platform/upstream/llvm.git
7 years ago[Inliner] Port all opt remarks to new streaming API
Adam Nemet [Tue, 27 Sep 2016 23:47:03 +0000 (23:47 +0000)]
[Inliner] Port all opt remarks to new streaming API

llvm-svn: 282559

7 years agoPass -S to opt in this test to avoid printing binary on mismatch
Adam Nemet [Tue, 27 Sep 2016 23:46:59 +0000 (23:46 +0000)]
Pass -S to opt in this test to avoid printing binary on mismatch

The purpose of the test is to verify diagnostics.

llvm-svn: 282558

7 years agoPut new warning in a diagnostic group.
Matthias Braun [Tue, 27 Sep 2016 23:44:38 +0000 (23:44 +0000)]
Put new warning in a diagnostic group.

The warning I added in r282426 should be a diagnostic group.

llvm-svn: 282557

7 years agoP0145R3 (C++17 evaluation order tweaks): evaluate the right-hand side of
Richard Smith [Tue, 27 Sep 2016 23:44:22 +0000 (23:44 +0000)]
P0145R3 (C++17 evaluation order tweaks): evaluate the right-hand side of
assignment and compound-assignment operators before the left-hand side. (Even
if it's an overloaded operator.)

This completes the implementation of P0145R3 + P0400R0 for all targets except
Windows, where the evaluation order guarantees for <<, >>, and ->* are
unimplementable as the ABI requires the function arguments are evaluated from
right to left (because parameter destructors are run from left to right in the
callee).

llvm-svn: 282556

7 years agoRevert r282547 and add test to show correct behavior.
Richard Trieu [Tue, 27 Sep 2016 23:44:07 +0000 (23:44 +0000)]
Revert r282547 and add test to show correct behavior.

llvm-svn: 282555

7 years ago[Coverage] The coverage region for switch covers the code after the switch.
Alex Lorenz [Tue, 27 Sep 2016 23:30:36 +0000 (23:30 +0000)]
[Coverage] The coverage region for switch covers the code after the switch.

This patch fixes a regression introduced in r262697 that changed the way the
coverage regions for switches are constructed. The PGO instrumentation counter
for a switch statement refers to the counter at the exit of the switch.
Therefore, the coverage region for the switch statement should cover the code
that comes after the switch, and not the switch statement itself.

rdar://28480997

Differential Revision: https://reviews.llvm.org/D24981

llvm-svn: 282554

7 years agoNext set of additional error checks for invalid Mach-O files for the
Kevin Enderby [Tue, 27 Sep 2016 23:24:13 +0000 (23:24 +0000)]
Next set of additional error checks for invalid Mach-O files for the
other load commands that use the MachO::dylinker_command type
but not used in llvm libObject code but used in llvm tool code.

This includes LC_ID_DYLINKER, LC_LOAD_DYLINKER
and LC_DYLD_ENVIRONMENT load commands.

llvm-svn: 282553

7 years ago[CMake] Force CMP0057 to NEW
Chris Bieneman [Tue, 27 Sep 2016 23:18:32 +0000 (23:18 +0000)]
[CMake] Force CMP0057 to NEW

Hans reported an issue with r282510 on the list. This should resolve the issue.

llvm-svn: 282552

7 years ago[LTO] Add an API to check if a symbol is a TLS one.
Davide Italiano [Tue, 27 Sep 2016 22:59:29 +0000 (22:59 +0000)]
[LTO] Add an API to check if a symbol is a TLS one.

Will be used in lld.

llvm-svn: 282551

7 years ago[AArch64][RegisterBankInfo] Switch to statically allocated ValueMapping.
Quentin Colombet [Tue, 27 Sep 2016 22:55:04 +0000 (22:55 +0000)]
[AArch64][RegisterBankInfo] Switch to statically allocated ValueMapping.

Another step toward TableGen'ed like structure for the RegisterBankInfo
of AArch64. By doing this, we also save a bit of compile time for the
exact same output.

llvm-svn: 282550

7 years ago[AArch64][RegisterBankInfo] Fix copy/paste in comments.
Quentin Colombet [Tue, 27 Sep 2016 22:54:57 +0000 (22:54 +0000)]
[AArch64][RegisterBankInfo] Fix copy/paste in comments.

NFC.

llvm-svn: 282549

7 years ago[sanitizer_common] Delete some copy/move methods in InternalScopedBuffer
Vedant Kumar [Tue, 27 Sep 2016 22:52:34 +0000 (22:52 +0000)]
[sanitizer_common] Delete some copy/move methods in InternalScopedBuffer

Differential Revision: https://reviews.llvm.org/D24811

llvm-svn: 282548

7 years agoFix defaulted member functions for templated classes.
Richard Trieu [Tue, 27 Sep 2016 22:28:59 +0000 (22:28 +0000)]
Fix defaulted member functions for templated classes.

In some cases, non-special member functions were being marked as being defaulted
in templated classes.  This can cause interactions with later code that expects
the default function to be one of the specific member functions.  Fix the check
so that templated class members are checked the same way as non-templated class
members are.

llvm-svn: 282547

7 years ago[x86] add folds for FP logic with vector zeros
Sanjay Patel [Tue, 27 Sep 2016 22:28:13 +0000 (22:28 +0000)]
[x86] add folds for FP logic with vector zeros

The 'or' case shows up in copysign. The copysign code also had
redundant checking for a scalar zero operand with 'and', so I
removed that.

