platform/upstream/llvm.git
8 years agoFix comment for MD5Hash.
Teresa Johnson [Tue, 9 Feb 2016 13:28:44 +0000 (13:28 +0000)]
Fix comment for MD5Hash.

llvm-svn: 260221

8 years ago[OPENMP] Fix test incompatibility with arm buildbots.
Alexey Bataev [Tue, 9 Feb 2016 12:16:42 +0000 (12:16 +0000)]
[OPENMP] Fix test incompatibility with arm buildbots.

llvm-svn: 260220

8 years agoFix possible OOB access found by buildbot
Alexey Bataev [Tue, 9 Feb 2016 11:01:58 +0000 (11:01 +0000)]
Fix possible OOB access found by buildbot

llvm-svn: 260219

8 years agoAdd AST matcher reference to documentation directory when building HTML docs.
Manuel Klimek [Tue, 9 Feb 2016 10:59:21 +0000 (10:59 +0000)]
Add AST matcher reference to documentation directory when building HTML docs.

llvm-svn: 260218

8 years ago[clang-tidy] Make readability-container-size-empty work with inline namespaces. Fix...
Gabor Horvath [Tue, 9 Feb 2016 10:20:48 +0000 (10:20 +0000)]
[clang-tidy] Make readability-container-size-empty work with inline namespaces. Fix PR25812.

llvm-svn: 260217

8 years agoFix a typo in an xfail decorator in TestExprsChar
Tamas Berghammer [Tue, 9 Feb 2016 10:10:42 +0000 (10:10 +0000)]
Fix a typo in an xfail decorator in TestExprsChar

llvm-svn: 260216

8 years ago[OPENMP] Allow to use the variable in the same 'threadprivate'
Alexey Bataev [Tue, 9 Feb 2016 09:41:42 +0000 (09:41 +0000)]
[OPENMP] Allow to use the variable in the same 'threadprivate'
directive.

llvm-svn: 260215

8 years agoDo not mark variable as threadprivate if it was marked already.
Alexey Bataev [Tue, 9 Feb 2016 09:41:33 +0000 (09:41 +0000)]
Do not mark variable as threadprivate if it was marked already.

llvm-svn: 260214

8 years ago[OPENMP] Allow to reference threadprivate variable in same directive.
Alexey Bataev [Tue, 9 Feb 2016 09:41:09 +0000 (09:41 +0000)]
[OPENMP] Allow to reference threadprivate variable in same directive.

llvm-svn: 260213

8 years ago[clang-tidy] Replace a custom ASTMatcher with a built-in one.
Gabor Horvath [Tue, 9 Feb 2016 09:26:11 +0000 (09:26 +0000)]
[clang-tidy] Replace a custom ASTMatcher with a built-in one.

llvm-svn: 260212

8 years agoFixed preprocessed output of the first token for pragmas.
Alexey Bataev [Tue, 9 Feb 2016 08:51:26 +0000 (08:51 +0000)]
Fixed preprocessed output of the first token for pragmas.
Clang did not expanded macros in the very first token of the pragmas
during preprocessed output

llvm-svn: 260211

8 years ago[X86][AVX2] Fix SIGN_EXTEND vector handling on AVX2 targets.
Simon Pilgrim [Tue, 9 Feb 2016 08:19:19 +0000 (08:19 +0000)]
[X86][AVX2] Fix SIGN_EXTEND vector handling on AVX2 targets.

On AVX2 target we are poorly legalizing SIGN_EXTEND ops for which the input's legalized type doesn't have the same number of elements as the destination, resulting in an ANY_EXTEND followed by a SIGN_EXTEND_INREG.

This patch uses the existing SIGN_EXTEND -> SIGN_EXTEND_VECTOR_INREG combine to extend the input to the size of the result and using SIGN_EXTEND_VECTOR_INREG instead.

Differential Revision: http://reviews.llvm.org/D16994

llvm-svn: 260210

8 years agolld/test/mach-o/dependency_info.yaml: Forgot to remove XFAIL:win32.
NAKAMURA Takumi [Tue, 9 Feb 2016 07:44:10 +0000 (07:44 +0000)]
lld/test/mach-o/dependency_info.yaml: Forgot to remove XFAIL:win32.

llvm-svn: 260209

8 years agolld/test/mach-o/dependency_info.yaml: Tweak for dos path.
NAKAMURA Takumi [Tue, 9 Feb 2016 07:30:24 +0000 (07:30 +0000)]
lld/test/mach-o/dependency_info.yaml: Tweak for dos path.

llvm-svn: 260208

8 years agocheck-lld: Introduce %python.
NAKAMURA Takumi [Tue, 9 Feb 2016 07:30:18 +0000 (07:30 +0000)]
check-lld: Introduce %python.

llvm-svn: 260207

8 years agoIntroduce the feature 'demangler' in check-lld. Mark lld/test/old-elf/X86_64/demangle...
NAKAMURA Takumi [Tue, 9 Feb 2016 07:30:11 +0000 (07:30 +0000)]
Introduce the feature 'demangler' in check-lld. Mark lld/test/old-elf/X86_64/demangle.test as REQUIRES:demangler.

llvm-svn: 260206

8 years agoDisable lld/test/old-elf/group-cmd-search.test on win32. "XFAIL:win32" was insufficient.
NAKAMURA Takumi [Tue, 9 Feb 2016 07:30:04 +0000 (07:30 +0000)]
Disable lld/test/old-elf/group-cmd-search.test on win32. "XFAIL:win32" was insufficient.

llvm-svn: 260205

8 years ago[GCC] GOMP_task: Change argument type of if_cond from int to bool
Jonas Hahnfeld [Tue, 9 Feb 2016 07:07:30 +0000 (07:07 +0000)]
[GCC] GOMP_task: Change argument type of if_cond from int to bool

(libgomp has bool as well)

This was causing a test failure in omp_test_if.c when building with GCC in
Debug mode. I have verified that GCC versions 4.9.2 and 5.3.0 now work and
compile-tested this change with clang 3.7.1 and Intel Compiler 16.0.

