platform/upstream/llvm.git
11 years agotsan: change fast state layout in preparation to dynamic traces
Dmitry Vyukov [Wed, 28 Nov 2012 10:49:27 +0000 (10:49 +0000)]
tsan: change fast state layout in preparation to dynamic traces

llvm-svn: 168784

11 years agotsan: move traces from tls into dedicated storage at fixed address
Dmitry Vyukov [Wed, 28 Nov 2012 10:35:31 +0000 (10:35 +0000)]
tsan: move traces from tls into dedicated storage at fixed address
helps to reduce tls size (it's weird to have multi-MB tls)
will help with dynamically adjustable trace size

llvm-svn: 168783

11 years ago[asan] Split AddressSanitizer into two passes (FunctionPass, ModulePass), Clang part.
Kostya Serebryany [Wed, 28 Nov 2012 10:32:11 +0000 (10:32 +0000)]
[asan] Split AddressSanitizer into two passes (FunctionPass, ModulePass), Clang part.

llvm-svn: 168782

11 years ago[asan] Split AddressSanitizer into two passes (FunctionPass, ModulePass), LLVM part...
Kostya Serebryany [Wed, 28 Nov 2012 10:31:36 +0000 (10:31 +0000)]
[asan] Split AddressSanitizer into two passes (FunctionPass, ModulePass), LLVM part. This requires a clang part which will follow.

llvm-svn: 168781

11 years agocmake: Fix installation of include files
Tobias Grosser [Wed, 28 Nov 2012 10:12:21 +0000 (10:12 +0000)]
cmake: Fix installation of include files

Install the configured Config/config.h file, rather than the
Config/config.h.cmake and Config/Config.h.in source files.

Contributed-by: Richard Membarth <richard.membarth@informatik.uni-erlangen.de>
llvm-svn: 168780

11 years agoAdd back support for reading and parsing 'deplibs'.
Bill Wendling [Wed, 28 Nov 2012 08:41:48 +0000 (08:41 +0000)]
Add back support for reading and parsing 'deplibs'.

This is for backwards compatibility for pre-3.x bc files. The code reads the
code, but does nothing with it.

llvm-svn: 168779

11 years agoRemove obsolete comment missed by r162937
Andy Gibbs [Wed, 28 Nov 2012 08:14:36 +0000 (08:14 +0000)]
Remove obsolete comment missed by r162937

llvm-svn: 168778

11 years agotsan: add description of memory layouts in different configs
Dmitry Vyukov [Wed, 28 Nov 2012 07:44:26 +0000 (07:44 +0000)]
tsan: add description of memory layouts in different configs

llvm-svn: 168777

11 years agoTeach Lexer::getSpelling about raw string literals. Specifically, if a raw
Richard Smith [Wed, 28 Nov 2012 07:29:00 +0000 (07:29 +0000)]
Teach Lexer::getSpelling about raw string literals. Specifically, if a raw
string literal needs cleaning (because it contains line-splicing in the
encoding prefix or in the ud-suffix), do not clean the section between the
double-quotes -- that's the "raw" bit!

llvm-svn: 168776

11 years agoStore on the CXXRecordDecl whether the class has, or would have, a copy
Richard Smith [Wed, 28 Nov 2012 06:23:12 +0000 (06:23 +0000)]
Store on the CXXRecordDecl whether the class has, or would have, a copy
constructor/assignment operator with a const-qualified parameter type. The
prior method for determining this incorrectly used overload resolution.

llvm-svn: 168775

11 years agoRemove workaround in RegionStore in r168741 since it is handled more generally by...
Ted Kremenek [Wed, 28 Nov 2012 05:36:28 +0000 (05:36 +0000)]
Remove workaround in RegionStore in r168741 since it is handled more generally by r168757.

llvm-svn: 168774

11 years agomisched: Analysis that partitions the DAG into subtrees.
Andrew Trick [Wed, 28 Nov 2012 05:13:28 +0000 (05:13 +0000)]
misched: Analysis that partitions the DAG into subtrees.

This is a simple, cheap infrastructure for analyzing the shape of a
DAG. It recognizes uniform DAGs that take the shape of bottom-up
subtrees, such as the included matrix multiplication example. This is
useful for heuristics that balance register pressure with ILP. Two
canonical expressions of the heuristic are implemented in scheduling
modes: -misched-ilpmin and -misched-ilpmax.

llvm-svn: 168773

11 years agomisched: rename ScheduleDAGILP to ScheduleDFS to prepare for other heuristics.
Andrew Trick [Wed, 28 Nov 2012 05:13:24 +0000 (05:13 +0000)]
misched: rename ScheduleDAGILP to ScheduleDFS to prepare for other heuristics.

llvm-svn: 168772

11 years agoIntroduce ASTUnresolvedSet, an UnresolvedSet-like class, whose contents are
Argyrios Kyrtzidis [Wed, 28 Nov 2012 03:56:16 +0000 (03:56 +0000)]
Introduce ASTUnresolvedSet, an UnresolvedSet-like class, whose contents are
allocated using the allocator associated with an ASTContext.

