platform/upstream/llvm.git
7 years ago[ELF] - Add support for locals list in version script.
George Rimar [Sat, 12 Nov 2016 07:04:15 +0000 (07:04 +0000)]
[ELF] - Add support for locals list in version script.

Previously we did not support anything except "local: *", patch changes that.

Actually GNU rules of proccessing wildcards are more complex than that (http://www.airs.com/blog/archives/300):
There are 2 iteration for wildcards, at first iteration "*" wildcards are ignored and handled at second iteration.

Since we previously decided not to implement such complex rules,
I suggest solution that is implemented in this patch. So for "local: *" case nothing changes,
but if we have wildcarded locals,
they are processed before wildcarded globals.

This should fix several FreeBSD ports, one of them is jpeg-turbo-1.5.1 and
currently blocks about 5k of ports.

Differential revision: https://reviews.llvm.org/D26395

llvm-svn: 286713

7 years ago[ELF] - Fixed comment. NFC.
George Rimar [Sat, 12 Nov 2016 06:27:42 +0000 (06:27 +0000)]
[ELF] - Fixed comment. NFC.

llvm-svn: 286712

7 years ago[AVX-512] Add unmasked version of shift by immediate and shift by single element...
Craig Topper [Sat, 12 Nov 2016 05:28:24 +0000 (05:28 +0000)]
[AVX-512] Add unmasked version of shift by immediate and shift by single element in XMM.

Summary:
This is the first step towards being able to add the avx512 shift by immediate intrinsics to InstCombineCalls where we aleady support the sse2 and avx2 intrinsics. We need to the unmasked versions so we can avoid having to teach InstCombineCalls that it would need to insert selects sometimes. Instead we'll just add the selects around the new instrinsics in the frontend.

This change should also enable the shift by i32 intrinsics to take a non-constant shift value just like the avx2 and sse intrinsics. This will enable us to fix PR30691 once we update clang.

Next I'll switch clang to use the new builtins. Then we'll come back to the backend and remove/autoupgrade the old intrinsics. Then I'll work on the same series for variable shifts.

Reviewers: RKSimon, zvi, delena

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D26333

llvm-svn: 286711

7 years agoCODE_OWNERS: Take ownership of the X86 backend.
Craig Topper [Sat, 12 Nov 2016 05:16:06 +0000 (05:16 +0000)]
CODE_OWNERS: Take ownership of the X86 backend.

llvm-svn: 286710

7 years ago[AVX-512] Add support for lowering shuffles to VALIGND/VALIGNQ
Craig Topper [Sat, 12 Nov 2016 05:05:27 +0000 (05:05 +0000)]
[AVX-512] Add support for lowering shuffles to VALIGND/VALIGNQ

Summary: VALIGND and VALIGNQ are similar to PALIGNR but instead of working on a 128-bit lane they work on the entire vector register. This change leverages the shuffle rotate detection code used for PALIGNR to detect these cases.

Reviewers: delena, RKSimon

Subscribers: Farhana, llvm-commits

Differential Revision: https://reviews.llvm.org/D26297

llvm-svn: 286709

7 years ago[asan] Fix strncmp and strncasecmp interceptors
Vitaly Buka [Sat, 12 Nov 2016 04:32:31 +0000 (04:32 +0000)]
[asan] Fix strncmp and strncasecmp interceptors

Summary:
In non-strict mode we will check memory access for both strings from beginning
to either:
  1. 0-char
  2. size
  3. different chars

In strict mode we will check from beginning to either:
  1. 0-char
  2. size

Previously in strict mode we always checked up to the 0-char.

Reviewers: kcc, eugenis

Subscribers: llvm-commits, kubabrecka

Differential Revision: https://reviews.llvm.org/D26574

llvm-svn: 286708

7 years agobuild: add a dependency on llvm-strings
Saleem Abdulrasool [Sat, 12 Nov 2016 03:45:21 +0000 (03:45 +0000)]
build: add a dependency on llvm-strings

Since we now have tests for llvm-strings, add a dependency on the tool.

llvm-svn: 286707

7 years agollvm-strings: ensure that the last string is correctly printed
Saleem Abdulrasool [Sat, 12 Nov 2016 03:39:21 +0000 (03:39 +0000)]
llvm-strings: ensure that the last string is correctly printed

We would ignore the last string that appeared if the file ended with a printable
character.  Ensure that we get the last string.

llvm-svn: 286706

7 years ago[OCaml] Adapt to the new attribute C API.
whitequark [Sat, 12 Nov 2016 03:38:30 +0000 (03:38 +0000)]
[OCaml] Adapt to the new attribute C API.

llvm-svn: 286705

7 years ago[C API] Fix several null pointer dereferences.
whitequark [Sat, 12 Nov 2016 03:38:23 +0000 (03:38 +0000)]
[C API] Fix several null pointer dereferences.

llvm-svn: 286704

7 years ago[libFuzzer] one more trophy
Kostya Serebryany [Sat, 12 Nov 2016 02:55:45 +0000 (02:55 +0000)]
[libFuzzer] one more trophy

llvm-svn: 286703

7 years ago[libFuzzer] use a valid ASCII string for a dummy seed corpus
Kostya Serebryany [Sat, 12 Nov 2016 02:27:21 +0000 (02:27 +0000)]
[libFuzzer] use a valid ASCII string for a dummy seed corpus

llvm-svn: 286702

7 years ago[ORC] Add a WrappedHandlerReturn type to map handler return types onto error
Lang Hames [Sat, 12 Nov 2016 02:19:31 +0000 (02:19 +0000)]
[ORC] Add a WrappedHandlerReturn type to map handler return types onto error
return types.