Differential Revision: http://reviews.llvm.org/D16921

llvm-svn: 260204

8 years ago[CMake] Updating caches README with explanations of useful cache files.
Chris Bieneman [Tue, 9 Feb 2016 06:49:08 +0000 (06:49 +0000)]
[CMake] Updating caches README with explanations of useful cache files.

This is in response to silvas' post-commit suggestion.

llvm-svn: 260203

8 years agoMinor updates to failing tests. NFC
Marshall Clow [Tue, 9 Feb 2016 06:38:56 +0000 (06:38 +0000)]
Minor updates to failing tests. NFC

llvm-svn: 260202

8 years ago[CMake] Providing a CMake cache for 3-stage builds
Chris Bieneman [Tue, 9 Feb 2016 06:01:47 +0000 (06:01 +0000)]
[CMake] Providing a CMake cache for 3-stage builds

This cache file can be used to generate a 3-stage clang build. You can configure using the following CMake command:

cmake -C <path to clang>/cmake/caches/3-stage.cmake -G Ninja <path to llvm>

You can then run "ninja stage3-clang" to build stage1, stage2 and stage3 clangs.

This is useful for finding non-determinism the compiler by verifying that stage2 and stage3 are identical.

llvm-svn: 260201

8 years agoAdd comments to some tests
Xinliang David Li [Tue, 9 Feb 2016 05:47:08 +0000 (05:47 +0000)]
Add comments to some tests

llvm-svn: 260200

8 years agoFix an issue where pressing CTRL+C in the interactive script interpreter causes LLDB...
Enrico Granata [Tue, 9 Feb 2016 05:46:47 +0000 (05:46 +0000)]
Fix an issue where pressing CTRL+C in the interactive script interpreter causes LLDB to crash

This is because PyThreadState_Get() assumes a non-NULL thread state and crashes otherwise; but PyThreadState_GET is just a shortcut (in non-Python-debugging builds) for the global variable that holds the thread state

The behavior of CTRL+C is slightly more erratic than one would like. CTRL+C in the middle of execution of Python code will cause that execution to be interrupted (e.g. time.sleep(1000)), but a CTRL+C at the prompt will just cause a KeyboardInterrupt and not exit the interpreter - worse, it will only trigger the exception once one presses ENTER.

None of this is optimal, of course, but I don't have a lot of time to appease the Python deities with the proper spells right now, and fixing the crasher is already a good thing in and of itself

llvm-svn: 260199

8 years agoFurther reduce test overhead
Xinliang David Li [Tue, 9 Feb 2016 05:36:57 +0000 (05:36 +0000)]
Further reduce test overhead

llvm-svn: 260198

8 years agoRefactor PGO function naming and MD5 hashing support out of ProfileData
Teresa Johnson [Tue, 9 Feb 2016 05:12:44 +0000 (05:12 +0000)]
Refactor PGO function naming and MD5 hashing support out of ProfileData

Summary:
Move the function renaming logic into the Function class, and the
MD5Hash routine into the MD5 header.

This will enable these routines to be shared with ThinLTO, which
will be changed to store the MD5 hash instead of full function name
in the combined index for significant size reductions. And using the same
function naming for locals in the function index facilitates future
integration with indirect call value profiles.

Reviewers: davidxl

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D17006

llvm-svn: 260197

8 years agoUse std::forward to make ErrorOr<T> constructible from a value that has a user-define...
Nick Lewycky [Tue, 9 Feb 2016 04:47:58 +0000 (04:47 +0000)]
Use std::forward to make ErrorOr<T> constructible from a value that has a user-defined conversion to T. No functionality change intended.

llvm-svn: 260196

8 years agoUse the reserved spellings for attributes
Saleem Abdulrasool [Tue, 9 Feb 2016 04:05:37 +0000 (04:05 +0000)]
Use the reserved spellings for attributes

Change the no_sanitize attribute to use the reserved spelling.

llvm-svn: 260195

8 years agoAvoid forcing emission of delayed dllexported classes on template instantiation
Reid Kleckner [Tue, 9 Feb 2016 02:51:17 +0000 (02:51 +0000)]
Avoid forcing emission of delayed dllexported classes on template instantiation

Fixes PR26490

llvm-svn: 260194

8 years ago[GMR/OperandBundles] Teach getModRefBehavior about operand bundles
Sanjoy Das [Tue, 9 Feb 2016 02:31:47 +0000 (02:31 +0000)]
[GMR/OperandBundles] Teach getModRefBehavior about operand bundles

In general, memory restrictions on a called function (e.g. readnone)
cannot be transferred to a CallSite that has operand bundles.  It is
possible to make this inference smarter, but lets fix the behavior to be
correct first.

llvm-svn: 260193

8 years ago[TestExprsChar] Fix a typo is failure archs list
Siva Chandra [Tue, 9 Feb 2016 02:18:50 +0000 (02:18 +0000)]
[TestExprsChar] Fix a typo is failure archs list

Reviewers: zturner

Subscribers: lldb-commits

Differential Revision: http://reviews.llvm.org/D17014

llvm-svn: 260192

8 years agoAdd cmdline options for LC_DATA_IN_CODE load command.
Pete Cooper [Tue, 9 Feb 2016 02:10:39 +0000 (02:10 +0000)]
Add cmdline options for LC_DATA_IN_CODE load command.

