platform/upstream/llvm.git
7 years agoAdd documentation for the transparent_union attribute
Alex Lorenz [Thu, 27 Oct 2016 13:51:30 +0000 (13:51 +0000)]
Add documentation for the transparent_union attribute

Differential Revision: https://reviews.llvm.org/D25995

llvm-svn: 285292

7 years agoFix warnings from python difflib.
Alexander Kornienko [Thu, 27 Oct 2016 13:46:49 +0000 (13:46 +0000)]
Fix warnings from python difflib.

llvm-svn: 285291

7 years agoUse fewer allocators.
Rafael Espindola [Thu, 27 Oct 2016 13:32:32 +0000 (13:32 +0000)]
Use fewer allocators.

Instead of having 3 section allocators per file, have 3 for all files.

This is a substantial performance improvement for some cases. Linking
chromium without gc speeds up by 1.065x.

This requires using _exit in fatal since we have to avoid destructing
an InputSection if fatal is called from the constructor.

Thanks to Rui for the suggestion.

llvm-svn: 285290

7 years ago[Sema] -Wunused-variable warning for array variables should behave
Alex Lorenz [Thu, 27 Oct 2016 13:30:51 +0000 (13:30 +0000)]
[Sema] -Wunused-variable warning for array variables should behave
similarly to scalar variables.

This commit makes the -Wunused-variable warning behaviour more consistent:
Now clang won't warn for array variables where it doesn't warn for scalar
variables.

rdar://24158862

Differential Revision: https://reviews.llvm.org/D25937

llvm-svn: 285289

7 years agoRevert r285285 "[Object/ELF] - Fixed behavior when SectionHeaderTable->sh_size is...
George Rimar [Thu, 27 Oct 2016 12:18:50 +0000 (12:18 +0000)]
Revert r285285 "[Object/ELF] - Fixed behavior when SectionHeaderTable->sh_size is too large."
It broke BB.

llvm-svn: 285288

7 years agoAdd const versions of Expr::getInits() and Expr::inits()
Erik Verbruggen [Thu, 27 Oct 2016 12:12:08 +0000 (12:12 +0000)]
Add const versions of Expr::getInits() and Expr::inits()

llvm-svn: 285287

7 years ago[SLP] Fix for PR30626: Compiler crash inside SLP Vectorizer.
Alexey Bataev [Thu, 27 Oct 2016 12:02:28 +0000 (12:02 +0000)]
[SLP] Fix for PR30626: Compiler crash inside SLP Vectorizer.

After successfull horizontal reduction vectorization attempt for PHI node
vectorizer tries to update root binary op by combining vectorized tree
and the ReductionPHI node. But during vectorization this ReductionPHI
can be vectorized itself and replaced by the `undef` value, while the
instruction itself is marked for deletion. This 'marked for deletion'
PHI node then can be used in new binary operation, causing "Use still
stuck around after Def is destroyed" crash upon PHI node deletion.

Also the test is fixed to make it perform actual testing.

Differential Revision: https://reviews.llvm.org/D25671

llvm-svn: 285286

7 years ago[Object/ELF] - Fixed behavior when SectionHeaderTable->sh_size is too large.
George Rimar [Thu, 27 Oct 2016 11:50:04 +0000 (11:50 +0000)]
[Object/ELF] - Fixed behavior when SectionHeaderTable->sh_size is too large.

Elf.h already has code checking that section table does not go past end of file.
Problem is that this check may not work on values greater than UINT64_MAX / Header->e_shentsize
because of calculation overflow.

Parch fixes the issue.

Differential revision: https://reviews.llvm.org/D25432

llvm-svn: 285285

7 years ago[Object/ELF] - Do not allow overflow when checking section size/offset.
George Rimar [Thu, 27 Oct 2016 11:44:56 +0000 (11:44 +0000)]
[Object/ELF] - Do not allow overflow when checking section size/offset.

Overflow was the reason of incorrect passing the check,
patch fixes the case.

Differentail revision: https://reviews.llvm.org/D25514

llvm-svn: 285284

7 years agoFixed a memory leak related to task dependencies.
Andrey Churbanov [Thu, 27 Oct 2016 11:43:07 +0000 (11:43 +0000)]
Fixed a memory leak related to task dependencies.

Differential Revision: http://reviews.llvm.org/D25504

Patch by Alex Duran.

llvm-svn: 285283

7 years ago[Object/ELF] - Do not crash if string table sh_size is equal to zero.
George Rimar [Thu, 27 Oct 2016 11:41:57 +0000 (11:41 +0000)]
[Object/ELF] - Do not crash if string table sh_size is equal to zero.

Revealed using "id_000038,sig_11,src_000015,op_havoc,rep_16" from PR30540,
when sh_size was 0, crash happened.

Differential revision: https://reviews.llvm.org/D25091

llvm-svn: 285282

7 years agoFix MSVC warning about missing 'this' from lambda default capture mode
Simon Pilgrim [Thu, 27 Oct 2016 10:51:29 +0000 (10:51 +0000)]
Fix MSVC warning about missing 'this' from lambda default capture mode

llvm-svn: 285281

7 years ago[ARM] Add newline char to test.
Sam Parker [Thu, 27 Oct 2016 10:43:02 +0000 (10:43 +0000)]
[ARM] Add newline char to test.

Missed a newline in the previous commit.

Differential Revision: https://reviews.llvm.org/D26027

llvm-svn: 285280

7 years ago[ELF] Synthetic symbol definitions for ARM static linking
Peter Smith [Thu, 27 Oct 2016 10:28:53 +0000 (10:28 +0000)]
[ELF] Synthetic symbol definitions for ARM static linking

When static linking in ARM (like Mips) __tls_get_addr is defined by
the library so we should not define it as a synthetic.