I'm not sure how to test vector 'and', 'andn', and 'xor' yet,
but it seems better to just include all of the logic ops since
we're fixing 'or' anyway.

llvm-svn: 282546

7 years agoShorten DiagnosticInfoOptimizationRemark* to OptimizationRemark*. NFC
Adam Nemet [Tue, 27 Sep 2016 22:19:29 +0000 (22:19 +0000)]
Shorten DiagnosticInfoOptimizationRemark* to OptimizationRemark*. NFC

With the new streaming interface in LLVM, these class names need to be
typed a lot and it's way too looong.

llvm-svn: 282545

7 years agoShorten DiagnosticInfoOptimizationRemark* to OptimizationRemark*. NFC
Adam Nemet [Tue, 27 Sep 2016 22:19:23 +0000 (22:19 +0000)]
Shorten DiagnosticInfoOptimizationRemark* to OptimizationRemark*. NFC

With the new streaming interface, these class names need to be typed a
lot and it's way too looong.

llvm-svn: 282544

7 years ago[TargetRegisterInfo, AArch64] Add target hook for isConstantPhysReg().
Geoff Berry [Tue, 27 Sep 2016 22:17:27 +0000 (22:17 +0000)]
[TargetRegisterInfo, AArch64] Add target hook for isConstantPhysReg().

Summary:
The current implementation of isConstantPhysReg() checks for defs of
physical registers to determine if they are constant.  Some
architectures (e.g. AArch64 XZR/WZR) have registers that are constant
and may be used as destinations to indicate the generated value is
discarded, preventing isConstantPhysReg() from returning true.  This
change adds a TargetRegisterInfo hook that overrides the no defs check
for cases such as this.

Reviewers: MatzeB, qcolombet, t.p.northover, jmolloy

Subscribers: junbuml, aemerson, mcrosier, rengolin

Differential Revision: https://reviews.llvm.org/D24570

llvm-svn: 282543

7 years ago[Inliner] Fold the analysis remark into the missed remark
Adam Nemet [Tue, 27 Sep 2016 21:58:17 +0000 (21:58 +0000)]
[Inliner] Fold the analysis remark into the missed remark

There is really no reason for these to be separate.

The vectorizer started this pretty bad tradition that the text of the
missed remarks is pretty meaningless, i.e. vectorization failed.  There,
you have to query analysis to get the full picture.

I think we should just explain the reason for missing the optimization
in the missed remark when possible.  Analysis remarks should provide
information that the pass gathers regardless whether the optimization is
passing or not.

llvm-svn: 282542

7 years ago[LoopSimplify] When simplifying phis in loop-simplify, do it only if it preserves...
Michael Zolotukhin [Tue, 27 Sep 2016 21:03:45 +0000 (21:03 +0000)]
[LoopSimplify] When simplifying phis in loop-simplify, do it only if it preserves LCSSA form.

llvm-svn: 282541

7 years agoAdapt to LLVM optimization remark interface change. NFC
Adam Nemet [Tue, 27 Sep 2016 20:55:12 +0000 (20:55 +0000)]
Adapt to LLVM optimization remark interface change. NFC

llvm-svn: 282540

7 years agoOutput optimization remarks in YAML
Adam Nemet [Tue, 27 Sep 2016 20:55:07 +0000 (20:55 +0000)]
Output optimization remarks in YAML

(Re-committed after moving the template specialization under the yaml
namespace.  GCC was complaining about this.)

This allows various presentation of this data using an external tool.
This was first recommended here[1].

As an example, consider this module:

  1 int foo();
  2 int bar();
  3
  4 int baz() {
  5   return foo() + bar();
  6 }

The inliner generates these missed-optimization remarks today (the
hotness information is pulled from PGO):

  remark: /tmp/s.c:5:10: foo will not be inlined into baz (hotness: 30)
  remark: /tmp/s.c:5:18: bar will not be inlined into baz (hotness: 30)

Now with -pass-remarks-output=<yaml-file>, we generate this YAML file:

  --- !Missed
  Pass:            inline
  Name:            NotInlined
  DebugLoc:        { File: /tmp/s.c, Line: 5, Column: 10 }
  Function:        baz
  Hotness:         30
  Args:
    - Callee: foo
    - String:  will not be inlined into
    - Caller: baz
  ...
  --- !Missed
  Pass:            inline
  Name:            NotInlined
  DebugLoc:        { File: /tmp/s.c, Line: 5, Column: 18 }
  Function:        baz
  Hotness:         30
  Args:
    - Callee: bar
    - String:  will not be inlined into
    - Caller: baz
  ...

This is a summary of the high-level decisions:

* There is a new streaming interface to emit optimization remarks.
E.g. for the inliner remark above:

   ORE.emit(DiagnosticInfoOptimizationRemarkMissed(
                DEBUG_TYPE, "NotInlined", &I)
            << NV("Callee", Callee) << " will not be inlined into "
            << NV("Caller", CS.getCaller()) << setIsVerbose());

NV stands for named value and allows the YAML client to process a remark
using its name (NotInlined) and the named arguments (Callee and Caller)
without parsing the text of the message.

Subsequent patches will update ORE users to use the new streaming API.

* I am using YAML I/O for writing the YAML file.  YAML I/O requires you
to specify reading and writing at once but reading is highly non-trivial
for some of the more complex LLVM types.  Since it's not clear that we
(ever) want to use LLVM to parse this YAML file, the code supports and
asserts that we're writing only.

On the other hand, I did experiment that the class hierarchy starting at
DiagnosticInfoOptimizationBase can be mapped back from YAML generated
here (see D24479).

* The YAML stream is stored in the LLVM context.

* In the example, we can probably further specify the IR value used,
i.e. print "Function" rather than "Value".

* As before hotness is computed in the analysis pass instead of
DiganosticInfo.  This avoids the layering problem since BFI is in
Analysis while DiagnosticInfo is in IR.

[1] https://reviews.llvm.org/D19678#419445

Differential Revision: https://reviews.llvm.org/D24587

llvm-svn: 282539

7 years agoSort headers
Adam Nemet [Tue, 27 Sep 2016 20:55:01 +0000 (20:55 +0000)]
Sort headers

llvm-svn: 282538

7 years agoUpdate FileSpec's interface to use StringRefs.
Zachary Turner [Tue, 27 Sep 2016 20:48:37 +0000 (20:48 +0000)]
Update FileSpec's interface to use StringRefs.

