some minor drive-by const correctness fixes and identifier updates as well. No functional changes.

llvm-svn: 209233

10 years agoRevert r209231, "Update AttributeReference.rst."
Peter Collingbourne [Tue, 20 May 2014 19:38:07 +0000 (19:38 +0000)]
Revert r209231, "Update AttributeReference.rst."

According to Aaron, this is being generated on the server now.

llvm-svn: 209232

10 years agoUpdate AttributeReference.rst.
Peter Collingbourne [Tue, 20 May 2014 19:27:39 +0000 (19:27 +0000)]
Update AttributeReference.rst.

llvm-svn: 209231

10 years ago[LSR] Canonicalize reg1 + ... + regN into reg1 + ... + 1*regN.
Quentin Colombet [Tue, 20 May 2014 19:25:04 +0000 (19:25 +0000)]
[LSR] Canonicalize reg1 + ... + regN into reg1 + ... + 1*regN.

This commit introduces a canonical representation for the formulae.
Basically, as soon as a formula has more that one base register, the scaled
register field is used for one of them. The register put into the scaled
register is preferably a loop variant.
The commit refactors how the formulae are built in order to produce such
representation.
This yields a more accurate, but still perfectible, cost model.

<rdar://problem/16731508>

llvm-svn: 209230

10 years agoFix testcase from r209228
Duncan P. N. Exon Smith [Tue, 20 May 2014 19:20:23 +0000 (19:20 +0000)]
Fix testcase from r209228

llvm-svn: 209229

10 years agoGlobalValue: Testcase for hidden visibility and local linkage
Duncan P. N. Exon Smith [Tue, 20 May 2014 19:04:31 +0000 (19:04 +0000)]
GlobalValue: Testcase for hidden visibility and local linkage

This is a testcase for r209227, a change in LLVM that automatically sets
visibility to default when the linkage is changed to local (rather than
asserting).

What this testcase triggers is hard to reproduce otherwise:  the
`GlobalValue` is created (with non-local linkage), the visibility is set
to hidden, and then the linkage is set to local.

PR19760

llvm-svn: 209228

10 years agoGlobalValue: Automatically reset visibility when setting local linkage
Duncan P. N. Exon Smith [Tue, 20 May 2014 19:00:58 +0000 (19:00 +0000)]
GlobalValue: Automatically reset visibility when setting local linkage

r208264 started asserting in `setLinkage()` and `setVisibility()` that
visibility and linkage are compatible.  There are a few places in clang
where visibility is set first, and then linkage later, so the assert
fires.  In `setLinkage()`, it's clear what the visibility *should* be,
so rather than updating all the call sites just automatically fix the
visibility.

The testcase for this is for *clang*, so it'll follow separately in cfe.

PR19760

llvm-svn: 209227

10 years agoFix test breakage introduced in r209223.
David Blaikie [Tue, 20 May 2014 18:36:35 +0000 (18:36 +0000)]
Fix test breakage introduced in r209223.

Oops, broke the broken enum constants again.

llvm-svn: 209226

10 years agoRewrite calculateDbgValueHistory to make it (hopefully) more transparent.
Alexey Samsonov [Tue, 20 May 2014 18:34:54 +0000 (18:34 +0000)]
Rewrite calculateDbgValueHistory to make it (hopefully) more transparent.

This change preserves the original algorithm of generating history
for user variables, but makes it more clear.

High-level description of algorithm:
Scan all the machine basic blocks and machine instructions in the order
they are emitted to the object file. Do the following:
1) If we see a DBG_VALUE instruction, add it to the history of the
corresponding user variable. Keep track of all user variables, whose
locations are described by a register.
2) If we see a regular instruction, look at all the registers it clobbers,
and terminate the location range for all variables described by these registers.
3) At the end of the basic block, terminate location ranges for all
user variables described by some register.

Although this change shouldn't be user-visible (the contents of .debug_loc section
should be the same), it changes some internal assumptions about the set
of instructions used to track the variable locations. Watching the bots.

llvm-svn: 209225

10 years agoFixed spelling.
Yaron Keren [Tue, 20 May 2014 18:23:05 +0000 (18:23 +0000)]
Fixed spelling.

llvm-svn: 209224

10 years agoPR19767: DebugInfo emission of pointer constants.
David Blaikie [Tue, 20 May 2014 18:21:51 +0000 (18:21 +0000)]
PR19767: DebugInfo emission of pointer constants.