Also added the defaults for whether to generate this load command, which
the cmdline options are able to override.

There was also a difference to ld64 which is fixed here in that ld64 will
generate an empty data in code command if requested.

rdar://problem/24472630

llvm-svn: 260191

8 years agoRemove TrailingObjects::operator delete. It's still suffering from
Richard Smith [Tue, 9 Feb 2016 02:09:16 +0000 (02:09 +0000)]
Remove TrailingObjects::operator delete. It's still suffering from
compiler-specific issues. Instead, repeat an 'operator delete' definition in
each derived class that is actually deleted, and give up on the static type
safety of an error when sized delete is accidentally used on a type derived
from TrailingObjects.

llvm-svn: 260190

8 years agoRemove 'llvm::TrailingObjects::operator delete', clang side.
Richard Smith [Tue, 9 Feb 2016 01:57:24 +0000 (01:57 +0000)]
Remove 'llvm::TrailingObjects::operator delete', clang side.

llvm-svn: 260189

8 years agoGenerate LC_FUNCTION_STARTS load command.
Pete Cooper [Tue, 9 Feb 2016 01:38:13 +0000 (01:38 +0000)]
Generate LC_FUNCTION_STARTS load command.

This load command generates data in the LINKEDIT section which
is a list of ULEB128 delta's to all of the functions in the __text section.

It is then 0 terminated and pointer aligned to pad.

ld64 exposes the -function-starts and no-function-starts cmdline options
to override behaviour from the defaults based on file types.

rdar://problem/24472630

llvm-svn: 260188

8 years agoFix the LLVM_ENABLE_MODULES build after adding TargetOpcodes.def in r259726.
David L Kreitzer [Tue, 9 Feb 2016 01:35:45 +0000 (01:35 +0000)]
Fix the LLVM_ENABLE_MODULES build after adding TargetOpcodes.def in r259726.

Differential Revision: http://reviews.llvm.org/D17005

llvm-svn: 260186

8 years agoWe should never dead strip mach header symbols.
Pete Cooper [Tue, 9 Feb 2016 01:34:13 +0000 (01:34 +0000)]
We should never dead strip mach header symbols.

This was a bug in our handling of these symbols compared to ld64.

Turns out that ld64 always marks these symbols as being not dead stripped.

llvm-svn: 260185

8 years agoReorder enum entry as its used for atom sorting.
Pete Cooper [Tue, 9 Feb 2016 01:27:30 +0000 (01:27 +0000)]
Reorder enum entry as its used for atom sorting.

The atom content type enum is used as a tie breaker to sort atoms.
In that case, we want MachHeader to be before typeCode as it really will
be before the code in the final executable.

Test case to follow in the next commit or two.

llvm-svn: 260184

8 years agoAdd an "addUsedAAAnalyses" helper function
Sanjoy Das [Tue, 9 Feb 2016 01:21:57 +0000 (01:21 +0000)]
Add an "addUsedAAAnalyses" helper function

Summary:
Passes that call `getAnalysisIfAvailable<T>` also need to call
`addUsedIfAvailable<T>` in `getAnalysisUsage` to indicate to the
legacy pass manager that it uses `T`.  This contract was being
violated by passes that used `createLegacyPMAAResults`.  This change
fixes this by exposing a helper in AliasAnalysis.h,
`addUsedAAAnalyses`, that is complementary to createLegacyPMAAResults
and does the right thing when called from `getAnalysisUsage`.

Reviewers: chandlerc

Subscribers: mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D17010

llvm-svn: 260183

8 years agoRemove SCEVAAWrapperPass from createLegacyPMAAResults; NFC
Sanjoy Das [Tue, 9 Feb 2016 01:21:50 +0000 (01:21 +0000)]
Remove SCEVAAWrapperPass from createLegacyPMAAResults; NFC

Summary:
createLegacyPMAAResults is only called by CGSCC and Module passes, so
the call to getAnalysisIfAvailable<SCEVAAWrapperPass>() never
succeeds (SCEVAAWrapperPass is a function pass).

Reviewers: chandlerc

Subscribers: mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D17009

llvm-svn: 260182

8 years agoFix undefined behavior when compiling in C++14 due to sized operator delete
Richard Smith [Tue, 9 Feb 2016 01:05:04 +0000 (01:05 +0000)]
Fix undefined behavior when compiling in C++14 due to sized operator delete
being called with the wrong size: convert CGFunctionInfo to use TrailingObjects
and ask TrailingObjects to provide a working 'operator delete' for us.

llvm-svn: 260181

8 years agoRe-commit r259942 (reverted in r260053) with a different workaround for the MSVC...
Richard Smith [Tue, 9 Feb 2016 01:03:42 +0000 (01:03 +0000)]
Re-commit r259942 (reverted in r260053) with a different workaround for the MSVC bug.