Use this inside CXXRecordDecl::DefinitionData instead of an UnresolvedSet to
avoid a potential memory leak.

rdar://12761275

llvm-svn: 168771

11 years agoDon't return a pointer to an UnresolvedSetImpl in the CXXRecordDecl interface,
Argyrios Kyrtzidis [Wed, 28 Nov 2012 03:56:09 +0000 (03:56 +0000)]
Don't return a pointer to an UnresolvedSetImpl in the CXXRecordDecl interface,
expose only the iterators instead.

llvm-svn: 168770

11 years agoC++ core issue 1344, PR10618: promote "addition of default argument makes this
Richard Smith [Wed, 28 Nov 2012 03:45:24 +0000 (03:45 +0000)]
C++ core issue 1344, PR10618: promote "addition of default argument makes this
a special member" diagnostic from warning to error, and fix the cases where it
produced diagnostics with incorrect wording.

We don't support this as an extension, and we ban it even in C++98 mode. This
breaks too much (for instance, the ABI-specified calling convention for a type
can change if it acquires a copy constructor through the addition of a default
argument).

llvm-svn: 168769

11 years agomisched: better alias analysis.
Andrew Trick [Wed, 28 Nov 2012 03:42:49 +0000 (03:42 +0000)]
misched: better alias analysis.

This fixes a hole in the "cheap" alias analysis logic implemented within
the DAG builder itself, regardless of whether proper alias analysis is
enabled. It now handles this pattern produced by LSR+CodeGenPrepare.

%sunkaddr1 = ptrtoint * %obj to i64
%sunkaddr2 = add i64 %sunkaddr1, %lsr.iv
%sunkaddr3 = inttoptr i64 %sunkaddr2 to i32*
store i32 %v, i32* %sunkaddr3

llvm-svn: 168768

11 years agomisched: Debug output fix. Use an always valid iterator.
Andrew Trick [Wed, 28 Nov 2012 03:42:47 +0000 (03:42 +0000)]
misched: Debug output fix. Use an always valid iterator.

llvm-svn: 168767

11 years agoIf Clang is looking for an Objective-C method on
Sean Callanan [Wed, 28 Nov 2012 03:23:20 +0000 (03:23 +0000)]
If Clang is looking for an Objective-C method on
a type, and we find it in the origin for that
type, don't look anywhere else; just report it.

<rdar://problem/12675970>

llvm-svn: 168766

11 years agoBBVectorize: Correctly merge SubclassOptionalData
Hal Finkel [Wed, 28 Nov 2012 03:04:10 +0000 (03:04 +0000)]
BBVectorize: Correctly merge SubclassOptionalData

When two instructions are combined into a vector instruction,
the resulting instruction must have the most-conservative flags.

llvm-svn: 168765

11 years agoAdd brief support for the fission .debug_info.dwo section for
Eric Christopher [Wed, 28 Nov 2012 02:49:38 +0000 (02:49 +0000)]
Add brief support for the fission .debug_info.dwo section for
ELF output.

llvm-svn: 168764

11 years agoAdd comments.
Eric Christopher [Wed, 28 Nov 2012 02:49:34 +0000 (02:49 +0000)]
Add comments.

llvm-svn: 168763

11 years agoRearrange ordering of sections.
Eric Christopher [Wed, 28 Nov 2012 02:49:32 +0000 (02:49 +0000)]
Rearrange ordering of sections.

llvm-svn: 168762

11 years agoMove and comment accessor routines.
Eric Christopher [Wed, 28 Nov 2012 02:49:28 +0000 (02:49 +0000)]
Move and comment accessor routines.

llvm-svn: 168761

11 years agoRemove all references to TargetInstrInfoImpl.
Jakob Stoklund Olesen [Wed, 28 Nov 2012 02:35:17 +0000 (02:35 +0000)]
Remove all references to TargetInstrInfoImpl.

This class has been merged into its super-class TargetInstrInfo.

llvm-svn: 168760

11 years agoMove the guts of TargetInstrInfoImpl into the TargetInstrInfo class.
Jakob Stoklund Olesen [Wed, 28 Nov 2012 02:35:13 +0000 (02:35 +0000)]
Move the guts of TargetInstrInfoImpl into the TargetInstrInfo class.

The *Impl class no longer serves a purpose now that the super-class
implementation is in CodeGen.

llvm-svn: 168759

11 years agoMove Target{Instr,Register}Info.cpp into lib/CodeGen.
Jakob Stoklund Olesen [Wed, 28 Nov 2012 02:35:09 +0000 (02:35 +0000)]
Move Target{Instr,Register}Info.cpp into lib/CodeGen.

The Target library is not allowed to depend on the large CodeGen
library, but the TRI and TII classes provide abstract interfaces that
require both caller and callee to link to CodeGen.

The implementation files for these classes provide default
implementations of some of the hooks. These methods may need to
reference CodeGen, so they belong in that library.

We already have a number of methods implemented in the
TargetInstrInfoImpl sub-class because of that. I will merge that class
into the parent next.

llvm-svn: 168758

11 years agoFix another false positive due to a CXX temporary object appearing in a C initializer.
Ted Kremenek [Wed, 28 Nov 2012 01:49:01 +0000 (01:49 +0000)]
Fix another false positive due to a CXX temporary object appearing in a C initializer.