In refactoring DwarfUnit::isUnsignedDIType I restricted it to only work
on values with signedness (unsigned or signed), asserting on anything
else (which did uncover some bugs). But it turns out that we do need to
emit constants of signless data, such as pointer constants - only null
pointer constants are known to need this so far, but it's conceivable
that there might be non-null pointer constants at some point (hardcoded
address offsets for device drivers?).

This patch just uses 'unsigned' for signless data such as pointer
constants. Arguably we could use signless representations
(DW_FORM_dataN) instead, allowing a trinary result from isUnsignedDIType
(signed, unsigned, signless), but this seems reasonable for now.

llvm-svn: 209223

10 years agoAvoids DCE on write_register
Renato Golin [Tue, 20 May 2014 17:40:03 +0000 (17:40 +0000)]
Avoids DCE on write_register

llvm-svn: 209222

10 years agoAdd a check for tautological bitwise comparisons to -Wtautological-compare.
Jordan Rose [Tue, 20 May 2014 17:31:11 +0000 (17:31 +0000)]
Add a check for tautological bitwise comparisons to -Wtautological-compare.

This catches issues like:

if ((x & 8) == 4) { ... }
if ((x | 4) != 3) { ... }

Patch by Anders Rönnholm!

llvm-svn: 209221

10 years agoMake the wording for the flatten docs a little clearer.
Peter Collingbourne [Tue, 20 May 2014 17:30:03 +0000 (17:30 +0000)]
Make the wording for the flatten docs a little clearer.

llvm-svn: 209220

10 years ago[PowerPC] PR19796: Also match ISD::TargetConstant in isIntS16Immediate
Adam Nemet [Tue, 20 May 2014 17:20:34 +0000 (17:20 +0000)]
[PowerPC] PR19796: Also match ISD::TargetConstant in isIntS16Immediate

The SplitIndexingFromLoad changes exposed a latent isel bug in the PowerPC64
backend.  We matched an immediate offset with STWX8 even though it only
supports register offset.

The culprit is the complex-pattern predicate, SelectAddrIdx, which decides
that if the offset is not ISD::Constant it must be a register.

Many thanks to Bill Schmidt for testing this.

llvm-svn: 209219

10 years agoFix testcase for case.
Eric Christopher [Tue, 20 May 2014 17:15:31 +0000 (17:15 +0000)]
Fix testcase for case.

llvm-svn: 209218

10 years agoImplement the flatten attribute.
Peter Collingbourne [Tue, 20 May 2014 17:12:51 +0000 (17:12 +0000)]
Implement the flatten attribute.

This is a GNU attribute that causes calls within the attributed function
to be inlined where possible. It is implemented by giving such calls the
alwaysinline attribute.

Differential Revision: http://reviews.llvm.org/D3816

llvm-svn: 209217

10 years agoClean up language and grammar.
Eric Christopher [Tue, 20 May 2014 17:11:11 +0000 (17:11 +0000)]
Clean up language and grammar.

Based on a patch by jfcaron3@gmail.com!
PR19806

llvm-svn: 209216

10 years agoClean up language and grammar.
Eric Christopher [Tue, 20 May 2014 17:10:39 +0000 (17:10 +0000)]
Clean up language and grammar.

Based on a patch by jfcaron3@gmail.com!
PR19806

llvm-svn: 209215

10 years agoGo via uintptr_t when casting away constness, otherwise GCC will warn
Joerg Sonnenberger [Tue, 20 May 2014 16:37:07 +0000 (16:37 +0000)]
Go via uintptr_t when casting away constness, otherwise GCC will warn
when using -Wcast-qual.

llvm-svn: 209214

10 years agoFix typos
Alp Toker [Tue, 20 May 2014 16:17:53 +0000 (16:17 +0000)]
Fix typos

llvm-svn: 209213

10 years agoclang/test/SemaCXX/err_init_conversion_failed.cpp: Tweak for i686-msvc.
NAKAMURA Takumi [Tue, 20 May 2014 15:44:42 +0000 (15:44 +0000)]
clang/test/SemaCXX/err_init_conversion_failed.cpp: Tweak for i686-msvc.