This fixes undefined behavior in C++14 due to the size of the object being
deleted being different from sizeof(dynamic type) when it is allocated with
trailing objects.

MSVC seems to have several bugs around using-declarations changing the access
of a member inherited from a base class, so use forwarding functions instead of
using-declarations to make TrailingObjects::operator delete accessible where
desired.

llvm-svn: 260180

8 years agoSimplify some expressions involving unique_ptr and ErrorOr
David Blaikie [Tue, 9 Feb 2016 01:02:24 +0000 (01:02 +0000)]
Simplify some expressions involving unique_ptr and ErrorOr

llvm-svn: 260179

8 years agoRemove skip and xfail decorators for target architecture.
Zachary Turner [Tue, 9 Feb 2016 00:36:34 +0000 (00:36 +0000)]
Remove skip and xfail decorators for target architecture.

This removes the following decorators:

* skipIfI386
* expectedFailureI386
* expectedFailurex86_64
* skipIfArch
* skipUnlessArch
* skipUnlessI386

And other related decorators.  All code using those decorators
is updated to use expectedFailureAll and skipIf

llvm-svn: 260178

8 years agoRemove skipIfHostPlatform and skipUnlessHostPlatform decorators.
Zachary Turner [Tue, 9 Feb 2016 00:36:27 +0000 (00:36 +0000)]
Remove skipIfHostPlatform and skipUnlessHostPlatform decorators.

llvm-svn: 260177

8 years agoRemove the skipUnlessArch decorator.
Zachary Turner [Tue, 9 Feb 2016 00:36:22 +0000 (00:36 +0000)]
Remove the skipUnlessArch decorator.

Convert everything over to using skipIf.

llvm-svn: 260176

8 years ago[asan] Fix build on Windows and PowerPC.
Evgeniy Stepanov [Tue, 9 Feb 2016 00:28:57 +0000 (00:28 +0000)]
[asan] Fix build on Windows and PowerPC.

llvm-svn: 260175

8 years agoThis patch is to fix PR26529 caused by r259736.
Wei Mi [Tue, 9 Feb 2016 00:07:08 +0000 (00:07 +0000)]
This patch is to fix PR26529 caused by r259736.

IndVarSimplify assumes scAddRecExpr to be expanded in literal form instead of
canonical form by calling disableCanonicalMode after it creates SCEVExpander.
When CanonicalMode is disabled, SCEVExpander::expand should always return PHI
node for scAddRecExpr. r259736 broke the assumption.

The fix is to let SCEVExpander::expand skip the reuse Value logic if
CanonicalMode is false.

In addition, Besides IndVarSimplify, LSR pass also calls disableCanonicalMode
before doing rewrite. We can remove the original check of LSRMode in reuse
Value logic and use CanonicalMode instead.

llvm-svn: 260174

8 years ago[llvm-nm] Remove excessive parenthesis, noticed by David Blaikie.
Davide Italiano [Mon, 8 Feb 2016 23:50:23 +0000 (23:50 +0000)]
[llvm-nm] Remove excessive parenthesis, noticed by David Blaikie.

llvm-svn: 260173

8 years agoRevert r260012 due to __gnu_cxx::hash_map breakage
Eric Fiselier [Mon, 8 Feb 2016 23:47:13 +0000 (23:47 +0000)]
Revert r260012 due to __gnu_cxx::hash_map breakage

llvm-svn: 260172

8 years ago[PGO] Revert r260146 as it breaks Darwin platforms.
Rong Xu [Mon, 8 Feb 2016 23:11:16 +0000 (23:11 +0000)]
[PGO] Revert r260146 as it breaks Darwin platforms.
r260146 | xur | 2016-02-08 13:07:46 -0800 (Mon, 08 Feb 2016) | 13 lines
[PGO] Differentiate Clang instrumentation and IR level instrumentation profiles

llvm-svn: 260170

8 years agoFactor out UnrollAnalyzer to Analysis, and add unit tests for it.
Michael Zolotukhin [Mon, 8 Feb 2016 23:03:59 +0000 (23:03 +0000)]
Factor out UnrollAnalyzer to Analysis, and add unit tests for it.

Summary:
Unrolling Analyzer is already pretty complicated, and it becomes harder and harder to exercise it with usual IR tests, as with them we can only check the final decision: whether the loop is unrolled or not. This change factors this framework out from LoopUnrollPass to analyses, which allows to use unit tests.
The change itself is supposed to be NFC, except adding a couple of tests.

I plan to add more tests as I add new functionality and find/fix bugs.

Reviewers: chandlerc, hfinkel, sanjoy

Subscribers: zzheng, sanjoy, llvm-commits

Differential Revision: http://reviews.llvm.org/D16623

llvm-svn: 260169

8 years ago[X86][SSE1] Add MOVLHPS/MOVHLPS lowering and memory folding support
Simon Pilgrim [Mon, 8 Feb 2016 23:03:46 +0000 (23:03 +0000)]
[X86][SSE1] Add MOVLHPS/MOVHLPS lowering and memory folding support

As discussed on PR26491, this patch adds support for lowering v4f32 shuffles to the MOVLHPS/MOVHLPS instructions. It also adds support for memory folding with their MOVLPS/MOVHPS load equivalents.