We also need to add __exidx_start and __exidx_end for the .ARM.exidx
section as the static libc library startup code is expecting them to
be defined by the default linker script for static linking on ARM.

Differential revision: https://reviews.llvm.org/D25978

llvm-svn: 285279

7 years ago[ARM] Predicate UMAAL selection on hasDSP.
Sam Parker [Thu, 27 Oct 2016 09:47:10 +0000 (09:47 +0000)]
[ARM] Predicate UMAAL selection on hasDSP.

UMAAL is a DSP instruction and it is not available on thumbv7m
(Cortex-M3) and thumbv6m (Cortex-M0+1) targets. Also fix wrong
CHECK prefix in longMAC.ll test.

Patch by Vadzim Dambrouski.

Differential Revision: https://reviews.llvm.org/D25890

llvm-svn: 285278

7 years agoEmpty the CtorLists/DtorLists once emitted.
Vassil Vassilev [Thu, 27 Oct 2016 09:12:20 +0000 (09:12 +0000)]
Empty the CtorLists/DtorLists once emitted.

This is essential when clang is running in incremental processing mode because
we don't want to reemit the 'tors over and over again.

Patch by Axel Naumann!

Reviewed by Richard Smith and me. (https://reviews.llvm.org/D25605)

llvm-svn: 285277

7 years ago[Xray] Don't generate output for xray tests
Benjamin Kramer [Thu, 27 Oct 2016 08:49:18 +0000 (08:49 +0000)]
[Xray] Don't generate output for xray tests

Patch by Sam McCall!

Differential Revision: http://reviews.llvm.org/D26024

llvm-svn: 285276

7 years agoMark invalid RecordDecls as completed.
Erik Verbruggen [Thu, 27 Oct 2016 08:37:14 +0000 (08:37 +0000)]
Mark invalid RecordDecls as completed.

Sema::ActOnTag creates TagDecls for records. However, if those record
declarations are invalid, and the parser is in C++ mode, it would
silently drop the TagDecl (and leave it as "beingDefined"). The problem
is that other code (e.g. the ASTWriter) will serialize all types, and
expects them to be complete. So, leaving them open would result in
failing asserts.

Fixes PR20320

Differential Revision: http://reviews.llvm.org/D21176

llvm-svn: 285275

7 years ago[AVR] Generate all of the TableGen files we need
Dylan McKay [Thu, 27 Oct 2016 08:20:47 +0000 (08:20 +0000)]
[AVR] Generate all of the TableGen files we need

This enables generation of all of the TableGen files that are used
downstream.

llvm-svn: 285274

7 years agoAMDGPU: Fix SILoadStoreOptimizer when writes cannot be merged due register dependencies
Nicolai Haehnle [Thu, 27 Oct 2016 08:15:07 +0000 (08:15 +0000)]
AMDGPU: Fix SILoadStoreOptimizer when writes cannot be merged due register dependencies

Summary:
When finding a match for a merge and collecting the instructions that must
be moved, keep in mind that the instruction we merge might actually use one
of the defs that are being moved.

Fixes piglit spec/arb_enhanced_layouts/execution/component-layout/vs-tcs-load-output[-indirect].

The fact that the ds_read in the test case is not eliminated suggests that
there might be another problem related to alias analysis, but that's a
separate problem: this pass should still work correctly even when earlier
optimization passes missed something or were disabled.

Reviewers: tstellarAMD, arsenm

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

Differential Revision: https://reviews.llvm.org/D25829

llvm-svn: 285273

7 years ago[AVR] Compile the disassembler
Dylan McKay [Thu, 27 Oct 2016 08:09:15 +0000 (08:09 +0000)]
[AVR] Compile the disassembler

This also updates references of 'TheAVRTarget' to the new
'getTheAVRTarget()' method.

llvm-svn: 285272

7 years ago[coroutines] Build fallthrough and set_exception statements.
Eric Fiselier [Thu, 27 Oct 2016 07:30:31 +0000 (07:30 +0000)]
[coroutines] Build fallthrough and set_exception statements.

Summary:
This patch adds semantic checking and building of the fall-through `co_return;` statement as well as the `p.set_exception(std::current_exception())` call for handling uncaught exceptions.

The fall-through statement is built and checked according to:
> [dcl.fct.def.coroutine]/4
> The unqualified-ids return_void and return_value are looked up in the scope of class P. If
> both are found, the program is ill-formed. If the unqualified-id return_void is found, flowing
> off the end of a coroutine is equivalent to a co_return with no operand. Otherwise, flowing off
> the end of a coroutine results in undefined behavior.

Similarly the `set_exception` call is only built when that unqualified-id is found in the scope of class P.

Additionally this patch adds fall-through warnings for non-void returning coroutines. Since it's surprising undefined behavior I thought it would be important to add the warning right away.

Reviewers: majnemer, GorNishanov, rsmith

Subscribers: mehdi_amini, cfe-commits

Differential Revision: https://reviews.llvm.org/D25349

llvm-svn: 285271

7 years ago[AVR] Add AVRISelDAGToDAG.cpp
Dylan McKay [Thu, 27 Oct 2016 07:03:47 +0000 (07:03 +0000)]
[AVR] Add AVRISelDAGToDAG.cpp

Summary: This pulls the AVR instruction selector in-tree.