Differential Revision: https://reviews.llvm.org/D24936

llvm-svn: 282537

7 years ago[cmake] Support overriding remaining HTML doc install directories
Michal Gorny [Tue, 27 Sep 2016 19:52:29 +0000 (19:52 +0000)]
[cmake] Support overriding remaining HTML doc install directories

Support overriding the Doxygen & OCamldoc install directories,
and provide a more FHS-compliant defaults for both of them. This extends
r282240 that added this override for Sphinx-built documentation.

LLVM_INSTALL_DOXYGEN_HTML_DIR and LLVM_INSTALL_OCAMLDOC_HTML_DIR are
added, to control the location where Doxygen-generated and
OCamldoc-generated HTML docs are installed appropriately. They both
specify CMake-style install paths, and therefore can either by relative
to the install prefix or absolute.

The new defaults are subdirectories of share/doc/llvm, and replace
the previous directories of 'docs/html' and 'docs/ocaml/html' that
resulted in creating invalid '/usr/docs' that furthermore lacked proper
namespacing for the LLVM package. The new defaults are consistent with
the ones used for Sphinx HTML documentation, differing only in the last
component. Since the 'html' subdirectory is already used for Sphinx
docs, the 'doxygen-html' and 'ocaml-html' directories are used instead.

Differential Revision: https://reviews.llvm.org/D24935

llvm-svn: 282536

7 years agoFix a typo, depricated -> deprecated
Martin Storsjo [Tue, 27 Sep 2016 19:45:30 +0000 (19:45 +0000)]
Fix a typo, depricated -> deprecated

Differential Revision: https://reviews.llvm.org/D22849

llvm-svn: 282534

7 years agoAdapt to LLVM EnableStatistics() change.
Matthias Braun [Tue, 27 Sep 2016 19:38:59 +0000 (19:38 +0000)]
Adapt to LLVM EnableStatistics() change.

llvm-svn: 282533

7 years agoStatistic: Bring back printing on exit by default
Matthias Braun [Tue, 27 Sep 2016 19:38:55 +0000 (19:38 +0000)]
Statistic: Bring back printing on exit by default

Turns out several external projects relied on llvm printing statistics
on exit. Let's go back to this behaviour by default and have an optional
parameter to disable it.

llvm-svn: 282532

7 years ago[COFF] Add support for IMAGE_REL_ARM_SECREL
Martin Storsjo [Tue, 27 Sep 2016 19:27:17 +0000 (19:27 +0000)]
[COFF] Add support for IMAGE_REL_ARM_SECREL

Handle this in the exact same way as IMAGE_REL_AMD64_SECREL
and IMAGE_REL_I386_SECREL.

Differential revision: https://reviews.llvm.org/D24608

llvm-svn: 282531

7 years agobuiltins: use thumb-2 if possible on comparesf2
Saleem Abdulrasool [Tue, 27 Sep 2016 19:12:31 +0000 (19:12 +0000)]
builtins: use thumb-2 if possible on comparesf2

This code can be built with thumb-2 like many of the other builtin routines.
Enable that here as well.

llvm-svn: 282530

7 years agoAdding a RegisterContextMinidump_x86_64 converter
Dimitar Vlahovski [Tue, 27 Sep 2016 19:05:55 +0000 (19:05 +0000)]
Adding a RegisterContextMinidump_x86_64 converter

Summary:
This is a register context converter from Minidump to Linux reg context.
This knows the layout of the register context in the Minidump file
(which is the same as in Windows FYI) and as a result emits a binary data
buffer that matches the Linux register context binary layout.
This way we can reuse the existing RegisterContextLinux_x86_64 and
RegisterContextCorePOSIX_x86_64 classes.

Reviewers: labath, zturner

Subscribers: beanz, mgorny, lldb-commits, amccarth

Differential Revision: https://reviews.llvm.org/D24919

llvm-svn: 282529

7 years ago[lit] Add instructions to run lit's test suite
Daniel Dunbar [Tue, 27 Sep 2016 18:58:50 +0000 (18:58 +0000)]
[lit] Add instructions to run lit's test suite

 - Patch by Brian Gesiak.

 - https://reviews.llvm.org/D24968

llvm-svn: 282525

7 years agoRevert r282483 - [cmake] Add linker option "-Wl,-z,defs" in standalone build
Michal Gorny [Tue, 27 Sep 2016 18:54:02 +0000 (18:54 +0000)]
Revert r282483 - [cmake] Add linker option "-Wl,-z,defs" in standalone build

Revert r282483 as it causes build failures due to missing symbols when
not linking to -lgcc_s (i.e. doing pure LLVM stack build). The patch can
be reintroduced when the build system is fixed to add all needed
libraries (libunwind, compiler-rt).

llvm-svn: 282524

7 years ago[llvm-cxxfilt] Use llvm::outs(). Simplify.
Davide Italiano [Tue, 27 Sep 2016 18:50:30 +0000 (18:50 +0000)]
[llvm-cxxfilt] Use llvm::outs(). Simplify.

This adds a dependency on Support/. As llvm-cxxfilt will grow
support for options this will be needed anyway.

llvm-svn: 282523

7 years ago[x86] use isNullFPConstant(); NFCI
Sanjay Patel [Tue, 27 Sep 2016 18:48:02 +0000 (18:48 +0000)]
[x86] use isNullFPConstant(); NFCI

Also, put the related FP logic functions together to see the similarities.

llvm-svn: 282522

7 years ago[DebugInfo] Add comments to phi dbg.value tracking code, NFC
Reid Kleckner [Tue, 27 Sep 2016 18:45:31 +0000 (18:45 +0000)]
[DebugInfo] Add comments to phi dbg.value tracking code, NFC

LLVM developers might be surprised to learn that there are blocks
without valid insertion points (catchswitch), so it seems worth calling
that out explicitly.  Also add a FIXME about what we should really be
doing if we ever need to make optimized Windows EH code debuggable.