The stop-gap here is to just drop such objects when processing the InitListExpr.
We still need a better solution.

Fixes <rdar://problem/12755044>.

llvm-svn: 168757

11 years agoobjective-C arc: Underline the selector when issuing
Fariborz Jahanian [Wed, 28 Nov 2012 01:27:44 +0000 (01:27 +0000)]
objective-C arc: Underline the selector when issuing
arc specific diagnostic on the selector. This is objc-arc
part of // rdar://11303469

llvm-svn: 168756

11 years agoMoving SectionMemoryManager to MCJIT to avoid cross dependency between JIT and Runtim...
Andrew Kaylor [Wed, 28 Nov 2012 01:02:06 +0000 (01:02 +0000)]
Moving SectionMemoryManager to MCJIT to avoid cross dependency between JIT and RuntimeDyld

llvm-svn: 168755

11 years ago<rdar://problem/12639603>
Greg Clayton [Wed, 28 Nov 2012 00:44:24 +0000 (00:44 +0000)]
<rdar://problem/12639603>

Simplify the logging on ObjectFile::~ObjectFile() to not access an classes above the object file (like the module) so we don't crash when logging object lifetimes. The log message contains the "this" pointer value which can be matched up with the constructor log.

llvm-svn: 168754

11 years agoSome grammar fixes
Eli Bendersky [Wed, 28 Nov 2012 00:27:25 +0000 (00:27 +0000)]
Some grammar fixes

llvm-svn: 168752

11 years agoRevert r168630, r168631, and r168633 as these are causing nightly test failures.
Chad Rosier [Wed, 28 Nov 2012 00:21:29 +0000 (00:21 +0000)]
Revert r168630, r168631, and r168633 as these are causing nightly test failures.

llvm-svn: 168751

11 years agolit: Bump the version to .3.
Daniel Dunbar [Wed, 28 Nov 2012 00:06:11 +0000 (00:06 +0000)]
lit: Bump the version to .3.

llvm-svn: 168750

11 years agotests/lit: Change test default parameters to assume local build.
Daniel Dunbar [Tue, 27 Nov 2012 23:56:28 +0000 (23:56 +0000)]
tests/lit: Change test default parameters to assume local build.
 - Also, support overriding them with lit parameters.

llvm-svn: 168749

11 years agoFixing a silly typo in the previous patch
Enrico Granata [Tue, 27 Nov 2012 23:50:00 +0000 (23:50 +0000)]
Fixing a silly typo in the previous patch

llvm-svn: 168748

11 years agoThis patch makes medium code model the default for 64-bit PowerPC ELF.
Bill Schmidt [Tue, 27 Nov 2012 23:36:26 +0000 (23:36 +0000)]
This patch makes medium code model the default for 64-bit PowerPC ELF.

When the CodeGenInfo is to be created for the PPC64 target machine,
a default code-model selection is converted to CodeModel::Medium
provided we are not targeting the Darwin OS.  Defaults for Darwin
are unaffected.

llvm-svn: 168747

11 years agoFixed the debugserver Xcode project to allow
Sean Callanan [Tue, 27 Nov 2012 23:34:41 +0000 (23:34 +0000)]
Fixed the debugserver Xcode project to allow
DebugClang builds of LLDB to build a properly
codesigned debugserver.  I did this by adding
a DebugClang configuration to debugserver that's
just a clone of the Debug configuration.

llvm-svn: 168746

11 years agodocs: [CMake] Add Xcode to the list of project formats CMake can generate.
NAKAMURA Takumi [Tue, 27 Nov 2012 23:34:28 +0000 (23:34 +0000)]
docs: [CMake] Add Xcode to the list of project formats CMake can generate.

Suggested by Sean McBride, thanks!

llvm-svn: 168745

11 years agoclang/www/get_started.html: s/cmake/CMake/
NAKAMURA Takumi [Tue, 27 Nov 2012 23:34:08 +0000 (23:34 +0000)]
clang/www/get_started.html: s/cmake/CMake/

Suggested by Sean McBride, thanks!

llvm-svn: 168744

11 years ago<rdar://problem/12754509>
Enrico Granata [Tue, 27 Nov 2012 23:28:32 +0000 (23:28 +0000)]
<rdar://problem/12754509>
Make sure that ValueObjectDynamicValue clears itself when no dynamic type information can be found
This behavior was supposed to be already happening (as per the comment lines)

llvm-svn: 168743

11 years agoProvide stop-gap solution to crash reported in PR 14436.
Ted Kremenek [Tue, 27 Nov 2012 23:05:37 +0000 (23:05 +0000)]
Provide stop-gap solution to crash reported in PR 14436.

This was also covered by <rdar://problem/12753384>.  The static analyzer
evaluates a CXXConstructExpr within an initializer expression and
RegionStore doesn't know how to handle the resulting CXXTempObjectRegion
that gets created.  We need a better solution than just dropping the
value, but we need to better understand how to implement the right
semantics here.