Reviewers: arsenm, kparzysz

Subscribers: llvm-commits, wdng, beanz, japaric, mgorny

Differential Revision: https://reviews.llvm.org/D25278

llvm-svn: 285270

7 years ago[AVR] Add the machine code emitter
Dylan McKay [Thu, 27 Oct 2016 06:56:46 +0000 (06:56 +0000)]
[AVR] Add the machine code emitter

Reviewers: arsenm, kparzysz

Subscribers: wdng, beanz, japaric, llvm-commits, mgorny

Differential Revision: https://reviews.llvm.org/D25388

llvm-svn: 285269

7 years ago[PPC] add vector byte reverse functions to altivec.h
Nemanja Ivanovic [Thu, 27 Oct 2016 06:23:57 +0000 (06:23 +0000)]
[PPC] add vector byte reverse functions to altivec.h

This patch corresponds to review https://reviews.llvm.org/D25915.
Committing on behalf of Sean Fertile.

llvm-svn: 285268

7 years ago[PowerPC] - No SExt/ZExt needed for count trailing zeros
Nemanja Ivanovic [Thu, 27 Oct 2016 05:17:58 +0000 (05:17 +0000)]
[PowerPC] - No SExt/ZExt needed for count trailing zeros

This patch corresponds to review:
https://reviews.llvm.org/D25896

It just eliminates the redundant ZExt after a count trailing zeros instruction.

llvm-svn: 285267

7 years ago[XRay] Check in Clang whether XRay supports the target when -fxray-instrument is...
Dean Michael Berris [Thu, 27 Oct 2016 04:56:14 +0000 (04:56 +0000)]
[XRay] Check in Clang whether XRay supports the target when -fxray-instrument is passed

Summary:
Added the code which explicitly emits an error in Clang in case
`-fxray-instrument` is passed, but XRay is not supported for the
selected target.

Reviewers: rsmith, aaron.ballman, rnk, dberris

Differential Revision: https://reviews.llvm.org/D24799

llvm-svn: 285266

7 years agoAtomics library: provide operations for __int128 when it is available.
Richard Smith [Thu, 27 Oct 2016 01:46:24 +0000 (01:46 +0000)]
Atomics library: provide operations for __int128 when it is available.

llvm-svn: 285265

7 years agoRemove check for -o option in offloading actions builder.
Samuel Antao [Thu, 27 Oct 2016 01:08:58 +0000 (01:08 +0000)]
Remove check for -o option in offloading actions builder.

This check is also present when jobs are built, so the offloading builder check is not needed anymore.

llvm-svn: 285264

7 years agoFix bug when compiling CUDA code with -emit-llvm and -o.
Samuel Antao [Thu, 27 Oct 2016 00:53:34 +0000 (00:53 +0000)]
Fix bug when compiling CUDA code with -emit-llvm and -o.

In this case the device code is not injected into an host action and therefore the
user should get an error as -o can't be used when generating two outputs.

llvm-svn: 285263

7 years ago[libFuzzer] speculatively trying to fix the Mac build; second attempt
Kostya Serebryany [Thu, 27 Oct 2016 00:36:38 +0000 (00:36 +0000)]
[libFuzzer] speculatively trying to fix the Mac build; second attempt

llvm-svn: 285262

7 years agoxray-extract.cc: Quick fix for mingw, to avoid errc::protocol_error.
NAKAMURA Takumi [Thu, 27 Oct 2016 00:34:24 +0000 (00:34 +0000)]
xray-extract.cc: Quick fix for mingw, to avoid errc::protocol_error.

errc::protocol_error is winsock-oriented and mingw doesn't include it. (MS does, though)

llvm-svn: 285261

7 years ago[libFuzzer] revert 285259 -- hit commit too soon
Kostya Serebryany [Thu, 27 Oct 2016 00:24:34 +0000 (00:24 +0000)]
[libFuzzer] revert 285259 -- hit commit too soon

llvm-svn: 285260

7 years ago[libFuzzer] speculatively trying to fix the Mac build
Kostya Serebryany [Thu, 27 Oct 2016 00:22:39 +0000 (00:22 +0000)]
[libFuzzer] speculatively trying to fix the Mac build

llvm-svn: 285259

7 years agoRefactor call emission to package the function pointer together with
John McCall [Wed, 26 Oct 2016 23:46:34 +0000 (23:46 +0000)]
Refactor call emission to package the function pointer together with
abstract information about the callee.  NFC.

The goal here is to make it easier to recognize indirect calls and
trigger additional logic in certain cases.  That logic will come in
a later patch; in the meantime, I felt that this was a significant
improvement to the code.

llvm-svn: 285258

7 years ago[IR] Retire unused getGEPReturnType overload. NFCI.
Davide Italiano [Wed, 26 Oct 2016 23:46:16 +0000 (23:46 +0000)]
[IR] Retire unused getGEPReturnType overload. NFCI.

llvm-svn: 285257

7 years ago[APFloat] Fix APFloat::getExactInverse when the input is nullptr. This is a regressio...
Tim Shen [Wed, 26 Oct 2016 23:31:41 +0000 (23:31 +0000)]
[APFloat] Fix APFloat::getExactInverse when the input is nullptr. This is a regression introduced by r285105.

Reviewers: kbarton, echristo, iteratee, eugenis

Subscribers: llvm-commits, mehdi_amini

Differential Revision: https://reviews.llvm.org/D26017

llvm-svn: 285256

7 years ago[sanitizers] Set Darwin specific linker and compiler flags for all tests
Anna Zaks [Wed, 26 Oct 2016 23:23:38 +0000 (23:23 +0000)]
[sanitizers] Set Darwin specific linker and compiler flags for all tests

Looks like we are missing these flags only in tsan and sanitizer-common.