This class allows user provided handlers to return either error-wrapped types
or plain types. In the latter case, the plain type is wrapped with a success
value of Error or Expected<T> type to fit it into the rest of the serialization
machinery.

This patch allows us to remove the RPC unit-test workaround added in r286646.

llvm-svn: 286701

7 years ago[analyzer] Update 'Automated' to 'Automatic' from r286694.
Devin Coughlin [Sat, 12 Nov 2016 01:50:04 +0000 (01:50 +0000)]
[analyzer] Update 'Automated' to 'Automatic' from r286694.

ARC is 'Automatic Reference Counting' and not 'Automated Reference Counting'.

llvm-svn: 286700

7 years ago[c++1z] Support constant folding for __builtin_strchr and __builtin_memchr.
Richard Smith [Sat, 12 Nov 2016 01:39:56 +0000 (01:39 +0000)]
[c++1z] Support constant folding for __builtin_strchr and __builtin_memchr.

llvm-svn: 286699

7 years agoRemove dead code.
Rui Ueyama [Sat, 12 Nov 2016 01:32:34 +0000 (01:32 +0000)]
Remove dead code.

llvm-svn: 286698

7 years ago[analyzer] Fix copy-pasta in NullableReturnedFromNonnullChecker checker name.
Devin Coughlin [Sat, 12 Nov 2016 01:23:01 +0000 (01:23 +0000)]
[analyzer] Fix copy-pasta in NullableReturnedFromNonnullChecker checker name.

The name of the NullableReturnedFromNonnullChecker in Checkers.td
was accidentally "NullablePassedToNonnull", which made it impossible
to explicitly turn the checker on.

rdar://problem/28354459

llvm-svn: 286697

7 years agoAdd instructions in clang get_started page about working with a monorepo
Mehdi Amini [Sat, 12 Nov 2016 01:18:17 +0000 (01:18 +0000)]
Add instructions in clang get_started page about working with a monorepo

llvm-svn: 286696

7 years agoImprove `git llvm push` to suggest `git pull` when applying patch fails
Mehdi Amini [Sat, 12 Nov 2016 01:17:59 +0000 (01:17 +0000)]
Improve `git llvm push` to suggest `git pull` when applying patch fails

Differential Revision: https://reviews.llvm.org/D26565

llvm-svn: 286695

7 years ago[analyzer] Improve misleading RetainCountChcker diagnostic under ARC
Devin Coughlin [Sat, 12 Nov 2016 01:03:06 +0000 (01:03 +0000)]
[analyzer] Improve misleading RetainCountChcker diagnostic under ARC

Under automated reference counting the analyzer treats a methods -- even those
starting with  'copy' and friends -- as returning an unowned value. This is
because ownership of CoreFoundation objects must be transferred to ARC
with __bridge_transfer or CFBridgingRelease() before being returned as
ARC-managed bridged objects.

Unfortunately this could lead to a poor diagnostic inside copy methods under
ARC where the analyzer would complain about a leak of a returned CF value inside
a method "whose name does not start with 'copy'" -- even though the name did
start with 'copy'.

This commit improves the diagnostic under ARC to say inside a method "returned
from a method managed by Automated Reference Counting".

rdar://problem/28849667

llvm-svn: 286694

7 years ago[tsan] Test that false races from ObjC's dealloc, .cxx_destruct, and initialize are...
Anna Zaks [Sat, 12 Nov 2016 00:46:07 +0000 (00:46 +0000)]
[tsan] Test that false races from ObjC's dealloc, .cxx_destruct, and initialize are ignored

Differential Revision: https://reviews.llvm.org/D26228

llvm-svn: 286693

7 years agoOne more set of changes to fix formatv() on linux.
Zachary Turner [Sat, 12 Nov 2016 00:35:58 +0000 (00:35 +0000)]
One more set of changes to fix formatv() on linux.

llvm-svn: 286692

7 years agoAdd missing #include.
Zachary Turner [Sat, 12 Nov 2016 00:30:37 +0000 (00:30 +0000)]
Add missing #include.

llvm-svn: 286691

7 years agoFix another problem with formatv().
Zachary Turner [Sat, 12 Nov 2016 00:28:10 +0000 (00:28 +0000)]
Fix another problem with formatv().

llvm-svn: 286690

7 years ago[libFuzzer] use less stack
Kostya Serebryany [Sat, 12 Nov 2016 00:24:35 +0000 (00:24 +0000)]
[libFuzzer] use less stack

llvm-svn: 286689

7 years agoRemove extra semicolon.
Rui Ueyama [Sat, 12 Nov 2016 00:23:32 +0000 (00:23 +0000)]
Remove extra semicolon.

llvm-svn: 286688

7 years agoAMDGPU/SI: Promote i16 = fp_[us]int f32 for VI
Tom Stellard [Sat, 12 Nov 2016 00:19:11 +0000 (00:19 +0000)]
AMDGPU/SI: Promote i16 = fp_[us]int f32 for VI

Summary: This fixes a regression caused by r286464.