For targeting i686-msvc, declarations are seen as thiscall like;

  void (template_test::S::*)(const int &) __attribute__((thiscall))
  void (template_test::S::*)(int) __attribute__((thiscall))

It didn't affect x86_64-msvc.

llvm-svn: 209212

10 years agoTemporarily revert: r209129 - [mips][mips64r6] Sorted *_ENC, *_DESC classes and tests
Daniel Sanders [Tue, 20 May 2014 14:46:24 +0000 (14:46 +0000)]
Temporarily revert: r209129 - [mips][mips64r6] Sorted *_ENC, *_DESC classes and tests

After discussion with Zoran, we have decided to temporarily revert this commit.
It's causing some difficult to resolve conflicts and we are under time pressure
to deliver an initial MIPS64r6 compiler.

We will re-apply an equivalent patch once the time pressure has passed.

llvm-svn: 209211

10 years ago[ASan Win] Simplify and improve the way we forward ASan interface calls from DLLs
Timur Iskhodzhanov [Tue, 20 May 2014 14:26:19 +0000 (14:26 +0000)]
[ASan Win] Simplify and improve the way we forward ASan interface calls from DLLs

Reviewed at http://reviews.llvm.org/D3848

llvm-svn: 209210

10 years agoNone of these attributes require FunctionTemplate to be explicitly listed as part...
Aaron Ballman [Tue, 20 May 2014 14:10:53 +0000 (14:10 +0000)]
None of these attributes require FunctionTemplate to be explicitly listed as part of their subject definition. FunctionTemplateDecls are not what the attribute appertains to in the first place -- it attaches to the underlying FunctionDecl.

The attribute emitter was using FunctionTemplate to map the diagnostic to "functions or methods", but that isn't a particularly clear diagnostic in these cases anyway (since they do not apply to ObjC methods). Updated the attribute emitter to remove custom logic for FunctionTemplateDecl, and updated the test cases for the change in diagnostic wording.

llvm-svn: 209209

10 years ago[ARM64] Port basic-a64-diagnostics.s over to ARM64
Bradley Smith [Tue, 20 May 2014 13:33:41 +0000 (13:33 +0000)]
[ARM64] Port basic-a64-diagnostics.s over to ARM64

llvm-svn: 209207

10 years agoTableGen: permit non-leaf ComplexPattern uses
Tim Northover [Tue, 20 May 2014 11:52:46 +0000 (11:52 +0000)]
TableGen: permit non-leaf ComplexPattern uses

This allows the results of a ComplexPattern check to be distributed to separate
named Operands, instead of the current system where all results must apply (and
match perfectly) with a single Operand.

For example, if "some_addrmode" is a ComplexPattern producing two results, you
can write:

   def : Pat<(load (some_addrmode GPR64:$base, imm:$offset)),
             (INST GPR64:$base, imm:$offset)>;

This should allow neater instruction definitions in TableGen that don't put all
possible aspects of addressing into a single operand, but are still usable with
relatively simple C++ CodeGen idioms.

llvm-svn: 209206

10 years agoclang-format: [JS] Understand top-level function literals properly.
Daniel Jasper [Tue, 20 May 2014 11:14:57 +0000 (11:14 +0000)]
clang-format: [JS] Understand top-level function literals properly.

llvm-svn: 209205

10 years agoAdd parentheses to suppress the gcc warning '-Wparentheses'.
Simon Atanasyan [Tue, 20 May 2014 10:23:04 +0000 (10:23 +0000)]
Add parentheses to suppress the gcc warning '-Wparentheses'.

No functional changes.

llvm-svn: 209203

10 years agoLegalizer: Make bswap promotion safe for vectors.
Benjamin Kramer [Tue, 20 May 2014 09:42:31 +0000 (09:42 +0000)]
Legalizer: Make bswap promotion safe for vectors.

llvm-svn: 209202

10 years ago[Mips] Add more relocation types and MIPS specific e_flags constants.
Simon Atanasyan [Tue, 20 May 2014 09:27:49 +0000 (09:27 +0000)]
[Mips] Add more relocation types and MIPS specific e_flags constants.

llvm-svn: 209201

10 years agoARMEB: Additional test files for ARM fixups
Christian Pirker [Tue, 20 May 2014 09:24:37 +0000 (09:24 +0000)]
ARMEB: Additional test files for ARM fixups

llvm-svn: 209200

10 years agoTableGen: convert InstAlias's Emit bit to an int.
Tim Northover [Tue, 20 May 2014 09:17:16 +0000 (09:17 +0000)]
TableGen: convert InstAlias's Emit bit to an int.