This results in linker warnings in some settings as it can cause the Unit
tests to be built with a different SDK version than that was used to build
the runtime. For example, we are not setting the minimal deployment target
on the tests but are setting the minimal deployment target for the sanitizer
library, which leads to the following warning on some bots: ld: warning:
object file (sanitizer_posix_test.cc.i386.o) was built for newer OSX version
(10.12) than being linked (10.11).

Differential Revision: https://reviews.llvm.org/D25860  https://reviews.llvm.org/D25352

llvm-svn: 285255

7 years agoUnconditionally pass `-lto_library` to the linker on Darwin
Mehdi Amini [Wed, 26 Oct 2016 23:23:08 +0000 (23:23 +0000)]
Unconditionally pass `-lto_library` to the linker on Darwin

We're only doing it with -flto currently, however it never "hurt"
to pass it, and users that are linking without -flto can get in
trouble if one of the dependency (a static library for instance)
contains bitcode.

Differential Revision: https://reviews.llvm.org/D25932

llvm-svn: 285254

7 years ago[analyzer] Report CFNumberGetValue API misuse
Anna Zaks [Wed, 26 Oct 2016 22:51:47 +0000 (22:51 +0000)]
[analyzer] Report CFNumberGetValue API misuse

This patch contains 2 improvements to the CFNumber checker:
 - Checking of CFNumberGetValue misuse.
 - Treating all CFNumber API misuse errors as non-fatal. (Previously we treated errors that could cause uninitialized memory as syncs and the truncation errors as non-fatal.)

This implements a subset of functionality from https://reviews.llvm.org/D17954.

Differential Revision: https://reviews.llvm.org/D25876

llvm-svn: 285253

7 years agollvm-objdump: Make some error messages more consistent
Justin Bogner [Wed, 26 Oct 2016 22:37:52 +0000 (22:37 +0000)]
llvm-objdump: Make some error messages more consistent

Most of the version of report_error were quoting the filename and
printing a colon between the file name and the error message, but this
one wasn't doing either of those. Fix the output to be more
consistent.

llvm-svn: 285252

7 years ago[CUDA] Move device placement new definitions into a wrapper header.
Justin Lebar [Wed, 26 Oct 2016 22:13:26 +0000 (22:13 +0000)]
[CUDA] Move device placement new definitions into a wrapper header.

Previously, these were always included -- after this change, you have to
 #include <new>, which is consistent with how things ought to work.

llvm-svn: 285251

7 years ago[CUDA] Switch cuda_wrappers/complex to use a proper include guard instead of #pragma...
Justin Lebar [Wed, 26 Oct 2016 22:13:20 +0000 (22:13 +0000)]
[CUDA] Switch cuda_wrappers/complex to use a proper include guard instead of #pragma once.

This is consistent with the rest of our internal headers.

llvm-svn: 285250

7 years ago[utils] Add a '--unified-report' option to the code coverage prep script
Vedant Kumar [Wed, 26 Oct 2016 22:07:39 +0000 (22:07 +0000)]
[utils] Add a '--unified-report' option to the code coverage prep script

In --unified-report mode, a single coverage report is prepared for all
specified binaries and written to *report_dir*. This mode is compatible
with all existing script options, including the --restrict mode which is
used to limit coverage reporting to certain files or directories.

This should not break any existing users of the script.

llvm-svn: 285249

7 years ago[utils] Use print_function in the code coverage prep script, NFC.
Vedant Kumar [Wed, 26 Oct 2016 22:07:37 +0000 (22:07 +0000)]
[utils] Use print_function in the code coverage prep script, NFC.

llvm-svn: 285248

7 years ago[utils] Add an '--only-merge' option to the code coverage prep script
Vedant Kumar [Wed, 26 Oct 2016 22:07:35 +0000 (22:07 +0000)]
[utils] Add an '--only-merge' option to the code coverage prep script

In --only-merge mode, the script terminates after the profile merging
step.  This makes the script less stateful: it's more natural to split
the merge out into a separate step instead of relying on the first
invocation of the script to do it.

This should not break any existing users of the script.

llvm-svn: 285247

7 years ago[AArch64] Create feature set for Samsung Exynos-M2
Evandro Menezes [Wed, 26 Oct 2016 22:06:20 +0000 (22:06 +0000)]
[AArch64] Create feature set for Samsung Exynos-M2

Since Exynos-M2 improved the FP square root unit a bit over the one in
Exynos-M1, it does not benefit from using the Newton series for such
operations.

llvm-svn: 285246

7 years agoFixing typos in __kmp_release_deps trace outputs
Jonathan Peyton [Wed, 26 Oct 2016 21:46:43 +0000 (21:46 +0000)]
Fixing typos in __kmp_release_deps trace outputs

Patch by Victor Campos

Differential Revision: https://reviews.llvm.org/D25972

llvm-svn: 285244

7 years agoUse getpagesize() instead of PAGE_SIZE macro when KMP_OS_LINUX is true
Jonathan Peyton [Wed, 26 Oct 2016 21:42:48 +0000 (21:42 +0000)]
Use getpagesize() instead of PAGE_SIZE macro when KMP_OS_LINUX is true

Patch by Victor Campos

Differential Revision: https://reviews.llvm.org/D26001

llvm-svn: 285243

7 years agoDebugInfo: fix incorrect alignment type (NFC)
Victor Leschuk [Wed, 26 Oct 2016 21:32:29 +0000 (21:32 +0000)]
DebugInfo: fix incorrect alignment type (NFC)

Change type of some missed DebugInfo-related alignment variables,
that are still uint64_t, to uint32_t.