Reviewers: arsenm

Subscribers: kzhuravl, wdng, nhaehnle, yaxunl, llvm-commits, tony-tye

Differential Revision: https://reviews.llvm.org/D26570

llvm-svn: 286687

7 years agoTry to fix build after llvm::formatv() patch.
Zachary Turner [Sat, 12 Nov 2016 00:18:42 +0000 (00:18 +0000)]
Try to fix build after llvm::formatv() patch.

llvm-svn: 286686

7 years ago[c++1z] std::byte didn't make it
Richard Smith [Sat, 12 Nov 2016 00:15:24 +0000 (00:15 +0000)]
[c++1z] std::byte didn't make it

llvm-svn: 286685

7 years agoEmit Section Contributions.
Rui Ueyama [Sat, 12 Nov 2016 00:00:51 +0000 (00:00 +0000)]
Emit Section Contributions.

Differential Revision: https://reviews.llvm.org/D26211

llvm-svn: 286684

7 years agoFix -Werror build with clang-cl.
Zachary Turner [Fri, 11 Nov 2016 23:58:11 +0000 (23:58 +0000)]
Fix -Werror build with clang-cl.

llvm-svn: 286683

7 years ago[Support] Introduce llvm::formatv() function.
Zachary Turner [Fri, 11 Nov 2016 23:57:40 +0000 (23:57 +0000)]
[Support] Introduce llvm::formatv() function.

This introduces a new type-safe general purpose formatting
library.  It provides compile-time type safety, does not require
a format specifier (since the type is deduced), and provides
mechanisms for extending the format capability to user defined
types, and overriding the formatting behavior for existing types.

This patch additionally adds documentation for the API to the
LLVM programmer's manual.

Mailing List Thread:
http://lists.llvm.org/pipermail/llvm-dev/2016-October/105836.html

Differential Revision: https://reviews.llvm.org/D25587

llvm-svn: 286682

7 years ago[c++1z] Two more features have been accepted.
Richard Smith [Fri, 11 Nov 2016 23:51:47 +0000 (23:51 +0000)]
[c++1z] Two more features have been accepted.

llvm-svn: 286681

7 years ago[index] Rename SymbolSubKind -> SymbolProperty, NFC.
Argyrios Kyrtzidis [Fri, 11 Nov 2016 23:49:55 +0000 (23:49 +0000)]
[index] Rename SymbolSubKind -> SymbolProperty, NFC.

This better reflects what it represents.

llvm-svn: 286680

7 years ago[c++1z] p0522 is approved.
Richard Smith [Fri, 11 Nov 2016 23:46:25 +0000 (23:46 +0000)]
[c++1z] p0522 is approved.

llvm-svn: 286679

7 years ago[c++1z] Add constant-folding support for strcmp, strncmp, and memcmp, to
Richard Smith [Fri, 11 Nov 2016 23:43:35 +0000 (23:43 +0000)]
[c++1z] Add constant-folding support for strcmp, strncmp, and memcmp, to
support constexpr char_traits.

llvm-svn: 286678

7 years agoDefine DbiStreamBuilder::addSectionContribs.
Rui Ueyama [Fri, 11 Nov 2016 23:41:13 +0000 (23:41 +0000)]
Define DbiStreamBuilder::addSectionContribs.

This patch defines a new function to add a SectionContribs stream
to a PDB file. Unlike SectionMap, SectionContribs contains a list
of input sections as opposed to output sections.

Note that this patch needs improving because currently we do not
set Module field in SectionContribs entries. In a follow-up patch,
I'll add Modules and then fix it after that.

Differential Revision: https://reviews.llvm.org/D26210

llvm-svn: 286677

7 years agoAMDGPU/SI: Fix visit order assumption in SIFixSGPRCopies
Tom Stellard [Fri, 11 Nov 2016 23:35:42 +0000 (23:35 +0000)]
AMDGPU/SI: Fix visit order assumption in SIFixSGPRCopies

Summary:
This pass was assuming that when a PHI instruction defined a register
used by another PHI instruction that the defining insstruction would
be legalized before the using instruction.

This assumption was causing the pass to not legalize some PHI nodes
within divergent flow-control.

This fixes a bug that was uncovered by r285762.

Reviewers: nhaehnle, arsenm

Subscribers: kzhuravl, wdng, nhaehnle, yaxunl, tony-tye, llvm-commits

Differential Revision: https://reviews.llvm.org/D26303

llvm-svn: 286676

7 years agoPrint correct directory in merge script.
Richard Trieu [Fri, 11 Nov 2016 23:26:28 +0000 (23:26 +0000)]
Print correct directory in merge script.

When providing the project directory to the merge script, print it out in the
commit instructions instead of the default project directory.

llvm-svn: 286675

7 years ago[tsan] Fix typo in test.
Evgeniy Stepanov [Fri, 11 Nov 2016 23:25:09 +0000 (23:25 +0000)]
[tsan] Fix typo in test.

llvm-svn: 286674

7 years agoDisable a test using posix_spawn on Android
Reid Kleckner [Fri, 11 Nov 2016 23:23:33 +0000 (23:23 +0000)]
Disable a test using posix_spawn on Android

llvm-svn: 286673

7 years ago[tsan][clang] Introduce a function attribute to disable TSan checking at run time
Anna Zaks [Fri, 11 Nov 2016 23:22:44 +0000 (23:22 +0000)]
[tsan][clang] Introduce a function attribute to disable TSan checking at run time

This introduces a function annotation that disables TSan checking for the
function at run time. The benefit over attribute((no_sanitize("thread")))
is that the accesses within the callees will also be suppressed.