When multiple aliases overlap, the correct string to print can often be
determined purely by considering the InstAlias declarations in some particular
order. This allows the user to specify that order manually when desired,
without resorting to hacking around with the default lexicographical order on
Record instantiation, which is error-prone and ugly.

I was also mistaken about "add w2, w3, w4" being the same as "add w2, w3, w4,
uxtw". That's only true if Rn is the stack pointer.

llvm-svn: 209199

10 years ago[X86] Tune LEA usage for Silvermont
Alexey Volkov [Tue, 20 May 2014 08:55:50 +0000 (08:55 +0000)]
[X86] Tune LEA usage for Silvermont

According to Intel Software Optimization Manual on Silvermont in some cases LEA
is better to be replaced with ADD instructions:
"The rule of thumb for ADDs and LEAs is that it is justified to use LEA
with a valid index and/or displacement for non-destructive destination purposes
(especially useful for stack offset cases), or to use a SCALE.
Otherwise, ADD(s) are preferable."

Differential Revision: http://reviews.llvm.org/D3826

llvm-svn: 209198

10 years ago[LV][REFACTOR] One more tiny fix for printing debug locations in loop vectorizer...
Zinovy Nis [Tue, 20 May 2014 08:26:20 +0000 (08:26 +0000)]
[LV][REFACTOR] One more tiny fix for printing debug locations in loop vectorizer. Now consistent with the remarks emitter.

Differential Revision: http://reviews.llvm.org/D3821

llvm-svn: 209197

10 years agoXCore target: sort typestring enum fields alphabetically
Robert Lytton [Tue, 20 May 2014 07:19:33 +0000 (07:19 +0000)]
XCore target: sort typestring enum fields alphabetically

llvm-svn: 209196

10 years agoSpeculative fix for Windows buildbot after r209138
Ben Langmuir [Tue, 20 May 2014 05:55:04 +0000 (05:55 +0000)]
Speculative fix for Windows buildbot after r209138

It appears that Windows doesn't like renaming over open files, which we
do in clearOutputFiles. The file being compiled should be safe to
removed, but this isn't very satisfying - we don't want to manually
manage the lifetime of files we cannot prove have no references.

llvm-svn: 209195

10 years agollvm-readobj: use range-based for loop
Saleem Abdulrasool [Tue, 20 May 2014 05:18:06 +0000 (05:18 +0000)]
llvm-readobj: use range-based for loop

Convert an additional site to a range based for loop.  NFC.

llvm-svn: 209194

10 years agoTeach isKnownNonNull that a nonnull return is not null. Add a test for this case...
Nick Lewycky [Tue, 20 May 2014 05:13:21 +0000 (05:13 +0000)]
Teach isKnownNonNull that a nonnull return is not null. Add a test for this case as well as the case of a nonnull attribute (already handled but not tested).

llvm-svn: 209193

10 years ago[C++11] Use 'nullptr'. Tooling edition.
Craig Topper [Tue, 20 May 2014 04:51:16 +0000 (04:51 +0000)]
[C++11] Use 'nullptr'. Tooling edition.

llvm-svn: 209192

10 years ago[C++11] Use 'nullptr'. Analysis edition.
Craig Topper [Tue, 20 May 2014 04:30:07 +0000 (04:30 +0000)]
[C++11] Use 'nullptr'. Analysis edition.

llvm-svn: 209191

10 years agoFix diagnostic message for member function pointer mismatches where one of the
Richard Trieu [Tue, 20 May 2014 04:10:24 +0000 (04:10 +0000)]
Fix diagnostic message for member function pointer mismatches where one of the
classes is a template argument.

llvm-svn: 209190

10 years agoDebugInfo: Emit function definitions within their namespace scope.
David Blaikie [Tue, 20 May 2014 03:23:24 +0000 (03:23 +0000)]
DebugInfo: Emit function definitions within their namespace scope.