Original change introduced in r284482.

llvm-svn: 285242

7 years agoHave on addRegular forward to another. NFC.
Rafael Espindola [Wed, 26 Oct 2016 20:57:14 +0000 (20:57 +0000)]
Have on addRegular forward to another. NFC.

Another step in making it easier to create symbols without an ELf_Sym.

llvm-svn: 285241

7 years agoRemove a constructor to simplify.
Rui Ueyama [Wed, 26 Oct 2016 20:44:32 +0000 (20:44 +0000)]
Remove a constructor to simplify.

llvm-svn: 285240

7 years ago[RecursiveASTVisitor] Visit the implicit expression of a CXXDefaultArgExpr
Malcolm Parsons [Wed, 26 Oct 2016 20:39:54 +0000 (20:39 +0000)]
[RecursiveASTVisitor] Visit the implicit expression of a CXXDefaultArgExpr

Summary:
The matcher
varDecl(hasDescendant(
    callExpr(hasDeclaration(functionDecl(unless(isNoThrow()))))))
didn't match calls from default arguments because the expression
for a CXXDefaultArgExpr was not visited.

Reviewers: klimek, jdennett, alexfh, aaron.ballman

Subscribers: aaron.ballman, cfe-commits

Differential Revision: https://reviews.llvm.org/D25992

llvm-svn: 285239

7 years agoMake the DefinedRegular constructors more regular. NFC.
Rafael Espindola [Wed, 26 Oct 2016 20:34:59 +0000 (20:34 +0000)]
Make the DefinedRegular constructors more regular. NFC.

All the global ones now just forward to the same constructor. This
makes it easier to construct them without creating a Elf_Sym.

llvm-svn: 285238

7 years ago[lit] Work around Windows MSys command line tokenization bug
Reid Kleckner [Wed, 26 Oct 2016 20:29:27 +0000 (20:29 +0000)]
[lit] Work around Windows MSys command line tokenization bug

Summary:
This will allow us to revert LLD r284768, which added spaces to get MSys
echo to print what we want.

Reviewers: ruiu, inglorion, rafael

Subscribers: modocache, llvm-commits

Differential Revision: https://reviews.llvm.org/D26009

llvm-svn: 285237

7 years agoFix build breakage.
Rui Ueyama [Wed, 26 Oct 2016 20:27:38 +0000 (20:27 +0000)]
Fix build breakage.

llvm-svn: 285236

7 years agoAdd comments.
Rui Ueyama [Wed, 26 Oct 2016 20:26:29 +0000 (20:26 +0000)]
Add comments.

llvm-svn: 285235

7 years agoCleanup nonportable behavior in tests for std::any
Casey Carter [Wed, 26 Oct 2016 20:18:13 +0000 (20:18 +0000)]
Cleanup nonportable behavior in tests for std::any

Fixes MS issues 63, 64, and 65.

test/std/utilities/any/any.class/any.cons/move.pass.cpp:
* "Moves are always destructive" is not a portable assumption; check with LIBCPP_ASSERT.

test/std/utilities/any/any.class/any.cons/value.pass.cpp:
* The standard does not forbid initializing std::any from any pointer-to-function type. Remove the non-conforming "DecayTag" test.

test/std/utilities/any/any.class/any.modifiers/swap.pass.cpp:
* Self-swap is not specified to perform no moves; check with LIBCPP_ASSERT.

Differential Revision: https://reviews.llvm.org/D26007

llvm-svn: 285234

7 years ago[PPC] Remove testcase from incorrect directory
Ehsan Amiri [Wed, 26 Oct 2016 20:16:59 +0000 (20:16 +0000)]
[PPC] Remove testcase from incorrect directory

During my last commit this testcase was put in an incorrect directory. Removing
it. Will put it in the right directory when I can verify everything is correct.

llvm-svn: 285233

7 years agoARM: don't rely on push/pop reglists being in order when folding SP adjust.
Tim Northover [Wed, 26 Oct 2016 20:01:00 +0000 (20:01 +0000)]
ARM: don't rely on push/pop reglists being in order when folding SP adjust.

It would be a very nice invariant to rely on, but unfortunately it doesn't
necessarily hold (and the causes of mis-sorted reglists appear to be quite
varied) so to be robust the frame lowering code can't assume that the first
register in the list is also the first one that actually gets pushed.

Should fix an issue where we were turning something like:

    push {r8, r4, r7, lr}
    sub sp, #24

into nonsense like:

    push {r2, r3, r4, r5, r6, r7, r8, r4, r7, lr}

llvm-svn: 285232

7 years agoDo not assume that FP vector operands are never legalized by expanding
Nemanja Ivanovic [Wed, 26 Oct 2016 19:51:35 +0000 (19:51 +0000)]
Do not assume that FP vector operands are never legalized by expanding

This patch ensures that if a floating point vector operand is legalized by
expanding, it is legalized through the stack rather than by calling
DAGTypeLegalizer::IntegerToVector which will cause a failure since the operand
is a non-integer type.