This first patch only really helps SSE1 targets as SSE2+ targets will widen the shuffle mask and use v2f64 equivalents (although they still combine to MOVLHPS/MOVHLPS for v2f64 splats). This will have to be addressed in a future patch, most likely when we add support for binary target shuffle combines.

Differential Revision: http://reviews.llvm.org/D16956

llvm-svn: 260168

8 years ago[asan] Fix Windows build.
Evgeniy Stepanov [Mon, 8 Feb 2016 23:01:06 +0000 (23:01 +0000)]
[asan] Fix Windows build.

llvm-svn: 260167

8 years ago[llvm-nm] Yet another attempt of simplifying code.
Davide Italiano [Mon, 8 Feb 2016 22:58:26 +0000 (22:58 +0000)]
[llvm-nm] Yet another attempt of simplifying code.

llvm-svn: 260166

8 years ago[asan] Remove leftover debug print statement.
Evgeniy Stepanov [Mon, 8 Feb 2016 22:54:46 +0000 (22:54 +0000)]
[asan] Remove leftover debug print statement.

llvm-svn: 260165

8 years ago[regalloc][WinEH] Do not mark intervals as not spillable if they contain a regmask
Andrew Kaylor [Mon, 8 Feb 2016 22:52:51 +0000 (22:52 +0000)]
[regalloc][WinEH] Do not mark intervals as not spillable if they contain a regmask

Differential Revision: http://reviews.llvm.org/D16831

llvm-svn: 260164

8 years ago[asan] Implement SEGV read vs write detection for ARM and AArch64.
Evgeniy Stepanov [Mon, 8 Feb 2016 22:50:25 +0000 (22:50 +0000)]
[asan] Implement SEGV read vs write detection for ARM and AArch64.

llvm-svn: 260163

8 years agollvm-cov: Fix reading gcov data that does not have function names
Justin Bogner [Mon, 8 Feb 2016 22:49:40 +0000 (22:49 +0000)]
llvm-cov: Fix reading gcov data that does not have function names

In order for recent gcov versions to read the coverage data, you have
to use UseCfgChecksum=true and FunctionNamesInData=false options for
coverage profiling pass. This is because gcov is expecting the
function section in .gcda to be exactly 3 words in size, containing
ident and two checksums.

While llvm-cov is compatible with UseCfgChecksum=true, it always
expects a function name in .gcda function sections (it's not
compatible with FunctionNamesInData=false). Thus it's currently
impossible to generate one set of coverage files that works with both
gcov and llvm-cov.

This change fixes the reading of coverage information to only read the
function name if it's present.

Patch by Arseny Kapoulkine. Thanks!

llvm-svn: 260162

8 years ago[PGO] Cover more cases of implicitly generated C++ methods
Xinliang David Li [Mon, 8 Feb 2016 22:41:37 +0000 (22:41 +0000)]
[PGO] Cover more cases of implicitly generated C++ methods

llvm-svn: 260161

8 years agoThis patch adds doxygen comments for all the intrinsincs in the header file pmmintrin...
Ekaterina Romanova [Mon, 8 Feb 2016 22:35:09 +0000 (22:35 +0000)]
This patch adds doxygen comments for all the intrinsincs in the header file pmmintrin.h. The doxygen comments are automatically generated based on Sony's intrinsics document.

Differential Revision: http://reviews.llvm.org/D16913

llvm-svn: 260160

8 years agoMake ParentMap work with explicit specializations of function templates.
Nico Weber [Mon, 8 Feb 2016 22:23:09 +0000 (22:23 +0000)]
Make ParentMap work with explicit specializations of function templates.

For an explicit specialization, we first build a FunctionDecl, and then
we call SubstDecl() on it to build a second FunctionDecl, which has the
first FunctionDecl as canonical decl.

The address of an explicit specialization of function template used to be the
canonical decl of the FunctionDecl.  This is different from all the other
DeduceTemplateArguments() calls in SemaOverload, and since the canonical decl
isn't visited by ParentMap while the redecl is, it also made ParentMap assert
when computing the parent of a address-of-explicit-specialization-fun-template.

To fix, remove the getCanonicalDecl() call.  No behavior difference for clang,
but it fixes an assert in ParentMap (which is e.g. used by libTooling).

llvm-svn: 260159

8 years ago[Concepts] Remove the IsConcept bit and associated member functions from VarDecl
Nathan Wilson [Mon, 8 Feb 2016 22:02:50 +0000 (22:02 +0000)]
[Concepts] Remove the IsConcept bit and associated member functions from VarDecl
because the information is now stored in TemplateDecl.

llvm-svn: 260155

8 years agocmake: Use "set" instead of "option" for LLVM_ENABLE_LTO
Justin Bogner [Mon, 8 Feb 2016 21:55:19 +0000 (21:55 +0000)]
cmake: Use "set" instead of "option" for LLVM_ENABLE_LTO

Apparently option is for bools and cmake-gui will display this
strangely with option.

Pointed out by edward-san - thanks!

llvm-svn: 260154

8 years agoMove includes inside guards. NFC.
Pete Cooper [Mon, 8 Feb 2016 21:50:45 +0000 (21:50 +0000)]
Move includes inside guards.  NFC.