Thanks to Jordan for his help diagnosing the behavior here.

llvm-svn: 168741

11 years agoobjective-C arc: load of a __weak object happens via call to
Fariborz Jahanian [Tue, 27 Nov 2012 23:02:53 +0000 (23:02 +0000)]
objective-C arc: load of a __weak object happens via call to
objc_loadWeak. This retains and autorelease the weakly-refereced
object. This hidden autorelease sometimes makes __weak variable alive even
after the weak reference is erased, because the object is still referenced
by an autorelease pool. This patch overcomes this behavior by loading a
weak object via call to objc_loadWeakRetained(), followng it by objc_release
at appropriate place, thereby removing the hidden autorelease. // rdar://10849570

llvm-svn: 168740

11 years agoFix comment formatting in RuntimeDyld.h
Andrew Kaylor [Tue, 27 Nov 2012 22:53:57 +0000 (22:53 +0000)]
Fix comment formatting in RuntimeDyld.h

llvm-svn: 168739

11 years ago[libclang] Remove WorkingDir field from CIndexer, it has been rendered useless.
Argyrios Kyrtzidis [Tue, 27 Nov 2012 22:49:42 +0000 (22:49 +0000)]
[libclang] Remove WorkingDir field from CIndexer, it has been rendered useless.
No functionality change.

llvm-svn: 168738

11 years agoAttempt to make the comments for dwarf debug look more like
Eric Christopher [Tue, 27 Nov 2012 22:43:45 +0000 (22:43 +0000)]
Attempt to make the comments for dwarf debug look more like
the coding standard would like.

llvm-svn: 168737

11 years agoReapply section moving, make sure string section is output last.
Eric Christopher [Tue, 27 Nov 2012 22:43:42 +0000 (22:43 +0000)]
Reapply section moving, make sure string section is output last.

llvm-svn: 168736

11 years ago[arm fast-isel] Appease the machine verifier by using the proper register
Chad Rosier [Tue, 27 Nov 2012 22:29:43 +0000 (22:29 +0000)]
[arm fast-isel] Appease the machine verifier by using the proper register
classes.  The vast majority of the remaining issues are due to uses of
invalid registers, which are defined by getRegForValue().  Those will be
a little more challenging to cleanup.
rdar://12719844

llvm-svn: 168735

11 years ago<rdar://problem/12636970>
Greg Clayton [Tue, 27 Nov 2012 22:18:23 +0000 (22:18 +0000)]
<rdar://problem/12636970>

Properly detect the if unnamed bitfields are supported by clang if the major calng version is higher than 425.

llvm-svn: 168734

11 years ago[arm fast-isel] Appease the machine verifier by using the proper register
Chad Rosier [Tue, 27 Nov 2012 22:12:11 +0000 (22:12 +0000)]
[arm fast-isel] Appease the machine verifier by using the proper register
classes.
rdar://12719844

llvm-svn: 168733

11 years agoMove PrettyStackTraceParserEntry to ParseAST.cpp
Nico Weber [Tue, 27 Nov 2012 21:57:34 +0000 (21:57 +0000)]
Move PrettyStackTraceParserEntry to ParseAST.cpp

r128056 moved PrettyStackTraceParserEntry construction from Parser.h
to ParseAST.cpp, so there's no need to keep this class in a header.

llvm-svn: 168731

11 years agoRemove unused internal linkage variable.
Richard Smith [Tue, 27 Nov 2012 21:51:36 +0000 (21:51 +0000)]
Remove unused internal linkage variable.

llvm-svn: 168729

11 years ago[arm fast-isel] Appease the machine verifier by using the proper register
Chad Rosier [Tue, 27 Nov 2012 21:46:46 +0000 (21:46 +0000)]
[arm fast-isel] Appease the machine verifier by using the proper register
classes.  Also a bit of cleanup.
rdar://12719844

llvm-svn: 168728

11 years agoMark expected failures on Linux (due to bugzilla #14437)
Daniel Malea [Tue, 27 Nov 2012 21:33:41 +0000 (21:33 +0000)]
Mark expected failures on Linux (due to bugzilla #14437)

llvm-svn: 168727

11 years agoAllow an ASTConsumer to selectively skip function bodies while parsing. Patch
Richard Smith [Tue, 27 Nov 2012 21:31:01 +0000 (21:31 +0000)]
Allow an ASTConsumer to selectively skip function bodies while parsing. Patch
by Olivier Goffart!

llvm-svn: 168726

11 years agoSimplify checking for whether we should implicitly declare special members and
Richard Smith [Tue, 27 Nov 2012 21:20:31 +0000 (21:20 +0000)]
Simplify checking for whether we should implicitly declare special members and
add some assertions. No functionality change.

llvm-svn: 168725

11 years agodo not execute the OpenMP tests when cloog is not found
Sebastian Pop [Tue, 27 Nov 2012 21:15:15 +0000 (21:15 +0000)]
do not execute the OpenMP tests when cloog is not found

llvm-svn: 168724

11 years agoAdd -verify-machineinstrs to these fast-isel test cases.
Chad Rosier [Tue, 27 Nov 2012 20:49:56 +0000 (20:49 +0000)]
Add -verify-machineinstrs to these fast-isel test cases.

llvm-svn: 168723

11 years agoModifying lli to use the SectionMemoryManager.
Andrew Kaylor [Tue, 27 Nov 2012 19:49:00 +0000 (19:49 +0000)]
Modifying lli to use the SectionMemoryManager.

