platform/upstream/llvm.git
7 years agoLiveIntervalAnalysis: Fix missing case in pruneSubRegValues()
Matthias Braun [Fri, 19 May 2017 00:18:03 +0000 (00:18 +0000)]
LiveIntervalAnalysis: Fix missing case in pruneSubRegValues()

pruneSubRegValues() needs to remove subregister ranges starting at
instructions that later get removed by eraseInstrs(). It missed to check
one case in which eraseInstrs() would remove an instruction.

Fixes http://llvm.org/PR32688

llvm-svn: 303396

7 years agoFix typo
Matthias Braun [Fri, 19 May 2017 00:17:59 +0000 (00:17 +0000)]
Fix typo

llvm-svn: 303395

7 years agoFix another warning.
Zachary Turner [Thu, 18 May 2017 23:30:51 +0000 (23:30 +0000)]
Fix another warning.

llvm-svn: 303394

7 years ago[NewGVN] Break infinite recursion in singleReachablePHIPath().
Davide Italiano [Thu, 18 May 2017 23:22:44 +0000 (23:22 +0000)]
[NewGVN] Break infinite recursion in singleReachablePHIPath().

We can have cycles between PHIs and this causes singleReachablePhi()
to call itself indefintely (until we run out of stack). The proper
solution would be that of computing SCCs, but it's not worth for
now, so just keep a visited set and give up when we find a cycle.
Thanks to Dan for the discussion/help with this.

Fixes PR33014.

llvm-svn: 303393

7 years ago[compiler-rt] Replace ifs with switch statement in IsHandledDeadlySignal
Vitaly Buka [Thu, 18 May 2017 23:13:22 +0000 (23:13 +0000)]
[compiler-rt] Replace ifs with switch statement in IsHandledDeadlySignal

Reviewers: eugenis

Subscribers: kubamracek, llvm-commits, dberris

Differential Revision: https://reviews.llvm.org/D33158

llvm-svn: 303392

7 years agoFix some build errors and warnings.
Zachary Turner [Thu, 18 May 2017 23:12:42 +0000 (23:12 +0000)]
Fix some build errors and warnings.

llvm-svn: 303391

7 years ago[CodeView] Raise the source to ID map out of the TypeStreamMerger.
Zachary Turner [Thu, 18 May 2017 23:04:08 +0000 (23:04 +0000)]
[CodeView] Raise the source to ID map out of the TypeStreamMerger.

This map will be needed to rewrite symbol streams after re-writing
the corresponding type streams.

llvm-svn: 303390

7 years ago[llvm-pdbdump] Add the ability to merge PDBs.
Zachary Turner [Thu, 18 May 2017 23:03:41 +0000 (23:03 +0000)]
[llvm-pdbdump] Add the ability to merge PDBs.

Merging PDBs is a feature that will be used heavily by
the linker.  The functionality already exists but does not
have deep test coverage because it's not easily exposed through
any tools.  This patch aims to address that by adding the
ability to merge PDBs via llvm-pdbdump.  It takes arbitrarily
many PDBs and outputs a single PDB.

Using this new functionality, a test is added for merging
type records.  Future patches will add the ability to merge
symbol records, module information, etc.

llvm-svn: 303389

7 years ago[CodeView] Provide a common interface for type collections.
Zachary Turner [Thu, 18 May 2017 23:03:06 +0000 (23:03 +0000)]
[CodeView] Provide a common interface for type collections.

Right now we have multiple notions of things that represent collections of
types. Most commonly used are TypeDatabase, which is supposed to keep
mappings from TypeIndex to type name when reading a type stream, which
happens when reading PDBs. And also TypeTableBuilder, which is used to
build up a collection of types dynamically which we will later serialize
(i.e. when writing PDBs).

But often you just want to do some operation on a collection of types, and
you may want to do the same operation on any kind of collection. For
example, you might want to merge two TypeTableBuilders or you might want
to merge two type streams that you loaded from various files.

This dichotomy between reading and writing is responsible for a lot of the
existing code duplication and overlapping responsibilities in the existing
CodeView library classes. For example, after building up a
TypeTableBuilder with a bunch of type records, if we want to dump it we
have to re-invent a bunch of extra glue because our dumper takes a
TypeDatabase or a CVTypeArray, which are both incompatible with
TypeTableBuilder.

This patch introduces an abstract base class called TypeCollection which
is shared between the various type collection like things. Wherever we
previously stored a TypeDatabase& in some common class, we now store a
TypeCollection&.

The advantage of this is that all the details of how the collection are
implemented, such as lazy deserialization of partial type streams, is
completely transparent and you can just treat any collection of types the
same regardless of where it came from.

Differential Revision: https://reviews.llvm.org/D33293

llvm-svn: 303388

7 years ago[InstCombine] add more tests for xor-of-icmps; NFC
Sanjay Patel [Thu, 18 May 2017 22:47:57 +0000 (22:47 +0000)]
[InstCombine] add more tests for xor-of-icmps; NFC

llvm-svn: 303387

7 years ago[ScopInfo] Fix r302231 to use logical or (||). NFC.
Michael Kruse [Thu, 18 May 2017 21:55:36 +0000 (21:55 +0000)]
[ScopInfo] Fix r302231 to use logical or (||). NFC.

In r302231 we mistakenly use bitwise or (|) instead of logical
or (||). This patch fixes that.

Contributed-by: Sameer AbuAsal <sabuasal@codeaurora.org>
Differential Revision: https://reviews.llvm.org/D33337

llvm-svn: 303386

7 years ago[NewGVN] Replace predicate info leftovers.
Davide Italiano [Thu, 18 May 2017 21:43:23 +0000 (21:43 +0000)]
[NewGVN] Replace predicate info leftovers.

This time with an additional fix, i.e. we remove the dead
@llvm.ssa.copy instruction.

llvm-svn: 303385

7 years agoUse linker script commands in writeMapFile.
Rafael Espindola [Thu, 18 May 2017 21:30:14 +0000 (21:30 +0000)]
Use linker script commands in writeMapFile.

This converts the last (chronologically) user of OutputSections to use
the linker script commands instead.

The idea is to convert all uses after fabricateDefaultCommands, so
that we have a single representation.

llvm-svn: 303384

7 years ago[ADT] Fix some Clang-tidy modernize-use-using warnings; other minor fixes (NFC).
Eugene Zelenko [Thu, 18 May 2017 21:27:11 +0000 (21:27 +0000)]
[ADT] Fix some Clang-tidy modernize-use-using warnings; other minor fixes (NFC).

llvm-svn: 303383

7 years agoclang-format: [JS] for await, and fix a crash with for loops.
Martin Probst [Thu, 18 May 2017 21:19:29 +0000 (21:19 +0000)]
clang-format: [JS] for await, and fix a crash with for loops.

Summary:
The syntax is actually `for await (const x of y)` (d'oh).
This also fixes a crash for `for` tokens not followed by additional tokens.