While I'm here, make auto usage more consistent with LLVM standards and
avoid an unecessary call to insertBefore.

llvm-svn: 282521

7 years ago[RDF] Add "dead" flag to node attributes
Krzysztof Parzyszek [Tue, 27 Sep 2016 18:24:33 +0000 (18:24 +0000)]
[RDF] Add "dead" flag to node attributes

llvm-svn: 282520

7 years ago[RDF] Special treatment of exception handling registers
Krzysztof Parzyszek [Tue, 27 Sep 2016 18:18:44 +0000 (18:18 +0000)]
[RDF] Special treatment of exception handling registers

A landing pad can have live-in registers that are defined by the runtime,
not the program (exception pointer register and exception selector
register). Make sure to recognize that case and not link these registers
with any defs in the program.
Each landing pad will have phi nodes added at the beginning to provide
definitions of these registers, but the uses of those phi nodes will not
have any reaching defs.

llvm-svn: 282519

7 years agoImprove CMake output of host and target triple
Chris Bieneman [Tue, 27 Sep 2016 18:08:40 +0000 (18:08 +0000)]
Improve CMake output of host and target triple

Summary:
The previous output was confusing as it would output "Taget triple:
x86_64-unknown-linux-gnu" even when LLVM_HOST_TRIPLE or
LLVM_DEFAULT_TARGET_TRIPLE were set on the CMake command line

Patch by: Alex Richardson!

Reviewers: beanz

Subscribers: Eugene.Zelenko

Differential Revision: https://reviews.llvm.org/D17067

llvm-svn: 282516

7 years ago[SCEV] Replace a struct with a function; NFC
Sanjoy Das [Tue, 27 Sep 2016 18:01:48 +0000 (18:01 +0000)]
[SCEV] Replace a struct with a function; NFC

We can do this now thanks to C++11 lambdas.

llvm-svn: 282515

7 years ago[SCEV] Use find instead of find_as; NFC
Sanjoy Das [Tue, 27 Sep 2016 18:01:46 +0000 (18:01 +0000)]
[SCEV] Use find instead of find_as; NFC

We don't need the extra generality here.

llvm-svn: 282514

7 years ago[SCEV] Reduce the scope of a struct; NFC
Sanjoy Das [Tue, 27 Sep 2016 18:01:44 +0000 (18:01 +0000)]
[SCEV] Reduce the scope of a struct; NFC

llvm-svn: 282513

7 years ago[SCEV] Remove custom RAII wrapper; NFC
Sanjoy Das [Tue, 27 Sep 2016 18:01:42 +0000 (18:01 +0000)]
[SCEV] Remove custom RAII wrapper; NFC

Instead use the pre-existing `scope_exit` class.

llvm-svn: 282512

7 years ago[SCEV] Make PendingLoopPredicates more frugal; NFCI
Sanjoy Das [Tue, 27 Sep 2016 18:01:38 +0000 (18:01 +0000)]
[SCEV] Make PendingLoopPredicates more frugal; NFCI

I don't expect `PendingLoopPredicates` to have very many
elements (e.g. when -O3'ing the sqlite3 amalgamation,
`PendingLoopPredicates` has at most 3 elements).  So now we use a
`SmallPtrSet` for it instead of the more heavyweight `DenseSet`.

llvm-svn: 282511

7 years ago[CMake] Use if(... IN_LIST ...) instead of list(FIND...)
Chris Bieneman [Tue, 27 Sep 2016 17:47:24 +0000 (17:47 +0000)]
[CMake] Use if(... IN_LIST ...) instead of list(FIND...)

NFC. This is just a little code cleanup to make things easier to read and understand.

llvm-svn: 282510

7 years agoMixed type atomic routines for unsigned integers.
Jonathan Peyton [Tue, 27 Sep 2016 17:38:48 +0000 (17:38 +0000)]
Mixed type atomic routines for unsigned integers.

New routines should be used for atomics like "<int>OP=<float>" when <int> is
unsigned. Using functions __kmpc_atomic_fixed<bits>_<op>_fp) produces incorrect
results

Differential Revision: https://reviews.llvm.org/D24756

llvm-svn: 282509

7 years agoconvert TestFatArchives.py over to no-debug-info test
Todd Fiala [Tue, 27 Sep 2016 17:17:21 +0000 (17:17 +0000)]
convert TestFatArchives.py over to no-debug-info test

We only use the .o-style debug info here regardless, so having
it run all three debuginfo styles was a waste.

This also strips out the custom build function and uses the
TestBase.build() method.

llvm-svn: 282508

7 years agoDisable monitor thread creation by default.
Jonathan Peyton [Tue, 27 Sep 2016 17:11:17 +0000 (17:11 +0000)]
Disable monitor thread creation by default.

This change set disables creation of the monitor thread by default.  The global
counter maintained by the monitor thread was replaced by logic that uses system
time directly, and cyclic yielding on Linux target was also removed since there
was no clear benefit of using it. Turning on KMP_USE_MONITOR variable (=1)
enables creation of monitor thread again if it is really necessary for some
reasons.

Differential Revision: https://reviews.llvm.org/D24739

llvm-svn: 282507

7 years agoPropagate DBG_VALUE entries when there are unvisited predecessors
Keith Walker [Tue, 27 Sep 2016 16:46:07 +0000 (16:46 +0000)]
Propagate DBG_VALUE entries when there are unvisited predecessors

Variables are sometimes missing their debug location information in
blocks in which the variables should be available. This would occur
when one or more predecessor blocks had not yet been visited by the
routine which propagated the information from predecessor blocks.

This is addressed by only considering predecessor blocks which have
already been visited.

The solution to this problem was suggested by Daniel Berlin on the
LLVM developer mailing list.

Differential Revision: https://reviews.llvm.org/D24927

llvm-svn: 282506

7 years agoUse xxhash for fast --build-id.
Rafael Espindola [Tue, 27 Sep 2016 16:43:49 +0000 (16:43 +0000)]
Use xxhash for fast --build-id.