The functionality of SectionMemoryManager is equivalent to the LLIMCJITMemoryManager being replaced except that it allocates memory as RW and later changes it to RX or R as needed.  The page permissions are set in the call to MCJIT::finalizeObject.

llvm-svn: 168722

11 years agoMoving SectionMemoryManager into RuntimeDyld and adding unit tests for it.
Andrew Kaylor [Tue, 27 Nov 2012 19:42:02 +0000 (19:42 +0000)]
Moving SectionMemoryManager into RuntimeDyld and adding unit tests for it.

The SectionMemoryManager now supports (and requires) applying section-specific page permissions.  Clients using this memory manager must call either MCJIT::finalizeObject() or SectionMemoryManager::applyPermissions() before executing JITed code.

See r168718 for changes from the previous implementation.

llvm-svn: 168721

11 years ago<rdar://problem/12759744> Provide physical memory distribution as part of profile...
Han Ming Ong [Tue, 27 Nov 2012 19:21:03 +0000 (19:21 +0000)]
<rdar://problem/12759744> Provide physical memory distribution as part of profile data

Make use of unix system calls to provide physical memory usage profile data.

llvm-svn: 168720

11 years agoModified depends() to recognize that when all levels are "=" and
Preston Briggs [Tue, 27 Nov 2012 19:12:26 +0000 (19:12 +0000)]
Modified depends() to recognize that when all levels are "=" and
there's no possible loo-independent dependence, then there's no
dependence.

Updated all test result appropriately.

llvm-svn: 168719

11 years agoImplementing page permission setting in MCJIT unit test SectionMemoryManager.cpp
Andrew Kaylor [Tue, 27 Nov 2012 19:00:17 +0000 (19:00 +0000)]
Implementing page permission setting in MCJIT unit test SectionMemoryManager.cpp

This commit is primarily here for the revision history.  I'm about to move the SectionMemoryManager into the RuntimeDyld library, but I wanted to check the changes in here so people could see the differences in the updated implementation.

llvm-svn: 168718

11 years agoCSE: allow PerformTrivialCoalescing to check copies across basic block
Manman Ren [Tue, 27 Nov 2012 18:58:41 +0000 (18:58 +0000)]
CSE: allow PerformTrivialCoalescing to check copies across basic block
boundaries.

Given the following case:
BB0
  %vreg1<def> = SUBrr %vreg0, %vreg7
  %vreg2<def> = COPY %vreg7
BB1
  %vreg10<def> = SUBrr %vreg0, %vreg2
We should be able to CSE between SUBrr in BB0 and SUBrr in BB1.

rdar://12462006

llvm-svn: 168717

11 years agoinstcombine: Don't replace all uses for instructions with no uses
Meador Inge [Tue, 27 Nov 2012 18:52:49 +0000 (18:52 +0000)]
instcombine: Don't replace all uses for instructions with no uses

My commit to migrate the printf simplifiers from the simplify-libcalls
in r168604 introduced a regression reported by Duncan [1].  The problem
is that in some cases the library call simplifier can return a new value
that has no uses and the new value's type is different than the old value's
type (which is fine because there are no uses).  The specific case that
triggered the bug looked something like:

   declare void @printf(i8*, ...)
   ...
   call void (i8*, ...)* @printf(i8* %fmt)

Which we want to optimized into:

   call i32 @putchar(i32 104)

However, the code was attempting to replace all uses of the printf with
the putchar and the types differ, hence a crash.  This is fixed by *just*
deleting the original instruction when there are no uses.  The old
simplify-libcalls pass is already doing something similar.

[1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-November/056338.html

llvm-svn: 168716

11 years agoFix type-o.
Howard Hinnant [Tue, 27 Nov 2012 18:52:32 +0000 (18:52 +0000)]
Fix type-o.

llvm-svn: 168715

11 years agomake IslAstInfo::printScop compatible with CloogInfo::printScop
Sebastian Pop [Tue, 27 Nov 2012 18:50:41 +0000 (18:50 +0000)]
make IslAstInfo::printScop compatible with CloogInfo::printScop

llvm-svn: 168714

11 years agoRemove by-chapter breakdown of what is implemented. The chart is now obsolete.
Howard Hinnant [Tue, 27 Nov 2012 18:35:09 +0000 (18:35 +0000)]
Remove by-chapter breakdown of what is implemented.  The chart is now obsolete.

llvm-svn: 168713

11 years agoRemove duplicated #includes.
Jakub Staszak [Tue, 27 Nov 2012 18:27:14 +0000 (18:27 +0000)]
Remove duplicated #includes.

llvm-svn: 168712

11 years agoSCEV: Even if the latch terminator is foldable we can't deduce the result of an unrel...
Benjamin Kramer [Tue, 27 Nov 2012 18:16:32 +0000 (18:16 +0000)]
SCEV: Even if the latch terminator is foldable we can't deduce the result of an unrelated condition with it.