Reviewers: djasper

Subscribers: cfe-commits, klimek

Differential Revision: https://reviews.llvm.org/D33329

llvm-svn: 303382

7 years ago[InstCombine] add helper to foldXorOfICmps(); NFCI
Sanjay Patel [Thu, 18 May 2017 20:53:16 +0000 (20:53 +0000)]
[InstCombine] add helper to foldXorOfICmps(); NFCI

Also, fix the old-style capitalization of the related functions
and move them to the 'private' section of the class since they
are just helpers of the visit* functions.

As shown in the post-commit comments for D32143, we are missing
folds for xor-of-icmps.

llvm-svn: 303381

7 years ago[scudo] lower quarantine default sizes
Kostya Kortchinsky [Thu, 18 May 2017 20:47:35 +0000 (20:47 +0000)]
[scudo] lower quarantine default sizes

Summary:
After discussing the current defaults with a couple of parties, the consensus
is that they are too high. 1Mb of quarantine has about a 4Mb impact on PSS, so
memory usage goes up quickly.
This is obviously configurable, but the default value should be more
"approachable", so both the global size and the thread local size are 1/4 of
what they used to be.

Reviewers: alekseyshl, kcc

Reviewed By: alekseyshl

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33321

llvm-svn: 303380

7 years ago[InstSimplify] Make m_Not work for xor -1, X
Craig Topper [Thu, 18 May 2017 20:27:32 +0000 (20:27 +0000)]
[InstSimplify] Make m_Not work for xor -1, X

Currently m_Not only works the canonical xor X, -1 form that InstCombine produces. InstSimplify can't rely on this canonicalization.

Differential Revision: https://reviews.llvm.org/D33331

llvm-svn: 303379

7 years agoSet IMAGE_DLL_CHARACTERISTICS_NO_BIND.
Rui Ueyama [Thu, 18 May 2017 20:26:58 +0000 (20:26 +0000)]
Set IMAGE_DLL_CHARACTERISTICS_NO_BIND.

Our output is not compatible with the Binding feature, so make it
explicit that.

Differential Revision: https://reviews.llvm.org/D33336

llvm-svn: 303378

7 years agoRevert r303375 "LLVM_FALLTHROUGH instead of fall-through comment."
Rui Ueyama [Thu, 18 May 2017 20:18:24 +0000 (20:18 +0000)]
Revert r303375 "LLVM_FALLTHROUGH instead of fall-through comment."

This reverts commit r303375 since it didn't compile.

llvm-svn: 303377

7 years ago[clangd] Make formatting.test asynchronous again.
Benjamin Kramer [Thu, 18 May 2017 20:10:04 +0000 (20:10 +0000)]
[clangd] Make formatting.test asynchronous again.

This test doesn't rely on the order of asynchronous messages, enable
threads so we have at least some coverage for those code paths.

llvm-svn: 303376

7 years agoLLVM_FALLTHROUGH instead of fall-through comment.
Galina Kistanova [Thu, 18 May 2017 20:01:52 +0000 (20:01 +0000)]
LLVM_FALLTHROUGH instead of fall-through comment.

llvm-svn: 303375

7 years agoMerge IAT and ILT.
Rui Ueyama [Thu, 18 May 2017 19:59:22 +0000 (19:59 +0000)]
Merge IAT and ILT.

Previously, LLD-produced executables had IAT (Import Address Table) and
ILT (Import Lookup Table) as separate chunks of data, although their
contents are identical. My interpretation of the COFF spec when I wrote
the COFF linker is that they need to be separate tables even though they
are the same.

But Peter found that the Windows loader is fine with executables in
which IAT and ILT are merged. This is a patch to merge IAT and ILT.
I confirmed that an lld-link self-hosted with this patch works fine.

Fixes https://bugs.llvm.org/show_bug.cgi?id=33064

Differential Revision: https://reviews.llvm.org/D33326

llvm-svn: 303374

7 years agoWhen we enter a module within a linkage specification, switch the linkage
Richard Smith [Thu, 18 May 2017 19:34:55 +0000 (19:34 +0000)]
When we enter a module within a linkage specification, switch the linkage
specification and the TU to the new module.

This is necessary to get the module ownership correct for entities that we
temporarily hang off the TranslationUnitDecl, such as template parameters and
function parameters.

llvm-svn: 303373

7 years agoFix the location of "missing ';'" suggestions after annotation tokens.
Richard Smith [Thu, 18 May 2017 19:21:48 +0000 (19:21 +0000)]
Fix the location of "missing ';'" suggestions after annotation tokens.

We were incorrectly setting PrevTokLocation to the first token in the
annotation token instead of the last when consuming it. To fix this without
adding a complex switch to the hot path through ConsumeToken, we now have a
ConsumeAnnotationToken function for consuming annotation tokens in addition
to the other Consume*Token special case functions.

llvm-svn: 303372

7 years agoCOFF: Replace DLLNames maps with vectors.
Peter Collingbourne [Thu, 18 May 2017 18:53:39 +0000 (18:53 +0000)]
COFF: Replace DLLNames maps with vectors.

The import lists are already binned by DLL name, so there's no need to
deduplicate here.

Differential Revision: https://reviews.llvm.org/D33330

llvm-svn: 303371

7 years agoCodeGen: Cast alloca to expected address space
Yaxun Liu [Thu, 18 May 2017 18:51:09 +0000 (18:51 +0000)]
CodeGen: Cast alloca to expected address space

Alloca always returns a pointer in alloca address space, which may
be different from the type defined by the language. For example,
in C++ the auto variables are in the default address space. Therefore
cast alloca to the expected address space when necessary.

Differential Revision: https://reviews.llvm.org/D32248

llvm-svn: 303370

7 years agoRevert r302938 "Add LiveRangeShrink pass to shrink live range within BB."
Hans Wennborg [Thu, 18 May 2017 18:50:05 +0000 (18:50 +0000)]
Revert r302938 "Add LiveRangeShrink pass to shrink live range within BB."

This also reverts follow-ups r303292 and r303298.

It broke some Chromium tests under MSan, and apparently also internal
tests at Google.

llvm-svn: 303369

7 years agoUse existing helper. NFC.
Rafael Espindola [Thu, 18 May 2017 18:24:11 +0000 (18:24 +0000)]
Use existing helper. NFC.

llvm-svn: 303368

7 years ago[InstSimplify] Add test cases for missing fold (A & B) | ~(A ^ B) -> ~(A ^ B).
Craig Topper [Thu, 18 May 2017 18:14:40 +0000 (18:14 +0000)]
[InstSimplify] Add test cases for missing fold (A & B) | ~(A ^ B) -> ~(A ^ B).

llvm-svn: 303367

7 years ago[index] Avoid one more crash caused by infinite recursion that happens when
Alex Lorenz [Thu, 18 May 2017 18:06:07 +0000 (18:06 +0000)]
[index] Avoid one more crash caused by infinite recursion that happens when
looking up a dependent name in a record that derives from itself

rdar://32273000

Differential Revision: https://reviews.llvm.org/D33324

llvm-svn: 303366

7 years agoReduce gcc-7 warnings by fall-through comments.
Galina Kistanova [Thu, 18 May 2017 17:53:47 +0000 (17:53 +0000)]
Reduce gcc-7 warnings by fall-through comments.