The motivation for this attribute is a guarantee given by the objective C
language that the calls to the reference count decrement and object
deallocation will be synchronized. To model this properly, we would need to
intercept all ref count decrement calls (which are very common in ObjC due
to use of ARC) and also every single message send. Instead, we propose to
just ignore all accesses made from within dealloc at run time. The main
downside is that this still does not introduce any synchronization, which
means we might still report false positives if the code that relies on this
synchronization is not executed from within dealloc. However, we have not
seen this in practice so far and think these cases will be very rare.

(This problem is similar in nature to https://reviews.llvm.org/D21609;
unfortunately, the same solution does not apply here.)

Differential Revision: https://reviews.llvm.org/D25857

llvm-svn: 286672

7 years ago[InstCombine] use dyn_cast rather isa+cast; NFC
Sanjay Patel [Fri, 11 Nov 2016 23:20:01 +0000 (23:20 +0000)]
[InstCombine] use dyn_cast rather isa+cast; NFC

Follow-up to r286664 cleanup as suggested by Eli. Thanks!

llvm-svn: 286671

7 years agoTread TSan LLVM flags to driver: unit tests.
Evgeniy Stepanov [Fri, 11 Nov 2016 23:17:49 +0000 (23:17 +0000)]
Tread TSan LLVM flags to driver: unit tests.

Summary: Unit tests for the new clang flags.

Reviewers: eugenis, dvyukov

Subscribers: kubabrecka, llvm-commits

Patch by Alex Shlyapnikov.

Differential Revision: https://reviews.llvm.org/D26462

llvm-svn: 286670

7 years agoTread TSan LLVM flags to driver: add TSan controlling flags to clang.
Evgeniy Stepanov [Fri, 11 Nov 2016 23:17:36 +0000 (23:17 +0000)]
Tread TSan LLVM flags to driver: add TSan controlling flags to clang.

Summary:
New clang flags, all default to true:
-f[no-]sanitize-thread-data-races
-f[no-]sanitize-thread-stack-traces
-f[no-]sanitize-thread-atomics

Reviewers: dvyukov, pcc, eugenis

Subscribers: pcc, cfe-commits

Patch by Alex Shlyapnikov.

Differential Revision: https://reviews.llvm.org/D26461

llvm-svn: 286669

7 years ago[InstCombine] update test to use FileCheck; NFC
Sanjay Patel [Fri, 11 Nov 2016 23:12:46 +0000 (23:12 +0000)]
[InstCombine] update test to use FileCheck; NFC

llvm-svn: 286668

7 years agoAdd missing REQUIRES.
Rafael Espindola [Fri, 11 Nov 2016 23:10:22 +0000 (23:10 +0000)]
Add missing REQUIRES.

llvm-svn: 286666

7 years ago[libFuzzer] do not initialize parts of TracePC -- let them be initialized by the...
Kostya Serebryany [Fri, 11 Nov 2016 23:06:53 +0000 (23:06 +0000)]
[libFuzzer] do not initialize parts of TracePC -- let them be initialized by the linker. Add no-msan attribute to the memcmp hook.

llvm-svn: 286665

7 years ago[InstCombine] clean up foldSelectOpOp(); NFC
Sanjay Patel [Fri, 11 Nov 2016 23:01:20 +0000 (23:01 +0000)]
[InstCombine] clean up foldSelectOpOp(); NFC

llvm-svn: 286664

7 years ago[tsan][llvm] Implement the function attribute to disable TSan checking at run time
Anna Zaks [Fri, 11 Nov 2016 23:01:02 +0000 (23:01 +0000)]
[tsan][llvm] Implement the function attribute to disable TSan checking at run time

This implements a function annotation that disables TSan checking for the
function at run time. The benefit over attribute((no_sanitize("thread")))
is that the accesses within the callees will also be suppressed.

The motivation for this attribute is a guarantee given by the objective C
language that the calls to the reference count decrement and object
deallocation will be synchronized. To model this properly, we would need
to intercept all ref count decrement calls (which are very common in ObjC
due to use of ARC) and also every single message send. Instead, we propose
to just ignore all accesses made from within dealloc at run time. The main
downside is that this still does not introduce any synchronization, which
means we might still report false positives if the code that relies on this
synchronization is not executed from within dealloc. However, we have not seen
this in practice so far and think these cases will be very rare.

Differential Revision: https://reviews.llvm.org/D25858

llvm-svn: 286663

7 years ago[LV] Stop saying "use -Rpass-analysis=loop-vectorize"
Adam Nemet [Fri, 11 Nov 2016 22:51:46 +0000 (22:51 +0000)]
[LV] Stop saying "use -Rpass-analysis=loop-vectorize"

This is PR28376.

Unfortunately given the current structure of optimization diagnostics we
lack the capability to tell whether the user has
passed -Rpass-analysis=loop-vectorize since this is local to the
front-end (BackendConsumer::OptimizationRemarkHandler).