These includes were before the #ifndef/#define guards.  Moving them to
where they should be.

llvm-svn: 260153

8 years ago[WebAssembly] Update the br_if instructions' operand orders to match the spec.
Dan Gohman [Mon, 8 Feb 2016 21:50:13 +0000 (21:50 +0000)]
[WebAssembly] Update the br_if instructions' operand orders to match the spec.

llvm-svn: 260152

8 years agorangify; NFC
Sanjay Patel [Mon, 8 Feb 2016 21:32:43 +0000 (21:32 +0000)]
rangify; NFC

llvm-svn: 260151

8 years ago[PGO] Update InstrProfData.inc to sync with llvm
Rong Xu [Mon, 8 Feb 2016 21:18:18 +0000 (21:18 +0000)]
[PGO] Update InstrProfData.inc to sync with llvm

Sync InstrProfData.inc with the one in llvm.

llvm-svn: 260148

8 years agoFix typo in asan_flags.
George Burgess IV [Mon, 8 Feb 2016 21:10:00 +0000 (21:10 +0000)]
Fix typo in asan_flags.

Patch by Michael Stapelberg.

llvm-svn: 260147

8 years ago[PGO] Differentiate Clang instrumentation and IR level instrumentation profiles
Rong Xu [Mon, 8 Feb 2016 21:07:46 +0000 (21:07 +0000)]
[PGO] Differentiate Clang instrumentation and IR level instrumentation profiles

This patch uses one bit in profile version to differentiate Clang
instrumentation and IR level instrumentation profiles.

PGOInstrumenation generates a COMDAT variable __llvm_profile_raw_version so
that the compiler runtime can set the right profile kind.
PGOInstrumenation now checks this bit to make sure it's an IR level
instrumentation profile.

Differential Revision: http://reviews.llvm.org/D15540

llvm-svn: 260146

8 years ago[x86] convert masked store of one element to scalar store
Sanjay Patel [Mon, 8 Feb 2016 21:05:08 +0000 (21:05 +0000)]
[x86] convert masked store of one element to scalar store

Another opportunity to reduce masked stores: in D16691, we decided not to attempt the 'one mask element is set'
transform in InstCombine, but this should be a win for any AVX machine.

Code comments note that this transform could be extended for other targets / cases.

Differential Revision: http://reviews.llvm.org/D16828

llvm-svn: 260145

8 years agocmake: Accept "thin" or "full" as arguments to -DLLVM_ENABLE_LTO
Justin Bogner [Mon, 8 Feb 2016 21:01:24 +0000 (21:01 +0000)]
cmake: Accept "thin" or "full" as arguments to -DLLVM_ENABLE_LTO

Mehdi suggested in a review of r259766 that it's also useful to easily
set the type of LTO. Augment the cmake variable to support that.

llvm-svn: 260143

8 years ago[PGO] Simpflify test and increase coverage
Xinliang David Li [Mon, 8 Feb 2016 20:46:26 +0000 (20:46 +0000)]
[PGO] Simpflify test and increase coverage

llvm-svn: 260142

8 years agoFix build bot failure
Xinliang David Li [Mon, 8 Feb 2016 20:08:21 +0000 (20:08 +0000)]
Fix build bot failure

llvm-svn: 260138

8 years agoAMDGPU/SI: Implement a work-around for smrd corrupting vccz bit
Tom Stellard [Mon, 8 Feb 2016 19:49:20 +0000 (19:49 +0000)]
AMDGPU/SI: Implement a work-around for smrd corrupting vccz bit

Summary:
We will hit this once we have enabled uniform branches.  The
smrd-vccz-bug.ll test will be added with the uniform branch commit.

Reviewers: mareko, arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D16725

llvm-svn: 260137

8 years agoAdd triple to h-registers-3.ll to make bots happy after r260133
Hans Wennborg [Mon, 8 Feb 2016 19:45:24 +0000 (19:45 +0000)]
Add triple to h-registers-3.ll to make bots happy after r260133

llvm-svn: 260136

8 years agoA number of improvements to decorator conditionals.
Zachary Turner [Mon, 8 Feb 2016 19:35:18 +0000 (19:35 +0000)]
A number of improvements to decorator conditionals.

* Change the `not_in` function to be called `no_match`.  This makes
  it clear that keyword arguments can be more than just lists.
* Change the name of `_check_list_or_lambda` to
  `_match_decorator_property`.  Again clarifying that decorator params
  are not always lists.
* Always use a regex match when matching strings.  This allows automatic
  support for regex matching on all decorator properties.  Also support
  compiled regex values.
* Fix a bug in the compiler check used by _decorateTest.  The two
  arguments were reversed, the condition was always wrong.
* Change one test that uses skipUnlessArch to use skipIf, to
  demonstrate that skipIf can now handle more scenarios.

Differential Revision: http://reviews.llvm.org/D16938

llvm-svn: 260135

8 years agoRemove expectedFailureWindows decorator.
Zachary Turner [Mon, 8 Feb 2016 19:34:59 +0000 (19:34 +0000)]
Remove expectedFailureWindows decorator.

expectedFailureWindows is equivalent to using the general
expectedFailureAll decorator with oslist="windows".  Additionally,
by moving towards these common decorators we can solve the issue
of having to support decorators that can be called with or without
arguments.  Once all decorators are always called with arguments,
and this is enforced by design (because you can't specify the condition
you're decorating for without passing an argument) the implementation
of the decorators can become much simpler

Differential Revision: http://reviews.llvm.org/D16936

llvm-svn: 260134

8 years ago[X86] Don't zero/sign-extend i1, i8, or i16 return values to 32 bits (PR22532)
Hans Wennborg [Mon, 8 Feb 2016 19:34:30 +0000 (19:34 +0000)]
[X86] Don't zero/sign-extend i1, i8, or i16 return values to 32 bits (PR22532)

This matches GCC and MSVC's behaviour, and saves on code size.