llvm-svn: 303365

7 years ago[InstCombine] move test and use better checks; NFC
Sanjay Patel [Thu, 18 May 2017 17:48:07 +0000 (17:48 +0000)]
[InstCombine] move test and use better checks; NFC

Previously, this test was checking for 'or i1', but that was actually matched by 'xor i1'.

llvm-svn: 303364

7 years agoUse a forward declaration. NFC.
Rafael Espindola [Thu, 18 May 2017 17:26:00 +0000 (17:26 +0000)]
Use a forward declaration. NFC.

llvm-svn: 303363

7 years ago[IR] De-virtualize ~Value to save a vptr
Reid Kleckner [Thu, 18 May 2017 17:24:10 +0000 (17:24 +0000)]
[IR] De-virtualize ~Value to save a vptr

Summary:
Implements PR889

Removing the virtual table pointer from Value saves 1% of RSS when doing
LTO of llc on Linux. The impact on time was positive, but too noisy to
conclusively say that performance improved. Here is a link to the
spreadsheet with the original data:

https://docs.google.com/spreadsheets/d/1F4FHir0qYnV0MEp2sYYp_BuvnJgWlWPhWOwZ6LbW7W4/edit?usp=sharing

This change makes it invalid to directly delete a Value, User, or
Instruction pointer. Instead, such code can be rewritten to a null check
and a call Value::deleteValue(). Value objects tend to have their
lifetimes managed through iplist, so for the most part, this isn't a big
deal.  However, there are some places where LLVM deletes values, and
those places had to be migrated to deleteValue.  I have also created
llvm::unique_value, which has a custom deleter, so it can be used in
place of std::unique_ptr<Value>.

I had to add the "DerivedUser" Deleter escape hatch for MemorySSA, which
derives from User outside of lib/IR. Code in IR cannot include MemorySSA
headers or call the MemoryAccess object destructors without introducing
a circular dependency, so we need some level of indirection.
Unfortunately, no class derived from User may have any virtual methods,
because adding a virtual method would break User::getHungOffOperands(),
which assumes that it can find the use list immediately prior to the
User object. I've added a static_assert to the appropriate OperandTraits
templates to help people avoid this trap.

Reviewers: chandlerc, mehdi_amini, pete, dberlin, george.burgess.iv

Reviewed By: chandlerc

Subscribers: krytarowski, eraman, george.burgess.iv, mzolotukhin, Prazek, nlewycky, hans, inglorion, pcc, tejohnson, dberlin, llvm-commits

Differential Revision: https://reviews.llvm.org/D31261

llvm-svn: 303362

7 years ago[LSR] Call canonicalize after we generate a new Formula in GenerateTruncates. Fix...
Wei Mi [Thu, 18 May 2017 17:21:22 +0000 (17:21 +0000)]
[LSR] Call canonicalize after we generate a new Formula in GenerateTruncates. Fix PR33077.

The testcase in PR33077 generates a LSR Use Formula with two SCEVAddRecExprs for the same
loop. Such uncommon formula will become non-canonical after GenerateTruncates adds sign
extension to the ScaledReg of the Formula, and it will break the assertion that every
Formula to be inserted is canonical.

The fix is to call canonicalize for the raw Formula generated by GenerateTruncates
before inserting it.

llvm-svn: 303361

7 years ago[LegacyPassManager] Remove TargetMachine constructors
Francis Visoiu Mistrih [Thu, 18 May 2017 17:21:13 +0000 (17:21 +0000)]
[LegacyPassManager] Remove TargetMachine constructors

This provides a new way to access the TargetMachine through
TargetPassConfig, as a dependency.

The patterns replaced here are:

* Passes handling a null TargetMachine call
  `getAnalysisIfAvailable<TargetPassConfig>`.

* Passes not handling a null TargetMachine
  `addRequired<TargetPassConfig>` and call
  `getAnalysis<TargetPassConfig>`.

* MachineFunctionPasses now use MF.getTarget().

* Remove all the TargetMachine constructors.
* Remove INITIALIZE_TM_PASS.

This fixes a crash when running `llc -start-before prologepilog`.

PEI needs StackProtector, which gets constructed without a TargetMachine
by the pass manager. The StackProtector pass doesn't handle the case
where there is no TargetMachine, so it segfaults.

Related to PR30324.

Differential Revision: https://reviews.llvm.org/D33222

llvm-svn: 303360

7 years ago[git-llvm] Don't attempt to propget files that don't exist yet in SVN
Reid Kleckner [Thu, 18 May 2017 17:17:17 +0000 (17:17 +0000)]
[git-llvm] Don't attempt to propget files that don't exist yet in SVN

svn propget will fail halfway through, and the patch will fail to apply.

llvm-svn: 303359

7 years ago[CodeGen] Propagate LValueBaseInfo instead of AlignmentSource
Krzysztof Parzyszek [Thu, 18 May 2017 17:07:11 +0000 (17:07 +0000)]
[CodeGen] Propagate LValueBaseInfo instead of AlignmentSource

The functions creating LValues propagated information about alignment
source. Extend the propagated data to also include information about
possible unrestricted aliasing. A new class LValueBaseInfo will
contain both AlignmentSource and MayAlias info.

This patch should not introduce any functional changes.

Differential Revision: https://reviews.llvm.org/D33284

llvm-svn: 303358

7 years agoUse make<> everywhere in COFF to make it consistent with ELF.
Rui Ueyama [Thu, 18 May 2017 17:03:49 +0000 (17:03 +0000)]
Use make<> everywhere in COFF to make it consistent with ELF.

We've been using make<> to allocate new objects in ELF. We have
the same function in COFF, but we didn't use it widely due to
negligence. This patch uses the function in COFF to close the gap
between ELF and COFF.

llvm-svn: 303357

7 years ago[Fortran Support] Change "global" pattern match to work for params
Siddharth Bhat [Thu, 18 May 2017 16:47:13 +0000 (16:47 +0000)]
[Fortran Support] Change "global" pattern match to work for params

Summary:
- Rename global / local naming convention that did not make much sense
to Visible / Invisible, where the visible refers to whether the ALLOCATE
call to the Fortran array is present in the current module or not.

- This match now works on both cross fortran module globals and on
parameters to functions since neither of them are necessarily allocated
at the point of their usage.

- Add testcase that matches against both a load and a store against
function parameters.

Differential Revision: https://reviews.llvm.org/D33190

llvm-svn: 303356

7 years agoAlternative way to detemplate GotSection.
Rafael Espindola [Thu, 18 May 2017 16:45:36 +0000 (16:45 +0000)]
Alternative way to detemplate GotSection.

GetSection is a template because write calls relocate.

relocate has two parts. The non alloc code really has to be a
template, as it is looking a raw input file data.

The alloc part is only a template because of getSize.

This patch folds the value of getSize early, detemplates
getRelocTargetVA and splits relocate into a templated non alloc case
and a regular function for the alloc case. This has the nice advantage
of making sure we collect all the information we need for relocations
before getting to InputSection::relocateNonAlloc.