This fixes PR 30715.

llvm-svn: 285231

7 years ago[PowerPC] Implement vector_insert_exp builtins - clang portion
Nemanja Ivanovic [Wed, 26 Oct 2016 19:27:11 +0000 (19:27 +0000)]
[PowerPC] Implement vector_insert_exp builtins - clang portion

This patch corresponds to review https://reviews.llvm.org/D25956.
Committing on behalf of Zaara Syeda.

llvm-svn: 285229

7 years agoSimplify `x >=u x >> y` and `x >=u x udiv y`
Sanjoy Das [Wed, 26 Oct 2016 19:18:43 +0000 (19:18 +0000)]
Simplify `x >=u x >> y` and `x >=u x udiv y`

Summary:
Extends InstSimplify to handle both `x >=u x >> y` and `x >=u x udiv y`.

This is a folloup of rL258422 and
https://github.com/rust-lang/rust/pull/30917 where llvm failed to
optimize away the bounds checking in a binary search.

Patch by Arthur Silva!

Reviewers: sanjoy

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D25941

llvm-svn: 285228

7 years agoRevert "[AliasSetTracker] Make AST smarter about intrinsics that don't actually affec...
Chad Rosier [Wed, 26 Oct 2016 19:18:19 +0000 (19:18 +0000)]
Revert "[AliasSetTracker] Make AST smarter about intrinsics that don't actually affect memory."

This reverts commit r285191.

LICM appears to rely on the Alias Set Tracker hitting lifetime markers to prevent
code from being moved outside of the original scope.

llvm-svn: 285227

7 years agoFix an issue where frame variable -s <varname> would not show the scope even though...
Enrico Granata [Wed, 26 Oct 2016 19:17:49 +0000 (19:17 +0000)]
Fix an issue where frame variable -s <varname> would not show the scope even though the user asked for it

Part of rdar://28434047

llvm-svn: 285226

7 years ago[PowerPC] Implement vec_insert_exp builtins - llvm portion
Nemanja Ivanovic [Wed, 26 Oct 2016 19:03:40 +0000 (19:03 +0000)]
[PowerPC] Implement vec_insert_exp builtins - llvm portion

This revision corresponds to review: https://reviews.llvm.org/D25957.
Committing on behalf of Zaara Syeda.

llvm-svn: 285225

7 years agoCall _exit.
Rafael Espindola [Wed, 26 Oct 2016 18:59:00 +0000 (18:59 +0000)]
Call _exit.

As the state of lld gets more complicated, shutting down gets more
expensive.

In a normal lld run we can just call _exit immediately after renaming
the temporary output file. We still want the ability to run a full
shutdown since that is useful for detecting memory leaks.

This patch adds a --full-shutdown flag and changes lit to use it.

llvm-svn: 285224

7 years ago[libFuzzer] simplify TracePC::HandleTrace even further. Also, when dealing with ...
Kostya Serebryany [Wed, 26 Oct 2016 18:52:04 +0000 (18:52 +0000)]
[libFuzzer] simplify TracePC::HandleTrace even further. Also, when dealing with -exit_on_src_pos, symbolize every PC only once

llvm-svn: 285223

7 years agoFix test from r285217.
Chad Rosier [Wed, 26 Oct 2016 18:49:16 +0000 (18:49 +0000)]
Fix test from r285217.

llvm-svn: 285222

7 years agoPass a InputSectionData to classoff.
Rafael Espindola [Wed, 26 Oct 2016 18:44:57 +0000 (18:44 +0000)]
Pass a InputSectionData to classoff.

This allows a non template class to hold input sections.

llvm-svn: 285221

7 years agoMove a helper function that doesn't depend on a class out of the class.
Rui Ueyama [Wed, 26 Oct 2016 18:28:08 +0000 (18:28 +0000)]
Move a helper function that doesn't depend on a class out of the class.

llvm-svn: 285220

7 years agoDefine a helper function to demangle symbols.
Rui Ueyama [Wed, 26 Oct 2016 18:28:06 +0000 (18:28 +0000)]
Define a helper function to demangle symbols.

llvm-svn: 285219

7 years ago[PPC] Implement vector reverse elements builtins (vec_reve)
Nemanja Ivanovic [Wed, 26 Oct 2016 18:25:45 +0000 (18:25 +0000)]
[PPC] Implement vector reverse elements builtins (vec_reve)

This patch corresponds to review https://reviews.llvm.org/D25906.
Committing on behalf of Tony Jiang.

llvm-svn: 285218

7 years ago[AArch64] Avoid materializing constant 1 when generating cneg instructions.
Chad Rosier [Wed, 26 Oct 2016 18:15:32 +0000 (18:15 +0000)]
[AArch64] Avoid materializing constant 1 when generating cneg instructions.

Instead of

 cmp w0, #1
 orr w8, wzr, #0x1
 cneg w0, w8, ne

we now generate

 cmp w0, #1
 csinv w0, w0, wzr, eq

PR28965

llvm-svn: 285217

7 years agoActually use = delete to mark constructors and operators we want to not exist. Just...
Enrico Granata [Wed, 26 Oct 2016 18:12:52 +0000 (18:12 +0000)]
Actually use = delete to mark constructors and operators we want to not exist. Just declaring them as private works, but it can confuse certain tools as it doesn't actually declare intent, and the C++11 way is more expressive anyway, so this is pure win

rdar://28960209

llvm-svn: 285216

7 years ago[WebAssembly] Update the README.txt.
Dan Gohman [Wed, 26 Oct 2016 17:44:09 +0000 (17:44 +0000)]
[WebAssembly] Update the README.txt.