We were already not extending i1 return values on x86_64 after r127766. This
takes that patch further by applying it to x86 target as well, and also for i8
and i16.

The ABI docs have been unclear about the required behaviour here. The new i386
psABI [1] clearly states (Table 2.4, page 14) that i1, i8, and i16 return
vales do not need to be extended beyond 8 bits. The x86_64 ABI doc is being
updated to say the same [2].

Differential Revision: http://reviews.llvm.org/D16907

 [1]. https://01.org/sites/default/files/file_attach/intel386-psabi-1.0.pdf
 [2]. https://groups.google.com/d/msg/x86-64-abi/E8O33onbnGQ/_RFWw_ixDQAJ

llvm-svn: 260133

8 years ago[asan] fix the Windows build (MSVC says: cannot specify explicit initializer for...
Kostya Serebryany [Mon, 8 Feb 2016 19:33:36 +0000 (19:33 +0000)]
[asan] fix the Windows build (MSVC says: cannot specify explicit initializer for arrays)

llvm-svn: 260132

8 years agoAArch64: match correct order in subtraction pattern.
Tim Northover [Mon, 8 Feb 2016 19:33:18 +0000 (19:33 +0000)]
AArch64: match correct order in subtraction pattern.

The accumulator in multiply-and-subtract instructions is actually subtracted
*from* so these patterns were computing the wrong value.

llvm-svn: 260131

8 years agofix typos; NFC
Sanjay Patel [Mon, 8 Feb 2016 19:27:33 +0000 (19:27 +0000)]
fix typos; NFC

llvm-svn: 260130

8 years ago[Coverage] Fix crash when handling certain macro expansions
Vedant Kumar [Mon, 8 Feb 2016 19:25:45 +0000 (19:25 +0000)]
[Coverage] Fix crash when handling certain macro expansions

When handling 'if' statements, we crash if the condition and the consequent
branch are spanned by a single macro expansion.

The crash occurs because of a sanity 'reset' in popRegions(): if an expansion
exactly spans an entire region, we set MostRecentLocation to the start of the
expansion (its 'include location'). This ensures we don't handleFileExit()
ourselves out of the expansion before we're done processing all of the regions
within it. This is tested in test/CoverageMapping/macro-expressions.c.

This causes a problem when an expansion spans both the condition and the
consequent branch of an 'if' statement. MostRecentLocation is updated to the
start of the 'if' statement in popRegions(), so the file for the expansion
isn't exited by the time we're done handling the statement. We then crash with
'fatal: File exit not handled before popRegions'.

The fix for this is to detect these kinds of expansions, and conservatively
update MostRecentLocation to the end of expansion region containing the
conditional. I've added tests to make sure we don't have the same problem with
other kinds of statements.

rdar://problem/23630316

Differential Revision: http://reviews.llvm.org/D16934

llvm-svn: 260129

8 years ago[asan] re-commit r259961, this time making the test Linux-Only
Kostya Serebryany [Mon, 8 Feb 2016 19:21:08 +0000 (19:21 +0000)]
[asan] re-commit r259961, this time making the test Linux-Only

llvm-svn: 260128

8 years agoSimplify test cases
Xinliang David Li [Mon, 8 Feb 2016 19:14:14 +0000 (19:14 +0000)]
Simplify test cases

llvm-svn: 260126

8 years agoSimplify this unittest.
Adrian Prantl [Mon, 8 Feb 2016 19:13:15 +0000 (19:13 +0000)]
Simplify this unittest.

Thanks to dblaikie for the suggestion!

llvm-svn: 260125

8 years agoRemove dead code.
Richard Smith [Mon, 8 Feb 2016 19:10:14 +0000 (19:10 +0000)]
Remove dead code.

llvm-svn: 260124

8 years agoAMDGPU: Remove bfi and bfm intrinsics
Matt Arsenault [Mon, 8 Feb 2016 19:06:01 +0000 (19:06 +0000)]
AMDGPU: Remove bfi and bfm intrinsics

Nothing is using them.

llvm-svn: 260123

8 years ago[ThinLTO] Remove imported available externally defs from comdats.
Teresa Johnson [Mon, 8 Feb 2016 18:47:20 +0000 (18:47 +0000)]
[ThinLTO] Remove imported available externally defs from comdats.

Summary:
Available externally definitions are considered declarations for the
linker and eventually dropped. As such they are not allowed to be
in comdats. Remove any such imported functions from comdats.

Reviewers: rafael

Subscribers: davidxl, llvm-commits, joker.eph

Differential Revision: http://reviews.llvm.org/D16120

llvm-svn: 260122

8 years agoFix windows build bot failure
Xinliang David Li [Mon, 8 Feb 2016 18:27:04 +0000 (18:27 +0000)]
Fix windows build bot failure

llvm-svn: 260120

8 years ago[PGO] Enable compression in pgo instrumentation
Xinliang David Li [Mon, 8 Feb 2016 18:14:02 +0000 (18:14 +0000)]
[PGO] Enable compression in pgo instrumentation

This reduces sizes of instrumented object files, final binaries,
process images, and raw profile data.