This workaround (presumably for ancient GDB) doesn't appear to be
required (GDB 7.5 seems to tolerate function definition DIEs in
namespace scope just fine).

llvm-svn: 209189

10 years agoFollow up to 209187, updating a test to use FileCheck. Needed to ignore an extra...
David Blaikie [Tue, 20 May 2014 02:40:34 +0000 (02:40 +0000)]
Follow up to 209187, updating a test to use FileCheck. Needed to ignore an extra DW_TAG_class_type

llvm-svn: 209188

10 years agoUpdate test/DebugInfo/2010-04-06-NestedFnDbgInfo.ll to use FileCheck.
David Blaikie [Tue, 20 May 2014 02:19:15 +0000 (02:19 +0000)]
Update test/DebugInfo/2010-04-06-NestedFnDbgInfo.ll to use FileCheck.

llvm-svn: 209187

10 years agoIgnore void returning overloaded functions fom -Wunused-comparison. PR19791.
Richard Trieu [Tue, 20 May 2014 01:34:43 +0000 (01:34 +0000)]
Ignore void returning overloaded functions fom -Wunused-comparison.  PR19791.

llvm-svn: 209186

10 years agoAdd 'nonnull', a new parameter and return attribute which indicates that the pointer...
Nick Lewycky [Tue, 20 May 2014 01:23:40 +0000 (01:23 +0000)]
Add 'nonnull', a new parameter and return attribute which indicates that the pointer is not null. Instcombine will elide comparisons between these and null. Patch by Luqman Aden!

llvm-svn: 209185

10 years agoARCMT/GC-check-warn-nsalloc.m: don't provide a diag group
Alp Toker [Mon, 19 May 2014 23:48:49 +0000 (23:48 +0000)]
ARCMT/GC-check-warn-nsalloc.m: don't provide a diag group

This diagnostic is now controlled solely by -no-ns-alloc-error thus matching
the original intended behaviour.

llvm-svn: 209184

10 years agoLTO: Add a testcase for linking modules with incompatible Debug Info
Adrian Prantl [Mon, 19 May 2014 23:41:25 +0000 (23:41 +0000)]
LTO: Add a testcase for linking modules with incompatible Debug Info
Versions.
rdar://problem/16926122

llvm-svn: 209183

10 years agoDemote the "Debug Info Version" module flag to llvm::Module::Warning
Adrian Prantl [Mon, 19 May 2014 23:40:06 +0000 (23:40 +0000)]
Demote the "Debug Info Version" module flag to llvm::Module::Warning
behavior on mismatch. The AutoUpgrader will drop incompatible debug info
any way and also emit a warning diagnostic for it.

rdar://problem/16926122

llvm-svn: 209182

10 years agoOK, NAKAMURA Takumi beat me to this change. So backing out my addition of
Kevin Enderby [Mon, 19 May 2014 23:26:51 +0000 (23:26 +0000)]
OK, NAKAMURA Takumi beat me to this change.  So backing out my addition of
llvm-size to the list so it only has one.  Sorry for the noise.

llvm-svn: 209181

10 years agoUsing SmallString and correct addr var
Renato Golin [Mon, 19 May 2014 23:25:25 +0000 (23:25 +0000)]
Using SmallString and correct addr var

llvm-svn: 209180

10 years agoDavid Blaikie pointed out that the test added in r209158 fails in the
Kevin Enderby [Mon, 19 May 2014 23:22:58 +0000 (23:22 +0000)]
David Blaikie pointed out that the test added in r209158 fails in the
check-llvm target under CMake.  And the test/CMakeFiles.txt needs to
have llvm-size added to the list.

llvm-svn: 209179

10 years agoDebugInfo: Assume all subprogram DIEs have been created before any abstract subprogra...
David Blaikie [Mon, 19 May 2014 23:16:19 +0000 (23:16 +0000)]
DebugInfo: Assume all subprogram DIEs have been created before any abstract subprograms are constructed.

Since we visit the whole list of subprograms for each CU at module
start, this is clearly true - don't test for the case, just assert it.