The speed improvements I got were:

firefox
  master 7.050784981
  patch  6.842361079 0.970439617353
chromium
  master 4.260626249
  patch  4.183148025 0.981815296749
chromium fast
  master 1.829028591
  patch  1.806439277 0.987649556649
the gold plugin
  master 0.336154128
  patch  0.331893374 0.987324998728
clang
  master 0.561869781
  patch  0.558640828 0.994253200458
llvm-as
  master 0.034025959
  patch  0.033984389 0.99877828572
the gold plugin fsds
  master 0.360710529
  patch  0.356483564 0.988281559145
clang fsds
  master 0.640518422
  patch  0.632329874 0.987215749432
llvm-as fsds
  master 0.031569416
  patch  0.030822055 0.976326423017
scylla
  master 3.154770529
  patch  3.11982016 0.988921422754

llvm-svn: 282505

7 years agoRevert "Adapt to LLVM optimization remark interface change. NFC"
Adam Nemet [Tue, 27 Sep 2016 16:39:27 +0000 (16:39 +0000)]
Revert "Adapt to LLVM optimization remark interface change. NFC"

This reverts commit r282500.

llvm-svn: 282504

7 years agoRevert "Output optimization remarks in YAML"
Adam Nemet [Tue, 27 Sep 2016 16:39:24 +0000 (16:39 +0000)]
Revert "Output optimization remarks in YAML"

This reverts commit r282499.

The GCC bots are failing

llvm-svn: 282503

7 years agoAdd llvm::join_items to StringExtras.
Zachary Turner [Tue, 27 Sep 2016 16:37:30 +0000 (16:37 +0000)]
Add llvm::join_items to StringExtras.

llvm::join_items is similar to llvm::join, which produces a string
by concatenating a sequence of values together separated by a
given separator.  But it differs in that the arguments to
llvm::join() are same-type members of a container, whereas the
arguments to llvm::join_items are arbitrary types passed into
a variadic template.  The only requirement on parameters to
llvm::join_items (including for the separator themselves) is
that they be implicitly convertible to std::string or have
an overload of std::string::operator+

Differential Revision: https://reviews.llvm.org/D24880

llvm-svn: 282502

7 years ago[lit] Fix refacto introduced by rL282479.
Daniel Dunbar [Tue, 27 Sep 2016 16:17:42 +0000 (16:17 +0000)]
[lit] Fix refacto introduced by rL282479.

llvm-svn: 282501

7 years agoAdapt to LLVM optimization remark interface change. NFC
Adam Nemet [Tue, 27 Sep 2016 16:15:21 +0000 (16:15 +0000)]
Adapt to LLVM optimization remark interface change. NFC

llvm-svn: 282500

7 years agoOutput optimization remarks in YAML
Adam Nemet [Tue, 27 Sep 2016 16:15:16 +0000 (16:15 +0000)]
Output optimization remarks in YAML

This allows various presentation of this data using an external tool.
This was first recommended here[1].

As an example, consider this module:

  1 int foo();
  2 int bar();
  3
  4 int baz() {
  5   return foo() + bar();
  6 }

The inliner generates these missed-optimization remarks today (the
hotness information is pulled from PGO):

  remark: /tmp/s.c:5:10: foo will not be inlined into baz (hotness: 30)
  remark: /tmp/s.c:5:18: bar will not be inlined into baz (hotness: 30)

Now with -pass-remarks-output=<yaml-file>, we generate this YAML file:

  --- !Missed
  Pass:            inline
  Name:            NotInlined
  DebugLoc:        { File: /tmp/s.c, Line: 5, Column: 10 }
  Function:        baz
  Hotness:         30
  Args:
    - Callee: foo
    - String:  will not be inlined into
    - Caller: baz
  ...
  --- !Missed
  Pass:            inline
  Name:            NotInlined
  DebugLoc:        { File: /tmp/s.c, Line: 5, Column: 18 }
  Function:        baz
  Hotness:         30
  Args:
    - Callee: bar
    - String:  will not be inlined into
    - Caller: baz
  ...

This is a summary of the high-level decisions:

* There is a new streaming interface to emit optimization remarks.
E.g. for the inliner remark above:

   ORE.emit(DiagnosticInfoOptimizationRemarkMissed(
                DEBUG_TYPE, "NotInlined", &I)
            << NV("Callee", Callee) << " will not be inlined into "
            << NV("Caller", CS.getCaller()) << setIsVerbose());

NV stands for named value and allows the YAML client to process a remark
using its name (NotInlined) and the named arguments (Callee and Caller)
without parsing the text of the message.

Subsequent patches will update ORE users to use the new streaming API.

* I am using YAML I/O for writing the YAML file.  YAML I/O requires you
to specify reading and writing at once but reading is highly non-trivial
for some of the more complex LLVM types.  Since it's not clear that we
(ever) want to use LLVM to parse this YAML file, the code supports and
asserts that we're writing only.

On the other hand, I did experiment that the class hierarchy starting at
DiagnosticInfoOptimizationBase can be mapped back from YAML generated
here (see D24479).

* The YAML stream is stored in the LLVM context.

* In the example, we can probably further specify the IR value used,
i.e. print "Function" rather than "Value".

* As before hotness is computed in the analysis pass instead of
DiganosticInfo.  This avoids the layering problem since BFI is in
Analysis while DiagnosticInfo is in IR.