Since we know got is alloc, it can just call the function directly and
avoid the template.

llvm-svn: 303355

7 years agoFix flag to start with 1 << 0. NFC.
Rafael Espindola [Thu, 18 May 2017 16:20:12 +0000 (16:20 +0000)]
Fix flag to start with 1 << 0. NFC.

Thanks to Andrew Ng for noticing it.

llvm-svn: 303354

7 years ago[clang-format] Fix MatchingOpeningBlockLineIndex computation
Krasimir Georgiev [Thu, 18 May 2017 15:16:24 +0000 (15:16 +0000)]
[clang-format] Fix MatchingOpeningBlockLineIndex computation

Summary:
Computed line index must be relative to the current 'parent' node, and
thus use CurrentLines instead of Lines.

Without this, a child line's MatchingOpeningBlockLineIndex is out of
range of the parent's list of line, which can cause crash or unexpected
behavior if this field is used in childs.

Contributed by @Typz!

Reviewers: krasimir, djasper

Reviewed By: krasimir

Subscribers: cfe-commits, klimek

Differential Revision: https://reviews.llvm.org/D32524

llvm-svn: 303353

7 years ago[COFF] Fix interaction between /DEBUG and /PDB
Zachary Turner [Thu, 18 May 2017 15:15:10 +0000 (15:15 +0000)]
[COFF] Fix interaction between /DEBUG and /PDB

When /DEBUG is not specified, /PDB should be ignored.  When
/DEBUG is specified, a PDB should be output regardless of
whether or not /PDB is specified.  /PDB just overrides the
default name.

This patch implements this behavior, and adds some tests, while
also removing a dead option /DEBUGPDB which was unused in any
code.

Differential Revision: https://reviews.llvm.org/D33302

llvm-svn: 303352

7 years agoFix some minor issues in PDB parsing library.
Zachary Turner [Thu, 18 May 2017 15:14:44 +0000 (15:14 +0000)]
Fix some minor issues in PDB parsing library.

1) Until now I'd never seen a valid PDB where the DBI stream and
   the PDB Stream disagreed on the "Age" field.  Because of that,
   we had code to assert that they matched.  Recently though I was
   given a PDB where they disagreed, so this assumption has proven
   to be incorrect.  Remove this check.

2) We were walking the entire list of hash values for types up front
   and then throwing away the values.  For large PDBs this was a
   significant slow down.  Remove this.

With this patch, I can dump the list of all compilands from a
1.5GB PDB file in just a few seconds.

llvm-svn: 303351

7 years ago[compiler-rt][cmake] Build unit tests conditionally with _FILE_OFFSET_BITS=64 and...
Simon Dardis [Thu, 18 May 2017 13:19:35 +0000 (13:19 +0000)]
[compiler-rt][cmake] Build unit tests conditionally with _FILE_OFFSET_BITS=64 and _LARGEFILE_SOURCE

The sanitizer library unit tests for libc can get a different definition
of 'struct stat' to what the sanitizer library is built with for certain
targets.

For MIPS the size element of 'struct stat' is after a macro guarded
explicit padding element.

This patch resolves any possible inconsistency by adding the same
_FILE_OFFSET_BITS=64 and _LARGE_SOURCE with the same
conditions as the sanitizer library to the build flags for the unit tests.

This resolves a recurring build failure on the MIPS buildbots due to
'struct stat' defintion differences.

Reviewers: slthakur

Differential Revision: https://reviews.llvm.org/D33131

llvm-svn: 303350

7 years ago[JumpThreading] Dont RAUW condition incorrectly
Anna Thomas [Thu, 18 May 2017 13:12:18 +0000 (13:12 +0000)]
[JumpThreading] Dont RAUW condition incorrectly

Summary:
We have a bug when RAUWing the condition if experimental.guard or assumes is a use of that
condition. This is because LazyValueInfo may have used the guards/assumes to identify the
value of the condition at the end of the block. RAUW replaces the uses
at the guard/assume as well as uses before the guard/assume. Both of
these are incorrect.
For now, disable RAUW for conditions and fix the logic as a next
step: https://reviews.llvm.org/D33257

Reviewers: sanjoy, reames, trentxintong

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33279

llvm-svn: 303349

7 years agoAdd Status -- llvm::Error glue
Pavel Labath [Thu, 18 May 2017 12:46:50 +0000 (12:46 +0000)]
Add Status -- llvm::Error glue

Summary:
This adds functions to convert between llvm::Error and Status classes.
Posix errors in Status are represented as llvm::ECError, and the rest as
llvm::StringError.

For the conversion from Error to Status, ECError is again represented as
a posix error in Status, while other errors are stored as generic errors
and only the string value is preserved.

Reviewers: zturner, jingham

Subscribers: lldb-commits

Differential Revision: https://reviews.llvm.org/D33241

llvm-svn: 303348

7 years ago[AMDGPU] SDWA operands should not intersect with potential MIs
Sam Kolton [Thu, 18 May 2017 12:12:03 +0000 (12:12 +0000)]
[AMDGPU] SDWA operands should not intersect with potential MIs

Summary:
There should be no intesection between SDWA operands and potential MIs. E.g.:
```
v_and_b32 v0, 0xff, v1 -> src:v1 sel:BYTE_0
v_and_b32 v2, 0xff, v0 -> src:v0 sel:BYTE_0
v_add_u32 v3, v4, v2
```
In that example it is possible that we would fold 2nd instruction into 3rd (v_add_u32_sdwa) and then try to fold 1st instruction into 2nd (that was already destroyed). So if SDWAOperand is also a potential MI then do not apply it.

Reviewers: vpykhtin, arsenm

Subscribers: kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye

Differential Revision: https://reviews.llvm.org/D32804

llvm-svn: 303347

7 years ago[MVT] add v1i1 MVT
Guy Blank [Thu, 18 May 2017 11:29:41 +0000 (11:29 +0000)]
[MVT] add v1i1 MVT

Adds the v1i1 MVT as a preparation for another commit (https://reviews.llvm.org/D32273)

Differential Revision: https://reviews.llvm.org/D32540

llvm-svn: 303346

7 years ago[GlobalISel][X86] G_ADD/G_SUB vector legalizer/selector support.
Igor Breger [Thu, 18 May 2017 11:10:56 +0000 (11:10 +0000)]
[GlobalISel][X86] G_ADD/G_SUB vector legalizer/selector support.

Summary: G_ADD/G_SUB vector legalizer/selector support.