A few old test cases seemed to have incomplete subprogram lists, but any
attempt to reproduce them shows full subprogram lists that even include
entities that have been completely inlined and the out of line
definition removed.

llvm-svn: 209178

10 years ago[CMake] Add llvm-size to check-llvm, to fix build since r209158.
NAKAMURA Takumi [Mon, 19 May 2014 23:12:43 +0000 (23:12 +0000)]
[CMake] Add llvm-size to check-llvm, to fix build since r209158.

llvm-svn: 209177

10 years ago[ARM64] Adds Cortex-A53 scheduling support for vector load/store post.
Chad Rosier [Mon, 19 May 2014 22:59:51 +0000 (22:59 +0000)]
[ARM64] Adds Cortex-A53 scheduling support for vector load/store post.
Patch by Dave Estes<cestes@codeaurora.org>!
PR19761 http://reviews.llvm.org/D3829

llvm-svn: 209176

10 years agoImprove test verification from r209172
Alp Toker [Mon, 19 May 2014 22:56:13 +0000 (22:56 +0000)]
Improve test verification from r209172

llvm-svn: 209175

10 years agoRemove unused method declaration
Matt Arsenault [Mon, 19 May 2014 22:55:35 +0000 (22:55 +0000)]
Remove unused method declaration

llvm-svn: 209174

10 years agoAdd documentation for llvm-dwarfdump tool
Alexey Samsonov [Mon, 19 May 2014 22:53:29 +0000 (22:53 +0000)]
Add documentation for llvm-dwarfdump tool

llvm-svn: 209173

10 years agoGet ARCMT/GC-check-warn-nsalloc.m working
Alp Toker [Mon, 19 May 2014 22:51:11 +0000 (22:51 +0000)]
Get ARCMT/GC-check-warn-nsalloc.m working

The -no-ns-alloc-error migration option now causes the diagnostic to be ignored
completely. If this isn't desired, the error can be downgraded to a warning
using the usual -Wno-error=arcmt-ns-alloc.

Note that we can't use -verify right now on this test because
VerifyDiagnosticConsumer gets confused by multiple SourceManager instances,
which is presumably the reason it was XFAILed in the first place and why the
regression wasn't detected. We'll grep instead for now.

llvm-svn: 209172

10 years agoFix usage of string when StringRef was needed
Renato Golin [Mon, 19 May 2014 22:36:19 +0000 (22:36 +0000)]
Fix usage of string when StringRef was needed

Also adding a variable to the test, so release bots match %1.
This should also calm the gdb buildbot.

.

llvm-svn: 209171

10 years agoAdded gdb remote protocol tests across all qRegisterInfo responses.
Todd Fiala [Mon, 19 May 2014 22:35:24 +0000 (22:35 +0000)]
Added gdb remote protocol tests across all qRegisterInfo responses.

Added support for gdb remote protocol capture/playback where there is a query/multiple-response
pattern.  The new playback entry supports:

- a general query command (key: next_query or query)
- an optional first-query command if that differs from the subsequent queries (key: first_query)
- an end regex for matching anything that would signify that the query/multi-response
  iteration has come to an end.  An assumption is that the end regex is not a content
  package we care about aside from ending the iteration. (key: end_regex)
- an optional 0-based index appended to the end of the query command
  (key: append_iteration_suffix), default: False.
- a key used to collect responses from the query.  Any response from the gdb remote
  that doesn't match the end-of-iteration regex is captured in the playback context
  dictionary using the key specified.  That key will be an array, where each array
  entry is one of the responses from the query/multi-response iteration. (key: save_key).
- a runaway response value, defaulting to 10k, where if this many responses is captured,
  assume the ending condition regex is invalid, or the debug monitor is doing something
  goofy that is going to blow out memory or time. (key: runaway_response_count, default: 10000)

See the lldbgdbserverutils.MultiResponseGdbRemoteEntry class for details.

A MultiResponseGdbRemoteEntry is added by adding an element to the GdbRemoteTestSequence
(via GdbRemoteTestSequence.add_log_lines), using a dictionary, where the "type" key
is set to "multi_response", and the rest of the keys in the dictionary entry are
set to the keys documented for MultiResponseGdbRemoteEntry.

Added helper functions to add the required entry to grab all qRegisterInfo responses.
Added another helper to parse the qRegisterInfo response packets into an array of
dictionaries, where each key:value in the dictionary comes from the register info
response packet.