Fixes PR14432.

llvm-svn: 168711

11 years agoX86: do not fold load instructions such as [V]MOVS[S|D] to other instructions
Manman Ren [Tue, 27 Nov 2012 18:09:26 +0000 (18:09 +0000)]
X86: do not fold load instructions such as [V]MOVS[S|D] to other instructions
when the destination register is wider than the memory load.

These load instructions load from m32 or m64 and set the upper bits to zero,
while the folded instructions may accept m128.

rdar://12721174

llvm-svn: 168710

11 years agoTest commit only modifying comments
Pedro Artigas [Tue, 27 Nov 2012 17:39:20 +0000 (17:39 +0000)]
Test commit only modifying comments

llvm-svn: 168709

11 years agoThis patch implements medium code model support for 64-bit PowerPC.
Bill Schmidt [Tue, 27 Nov 2012 17:35:46 +0000 (17:35 +0000)]
This patch implements medium code model support for 64-bit PowerPC.

The default for 64-bit PowerPC is small code model, in which TOC entries
must be addressable using a 16-bit offset from the TOC pointer.  Additionally,
only TOC entries are addressed via the TOC pointer.

With medium code model, TOC entries and data sections can all be addressed
via the TOC pointer using a 32-bit offset.  Cooperation with the linker
allows 16-bit offsets to be used when these are sufficient, reducing the
number of extra instructions that need to be executed.  Medium code model
also does not generate explicit TOC entries in ".section toc" for variables
that are wholly internal to the compilation unit.

Consider a load of an external 4-byte integer.  With small code model, the
compiler generates:

ld 3, .LC1@toc(2)
lwz 4, 0(3)

.section .toc,"aw",@progbits
.LC1:
.tc ei[TC],ei

With medium model, it instead generates:

addis 3, 2, .LC1@toc@ha
ld 3, .LC1@toc@l(3)
lwz 4, 0(3)

.section .toc,"aw",@progbits
.LC1:
.tc ei[TC],ei

Here .LC1@toc@ha is a relocation requesting the upper 16 bits of the
32-bit offset of ei's TOC entry from the TOC base pointer.  Similarly,
.LC1@toc@l is a relocation requesting the lower 16 bits.  Note that if
the linker determines that ei's TOC entry is within a 16-bit offset of
the TOC base pointer, it will replace the "addis" with a "nop", and
replace the "ld" with the identical "ld" instruction from the small
code model example.

Consider next a load of a function-scope static integer.  For small code
model, the compiler generates:

ld 3, .LC1@toc(2)
lwz 4, 0(3)

.section .toc,"aw",@progbits
.LC1:
.tc test_fn_static.si[TC],test_fn_static.si
.type test_fn_static.si,@object
.local test_fn_static.si
.comm test_fn_static.si,4,4

For medium code model, the compiler generates:

addis 3, 2, test_fn_static.si@toc@ha
addi 3, 3, test_fn_static.si@toc@l
lwz 4, 0(3)

.type test_fn_static.si,@object
.local test_fn_static.si
.comm test_fn_static.si,4,4

Again, the linker may replace the "addis" with a "nop", calculating only
a 16-bit offset when this is sufficient.

Note that it would be more efficient for the compiler to generate:

addis 3, 2, test_fn_static.si@toc@ha
        lwz 4, test_fn_static.si@toc@l(3)

The current patch does not perform this optimization yet.  This will be
addressed as a peephole optimization in a later patch.

For the moment, the default code model for 64-bit PowerPC will remain the
small code model.  We plan to eventually change the default to medium code
model, which matches current upstream GCC behavior.  Note that the different
code models are ABI-compatible, so code compiled with different models will
be linked and execute correctly.

I've tested the regression suite and the application/benchmark test suite in
two ways:  Once with the patch as submitted here, and once with additional
logic to force medium code model as the default.  The tests all compile
cleanly, with one exception.  The mandel-2 application test fails due to an
unrelated ABI compatibility with passing complex numbers.  It just so happens
that small code model was incredibly lucky, in that temporary values in
floating-point registers held the expected values needed by the external
library routine that was called incorrectly.  My current thought is to correct
the ABI problems with _Complex before making medium code model the default,
to avoid introducing this "regression."

Here are a few comments on how the patch works, since the selection code
can be difficult to follow:

The existing logic for small code model defines three pseudo-instructions:
LDtoc for most uses, LDtocJTI for jump table addresses, and LDtocCPT for
constant pool addresses.  These are expanded by SelectCodeCommon().  The
pseudo-instruction approach doesn't work for medium code model, because
we need to generate two instructions when we match the same pattern.
Instead, new logic in PPCDAGToDAGISel::Select() intercepts the TOC_ENTRY
node for medium code model, and generates an ADDIStocHA followed by either
a LDtocL or an ADDItocL.  These new node types correspond naturally to
the sequences described above.