So rather than printing this even if the user has already
passed -Rpass-analysis, this patch just punts and stops recommending
this option.  I don't think that getting this right is worth the
complexity.

Differential Revision: https://reviews.llvm.org/D26563

llvm-svn: 286662

7 years ago[RPC] Add const qualifier to MemberFnWrapper to make buildbots happy.
Lang Hames [Fri, 11 Nov 2016 22:50:16 +0000 (22:50 +0000)]
[RPC] Add const qualifier to MemberFnWrapper to make buildbots happy.

This is a temporary fix: The right solution is to make sure addHandler can
support mutable lambdas. I'll add that in a follow-up patch.

llvm-svn: 286661

7 years ago[cxx_status] Prepare features list for WG21 motions.
Richard Smith [Fri, 11 Nov 2016 22:48:43 +0000 (22:48 +0000)]
[cxx_status] Prepare features list for WG21 motions.

llvm-svn: 286660

7 years agoDon't sort sections in -r links.
Rafael Espindola [Fri, 11 Nov 2016 22:43:27 +0000 (22:43 +0000)]
Don't sort sections in -r links.

Unlike gold, bfd, gas or MC we were putting exidx sections first since
they are ro.

The spec doesn't explicitly say that they must come after, but it is
definitely more convenient for the consumer, matches other producers
and matches other areas in ELF (like SHT_GROUP) where sections are
ordered in a natural way.

llvm-svn: 286659

7 years ago[tsan] Expose __tsan_ignore_thread_begin and __tsan_ignore_thread_end in TSan interface
Anna Zaks [Fri, 11 Nov 2016 22:43:12 +0000 (22:43 +0000)]
[tsan] Expose __tsan_ignore_thread_begin and __tsan_ignore_thread_end in TSan interface

This patch is needed to implement the function attribute that disable TSan checking at run time.

Differential Revision: https://reviews.llvm.org/D25859

llvm-svn: 286658

7 years agoFix typo in comment.
Rui Ueyama [Fri, 11 Nov 2016 22:41:47 +0000 (22:41 +0000)]
Fix typo in comment.

llvm-svn: 286657

7 years agoRevert "(origin/master, origin/HEAD) MachineScheduler/ScheduleDAG: Add support to...
Matthias Braun [Fri, 11 Nov 2016 22:39:50 +0000 (22:39 +0000)]
Revert "(origin/master, origin/HEAD) MachineScheduler/ScheduleDAG: Add support to skipping a node."

Revert accidentally committed change.

This reverts commit r286655.

llvm-svn: 286656

7 years agoMachineScheduler/ScheduleDAG: Add support to skipping a node.
Matthias Braun [Fri, 11 Nov 2016 22:37:34 +0000 (22:37 +0000)]
MachineScheduler/ScheduleDAG: Add support to skipping a node.

The DAG mutators in the scheduler cannot really remove DAG nodes as
additional anlysis information such as ScheduleDAGToplogicalSort are
already computed at this point and rely on a fixed number of DAG nodes.

Alleviate the missing removal with a new flag: Setting the new skip
flag on a node ignores it during scheduling.

llvm-svn: 286655

7 years agoScheduleDAGInstrs: Move VRegUses to ScheduleDAGMILive; NFCI
Matthias Braun [Fri, 11 Nov 2016 22:37:31 +0000 (22:37 +0000)]
ScheduleDAGInstrs: Move VRegUses to ScheduleDAGMILive; NFCI

Push VRegUses/collectVRegUses() down the class hierarchy towards its
only user ScheduleDAGMILive.

NFCI: The initialization of the map happens at a later point but that
should not matter.

This is in preparation to allow DAG mutators to merge nodes, which
relies on this map getting computed later.

llvm-svn: 286654

7 years agoMachineScheduler: Dump EntrySU/ExitSU if possible
Matthias Braun [Fri, 11 Nov 2016 22:37:28 +0000 (22:37 +0000)]
MachineScheduler: Dump EntrySU/ExitSU if possible

llvm-svn: 286653

7 years agoScheduleDAG: Identify EntrySU/ExitSU when dumping node ids
Matthias Braun [Fri, 11 Nov 2016 22:37:26 +0000 (22:37 +0000)]
ScheduleDAG: Identify EntrySU/ExitSU when dumping node ids

llvm-svn: 286652

7 years ago[ASTMatchers] Fix a typo in cStyleCastExpr() HTML docs as well. NFC.
Artem Dergachev [Fri, 11 Nov 2016 22:34:53 +0000 (22:34 +0000)]
[ASTMatchers] Fix a typo in cStyleCastExpr() HTML docs as well. NFC.

llvm-svn: 286651

7 years ago[PowerPC] Implement remaining permute builtins in altivec.h - Clang portion
Nemanja Ivanovic [Fri, 11 Nov 2016 22:34:44 +0000 (22:34 +0000)]
[PowerPC] Implement remaining permute builtins in altivec.h - Clang portion

This patch corresponds to review:
https://reviews.llvm.org/D26479

It adds the remaining vector permute/rotate builtins to altivec.h.

llvm-svn: 286650

7 years agoFix wrong formatting in lld introduced in r286561 (NFC)
Mehdi Amini [Fri, 11 Nov 2016 22:28:19 +0000 (22:28 +0000)]
Fix wrong formatting in lld introduced in r286561 (NFC)

Pointed out by Davide.

llvm-svn: 286649

7 years agoFunctionComparator: don't rely on argument evaluation order.
Erik Eckstein [Fri, 11 Nov 2016 22:21:39 +0000 (22:21 +0000)]
FunctionComparator: don't rely on argument evaluation order.