Added a test to verify that a generic register exists for the program counter,
frame pointer, stack pointer and cpu flags across all register info responses.

Added a test to verify that at least one register set exists across all register
info responses.

llvm-svn: 209170

10 years ago[modules] Fix ODR violation: there's another clang::TypeLocVisitor elsewhere.
Richard Smith [Mon, 19 May 2014 22:34:38 +0000 (22:34 +0000)]
[modules] Fix ODR violation: there's another clang::TypeLocVisitor elsewhere.

llvm-svn: 209169

10 years ago[module] Remove a header that isn't used and doesn't compile.
Richard Smith [Mon, 19 May 2014 22:20:41 +0000 (22:20 +0000)]
[module] Remove a header that isn't used and doesn't compile.

llvm-svn: 209168

10 years agoImplement the no_split_stack attribute.
Peter Collingbourne [Mon, 19 May 2014 22:14:34 +0000 (22:14 +0000)]
Implement the no_split_stack attribute.

This is a GNU attribute that allows split stacks to be turned off on a
per-function basis.

Differential Revision: http://reviews.llvm.org/D3817

llvm-svn: 209167

10 years agoDebugInfo: Don't include DW_AT_inline on each abstract definition multiple times.
David Blaikie [Mon, 19 May 2014 22:07:16 +0000 (22:07 +0000)]
DebugInfo: Don't include DW_AT_inline on each abstract definition multiple times.

When I refactored this in r208636 I accidentally caused this to be added
multiple times to each abstract subprogram (not accounting for the
deduplicating effect of the InlinedSubprogramDIEs set).

This got better in r208798 when the abstract definitions got the
attribute added to them at construction time, but still had the
redundant copies introduced in r208636.

This commit removes those excess DW_AT_inlines and relies solely on the
insertion in r208798.

llvm-svn: 209166

10 years agoDebugInfo: Fix missing inlined_subroutines caused by r208748.
David Blaikie [Mon, 19 May 2014 21:54:31 +0000 (21:54 +0000)]
DebugInfo: Fix missing inlined_subroutines caused by r208748.

The check in DwarfDebug::constructScopeDIE was meant to consider inlined
subroutines as any non-top-level scope that was a subprogram. Instead of
checking "not top level scope" it was checking if the /subprogram's/
scope was non-top-level.

Fix this and beef up a test case to demonstrate some of the missing
inlined_subroutines are no longer missing.

In the course of fixing this I also found that r208748 (with this fix)
found one /extra/ inlined_subroutine in concrete_out_of_line.ll due to
two inlined_subroutines having the same inlinedAt location. The previous
implementation was collapsing these into a single inlined subroutine.

I'm not sure what the original code was that created this .ll file so
I'm not sure if this actually happens in practice today. Since we
deliberately include column information to disambiguate two calls on the
same line, that may've addressed this bug in the frontend, but it's good
to know that workaround isn't necessary for this particular case
anymore.

llvm-svn: 209165

10 years agoFix typos.
Eric Christopher [Mon, 19 May 2014 21:18:47 +0000 (21:18 +0000)]
Fix typos.

llvm-svn: 209164

10 years agocmake: Remove -D NDEBUG from CFLAGS as well as CXXFLAGS
Reid Kleckner [Mon, 19 May 2014 21:13:41 +0000 (21:13 +0000)]
cmake: Remove -D NDEBUG from CFLAGS as well as CXXFLAGS

This silences ~7 warnings on .c files in the LLVM build.

llvm-svn: 209163

10 years ago[ConstantHoisting][X86] Change the cost model to never hoist constants for types...
Juergen Ributzka [Mon, 19 May 2014 21:00:53 +0000 (21:00 +0000)]
[ConstantHoisting][X86] Change the cost model to never hoist constants for types larger than i128.

Currently the X86 backend doesn't support types larger than i128 very well. For
example an i192 multiply will assert in codegen when the 2nd argument is a constant and the constant got hoisted.

This fix changes the cost model to never hoist constants for types larger than
i128. Once the codegen issues have been resolved, the cost model can be updated
to allow also larger types.