The addis/ld sequence is generated for the following cases:
 * Jump table addresses
 * Function addresses
 * External global variables
 * Tentative definitions of global variables (common linkage)

The addis/addi sequence is generated for the following cases:
 * Constant pool entries
 * File-scope static global variables
 * Function-scope static variables

Expanding to the two-instruction sequences at select time exposes the
instructions to subsequent optimization, particularly scheduling.

The rest of the processing occurs at assembly time, in
PPCAsmPrinter::EmitInstruction.  Each of the instructions is converted to
a "real" PowerPC instruction.  When a TOC entry needs to be created, this
is done here in the same manner as for the existing LDtoc, LDtocJTI, and
LDtocCPT pseudo-instructions (I factored out a new routine to handle this).

I had originally thought that if a TOC entry was needed for LDtocL or
ADDItocL, it would already have been generated for the previous ADDIStocHA.
However, at higher optimization levels, the ADDIStocHA may appear in a
different block, which may be assembled textually following the block
containing the LDtocL or ADDItocL.  So it is necessary to include the
possibility of creating a new TOC entry for those two instructions.

Note that for LDtocL, we generate a new form of LD called LDrs.  This
allows specifying the @toc@l relocation for the offset field of the LD
instruction (i.e., the offset is replaced by a SymbolLo relocation).
When the peephole optimization described above is added, we will need
to do similar things for all immediate-form load and store operations.

The seven "mcm-n.ll" test cases are kept separate because otherwise the
intermingling of various TOC entries and so forth makes the tests fragile
and hard to understand.

The above assumes use of an external assembler.  For use of the
integrated assembler, new relocations are added and used by
PPCELFObjectWriter.  Testing is done with "mcm-obj.ll", which tests for
proper generation of the various relocations for the same sequences
tested with the external assembler.

llvm-svn: 168708

11 years agoRemove an extra semicolon.
Chad Rosier [Tue, 27 Nov 2012 17:31:26 +0000 (17:31 +0000)]
Remove an extra semicolon.

llvm-svn: 168707

11 years agoFix examples.
Rafael Espindola [Tue, 27 Nov 2012 16:16:02 +0000 (16:16 +0000)]
Fix examples.

llvm-svn: 168705

11 years agoNever use .lcomm on platforms where it does not accept an alignment
Ulrich Weigand [Tue, 27 Nov 2012 16:11:16 +0000 (16:11 +0000)]
Never use .lcomm on platforms where it does not accept an alignment
argument.  Instead, use a pair of .local and .comm directives.

This avoids spurious differences between binaries built by the
integrated assembler vs. those built by the external assembler,
since the external assembler may impose alignment requirements
on .lcomm symbols where the integrated assembler does not.

llvm-svn: 168704

11 years agoRemove some dead code. CLANG_IS_PRODUCTION is now just a build flag and
Rafael Espindola [Tue, 27 Nov 2012 16:10:37 +0000 (16:10 +0000)]
Remove some dead code. CLANG_IS_PRODUCTION is now just a build flag and
is not used in any #ifdef.

llvm-svn: 168703

11 years agoMove sprintf simplifier tests to test/Transforms/InstCombine
Meador Inge [Tue, 27 Nov 2012 15:35:58 +0000 (15:35 +0000)]
Move sprintf simplifier tests to test/Transforms/InstCombine

The tests from SPrintF.ll should have been migrated to sprintf-1.ll in
r168677, but I forgot to do it.

llvm-svn: 168702

11 years agoAdd -fsanitize=integer for reporting suspicious integer behaviors.
Will Dietz [Tue, 27 Nov 2012 15:01:55 +0000 (15:01 +0000)]
Add -fsanitize=integer for reporting suspicious integer behaviors.

Introduces new sanitizer "unsigned-integer-overflow".

llvm-svn: 168701

11 years agoubsan: Support unsigned overflows, and divide-by-zero int/float split.
Will Dietz [Tue, 27 Nov 2012 15:01:43 +0000 (15:01 +0000)]
ubsan: Support unsigned overflows, and divide-by-zero int/float split.

llvm-svn: 168700

11 years agoRemove outdated FIXME; should have removed that in r160782
Timur Iskhodzhanov [Tue, 27 Nov 2012 12:55:47 +0000 (12:55 +0000)]
Remove outdated FIXME; should have removed that in r160782

llvm-svn: 168698

11 years agotsan: fix macro mess
Dmitry Vyukov [Tue, 27 Nov 2012 12:51:16 +0000 (12:51 +0000)]
tsan: fix macro mess

llvm-svn: 168697

11 years agoRemove the dependent libraries feature.
Bill Wendling [Tue, 27 Nov 2012 09:55:56 +0000 (09:55 +0000)]
Remove the dependent libraries feature.