This is a follow-up on the recent refactoring of the FunctionMerge pass.
It should fix a fail of the new FunctionComparator unittest whe compiling with MSVC.

llvm-svn: 286648

7 years agoFix static initialization order fiasco in MCTests
Mehdi Amini [Fri, 11 Nov 2016 22:18:42 +0000 (22:18 +0000)]
Fix static initialization order fiasco in MCTests

Reported by Kostya on llvm-dev, uncovered by an ASAN bot

llvm-svn: 286647

7 years ago[ORC] Temporarily fix the RPCUtils unit test by explicitly specifying a handler
Lang Hames [Fri, 11 Nov 2016 22:16:10 +0000 (22:16 +0000)]
[ORC] Temporarily fix the RPCUtils unit test by explicitly specifying a handler
return type.

This should be fixed permanently by having the RPCUtils header recognize the
ErrorSuccess type. I'll commit that in a follow up patch.

llvm-svn: 286646

7 years agoNFC ProgrammersManual fix
Piotr Padlewski [Fri, 11 Nov 2016 22:12:15 +0000 (22:12 +0000)]
NFC ProgrammersManual fix

llvm-svn: 286645

7 years agoSimplify code and address review comments (NFC)
Adrian Prantl [Fri, 11 Nov 2016 22:09:25 +0000 (22:09 +0000)]
Simplify code and address review comments (NFC)

llvm-svn: 286644

7 years agoMake test more strict. NFC.
Rafael Espindola [Fri, 11 Nov 2016 22:07:15 +0000 (22:07 +0000)]
Make test more strict. NFC.

It was non-obvious that a CHECK was not matching the following section.

llvm-svn: 286643

7 years ago[Orc] Update the BuildingAJIT Chapter 5 server class for the recent RPC changes.
Lang Hames [Fri, 11 Nov 2016 21:55:25 +0000 (21:55 +0000)]
[Orc] Update the BuildingAJIT Chapter 5 server class for the recent RPC changes.

llvm-svn: 286642

7 years agoRemove double setting of invalid flag.
Richard Trieu [Fri, 11 Nov 2016 21:50:39 +0000 (21:50 +0000)]
Remove double setting of invalid flag.

In r286630, Decl::setInvalidDecl will automatically set the invalid flag for
BindingDecl for children in invalid DecompositionDecl.  It no longer is
necessary to do a separate setInvalidDecl when finalizing a BindingDecl.

llvm-svn: 286641

7 years agoFix a reference-to-temporary introduced in r286607.
Adrian Prantl [Fri, 11 Nov 2016 21:48:09 +0000 (21:48 +0000)]
Fix a reference-to-temporary introduced in r286607.

llvm-svn: 286640

7 years ago[ORC] Re-apply 286620 with fixes for the ErrorSuccess class.
Lang Hames [Fri, 11 Nov 2016 21:42:09 +0000 (21:42 +0000)]
[ORC] Re-apply 286620 with fixes for the ErrorSuccess class.

llvm-svn: 286639

7 years ago[PowerPC] Add remaining vector permute builtins in altivec.h - LLVM portion
Nemanja Ivanovic [Fri, 11 Nov 2016 21:42:01 +0000 (21:42 +0000)]
[PowerPC] Add remaining vector permute builtins in altivec.h - LLVM portion

This patch corresponds to review:
https://reviews.llvm.org/D26480

Adds all the intrinsics used for various permute builtins that will
be added to altivec.h.

llvm-svn: 286638

7 years ago[cfi] Test case for weak symbol handling.
Evgeniy Stepanov [Fri, 11 Nov 2016 21:39:35 +0000 (21:39 +0000)]
[cfi] Test case for weak symbol handling.

llvm-svn: 286637

7 years ago[cfi] Fix weak functions handling.
Evgeniy Stepanov [Fri, 11 Nov 2016 21:39:26 +0000 (21:39 +0000)]
[cfi] Fix weak functions handling.

When a function pointer is replaced with a jumptable pointer, special
case is needed to preserve the semantics of extern_weak functions.
Since a jumptable entry can not be extern_weak, we emulate that
behaviour by replacing all references to F (the extern_weak function)
with the following expression: F != nullptr ? JumpTablePtr : nullptr.

Extra special care is needed for global initializers, since most (or
probably all) backends can not lower an initializer that includes
this kind of constant expression. Initializers like that are replaced
with a global constructor (i.e. a runtime initializer).

llvm-svn: 286636

7 years agoAvoid a crash with -r and .comment.
Rafael Espindola [Fri, 11 Nov 2016 21:36:25 +0000 (21:36 +0000)]
Avoid a crash with -r and .comment.

We would create a MergeInputSection for the synthetic .comment and
crash trying to add it to a regular output section.