This is related to <rdar://problem/16954938>

llvm-svn: 209162

10 years agoIf two sibling modules declare the same entity, and we indirectly pull a
Richard Smith [Mon, 19 May 2014 20:59:20 +0000 (20:59 +0000)]
If two sibling modules declare the same entity, and we indirectly pull a
declaration of that entity in from one of those modules, keep track of the fact
that we've not completed the redeclaration chain yet so that we can pull the
remaining declarations in from the other module if they're needed.

llvm-svn: 209161

10 years agoDon't allow two threads to both be in SBDebugger::Create() due to threading issues...
Greg Clayton [Mon, 19 May 2014 20:42:14 +0000 (20:42 +0000)]
Don't allow two threads to both be in SBDebugger::Create() due to threading issues in FormatManager.

<rdar://problem/16937126>

llvm-svn: 209160

10 years ago[X86] Add ISel patterns to improve the selection of TZCNT and LZCNT.
Andrea Di Biagio [Mon, 19 May 2014 20:38:59 +0000 (20:38 +0000)]
[X86] Add ISel patterns to improve the selection of TZCNT and LZCNT.

Instructions TZCNT (requires BMI1) and LZCNT (requires LZCNT), always
provide the operand size as output if the input operand is zero.

We can take advantage of this knowledge during instruction selection
stage in order to simplify a few corner case.

llvm-svn: 209159

10 years agoImplement MachOObjectFile::isSectionData() and MachOObjectFile::isSectionBSS
Kevin Enderby [Mon, 19 May 2014 20:36:02 +0000 (20:36 +0000)]
Implement MachOObjectFile::isSectionData() and MachOObjectFile::isSectionBSS
so that llvm-size will total up all the sections in the Berkeley format.  This
allows for rough categorizations for Mach-O sections.  And allows the total of
llvm-size’s Berkeley and System V formats to be the same.

llvm-svn: 209158

10 years agoAllow dllimport on function definitions when they're template instantiations
Hans Wennborg [Mon, 19 May 2014 20:14:13 +0000 (20:14 +0000)]
Allow dllimport on function definitions when they're template instantiations

llvm-svn: 209157

10 years agoAdded more insertps optimizations
Filipe Cabecinhas [Mon, 19 May 2014 19:45:57 +0000 (19:45 +0000)]
Added more insertps optimizations

Summary:
When inserting an element that's coming from a vector load or a broadcast
of a vector (or scalar) load, combine the load into the insertps
instruction.
Added PerformINSERTPSCombine for the case where we need to fix the load
(load of a vector + insertps with a non-zero CountS).
Added patterns for the broadcasts.

Also added tests for SSE4.1, AVX, and AVX2.

Reviewers: delena, nadav, craig.topper

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D3581

llvm-svn: 209156

10 years agoFix testers by removing dubious testcase for r209154.
Lang Hames [Mon, 19 May 2014 19:38:48 +0000 (19:38 +0000)]
Fix testers by removing dubious testcase for r209154.

It turns out that not all the world is x86-64. Who knew?

I'll get to work on a more appropriate test case for this patch.

llvm-svn: 209155

10 years ago[RuntimeDyld] Fix x86-64 MachO GOT relocation handling.
Lang Hames [Mon, 19 May 2014 19:21:25 +0000 (19:21 +0000)]
[RuntimeDyld] Fix x86-64 MachO GOT relocation handling.

For GOT relocations the addend should modify the offset to the
GOT entry, not the value of the entry itself. Teach RuntimeDyldMachO
to do The Right Thing here.

Fixes <rdar://problem/16961886>.

llvm-svn: 209154

10 years ago[Mips] Show warning if the linker cannot find a pair for a R_MIPS_HI16
Simon Atanasyan [Mon, 19 May 2014 18:48:57 +0000 (18:48 +0000)]
[Mips] Show warning if the linker cannot find a pair for a R_MIPS_HI16
relocation. In fact this case violates ABI but sometimes compilers might
produce such code.

llvm-svn: 209153

10 years agoKill symbolization functionality in llvm-dwarfdump. We have llvm-symbolizer for that.
Alexey Samsonov [Mon, 19 May 2014 18:45:32 +0000 (18:45 +0000)]
Kill symbolization functionality in llvm-dwarfdump. We have llvm-symbolizer for that.

llvm-svn: 209152