[1] https://reviews.llvm.org/D19678#419445

Differential Revision: https://reviews.llvm.org/D24587

llvm-svn: 282499

7 years agoSort headers
Adam Nemet [Tue, 27 Sep 2016 16:15:11 +0000 (16:15 +0000)]
Sort headers

llvm-svn: 282498

7 years ago[ELF] - Fixed string encoding in test. NFC.
George Rimar [Tue, 27 Sep 2016 15:58:31 +0000 (15:58 +0000)]
[ELF] - Fixed string encoding in test. NFC.

llvm-svn: 282497

7 years agoxfail TestExec.py on macOS
Todd Fiala [Tue, 27 Sep 2016 15:57:12 +0000 (15:57 +0000)]
xfail TestExec.py on macOS

Tracked by:
rdar://28476369

llvm-svn: 282496

7 years ago[ELF] - Fixed linkage error when using -g --gc-sections together.
George Rimar [Tue, 27 Sep 2016 15:55:29 +0000 (15:55 +0000)]
[ELF] - Fixed linkage error when using -g --gc-sections together.

r282444 introduced new issue, sample program below
fails to link on

assert(Piece.Live);
int main() { return 0; }
clang test.cpp -c -o out.o -g
ld.lld -flavor gnu --gc-sections out.o -o out

Problem is that .debug_info contains relocations to .debug_str:
Section (7) .rela.debug_info {
..

0xC R_X86_64_32 .debug_str 0x0
0x12 R_X86_64_32 .debug_str 0x37
..
But we do not preserve .debug_str in a right way now.

To fix this we should ignore relocations from non-allocatable sections to allocatable
to allow GC work at full power, but still should proccess relocations from non-allocatable to non-allocatable sections
as usual to mark some parts of debug sections alive to keep them so we do not end
up with such assert when trying to access dead pieces. That looks like what gold/ld do, they do
not strip .debug_str section from what I saw using sample provided.

Thanks to Evgeny Leviant for suggestions about how to fix this.

Differential revision: https://reviews.llvm.org/D24967

llvm-svn: 282495

7 years agoproject_id is from another era in phabricator land and does not provide any value.
Manuel Klimek [Tue, 27 Sep 2016 15:47:29 +0000 (15:47 +0000)]
project_id is from another era in phabricator land and does not provide any value.

Patch by Eitan Adler.

llvm-svn: 282494

7 years agoAdd xxhash to llvm.
Rafael Espindola [Tue, 27 Sep 2016 15:45:57 +0000 (15:45 +0000)]
Add xxhash to llvm.

It will be used for fast fingerprinting in lld at least.

llvm-svn: 282493

7 years agoUpdate to commit r282488, fix the buildboot failure.
Ayman Musa [Tue, 27 Sep 2016 15:37:31 +0000 (15:37 +0000)]
Update to commit r282488, fix the buildboot failure.

llvm-svn: 282492

7 years ago[SE] Change CoreTests target name
Jason Henline [Tue, 27 Sep 2016 15:32:52 +0000 (15:32 +0000)]
[SE] Change CoreTests target name

Summary:
Call it StreamExecutorCoreTests in order to prevent collision with
targets from other modules.

Reviewers: jlebar, jprice

Subscribers: beanz, mgorny, jlebar, parallel_libs-commits

Differential Revision: https://reviews.llvm.org/D24949

llvm-svn: 282491

7 years ago[docs] Fix naming style in the example
Alexander Kornienko [Tue, 27 Sep 2016 14:49:45 +0000 (14:49 +0000)]
[docs] Fix naming style in the example

llvm-svn: 282490

7 years ago[AMDGPU] Enable changing instprinter's behavior based on the per-function
Konstantin Zhuravlyov [Tue, 27 Sep 2016 14:42:48 +0000 (14:42 +0000)]
[AMDGPU] Enable changing instprinter's behavior based on the per-function
subtarget

This is a prerequisite for coming waitcnt changes

Differential Revision: https://reviews.llvm.org/D24939

llvm-svn: 282489

7 years ago[avx512] Add aliases to some missing avx512 intrinsics.
Ayman Musa [Tue, 27 Sep 2016 14:06:32 +0000 (14:06 +0000)]
[avx512] Add aliases to some missing avx512 intrinsics.

Differential Revision:https: //reviews.llvm.org/D24961

llvm-svn: 282488

7 years ago[mips] Disable tail calls temporarily
Simon Dardis [Tue, 27 Sep 2016 13:15:54 +0000 (13:15 +0000)]
[mips] Disable tail calls temporarily

Disable tail calls while the remaining bugs are fixed. Enable only for tests.

Reviewers: vkalintiris

Differential Review: https://reviews.llvm.org/D24912

llvm-svn: 282487

7 years agoWorkaround ASTMatcher crashes. Added some more test cases.
Eric Liu [Tue, 27 Sep 2016 12:54:48 +0000 (12:54 +0000)]
Workaround ASTMatcher crashes. Added some more test cases.

Summary:
- UsingDecl matcher crashed when `UsingShadowDecl` has no parent map. Workaround by moving parent check into `UsingDecl`.
- FunctionDecl matcher crashed when there is a lambda defined in parameter list (also due to no parent map).
  Workaround by putting `unless(cxxMethodDecl())` before parent check.

Reviewers: klimek, sbenza, aaron.ballman, hokein

Subscribers: aaron.ballman, cfe-commits

Differential Revision: https://reviews.llvm.org/D24862

llvm-svn: 282486

7 years ago[mips] Add rsqrt, recip for MIPS
Simon Dardis [Tue, 27 Sep 2016 12:25:15 +0000 (12:25 +0000)]
[mips] Add rsqrt, recip for MIPS

Add rsqrt.[ds], recip.[ds] for MIPS. Correct the microMIPS definitions for
architecture support and register usage.

Reviewers: vkalintiris, zoran.jovanoic

Differential Review: https://reviews.llvm.org/D24499

llvm-svn: 282485

7 years agoSilencing a Sphinx diagnostic with options, again.
Aaron Ballman [Tue, 27 Sep 2016 12:17:05 +0000 (12:17 +0000)]
Silencing a Sphinx diagnostic with options, again.