With this we just don't add the synthetic section with -r. That is
consistent with gold that doesn't create .note.gnu.gold-version with
-r.

llvm-svn: 286635

7 years agoFix build since llvm r286566 and require at least llvm 4.0
Tom Stellard [Fri, 11 Nov 2016 21:34:47 +0000 (21:34 +0000)]
Fix build since llvm r286566 and require at least llvm 4.0

llvm-svn: 286634

7 years ago[analyzer] Teach RetainCountChecker about VTCompressionSessionEncodeFrame()
Devin Coughlin [Fri, 11 Nov 2016 21:31:38 +0000 (21:31 +0000)]
[analyzer] Teach RetainCountChecker about VTCompressionSessionEncodeFrame()

The context argument passed to VideoToolbox's
VTCompressionSessionEncodeFrame() function is ultimately passed to a callback
supplied when creating the compression session and so may be freed by that
callback.  To suppress false positives in this case, teach the retain count
checker to stop tracking that argument.

This isn't suppressed by the usual callback context mechanism because the call
to VTCompressionSessionEncodeFrame() doesn't include the callback itself.

rdar://problem/27685213

llvm-svn: 286633

7 years agoMake the FunctionComparator of the MergeFunctions pass a stand-alone utility.
Erik Eckstein [Fri, 11 Nov 2016 21:15:13 +0000 (21:15 +0000)]
Make the FunctionComparator of the MergeFunctions pass a stand-alone utility.

This is pure refactoring. NFC.

This change moves the FunctionComparator (together with the GlobalNumberState
utility) in to a separate file so that it can be used by other passes.
For example, the SwiftMergeFunctions pass in the Swift compiler:
https://github.com/apple/swift/blob/master/lib/LLVMPasses/LLVMMergeFunctions.cpp

Details of the change:

*) The big part is just moving code out of MergeFunctions.cpp into FunctionComparator.h/cpp
*) Make FunctionComparator member functions protected (instead of private)
   so that a derived comparator class can use them.

Following refactoring helps to share code between the base FunctionComparator
class and a derived class:

*) Add a beginCompare() function
*) Move some basic function property comparisons into a separate function compareSignature()
*) Do the GEP comparison inside cmpOperations() which now has a new
   needToCmpOperands reference parameter

https://reviews.llvm.org/D25385

llvm-svn: 286632

7 years agoRemove weak-linked symbols for SBBreakpointListImpl
Todd Fiala [Fri, 11 Nov 2016 21:06:40 +0000 (21:06 +0000)]
Remove weak-linked symbols for SBBreakpointListImpl

Summary:
Similar to SBStructuredData's Impl class, SBBreakpointListImpl was
getting weak-link exported in the lldb namespace. This change list fixes
that by moving out of the lldb public namespace, which removes it from
public export visibility.

Fixes:
rdar://28960344

Reviewers: jingham

Subscribers: lldb-commits

Differential Revision: https://reviews.llvm.org/D26553

llvm-svn: 286631

7 years agoWhen a DecompositionDecl is marked invalid, also set the child BindingDecl's to
Richard Trieu [Fri, 11 Nov 2016 20:51:04 +0000 (20:51 +0000)]
When a DecompositionDecl is marked invalid, also set the child BindingDecl's to
invalid.

llvm-svn: 286630

7 years agoFix -Wpessimizing-move warning.
Rui Ueyama [Fri, 11 Nov 2016 20:39:02 +0000 (20:39 +0000)]
Fix -Wpessimizing-move warning.

llvm-svn: 286629

7 years ago[ASTMatchers] Fix a typo in cStyleCastExpr() docs. NFC.
Artem Dergachev [Fri, 11 Nov 2016 20:29:59 +0000 (20:29 +0000)]
[ASTMatchers] Fix a typo in cStyleCastExpr() docs. NFC.

llvm-svn: 286628

7 years ago[PowerPC] Add vector conversion builtins to altivec.h - clang portion
Nemanja Ivanovic [Fri, 11 Nov 2016 19:56:17 +0000 (19:56 +0000)]
[PowerPC] Add vector conversion builtins to altivec.h - clang portion

This patch corresponds to review:
https://reviews.llvm.org/D26308

It adds a number of vector type conversion builtins to altivec.h.

llvm-svn: 286627

7 years agoFixed the lost FastMathFlags for FCmp operations in SLPVectorizer.
Vyacheslav Klochkov [Fri, 11 Nov 2016 19:55:29 +0000 (19:55 +0000)]
Fixed the lost FastMathFlags for FCmp operations in SLPVectorizer.
Reviewer: Michael Zolotukhin.
Differential Revision: https://reviews.llvm.org/D26543

llvm-svn: 286626

7 years ago[AArch64] Update a FIXME comment to reflect current state. NFC.
Chad Rosier [Fri, 11 Nov 2016 19:52:45 +0000 (19:52 +0000)]
[AArch64] Update a FIXME comment to reflect current state. NFC.

llvm-svn: 286625

7 years agoBitcode: Change getModuleSummaryIndex() to return an llvm::Expected.
Peter Collingbourne [Fri, 11 Nov 2016 19:50:39 +0000 (19:50 +0000)]
Bitcode: Change getModuleSummaryIndex() to return an llvm::Expected.

Differential Revision: https://reviews.llvm.org/D26539

llvm-svn: 286624

7 years agoBitcode: Clean up error handling for certain bitcode query functions.
Peter Collingbourne [Fri, 11 Nov 2016 19:50:24 +0000 (19:50 +0000)]
Bitcode: Clean up error handling for certain bitcode query functions.