Reviewers: zvi, guyblank

Reviewed By: guyblank

Subscribers: rovka, llvm-commits, kristof.beyls

Differential Revision: https://reviews.llvm.org/D33232

llvm-svn: 303345

7 years agoFix 'not all control paths return a value' warning on windows buildbots.
Simon Pilgrim [Thu, 18 May 2017 10:48:23 +0000 (10:48 +0000)]
Fix 'not all control paths return a value' warning on windows buildbots.

llvm-svn: 303344

7 years ago[index] Record references to class receivers used in property references
Alex Lorenz [Thu, 18 May 2017 10:43:11 +0000 (10:43 +0000)]
[index] Record references to class receivers used in property references

rdar://32250025

llvm-svn: 303343

7 years ago[X86][AVX512] Add 512-bit vector ctpop costs + tests
Simon Pilgrim [Thu, 18 May 2017 10:42:34 +0000 (10:42 +0000)]
[X86][AVX512] Add 512-bit vector ctpop costs + tests

llvm-svn: 303342

7 years agoRe-commit: [globalisel][tablegen] Import rules containing intrinsic_wo_chain.
Daniel Sanders [Thu, 18 May 2017 10:33:36 +0000 (10:33 +0000)]
Re-commit: [globalisel][tablegen] Import rules containing intrinsic_wo_chain.

Summary:
As of this patch, 1018 out of 3938 rules are currently imported.

Depends on D32275

Reviewers: qcolombet, kristof.beyls, rovka, t.p.northover, ab, aditya_nandakumar

Reviewed By: qcolombet

Subscribers: dberris, igorb, llvm-commits

Differential Revision: https://reviews.llvm.org/D32278

The previous commit failed on test-suite/Bitcode/simd_ops/AArch64_halide_runtime.bc
because isImmOperandEqual() assumed MO was a register operand and that's not
always true.

llvm-svn: 303341

7 years ago[X86] Add explicit triple to test invocation
Zvi Rackover [Thu, 18 May 2017 09:32:56 +0000 (09:32 +0000)]
[X86] Add explicit triple to test invocation

llvm-svn: 303340

7 years agoRevert r302781 and subsequent attempts to disable part of it.
Daniel Jasper [Thu, 18 May 2017 09:31:37 +0000 (09:31 +0000)]
Revert r302781 and subsequent attempts to disable part of it.

The Msan unit tests are still broken and by this point, I think we
should start over.

llvm-svn: 303339

7 years agoFixup r303324 - temporary disable stndup interceptor, due to r302781 being buggy
Diana Picus [Thu, 18 May 2017 09:12:27 +0000 (09:12 +0000)]
Fixup r303324 - temporary disable stndup interceptor, due to r302781 being buggy

r303324 missed one of the tests added by r302781. This commit applies
the same fix as r303324 to the missed test (strndup.cc).

llvm-svn: 303338

7 years ago[ELF] Support R_ARM_SBREL32 Relocation
Peter Smith [Thu, 18 May 2017 09:12:21 +0000 (09:12 +0000)]
[ELF] Support R_ARM_SBREL32 Relocation

This change adds support for the R_ARM_SBREL32 relocation. The relocation
is a base relative relocation that is produced by clang/llvm when -frwpi
is used. The use case for the -frwpi option is position independent data
for embedded systems that do not have a GOT. With -frwpi all data is
accessed via an offset from a base register (usually r9), where r9 is set
at run time to where the data has been loaded. The base of the data is
known as the static base.

The ARM ABI defines the static base as:
B(S) is the addressing origin of the output segment defining the symbol S.
The origin is not required to be the base address of the segment. For
simplicity we choose to use the base address of the segment.

The ARM procedure call standard only defines a read write variant using
R_ARM_SBREL32 relocations. The read-only data is accessed via pc-relative
offsets from the code, this is implemented in clang as -fropi.

Fixes PR32924

Differential Revision: https://reviews.llvm.org/D33280

llvm-svn: 303337

7 years ago[SCEV][NFC] Remove duplication of isLoopInvariant code
Max Kazantsev [Thu, 18 May 2017 08:26:41 +0000 (08:26 +0000)]
[SCEV][NFC] Remove duplication of isLoopInvariant code

Replace two places that duplicate the code of isLoopInvariant method with
the invocation of this method.

Differential Revision: https://reviews.llvm.org/D33313

llvm-svn: 303336

7 years ago[DWARF] - Simplify RelocVisitor implementation.
George Rimar [Thu, 18 May 2017 08:25:11 +0000 (08:25 +0000)]
[DWARF] - Simplify RelocVisitor implementation.

We do not need to store relocation width field.
Patch removes relative code, that simplifies implementation.

Differential revision: https://reviews.llvm.org/D33274

llvm-svn: 303335

7 years ago[lib/Object] - Fix build bot after r303331 "[lib/Object] - Minor API update for llvm...
George Rimar [Thu, 18 May 2017 08:14:43 +0000 (08:14 +0000)]
[lib/Object] - Fix build bot after r303331 "[lib/Object] - Minor API update for llvm::Decompressor".

Error was:
Decompressor.h:33:28: error: extra qualification â€˜llvm::object::Decompressor::’ on member â€˜resizeAndDecompress’ [-fpermissive]
   template <class T> Error Decompressor::resizeAndDecompress(T &Out) {

llvm-svn: 303334

7 years ago[X86] Replace slow LEA instructions in X86
Lama Saba [Thu, 18 May 2017 08:11:50 +0000 (08:11 +0000)]
[X86] Replace slow LEA instructions in X86

  According to Intel's Optimization Reference Manual for SNB+:
  " For LEA instructions with three source operands and some specific situations, instruction latency has increased to 3 cycles, and must
    dispatch via port 1:
  - LEA that has all three source operands: base, index, and offset
  - LEA that uses base and index registers where the base is EBP, RBP,or R13
  - LEA that uses RIP relative addressing mode
  - LEA that uses 16-bit addressing mode "
  This patch currently handles the first 2 cases only.

Differential Revision: https://reviews.llvm.org/D32277

llvm-svn: 303333

7 years ago[clang-format] Make NoLineBreakFormatter respect MustBreakBefore
Krasimir Georgiev [Thu, 18 May 2017 08:07:52 +0000 (08:07 +0000)]
[clang-format] Make NoLineBreakFormatter respect MustBreakBefore

Summary:
This patch makes NoLineBreakFormatter to insert a break before tokens where
MustBreakBefore is true.

Reviewers: djasper

Reviewed By: djasper

Subscribers: cfe-commits, klimek

Differential Revision: https://reviews.llvm.org/D33238

llvm-svn: 303332

7 years ago[lib/Object] - Minor API update for llvm::Decompressor.
George Rimar [Thu, 18 May 2017 08:00:01 +0000 (08:00 +0000)]
[lib/Object] - Minor API update for llvm::Decompressor.

I revisited Decompressor API (issue with it was triggered during D32865 review)
and found it is probably provides more then we really need.

Issue was about next method's signature:

Error decompress(SmallString<32> &Out);
It is too strict. At first I wanted to change it to decompress(SmallVectorImpl<char> &Out),
but then found it is still not flexible because sticks to SmallVector.

During reviews was suggested to use templating to simplify code. Patch do that.

Differential revision: https://reviews.llvm.org/D33200

llvm-svn: 303331

7 years agoclang-format: fix prefix for doxygen comments after member
Krasimir Georgiev [Thu, 18 May 2017 07:36:21 +0000 (07:36 +0000)]
clang-format: fix prefix for doxygen comments after member

Summary:
Doxygen supports putting documentation blocks after member, by adding
an additional < marker in the comment block. This patch makes sure
this marker is used in lines which are introduced by breaking the
comment.

  int foo; ///< Some very long comment.

becomes:

  int foo; ///< Some very long
           ///< comment.