Update the README.txt with newer information, add a link to the Emscripten
page explaining the current easiest way to use the LLVM wasm backend, and
mention that other ways of using the LLVM wasm backend are in development.

llvm-svn: 285215

7 years ago[MC] Fix comma typo in .loc parsing
Nirav Dave [Wed, 26 Oct 2016 17:28:58 +0000 (17:28 +0000)]
[MC] Fix comma typo in .loc parsing

llvm-svn: 285214

7 years agoSilence unused parameter warnings in archetypes.hpp
Casey Carter [Wed, 26 Oct 2016 17:22:25 +0000 (17:22 +0000)]
Silence unused parameter warnings in archetypes.hpp
Reviewed at: https://reviews.llvm.org/D25958

llvm-svn: 285213

7 years agoReapply: "Remove debug location from common tail when tail-merging"
Robert Lougher [Wed, 26 Oct 2016 17:01:47 +0000 (17:01 +0000)]
Reapply: "Remove debug location from common tail when tail-merging"

This reapplies revision 285093.  Original commit message:

The branch folding pass tail merges blocks into a common-tail.  However, the
tail retains the debug information from one of the original inputs to the
merge (chosen randomly).  This is a problem for sampled-based PGO, as hits
on the common-tail will be attributed to whichever block was chosen,
irrespective of which path was actually taken to the common-tail.

This patch fixes the issue by nulling the debug location for the common-tail.

Differential Revision: https://reviews.llvm.org/D25742

llvm-svn: 285212

7 years agoAMDGPU: Add missing ISA versions gfx7.x.x and 8.x.x.
Yaxun Liu [Wed, 26 Oct 2016 16:40:21 +0000 (16:40 +0000)]
AMDGPU: Add missing ISA versions gfx7.x.x and 8.x.x.

Patch by Laurent Morichetti.

Differential Revision: https://reviews.llvm.org/D25920

llvm-svn: 285211

7 years agoAMDGPU: Refactor processor definition to use ISA version features
Yaxun Liu [Wed, 26 Oct 2016 16:37:56 +0000 (16:37 +0000)]
AMDGPU: Refactor processor definition to use ISA version features

Add missing ISA versions 7.0.2/8.0.4/8.1.0. to backend.

Refactor processor definition to use ISA version features.

Fixed ISA version for stoney.

Based on Laurent Morichetti's patch.

Differential Revision: https://reviews.llvm.org/D25919

llvm-svn: 285210

7 years ago[scudo] Lay the foundation for 32-bit support
Kostya Kortchinsky [Wed, 26 Oct 2016 16:16:58 +0000 (16:16 +0000)]
[scudo] Lay the foundation for 32-bit support

Summary:
In order to support 32-bit platforms, we have to make some adjustments in
multiple locations, one of them being the Scudo chunk header. For it to fit on
64 bits (as a reminder, on x64 it's 128 bits), I had to crunch the space taken
by some of the fields. In order to keep the offset field small, the secondary
allocator was changed to accomodate aligned allocations for larger alignments,
hence making the offset constant for chunks serviced by it.

The resulting header candidate has been added, and further modifications to
allow 32-bit support will follow.

Another notable change is the addition of MaybeStartBackgroudThread() to allow
release of the memory to the OS.

Reviewers: kcc

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D25688

llvm-svn: 285209

7 years ago[ubsan] Fix vptr.cpp test to be more resilient. NFC.
Robert Lougher [Wed, 26 Oct 2016 16:02:36 +0000 (16:02 +0000)]
[ubsan] Fix vptr.cpp test to be more resilient. NFC.

The test contains a switch statement in which two of the cases are
tail-merged, with the call to __ubsan_handle_dynamic_type_cache_miss_abort
in the common tail. When tail-merging occurs, the debug location of the
tail is randomly taken from one of the merge inputs.  Luckily for the test,
the expected line number in the check is the one which is chosen by the
tail-merge.  However, if the switch cases are re-ordered the test will
fail.

This patch disables tail-merge, making the test resilient to changes
in tail-merge, and unblocking review D25742.  It does not change the
semantics of the test.

llvm-svn: 285208

7 years agoIntroduce updateDiscriminator interface to DILocation to make it cleaner assigning...
Dehao Chen [Wed, 26 Oct 2016 15:48:45 +0000 (15:48 +0000)]
Introduce updateDiscriminator interface to DILocation to make it cleaner assigning discriminators.

Summary: This patch introduces updateDiscriminator to DILocation so that it can be directly called by AddDiscriminator. It also makes it easier to update the discriminator later.

Reviewers: dnovillo, dblaikie, aprantl, echristo

Subscribers: mehdi_amini, llvm-commits

Differential Revision: https://reviews.llvm.org/D25959

llvm-svn: 285207

7 years ago[CMake] Adding example distribution CMake cache files
Chris Bieneman [Wed, 26 Oct 2016 15:41:38 +0000 (15:41 +0000)]
[CMake] Adding example distribution CMake cache files

These cache file are provided as an example of how to set up simple multi-stage CMake builds. I have a batch of documentation updates for LLVM.org which reference these files.

llvm-svn: 285206

7 years agoReduce the number of allocators.
Rafael Espindola [Wed, 26 Oct 2016 15:34:24 +0000 (15:34 +0000)]
Reduce the number of allocators.

We used to have one allocator per file, which reduces the advantage of
using an allocator in the first place.

This is a small speed up is most cases. The largest speedup was in
1.014X in chromium no-gc. The largest slowdown was scylla at 1.003X.

llvm-svn: 285205

7 years ago[ARM RT] Fix broken clear_cache debug build on ARM
Renato Golin [Wed, 26 Oct 2016 15:20:33 +0000 (15:20 +0000)]
[ARM RT] Fix broken clear_cache debug build on ARM

clear_cache is using R7 for the SVC call and that's the frame pointer in
GCC, which is only disabled on -O2/3, so Release builds finish, Debug don't.