The functions getBitcodeTargetTriple(), isBitcodeContainingObjCCategory(),
getBitcodeProducerString() and hasGlobalValueSummary() now return errors
via their return value rather than via the diagnostic handler.

To make this work, re-implement these functions using non-member functions
so that they can be used without the LLVMContext required by BitcodeReader.

Differential Revision: https://reviews.llvm.org/D26532

llvm-svn: 286623

7 years agoBitcode: Prepare to move bitcode readers to free functions.
Peter Collingbourne [Fri, 11 Nov 2016 19:50:10 +0000 (19:50 +0000)]
Bitcode: Prepare to move bitcode readers to free functions.

Make initStream() a free function, and change BitcodeReaderBase ctor to take
a BitstreamCursor.

llvm-svn: 286622

7 years ago[ORC] Revert r286620 while I investigate a bot failure.
Lang Hames [Fri, 11 Nov 2016 19:46:46 +0000 (19:46 +0000)]
[ORC] Revert r286620 while I investigate a bot failure.

llvm-svn: 286621

7 years ago[ORC] Refactor the ORC RPC utilities to add some new features.
Lang Hames [Fri, 11 Nov 2016 19:42:44 +0000 (19:42 +0000)]
[ORC] Refactor the ORC RPC utilities to add some new features.

(1) Add support for function key negotiation.

The previous version of the RPC required both sides to maintain the same
enumeration for functions in the API. This means that any version skew between
the client and server would result in communication failure.

With this version of the patch functions (and serializable types) are defined
with string names, and the derived function signature strings are used to
negotiate the actual function keys (which are used for efficient call
serialization). This allows clients to connect to any server that supports a
superset of the API (based on the function signatures it supports).

(2) Add a callAsync primitive.

The callAsync primitive can be used to install a return value handler that will
run as soon as the RPC function's return value is sent back from the remote.

(3) Launch policies for RPC function handlers.

The new addHandler method, which installs handlers for RPC functions, takes two
arguments: (1) the handler itself, and (2) an optional "launch policy". When the
RPC function is called, the launch policy (if present) is invoked to actually
launch the handler. This allows the handler to be spawned on a background
thread, or added to a work list. If no launch policy is used, the handler is run
on the server thread itself. This should only be used for short-running
handlers, or entirely synchronous RPC APIs.

(4) Zero cost cross type serialization.

You can now define serialization from any type to a different "wire" type. For
example, this allows you to call an RPC function that's defined to take a
std::string while passing a StringRef argument. If a serializer from StringRef
to std::string has been defined for the channel type this will be used to
serialize the argument without having to construct a std::string instance.

This allows buffer reference types to be used as arguments to RPC calls without
requiring a copy of the buffer to be made.

llvm-svn: 286620

7 years ago[InstCombine] add tests to show size-increasing select transforms
Sanjay Patel [Fri, 11 Nov 2016 19:37:54 +0000 (19:37 +0000)]
[InstCombine] add tests to show size-increasing select transforms

llvm-svn: 286619

7 years agoAdd a test that uses coverage and printf from a DLL
Reid Kleckner [Fri, 11 Nov 2016 19:27:52 +0000 (19:27 +0000)]
Add a test that uses coverage and printf from a DLL

This test fails without LLVM r286615

llvm-svn: 286618

7 years ago[AArch64] Add test to show narrow zero store merging is disabled with strict align...
Chad Rosier [Fri, 11 Nov 2016 19:25:48 +0000 (19:25 +0000)]
[AArch64] Add test to show narrow zero store merging is disabled with strict align. NFC.

llvm-svn: 286617

7 years ago[AArch64] Fix bugs in isel lowering replaceSplatVectorStore.
Geoff Berry [Fri, 11 Nov 2016 19:25:20 +0000 (19:25 +0000)]
[AArch64] Fix bugs in isel lowering replaceSplatVectorStore.

Summary:
Fix off-by-one indexing error in loop checking that inserted value was a
splat vector.

Add code to check that INSERT_VECTOR_ELT nodes constructing the splat
vector have the expected constant index values.

Reviewers: t.p.northover, jmolloy, mcrosier

Subscribers: aemerson, llvm-commits, rengolin

Differential Revision: https://reviews.llvm.org/D26409

llvm-svn: 286616

7 years ago[sancov] Don't instrument MSVC CRT stdio config helpers
Reid Kleckner [Fri, 11 Nov 2016 19:18:45 +0000 (19:18 +0000)]
[sancov] Don't instrument MSVC CRT stdio config helpers

They get called before initialization, which is a problem for winasan.

Test coming in compiler-rt.

llvm-svn: 286615

7 years ago[CMake] Check runtimes subdir when looking for libcxx and libuwind
Petr Hosek [Fri, 11 Nov 2016 19:12:58 +0000 (19:12 +0000)]
[CMake] Check runtimes subdir when looking for libcxx and libuwind

The runtimes subdir is the new location for runtimes, we should
include it when looking for libcxx and libunwind headers.

Differential Revision: https://reviews.llvm.org/D26362

llvm-svn: 286614

7 years ago[cfi] Enable cfi-icall on ARM and AArch64.
Evgeniy Stepanov [Fri, 11 Nov 2016 18:49:49 +0000 (18:49 +0000)]
[cfi] Enable cfi-icall on ARM and AArch64.

llvm-svn: 286613