The format of the indexed profile data remain the same.

Differential Revision: http://reviews.llvm.org/D16388

llvm-svn: 260118

8 years ago[PGO] Enable compression in pgo instrumentation
Xinliang David Li [Mon, 8 Feb 2016 18:13:49 +0000 (18:13 +0000)]
[PGO] Enable compression in pgo instrumentation

This reduces sizes of instrumented object files, final binaries,
process images, and raw profile data.

The format of the indexed profile data remain the same.

Differential Revision: http://reviews.llvm.org/D16388

llvm-svn: 260117

8 years agoFix missing space (NFC)
Rong Xu [Mon, 8 Feb 2016 18:12:41 +0000 (18:12 +0000)]
Fix missing space (NFC)

Fixed the inconsistently placed : (missing space) introduced in in r259811.

llvm-svn: 260116

8 years agoClean up a test; get rid of hard-wired char/wchar_t code for template fns that take...
Marshall Clow [Mon, 8 Feb 2016 17:38:23 +0000 (17:38 +0000)]
Clean up a test; get rid of hard-wired char/wchar_t code for template fns that take any char type. Prep work for PR#26503

llvm-svn: 260115

8 years agomath: Add frexp ported from amd-builtins
Aaron Watry [Mon, 8 Feb 2016 17:07:21 +0000 (17:07 +0000)]
math: Add frexp ported from amd-builtins

The float implementation is almost a direct port from the amd-builtins,
but instead of just having a scalar and float4 implementation, it has
a scalar and arbitrary width vector implementation.

The double scalar is also a direct port from AMD's builtin release.

The double vector implementation copies the logic in the float vector
implementation using the values from the double scalar version.

Both have been tested in piglit using tests sent to that project's
mailing list.

Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 260114

8 years agoUse llvm::TempDIScope instead of manually deleting a temporary MDNode.
Adrian Prantl [Mon, 8 Feb 2016 17:03:28 +0000 (17:03 +0000)]
Use llvm::TempDIScope instead of manually deleting a temporary MDNode.

llvm-svn: 260113

8 years ago[SCEV][LAA] Re-commit r260085 and r260086, this time with a fix for the memory
Silviu Baranga [Mon, 8 Feb 2016 17:02:45 +0000 (17:02 +0000)]
[SCEV][LAA] Re-commit r260085 and r260086, this time with a fix for the memory
sanitizer issue. The PredicatedScalarEvolution's copy constructor
wasn't copying the Generation value, and was leaving it un-initialized.

Original commit message:

[SCEV][LAA] Add no wrap SCEV predicates and use use them to improve strided pointer detection

Summary:
This change adds no wrap SCEV predicates with:
  - support for runtime checking
  - support for expression rewriting:
      (sext ({x,+,y}) -> {sext(x),+,sext(y)}
      (zext ({x,+,y}) -> {zext(x),+,sext(y)}

Note that we are sign extending the increment of the SCEV, even for
the zext case. This is needed to cover the fairly common case where y would
be a (small) negative integer. In order to do this, this change adds two new
flags: nusw and nssw that are applicable to AddRecExprs and permit the
transformations above.

We also change isStridedPtr in LAA to be able to make use of
these predicates. With this feature we should now always be able to
work around overflow issues in the dependence analysis.

Reviewers: mzolotukhin, sanjoy, anemet

Subscribers: mzolotukhin, sanjoy, llvm-commits, rengolin, jmolloy, hfinkel

Differential Revision: http://reviews.llvm.org/D15412

llvm-svn: 260112

8 years agoAdd a unit test for r259973.
Adrian Prantl [Mon, 8 Feb 2016 17:02:34 +0000 (17:02 +0000)]
Add a unit test for r259973.

llvm-svn: 260111

8 years ago[JumpThreading] Change a return of ComputeValueKnownInPredecessors()
Haicheng Wu [Mon, 8 Feb 2016 17:00:39 +0000 (17:00 +0000)]
[JumpThreading] Change a return of ComputeValueKnownInPredecessors()

Change a return statement of ComputeValueKnownInPredecessors() to be the same as
the rest return statements of the function. Otherwise, it might return true with
an empty Result when the current basic block has no predecessors and trigger the
first assert of JumpThreading::ProcessThreadableEdges().

llvm-svn: 260110

8 years agoSelectionDAG: Lower some range metadata to AssertZext
Matt Arsenault [Mon, 8 Feb 2016 16:28:19 +0000 (16:28 +0000)]
SelectionDAG: Lower some range metadata to AssertZext

If a range has a lower bound of 0, add an AssertZext from the
nearest floor power of two.

This allows operations with some workitem intrinsics with known
maximum ranges to use fast 24-bit multiplies.

llvm-svn: 260109

8 years ago[clang-tidy] Some improvements in 'misc-definitions-in-headers' check.
Haojian Wu [Mon, 8 Feb 2016 16:05:39 +0000 (16:05 +0000)]
[clang-tidy] Some improvements in 'misc-definitions-in-headers' check.

Reviewers: alexfh

Subscribers: cfe-commits

Differential Revision: http://reviews.llvm.org/D16979

llvm-svn: 260108