Fixes PR30797.

llvm-svn: 285204

7 years agoReapply "AMDGPU: Don't use offen if it is 0"
Matt Arsenault [Wed, 26 Oct 2016 15:08:16 +0000 (15:08 +0000)]
Reapply "AMDGPU: Don't use offen if it is 0"

This reverts r283003

llvm-svn: 285203

7 years agoAMDGPU: Fix counting si_mask_branch as 4 bytes
Matt Arsenault [Wed, 26 Oct 2016 14:53:54 +0000 (14:53 +0000)]
AMDGPU: Fix counting si_mask_branch as 4 bytes

llvm-svn: 285202

7 years agoFix nondeterministic output in local stack slot alloc pass
Matt Arsenault [Wed, 26 Oct 2016 14:53:50 +0000 (14:53 +0000)]
Fix nondeterministic output in local stack slot alloc pass

This finds all of the references to a frame index in a function, and
sorts by the offset. If multiple instructions use the same offset,
nothing was breaking the tie for sorting.

This avoids the test failures the reverted r282999 introduced.

llvm-svn: 285201

7 years ago[InstCombine] clean up commonCastTransforms; NFC
Sanjay Patel [Wed, 26 Oct 2016 14:52:35 +0000 (14:52 +0000)]
[InstCombine] clean up commonCastTransforms; NFC

1. Use 'auto' with dyn_cast.
2. Variables start with a capital letter.
3. Use proper punctuation in comments.

llvm-svn: 285200

7 years agoLegalizeDAG: Support promoting [US]DIV and [US]REM operations
Tom Stellard [Wed, 26 Oct 2016 14:52:25 +0000 (14:52 +0000)]
LegalizeDAG: Support promoting [US]DIV and [US]REM operations

Summary:
AMDGPU will need this one i16 is added as a legal type.  This is tested by:

test/CodeGen/AMDGPU/sdiv.ll
test/CodeGen/AMDGPU/sdivrem24.ll
test/CodeGen/AMDGPU/udiv.ll
test/CodeGen/AMDGPU/udivrem24.ll

Reviewers: bogner, efriedma

Subscribers: efriedma, wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D25699

llvm-svn: 285199

7 years agoAMDGPU/SI: Don't emit multi-dword flat memory ops when they might access scratch
Tom Stellard [Wed, 26 Oct 2016 14:38:47 +0000 (14:38 +0000)]
AMDGPU/SI: Don't emit multi-dword flat memory ops when they might access scratch

Summary:
A single flat memory operations that might access the scratch buffer
can only access MaxPrivateElementSize bytes.

Reviewers: arsenm

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

Differential Revision: https://reviews.llvm.org/D25788

llvm-svn: 285198

7 years agoAMDGPU/SI: Remove unnecessary run lines from test
Tom Stellard [Wed, 26 Oct 2016 14:21:09 +0000 (14:21 +0000)]
AMDGPU/SI: Remove unnecessary run lines from test

Summary:
This test had run lines disabling/enabling the promote alloca pass, but
enabling/disabling promote alloca had no impact on the output.

Reviewers: arsenm

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

Differential Revision: https://reviews.llvm.org/D25787

llvm-svn: 285197

7 years ago[X86] AVX512 fallback for floating-point scalar selects
Zvi Rackover [Wed, 26 Oct 2016 14:12:46 +0000 (14:12 +0000)]
[X86] AVX512 fallback for floating-point scalar selects

Summary:
In the case where of 'select i1 , f32, f32' or select i1, f64, f64 prefer lowering to masked-moves over branches.

Fixes pr30561

Reviewers: igorb, aymanmus, delena

Differential Revision: https://reviews.llvm.org/D25310

llvm-svn: 285196

7 years ago[InstCombine] consolidate zext tests and auto-generate checks; NFC
Sanjay Patel [Wed, 26 Oct 2016 14:08:49 +0000 (14:08 +0000)]
[InstCombine] consolidate zext tests and auto-generate checks; NFC

llvm-svn: 285195

7 years ago[InstCombine] auto-generate better checks; NFC
Sanjay Patel [Wed, 26 Oct 2016 13:58:22 +0000 (13:58 +0000)]
[InstCombine] auto-generate better checks; NFC

llvm-svn: 285194

7 years ago[PP] Replace some index based for loops with range based ones
Erik Verbruggen [Wed, 26 Oct 2016 13:06:13 +0000 (13:06 +0000)]
[PP] Replace some index based for loops with range based ones

While in the area, also change some unsigned variables to size_t, and
introduce an LLVM_FALLTHROUGH instead of a comment stating that.

Differential Revision: http://reviews.llvm.org/D25982

llvm-svn: 285193

7 years agoFix use-after-scope in ASTContext.
Benjamin Kramer [Wed, 26 Oct 2016 12:51:45 +0000 (12:51 +0000)]
Fix use-after-scope in ASTContext.

Extend lifetime of ExceptionTypeStorage, as it is referenced by
CanonicalEPI and used outside the block (ExceptionSpec.Exceptions is an
ArrayRef)

Patch by Sam McCall!

Differential Revision: https://reviews.llvm.org/D25983

llvm-svn: 285192

7 years ago[AliasSetTracker] Make AST smarter about intrinsics that don't actually affect memory.
Chad Rosier [Wed, 26 Oct 2016 12:42:11 +0000 (12:42 +0000)]
[AliasSetTracker] Make AST smarter about intrinsics that don't actually affect memory.

Differential Revision: https://reviews.llvm.org/D25969

llvm-svn: 285191