Warning, treated as error:
/opt/llvm/build.attributes.src/tools/clang/docs/CommandGuide/clang.rst:413: WARNING: unknown option: -save-stats=cwd

llvm-svn: 282484

7 years ago[cmake] Add linker option "-Wl,-z,defs" in standalone build
Michal Gorny [Tue, 27 Sep 2016 12:15:35 +0000 (12:15 +0000)]
[cmake] Add linker option "-Wl,-z,defs" in standalone build

Add the "-Wl,-z,defs" linker option that is used to prevent
underlinking. It is already used by LLVM itself but does not get
propagated into stand-alone build of libc++. This patch ensures
that the option is passed in independently of whether libc++ is built
in-tree or out-of-tree.

Patch by Lei Zhang.

Differential Revision: https://reviews.llvm.org/D24119

llvm-svn: 282483

7 years ago[docs] Make WritingAnLLVMPass.rst up-to-date with current state of things
Andrey Bokhanko [Tue, 27 Sep 2016 12:07:21 +0000 (12:07 +0000)]
[docs] Make WritingAnLLVMPass.rst up-to-date with current state of things

This patch updates WritingAnLLVMPass.rst to make it in line with current state of things.

Specifically:

* Makefile instructions replaced with CMake ones
* Filenames replaced with correct ones
* Example reformatted a bit to make it less confusing and more conforming to LLVM Coding Standards
* opt tool output updated with what it actually prints nowdays
* "gcse" (which doesn't exist anymore) replaced with "gvn" (which still does)

Differential Revision: https://reviews.llvm.org/D24233

llvm-svn: 282482

7 years ago[Power9] Builtins for ELF v.2 ABI conformance - front end portion
Nemanja Ivanovic [Tue, 27 Sep 2016 10:45:22 +0000 (10:45 +0000)]
[Power9] Builtins for ELF v.2 ABI conformance - front end portion

This patch corresponds to review:
https://reviews.llvm.org/D24397

It adds the __POWER9_VECTOR__ macro and the -mpower9-vector option along with
a number of altivec.h functions (refer to the code review for a list).

llvm-svn: 282481

7 years ago[include-fixer] Add customized editor settings documents.
Haojian Wu [Tue, 27 Sep 2016 10:43:38 +0000 (10:43 +0000)]
[include-fixer] Add customized editor settings documents.

Reviewers: bkramer

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D24719

llvm-svn: 282480

7 years agoTrying to fix lldb build breakage probably caused by rL282452
Dimitar Vlahovski [Tue, 27 Sep 2016 10:34:43 +0000 (10:34 +0000)]
Trying to fix lldb build breakage probably caused by rL282452

llvm-svn: 282479

7 years ago[Power9] Builtins for ELF v.2 API conformance - back end portion
Nemanja Ivanovic [Tue, 27 Sep 2016 08:42:12 +0000 (08:42 +0000)]
[Power9] Builtins for ELF v.2 API conformance - back end portion

This patch corresponds to review:
https://reviews.llvm.org/D24396

This patch adds support for the "vector count trailing zeroes",
"vector compare not equal" and "vector compare not equal or zero instructions"
as well as "scalar count trailing zeroes" instructions. It also changes the
vector negation to use XXLNOR (when VSX is enabled) so as not to increase
register pressure (previously this was done with a splat immediate of all
ones followed by an XXLXOR). This was done because the altivec.h
builtins (patch to follow) use vector negation and the use of an additional
register for the splat immediate is not optimal.

llvm-svn: 282478

7 years ago[clang-move] Use isStaticStorageClass matcher.
Haojian Wu [Tue, 27 Sep 2016 08:01:04 +0000 (08:01 +0000)]
[clang-move] Use isStaticStorageClass matcher.

llvm-svn: 282477

7 years ago[clang-tidy] Use isStaticStorageClass ast matcher.
Haojian Wu [Tue, 27 Sep 2016 07:58:52 +0000 (07:58 +0000)]
[clang-tidy] Use isStaticStorageClass ast matcher.

llvm-svn: 282476

7 years ago[cmake] Strip possibly-inherited compiler flags in in-tree build only
Michal Gorny [Tue, 27 Sep 2016 07:55:26 +0000 (07:55 +0000)]
[cmake] Strip possibly-inherited compiler flags in in-tree build only

Strip the set of flags (including debug defs, -m32) that could
be inherited from top-level LLVM build only when in-tree build is
performed. This prevents libcxx from confusingly and undesiredly
stripping user-supplied flags e.g. when performing packaging system
controlled multi-ABI build.

Otherwise, in order to perform 32-bit builds the build scripts would
have to use LIBCXX_BUILD_32_BITS. However, -m32 is only one of the many
different ABI flags for different targets, and it really makes no sense
to add separate CMake options for each possible -m* flag and then keep
a mapping from well-known flags to the custom CMake options.

Differential Revision: https://reviews.llvm.org/D24809

llvm-svn: 282475

7 years ago[ASTMatcher] Clarify isStaticStorageClass and hasStaticStorageDuration documents.
Haojian Wu [Tue, 27 Sep 2016 07:53:20 +0000 (07:53 +0000)]
[ASTMatcher] Clarify isStaticStorageClass and hasStaticStorageDuration documents.

Reviewers: aaron.ballman

Subscribers: klimek, cfe-commits

Differential Revision: https://reviews.llvm.org/D24928

llvm-svn: 282474

7 years ago[X86] Add test case for PR30511 and r282341.
Craig Topper [Tue, 27 Sep 2016 06:44:30 +0000 (06:44 +0000)]
[X86] Add test case for PR30511 and r282341.

llvm-svn: 282473

7 years ago[X86] Expand all-ones-vector test to cover 256-bit and 512-bit vectors.
Craig Topper [Tue, 27 Sep 2016 06:44:27 +0000 (06:44 +0000)]
[X86] Expand all-ones-vector test to cover 256-bit and 512-bit vectors.

llvm-svn: 282472

7 years ago[X86] Use std::max to calculate alignment instead of assuming RC->getSize() will...
Craig Topper [Tue, 27 Sep 2016 06:44:25 +0000 (06:44 +0000)]
[X86] Use std::max to calculate alignment instead of assuming RC->getSize() will not return a value greater than 32. I think it theoretically could be 64 for AVX-512.

llvm-svn: 282471

7 years ago[libc++abi] Default to DLL annotations on Windows
Shoaib Meenai [Tue, 27 Sep 2016 03:44:09 +0000 (03:44 +0000)]
[libc++abi] Default to DLL annotations on Windows

`__declspec(dllexport)` and `__declspec(dllimport)` should only be used
when building libc++abi as a DLL, but that's the more common use case,
so default to adding the annotations and add an option to opt out.