Contributed by @Typz!

Reviewers: krasimir

Reviewed By: krasimir

Subscribers: djasper, klimek, cfe-commits

Differential Revision: https://reviews.llvm.org/D33282

llvm-svn: 303330

7 years agoFix buildbot failure after rL303327: [BPI] Reduce the probability of unreachable...
Serguei Katkov [Thu, 18 May 2017 07:20:52 +0000 (07:20 +0000)]
Fix buildbot failure after rL303327: [BPI] Reduce the probability of unreachable edge to minimal value greater than 0.

One more test is updated to meet new branch probability for unreachable branches.

llvm-svn: 303329

7 years ago[X86] Adding tests for scalar bitcasts from vsetcc. NFC.
Zvi Rackover [Thu, 18 May 2017 07:04:48 +0000 (07:04 +0000)]
[X86] Adding tests for scalar bitcasts from vsetcc. NFC.

llvm-svn: 303328

7 years ago[BPI] Reduce the probability of unreachable edge to minimal value greater than 0
Serguei Katkov [Thu, 18 May 2017 06:11:56 +0000 (06:11 +0000)]
[BPI] Reduce the probability of unreachable edge to minimal value greater than 0

The probability of edge coming to unreachable block should be as low as possible.
The change reduces the probability to minimal value greater than zero.

The bug https://bugs.llvm.org/show_bug.cgi?id=32214 show the example when
the probability of edge coming to unreachable block is greater than for edge
coming to out of the loop and it causes incorrect loop rotation.

Please note that with this change the behavior of unreachable heuristic is a bit different
than others. Specifically, before this change the sum of probabilities
coming to unreachable blocks have the same weight for all branches
(it was just split over all edges of this block coming to unreachable blocks).
With this change it might be slightly different but not to much due to probability of
taken branch to unreachable block is really small.

Reviewers: chandlerc, sanjoy, vsk, congh, junbuml, davidxl, dexonsmith
Reviewed By: chandlerc, dexonsmith
Subscribers: reames, llvm-commits
Differential Revision: https://reviews.llvm.org/D30633

llvm-svn: 303327

7 years ago[ThinLTO] Do not assert when adding a module with a different but
Akira Hatanaka [Thu, 18 May 2017 03:52:29 +0000 (03:52 +0000)]
[ThinLTO] Do not assert when adding a module with a different but
compatible target triple

Currently, an assertion fails in ThinLTOCodeGenerator::addModule when
the target triple of the module being added doesn't match that of the
one stored in TMBuilder. This patch relaxes the constraint and makes
changes to allow target triples that only differ in their version
numbers on Apple platforms, similarly to what r228999 did.

rdar://problem/30133904

Differential Revision: https://reviews.llvm.org/D33291

llvm-svn: 303326

7 years agoFix an assertion failure in FormatASTNodeDiagnosticArgument.
Alexander Kornienko [Thu, 18 May 2017 03:02:15 +0000 (03:02 +0000)]
Fix an assertion failure in FormatASTNodeDiagnosticArgument.

Summary:
The test being added in this patch used to cause an assertion failure:

/build/./bin/clang -cc1 -internal-isystem /build/lib/clang/5.0.0/include -nostdsysteminc -verify -fsyntax-only -std=c++11 -Wshadow-all /src/tools/clang/test/SemaCXX/warn-shadow.cpp
--
Exit Code: 134

Command Output (stderr):
--
clang: /src/tools/clang/lib/AST/ASTDiagnostic.cpp:424: void clang::FormatASTNodeDiagnosticArgument(DiagnosticsEngine::ArgumentKind, intptr_t, llvm::StringRef, llvm::StringRef, ArrayRef<DiagnosticsEngine::ArgumentValue>, SmallVectorImpl<char> &, void *, ArrayRef<intptr_t>): Assertion `isa<NamedDecl>(DC) && "Expected a NamedDecl"' failed.
#0 0x0000000001c7a1b4 PrintStackTraceSignalHandler(void*) (/build/./bin/clang+0x1c7a1b4)
#1 0x0000000001c7a4e6 SignalHandler(int) (/build/./bin/clang+0x1c7a4e6)
#2 0x00007f30880078d0 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0xf8d0)
#3 0x00007f3087054067 gsignal (/lib/x86_64-linux-gnu/libc.so.6+0x35067)
#4 0x00007f3087055448 abort (/lib/x86_64-linux-gnu/libc.so.6+0x36448)
#5 0x00007f308704d266 (/lib/x86_64-linux-gnu/libc.so.6+0x2e266)
#6 0x00007f308704d312 (/lib/x86_64-linux-gnu/libc.so.6+0x2e312)
#7 0x00000000035b7f22 clang::FormatASTNodeDiagnosticArgument(clang::DiagnosticsEngine::ArgumentKind, long, llvm::StringRef, llvm::StringRef, llvm::ArrayRef<std::pair<clang::DiagnosticsEngine::ArgumentKind, long> >, llvm::SmallVectorImpl<char>&, void*, llvm::ArrayRef<long>) (/build/
./bin/clang+0x35b7f22)
#8 0x0000000001ddbae4 clang::Diagnostic::FormatDiagnostic(char const*, char const*, llvm::SmallVectorImpl<char>&) const (/build/./bin/clang+0x1ddbae4)
#9 0x0000000001ddb323 clang::Diagnostic::FormatDiagnostic(char const*, char const*, llvm::SmallVectorImpl<char>&) const (/build/./bin/clang+0x1ddb323)
#10 0x00000000022878a4 clang::TextDiagnosticBuffer::HandleDiagnostic(clang::DiagnosticsEngine::Level, clang::Diagnostic const&) (/build/./bin/clang+0x22878a4)
#11 0x0000000001ddf387 clang::DiagnosticIDs::ProcessDiag(clang::DiagnosticsEngine&) const (/build/./bin/clang+0x1ddf387)
#12 0x0000000001dd9dea clang::DiagnosticsEngine::EmitCurrentDiagnostic(bool) (/build/./bin/clang+0x1dd9dea)
#13 0x0000000002cad00c clang::Sema::EmitCurrentDiagnostic(unsigned int) (/build/./bin/clang+0x2cad00c)
#14 0x0000000002d91cd2 clang::Sema::CheckShadow(clang::NamedDecl*, clang::NamedDecl*, clang::LookupResult const&) (/build/./bin/clang+0x2d91cd2)

Stack dump:
0.      Program arguments: /build/./bin/clang -cc1 -internal-isystem /build/lib/clang/5.0.0/include -nostdsysteminc -verify -fsyntax-only -std=c++11 -Wshadow-all /src/tools/clang/test/SemaCXX/warn-shadow.cpp
1.      /src/tools/clang/test/SemaCXX/warn-shadow.cpp:214:23: current parser token ';'
2.      /src/tools/clang/test/SemaCXX/warn-shadow.cpp:213:26: parsing function body 'handleLinkageSpec'
3.      /src/tools/clang/test/SemaCXX/warn-shadow.cpp:213:26: in compound statement ('{}')
/build/tools/clang/test/SemaCXX/Output/warn-shadow.cpp.script: line 1: 15595 Aborted                 (core dumped) /build/./bin/clang -cc1 -internal-isystem /build/lib/clang/5.0.0/include -nostdsysteminc -verify -fsyntax-only -std=c++11 -Wshadow-all /src/tools/clang/test/SemaCXX/warn-shadow.cpp