The dependent libraries feature was never used and has bit-rotted. Remove it.

llvm-svn: 168694

11 years agotsan: fix compilation for dead old compilers (why we are supporting them at all?..)
Dmitry Vyukov [Tue, 27 Nov 2012 09:35:44 +0000 (09:35 +0000)]
tsan: fix compilation for dead old compilers (why we are supporting them at all?..)

llvm-svn: 168693

11 years agotsan: add memory range access functions to public iface
Dmitry Vyukov [Tue, 27 Nov 2012 08:41:39 +0000 (08:41 +0000)]
tsan: add memory range access functions to public iface

llvm-svn: 168692

11 years agollvm/test/Transforms/SimplifyLibCalls: FileCheck-ize 3 tests.
NAKAMURA Takumi [Tue, 27 Nov 2012 08:18:23 +0000 (08:18 +0000)]
llvm/test/Transforms/SimplifyLibCalls: FileCheck-ize 3 tests.

llvm-svn: 168691

11 years agollvm/test/Transforms/SimplifyLibCalls/SPrintF.ll: Handle @sprintf() with -instcombine...
NAKAMURA Takumi [Tue, 27 Nov 2012 08:18:15 +0000 (08:18 +0000)]
llvm/test/Transforms/SimplifyLibCalls/SPrintF.ll: Handle @sprintf() with -instcombine, not -simplify-libcalls.

llvm-svn: 168690

11 years agollvm/test/Transforms/SimplifyLibCalls/SPrintF.ll: Fix datalayout since r168516.
NAKAMURA Takumi [Tue, 27 Nov 2012 08:18:08 +0000 (08:18 +0000)]
llvm/test/Transforms/SimplifyLibCalls/SPrintF.ll: Fix datalayout since r168516.

llvm-svn: 168689

11 years agoTrailing linefeeds.
NAKAMURA Takumi [Tue, 27 Nov 2012 08:17:58 +0000 (08:17 +0000)]
Trailing linefeeds.

llvm-svn: 168688

11 years agoRevert accidental commit.
Craig Topper [Tue, 27 Nov 2012 08:17:04 +0000 (08:17 +0000)]
Revert accidental commit.

llvm-svn: 168687

11 years agoMake PrintReg constructor explicit to prevent weird implicit conversions from acciden...
Craig Topper [Tue, 27 Nov 2012 08:14:24 +0000 (08:14 +0000)]
Make PrintReg constructor explicit to prevent weird implicit conversions from accidentally being triggered.

llvm-svn: 168686

11 years agoAdd ENABLE_CXX11 and ENABLE_WERROR to Makefile.llvm.rules for sample project. They...
Craig Topper [Tue, 27 Nov 2012 08:12:24 +0000 (08:12 +0000)]
Add ENABLE_CXX11 and ENABLE_WERROR to Makefile.llvm.rules for sample project. They were previously added to Makefile.llvm.config.in but the consumption was missing

llvm-svn: 168685

11 years agotsan: instrument atomic nand operation
Dmitry Vyukov [Tue, 27 Nov 2012 08:09:25 +0000 (08:09 +0000)]
tsan: instrument atomic nand operation

llvm-svn: 168684

11 years agotsan: add 128-bit atomic operations
Dmitry Vyukov [Tue, 27 Nov 2012 07:41:27 +0000 (07:41 +0000)]
tsan: add 128-bit atomic operations

llvm-svn: 168683

11 years agotsan: refactor atomic operations implementation
Dmitry Vyukov [Tue, 27 Nov 2012 07:25:50 +0000 (07:25 +0000)]
tsan: refactor atomic operations implementation
do the atomic operation under the sync object mutex
make acquire/release sync atomic with the operation itself
combine acquire and release into a single acq_rel operation

llvm-svn: 168682

11 years agoAdd test cases for r168417.
Craig Topper [Tue, 27 Nov 2012 07:19:54 +0000 (07:19 +0000)]
Add test cases for r168417.

llvm-svn: 168681

11 years agoRevert rearrangement of debug info sections to unblock the bots
Eric Christopher [Tue, 27 Nov 2012 06:49:23 +0000 (06:49 +0000)]
Revert rearrangement of debug info sections to unblock the bots
and O0 + debug codegen.

llvm-svn: 168680

11 years agotest/Transforms/SimplifyLibCalls/SPrintF.ll: Suppress this for now. r168677 unveiled...
NAKAMURA Takumi [Tue, 27 Nov 2012 06:42:48 +0000 (06:42 +0000)]
test/Transforms/SimplifyLibCalls/SPrintF.ll: Suppress this for now. r168677 unveiled another failure.

FYI, this test makes no sense with "not grep"... I saw "assertion failure" in stderr.

llvm-svn: 168679

11 years agoModify depends(Src, Dst, PossiblyLoopIndependent).
Preston Briggs [Tue, 27 Nov 2012 06:41:46 +0000 (06:41 +0000)]
Modify depends(Src, Dst, PossiblyLoopIndependent).
If the Src and Dst are the same instruction,
no loop-independent dependence is possible,
so we force the PossiblyLoopIndependent flag to false.

The test case results are updated appropriately.

llvm-svn: 168678

11 years agoinstcombine: Migrate sprintf optimizations
Meador Inge [Tue, 27 Nov 2012 05:57:54 +0000 (05:57 +0000)]
instcombine: Migrate sprintf optimizations

This patch migrates the sprintf optimizations from the simplify-libcalls
pass into the instcombine library call simplifier.

llvm-svn: 168677