Similar to r282449, which made the corresponding change for libc++.

Differential Revision: https://reviews.llvm.org/D24945

llvm-svn: 282470

7 years ago[libFuzzer] run re2 test in 8 threads by default
Kostya Serebryany [Tue, 27 Sep 2016 03:33:57 +0000 (03:33 +0000)]
[libFuzzer] run re2 test in 8 threads by default

llvm-svn: 282469

7 years agoFix possible division by zero
Eric Fiselier [Tue, 27 Sep 2016 02:13:27 +0000 (02:13 +0000)]
Fix possible division by zero

llvm-svn: 282468

7 years ago[sanitizer-coverage] fix a bug in trace-gep
Kostya Serebryany [Tue, 27 Sep 2016 01:55:08 +0000 (01:55 +0000)]
[sanitizer-coverage] fix a bug in trace-gep

llvm-svn: 282467

7 years agoRemove out of date items in TODO.txt
Eric Fiselier [Tue, 27 Sep 2016 01:28:47 +0000 (01:28 +0000)]
Remove out of date items in TODO.txt

llvm-svn: 282466

7 years ago[sanitizer-coverage] don't emit the CTOR function if nothing has been instrumented
Kostya Serebryany [Tue, 27 Sep 2016 01:08:33 +0000 (01:08 +0000)]
[sanitizer-coverage] don't emit the CTOR function if nothing has been instrumented

llvm-svn: 282465

7 years agoRemove default argument from lambda to appease old MSVC.
Richard Smith [Tue, 27 Sep 2016 00:53:24 +0000 (00:53 +0000)]
Remove default argument from lambda to appease old MSVC.

llvm-svn: 282464

7 years agoDefend test against differences between 32-bit and 64-bit MSABI manglings.
Richard Smith [Tue, 27 Sep 2016 00:52:29 +0000 (00:52 +0000)]
Defend test against differences between 32-bit and 64-bit MSABI manglings.

llvm-svn: 282463

7 years agoRevert r277556. Add -lowertypetests-bitsets-level to control bitsets generation
Ivan Krasin [Tue, 27 Sep 2016 00:29:53 +0000 (00:29 +0000)]
Revert r277556. Add -lowertypetests-bitsets-level to control bitsets generation

Summary:
We don't currently need this facility for CFI. Disabling individual hot methods proved
to be a better strategy in Chrome.

Also, the design of the feature is suboptimal, as pointed out by Peter Collingbourne.

Reviewers: pcc

Subscribers: kcc

Differential Revision: https://reviews.llvm.org/D24948

llvm-svn: 282461

7 years ago[libFuzzer] add a test based on openssl-1.0.1f (finds heartbleed)
Kostya Serebryany [Tue, 27 Sep 2016 00:27:40 +0000 (00:27 +0000)]
[libFuzzer] add a test based on openssl-1.0.1f (finds heartbleed)

llvm-svn: 282460

7 years ago[libFuzzer] add -exit_on_src_pos to test libFuzzer itself, add a test script for...
Kostya Serebryany [Tue, 27 Sep 2016 00:10:20 +0000 (00:10 +0000)]
[libFuzzer] add -exit_on_src_pos to test libFuzzer itself, add a test script for RE2 that uses this flag

llvm-svn: 282458

7 years agoP0145R3 (C++17 evaluation order tweaks): evaluate the base expression before
Richard Smith [Mon, 26 Sep 2016 23:56:57 +0000 (23:56 +0000)]
P0145R3 (C++17 evaluation order tweaks): evaluate the base expression before
the pointer-to-member expression in calls through .* and ->* expressions.

llvm-svn: 282457

7 years agoLowerTypeTests: Remove unused variable.
Peter Collingbourne [Mon, 26 Sep 2016 23:56:17 +0000 (23:56 +0000)]
LowerTypeTests: Remove unused variable.

llvm-svn: 282456

7 years agoLowerTypeTests: Create LowerTypeTestsModule class and move implementation there....
Peter Collingbourne [Mon, 26 Sep 2016 23:54:39 +0000 (23:54 +0000)]
LowerTypeTests: Create LowerTypeTestsModule class and move implementation there. Related simplifications.

llvm-svn: 282455

7 years agoCreate FileOutputBuffer lazily.
Rui Ueyama [Mon, 26 Sep 2016 23:53:55 +0000 (23:53 +0000)]
Create FileOutputBuffer lazily.

So that it is clear that FileOutputBuffer does not depend on
PDB file builder. Eventually we will have to to get the file size
info from the file builder to create a file with the exact size.
NFC.

llvm-svn: 282454

7 years agoP0145R3 (C++17 evaluation order tweaks): consistently emit the LHS of array
Richard Smith [Mon, 26 Sep 2016 23:49:47 +0000 (23:49 +0000)]
P0145R3 (C++17 evaluation order tweaks): consistently emit the LHS of array
subscripting before the RHS, regardless of which is the base and which is the
index.

llvm-svn: 282453

7 years ago[lit] Add a --max-failures option.
Daniel Dunbar [Mon, 26 Sep 2016 23:38:23 +0000 (23:38 +0000)]
[lit] Add a --max-failures option.

 - This is primarily useful as a "fail fast" mode for lit, where it will stop
   running tests after the first failure.

 - Patch by Max Moiseev.

llvm-svn: 282452