Reviewers: rsmith

Reviewed By: rsmith

Subscribers: krytarowski, cfe-commits

Differential Revision: https://reviews.llvm.org/D33207

llvm-svn: 303325

7 years agotemporary disable stndup interceptor, due to r302781 being buggy
Kostya Serebryany [Thu, 18 May 2017 03:00:07 +0000 (03:00 +0000)]
temporary disable stndup interceptor, due to r302781 being buggy

llvm-svn: 303324

7 years ago[Target/X86] Remove unneeded return. NFCI.
Davide Italiano [Thu, 18 May 2017 02:36:42 +0000 (02:36 +0000)]
[Target/X86] Remove unneeded return. NFCI.

llvm-svn: 303323

7 years ago[modules] Switch from inferring owning modules based on source location to
Richard Smith [Thu, 18 May 2017 02:29:20 +0000 (02:29 +0000)]
[modules] Switch from inferring owning modules based on source location to
inferring based on the current module at the point of creation.

This should result in no functional change except when building a preprocessed
module (or more generally when using #pragma clang module begin/end to switch
module in the middle of a file), in which case it allows us to correctly track
the owning module for declarations. We can't map from FileID to module in the
preprocessed module case, since all modules would have the same FileID.

There are still a couple of remaining places that try to infer a module from a
source location; I'll clean those up in follow-up changes.

llvm-svn: 303322

7 years ago[clang-tidy] Optimize GlobList::contains
Alexander Kornienko [Thu, 18 May 2017 01:13:51 +0000 (01:13 +0000)]
[clang-tidy] Optimize GlobList::contains

With large lists of checks and large number of warnings GlobList::contains
starts being ridiculously CPU hungry, since it runs regexp match per glob.
Caching results of glob matching in a StringMap significantly speeds up check
filtering even for small GlobLists.

/tmp/q.cc:

void f() {
  int I;
  {int I;}
  {int I;}
  {int I;}
  ... // 200k times
}

Before the patch:

GlobList with 2 entries:
  $ time clang-tidy-old -checks=-*,modernize-use-override /tmp/q.cc -- -Wshadow
  200000 warnings generated.
  Suppressed 200000 warnings (200000 with check filters).

  real    0m3.826s
  user    0m3.176s
  sys     0m0.504s

GlobList with 28 entries:
  $ time clang-tidy-old -checks=-*,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,modernize-use-override /tmp/q.cc -- -Wshadow
  200000 warnings generated.
  Suppressed 200000 warnings (200000 with check filters).

  real    0m5.000s
  user    0m4.744s
  sys     0m0.060s

GlobList with 158 entries:
  $ time clang-tidy-old -checks=-*,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,modernize-use-override /tmp/q.cc -- -Wshadow
  200000 warnings generated.
  Suppressed 200000 warnings (200000 with check filters).

  real    0m13.920s
  user    0m13.636s
  sys     0m0.104s

With the patch runtime is practically independent from the length of the GlobList:
  $ time clang-tidy-new -checks=-*,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,modernize-use-override /tmp/q.cc -- -Wshadow
  200000 warnings generated.
  Suppressed 200000 warnings (200000 with check filters).

  real    0m2.300s
  user    0m2.104s
  sys     0m0.044s

llvm-svn: 303321

7 years ago[Statistics] Use the new Statistic::updateMax to atomically calculate a maximum value...
Craig Topper [Thu, 18 May 2017 01:11:52 +0000 (01:11 +0000)]
[Statistics] Use the new Statistic::updateMax to atomically calculate a maximum value statistic.

llvm-svn: 303320

7 years agoUpdate three tests I missed in r302979 and r302990
Justin Bogner [Thu, 18 May 2017 00:58:06 +0000 (00:58 +0000)]
Update three tests I missed in r302979 and r302990

llvm-svn: 303319

7 years ago[Statistics] Add a method to atomically update a statistic that contains a maximum
Craig Topper [Thu, 18 May 2017 00:51:39 +0000 (00:51 +0000)]
[Statistics] Add a method to atomically update a statistic that contains a maximum

Summary:
There are several places in the codebase that try to calculate a maximum value in a Statistic object. We currently do this in one of two ways:

  MaxNumFoo = std::max(MaxNumFoo, NumFoo);

or

  MaxNumFoo = (MaxNumFoo > NumFoo) ? MaxNumFoo : NumFoo;

The first version reads from MaxNumFoo one time and uncontionally rwrites to it. The second version possibly reads it twice depending on the result of the first compare.  But we have no way of knowing if the value was changed by another thread between the reads and the writes.

This patch adds a method to the Statistic object that can ensure that we only store if our value is the max and the previous max didn't change after we read it. If it changed we'll recheck if our value should still be the max or not and try again.

This spawned from an audit I'm trying to do of all places we uses the implicit conversion to unsigned on the Statistics objects. See my previous thread on llvm-dev https://groups.google.com/forum/#!topic/llvm-dev/yfvxiorKrDQ

Reviewers: dberlin, chandlerc, hfinkel, dblaikie

Reviewed By: chandlerc

Subscribers: llvm-commits, sanjoy

Differential Revision: https://reviews.llvm.org/D33301

llvm-svn: 303318

7 years agoThe constant expression evaluator should examine function arguments for non-constexpr...
Nick Lewycky [Wed, 17 May 2017 23:56:54 +0000 (23:56 +0000)]
The constant expression evaluator should examine function arguments for non-constexpr function calls unless the EvalInfo says to stop.

llvm-svn: 303317

7 years agoCodeGen: BlockPlacement: Add Message strings to asserts. NFC
Kyle Butt [Wed, 17 May 2017 23:44:41 +0000 (23:44 +0000)]
CodeGen: BlockPlacement: Add Message strings to asserts. NFC

Add message strings to all the unlabeled asserts in the file.

Differential Revision: https://reviews.llvm.org/D33078

llvm-svn: 303316

7 years ago[InstCombine] add test for xor-of-icmps; NFC
Sanjay Patel [Wed, 17 May 2017 23:22:52 +0000 (23:22 +0000)]
[InstCombine] add test for xor-of-icmps; NFC

This is another form of the problem discussed in D32143.

llvm-svn: 303315

7 years ago[Statistics] Use Statistic::operator+= instead of adding and assigning separately.
Craig Topper [Wed, 17 May 2017 23:22:10 +0000 (23:22 +0000)]
[Statistics] Use Statistic::operator+= instead of adding and assigning separately.

I believe this technically fixes a multithreaded race condition in this code. But my primary concern was as part of looking at removing the ability to treat Statistics like a plain unsigned. There are many weird operations on Statistics in the codebase.

llvm-svn: 303314

7 years agoRevert "[globalisel][tablegen] Import rules containing intrinsic_wo_chain."
Quentin Colombet [Wed, 17 May 2017 23:17:29 +0000 (23:17 +0000)]
Revert "[globalisel][tablegen] Import rules containing intrinsic_wo_chain."

This reverts commit r303259.

This breaks the GISel bot:
http://lab.llvm.org:8080/green/job/Compiler_Verifiers_GlobalISEL/5163/consoleFull#-134276167849ba4694-19c4-4d7e-bec5-911270d8a58c

llvm-svn: 303313

7 years ago[InstCombine] handle icmp i1 X, C early to avoid creating an unknown pattern
Sanjay Patel [Wed, 17 May 2017 22:29:40 +0000 (22:29 +0000)]
[InstCombine] handle icmp i1 X, C early to avoid creating an unknown pattern

The missing optimization for xor-of-icmps still needs to be added, but by
being more efficient (not generating unnecessary logic ops with constants)
we avoid the bug.

See discussion in post-commit comments:
https://reviews.llvm.org/D32143

llvm-svn: 303312

7 years agoAttempt to pacify ASan and UBSan reports in CrashRecovery tests
Reid Kleckner [Wed, 17 May 2017 22:23:20 +0000 (22:23 +0000)]
Attempt to pacify ASan and UBSan reports in CrashRecovery tests

llvm-svn: 303311

7 years ago[InstCombine] add test for missing icmp bool fold; NFC
Sanjay Patel [Wed, 17 May 2017 22:20:02 +0000 (22:20 +0000)]
[InstCombine] add test for missing icmp bool fold; NFC

llvm-svn: 303310

7 years ago[InstCombine] move icmp bool canonicalizations to helper; NFC
Sanjay Patel [Wed, 17 May 2017 22:15:07 +0000 (22:15 +0000)]
[InstCombine] move icmp bool canonicalizations to helper; NFC

As noted in the post-commit comments in D32143, we should be
catching the constant operand cases sooner to be more efficient
and less likely to expose a missing fold.

llvm-svn: 303309

7 years agoAMDGPU: Start defining a calling convention
Matt Arsenault [Wed, 17 May 2017 21:56:25 +0000 (21:56 +0000)]
AMDGPU: Start defining a calling convention

Partially implement callee-side for arguments and return values.
byval doesn't work properly, and most likely sret or other on-stack
return values most as well.

llvm-svn: 303308

7 years agoCodeGen: Power: Add lowering for shifts of v1i128.
Kyle Butt [Wed, 17 May 2017 21:54:41 +0000 (21:54 +0000)]
CodeGen: Power: Add lowering for shifts of v1i128.

When legalizing vector operations on vNi128, they will be split to v1i128
because that is a legal type on ppc64, but then the compiler will crash in
selection dag because it fails to select for these operations. This patch fixes
shift operations. Logical shift right and left shift can be performed in the
vector unit, but algebraic shift right requires being split.

Differential Revision: https://reviews.llvm.org/D32774

llvm-svn: 303307

7 years agoFix PR33028
Michael Liao [Wed, 17 May 2017 21:48:00 +0000 (21:48 +0000)]
Fix PR33028

- '-verify-mahcineinstrs' starts to complain allocatable live-in physical
  registers on non-entry or non-landing-pad basic blocks.
- Refactor the XBEGIN translation to define EAX on a dedicated fallback code
  path due to XABORT. Add a pseudo instruction to define EAX explicitly to
  avoid add physical register live-in.

Differential Revision: https://reviews.llvm.org/D33168

llvm-svn: 303306

7 years agoAMDGPU: Remove old intrinsic uses
Matt Arsenault [Wed, 17 May 2017 21:38:21 +0000 (21:38 +0000)]
AMDGPU: Remove old intrinsic uses

llvm-svn: 303305

7 years agoRe-submit r303225: Garbage collect dllimported symbols.
Rui Ueyama [Wed, 17 May 2017 21:36:08 +0000 (21:36 +0000)]
Re-submit r303225: Garbage collect dllimported symbols.

This reverts re-submits r303225 which was reverted in r303270 because it
broke the sanitizer-windows bot.

The reason of the failure is that we were writing dead symbols to the
symbol table. I fixed the issue.

llvm-svn: 303304

7 years agoAMDGPU: Expand frame indexes to be relative to scratch wave offset
Matt Arsenault [Wed, 17 May 2017 21:23:14 +0000 (21:23 +0000)]
AMDGPU: Expand frame indexes to be relative to scratch wave offset

In order for an arbitrary callee to access an object
in a caller's stack frame, the 32-bit offset used as
the private pointer needs to be relative to the kernel's
scratch wave offset register.

Convert to this by finding the difference from the current
stack frame and scaling by the wavefront size.

llvm-svn: 303303

7 years ago[XRay] Fix __xray_function_address on PPC reguarding local entry points.
Tim Shen [Wed, 17 May 2017 21:20:00 +0000 (21:20 +0000)]
[XRay] Fix __xray_function_address on PPC reguarding local entry points.

Reviewers: echristo, dberris

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33266

llvm-svn: 303302

7 years agoAMDGPU: Change mubuf soffset register when SP relative
Matt Arsenault [Wed, 17 May 2017 21:02:58 +0000 (21:02 +0000)]
AMDGPU: Change mubuf soffset register when SP relative

Check the MachinePointerInfo for whether the access is
supposed to be relative to the stack pointer.

No tests because this is used in later commits implementing
calls.

llvm-svn: 303301

7 years ago[X86][AVX512] Add 512-bit vector ctlz costs + tests
Simon Pilgrim [Wed, 17 May 2017 21:02:18 +0000 (21:02 +0000)]
[X86][AVX512] Add 512-bit vector ctlz costs + tests

llvm-svn: 303300

7 years ago[llvm-pdbdump] in yaml2pdb, generate default output filename if none given
Bob Haarman [Wed, 17 May 2017 20:46:48 +0000 (20:46 +0000)]
[llvm-pdbdump] in yaml2pdb, generate default output filename if none given

Summary:
llvm-pdbdump yaml2pdb used to fail with a misleading error
message ("An I/O error occurred on the file system") if no output file
was specified. This change adds an assert to PDBFileBuilder to check
that an output file name is specified, and makes llvm-pdbdump generate
an output file name based on the input file name if no output file
name is explicitly specified.

Reviewers: amccarth, zturner

Reviewed By: zturner

Subscribers: fhahn, llvm-commits

Differential Revision: https://reviews.llvm.org/D33296

llvm-svn: 303299

7 years agoupdate the test that should have been updated in r303292. (NFC)
Dehao Chen [Wed, 17 May 2017 20:44:08 +0000 (20:44 +0000)]
update the test that should have been updated in r303292. (NFC)

llvm-svn: 303298

7 years agoAdd some helpers for manipulating BinaryStreamRefs.
Zachary Turner [Wed, 17 May 2017 20:42:52 +0000 (20:42 +0000)]
Add some helpers for manipulating BinaryStreamRefs.

llvm-svn: 303297