platform/upstream/llvm.git
7 years agoCleanup _LIBCPP_HAS_NO_<c++11-feature> in std::unordered_map and std::unordered_multimap
Eric Fiselier [Tue, 18 Apr 2017 22:50:56 +0000 (22:50 +0000)]
Cleanup _LIBCPP_HAS_NO_<c++11-feature> in std::unordered_map and std::unordered_multimap

This completes the cleanup of the containers, at least within the tests.

llvm-svn: 300620

7 years agoCleanup _LIBCPP_HAS_NO_<c++11-feature> macros in std::unordered_set and std::unordere...
Eric Fiselier [Tue, 18 Apr 2017 22:37:32 +0000 (22:37 +0000)]
Cleanup _LIBCPP_HAS_NO_<c++11-feature> macros in std::unordered_set and std::unordered_multiset

llvm-svn: 300619

7 years agoDefine HAVE_SIGACTION to 1 in Xcode build
Chris Bieneman [Tue, 18 Apr 2017 22:37:00 +0000 (22:37 +0000)]
Define HAVE_SIGACTION to 1 in Xcode build

This is needed to make the Xcode project build since it doesn't have auto-generated Config header.

llvm-svn: 300618

7 years ago[x86] add tests for potential andn optimization; NFC
Sanjay Patel [Tue, 18 Apr 2017 22:36:59 +0000 (22:36 +0000)]
[x86] add tests for potential andn optimization; NFC

llvm-svn: 300617

7 years ago[coff] fix test for msvclto
Bob Haarman [Tue, 18 Apr 2017 22:29:36 +0000 (22:29 +0000)]
[coff] fix test for msvclto

llvm-svn: 300616

7 years agoInclude time.h, and fix a Darwin warning
Chris Bieneman [Tue, 18 Apr 2017 22:11:13 +0000 (22:11 +0000)]
Include time.h, and fix a Darwin warning

This is a little more cleanup from r300579.

llvm-svn: 300615

7 years agoFix crash in AttributeList::addAttributes, add test
Reid Kleckner [Tue, 18 Apr 2017 22:10:18 +0000 (22:10 +0000)]
Fix crash in AttributeList::addAttributes, add test

llvm-svn: 300614

7 years agoAdd a getPointerOperandType() helper to LoadInst and StoreInst; NFC
Sanjoy Das [Tue, 18 Apr 2017 22:00:54 +0000 (22:00 +0000)]
Add a getPointerOperandType() helper to LoadInst and StoreInst; NFC

I will use this in a later change.

llvm-svn: 300613

7 years ago[coff] use newlines instead of spaces as separators in msvclto response file
Bob Haarman [Tue, 18 Apr 2017 22:00:29 +0000 (22:00 +0000)]
[coff] use newlines instead of spaces as separators in msvclto response file

Summary:
Fixes PR32689: /msvclto creates response files with lines
that are too long for msvc's linker (LNK1170).

Reviewers: hans, rnk, ruiu

Subscribers: llvm-commits, mehdi_amini

Differential Revision: https://reviews.llvm.org/D32185

llvm-svn: 300612

7 years ago[modules-ts] Fold together -x c++ and -x c++-module at -cc1 level.
Richard Smith [Tue, 18 Apr 2017 21:55:37 +0000 (21:55 +0000)]
[modules-ts] Fold together -x c++ and -x c++-module at -cc1 level.

The driver needs to know whether it's building a module interface or
implementation unit because it affects which outputs it produces and how it
builds the command pipeline. But the frontend doesn't need to know and should
not care: all it needs to know is what action it is being asked to perform on
the input.

(This is in preparation for permitting -emit-obj to be used on a module
interface unit to produce object code without going via a "full" PCM file.)

llvm-svn: 300611

7 years agoFix Windows bot failure
Chris Bieneman [Tue, 18 Apr 2017 21:47:50 +0000 (21:47 +0000)]
Fix Windows bot failure

timespec is not available on Windows, and we should use size_t instead of nfds_t.

llvm-svn: 300610

7 years agoDo not warn about whitespace between ??/ trigraph and newline in line comments if...
Richard Smith [Tue, 18 Apr 2017 21:45:04 +0000 (21:45 +0000)]
Do not warn about whitespace between ??/ trigraph and newline in line comments if trigraphs are disabled in the current language.

llvm-svn: 300609

7 years ago[MemoryBuiltins] Add isMallocOrCallocLikeFn so BasicAA can check for both at the...
Craig Topper [Tue, 18 Apr 2017 21:43:46 +0000 (21:43 +0000)]
[MemoryBuiltins] Add isMallocOrCallocLikeFn so BasicAA can check for both at the same time

BasicAA wants to know if a function is either a malloc or calloc like function. Currently we have to check both separately. This means both calls check if its an intrinsic, query TLI, check the nobuiltin attribute, scan the AllocationFnData, etc.

This patch adds a isMallocOrCallocLikeFn so we can go through all of the checks once per call.

This also changes the one other location I saw that called both together.

Differential Revision: https://reviews.llvm.org/D32188

llvm-svn: 300608

7 years ago[LoopReroll] Prefer hasNUses/hasNUses or more as they're cheaper. NFCI.
Davide Italiano [Tue, 18 Apr 2017 21:42:21 +0000 (21:42 +0000)]
[LoopReroll] Prefer hasNUses/hasNUses or more as they're cheaper. NFCI.

llvm-svn: 300607

7 years agoFixing error on Android build (-Werror)
Chris Bieneman [Tue, 18 Apr 2017 21:35:26 +0000 (21:35 +0000)]
Fixing error on Android build (-Werror)

This is fallout from r300579.

llvm-svn: 300606

7 years agoRemoving unused include
Chris Bieneman [Tue, 18 Apr 2017 21:23:55 +0000 (21:23 +0000)]
Removing unused include

This is causing the Windows bot failures.

llvm-svn: 300605

7 years agoCleanup _LIBCPP_HAS_NO_<c++11-feature> for std::queue and std::priority_queue.
Eric Fiselier [Tue, 18 Apr 2017 21:23:18 +0000 (21:23 +0000)]
Cleanup _LIBCPP_HAS_NO_<c++11-feature> for std::queue and std::priority_queue.

llvm-svn: 300604

7 years agoDAG: Make mayBeEmittedAsTailCall parameter const
Matt Arsenault [Tue, 18 Apr 2017 21:16:46 +0000 (21:16 +0000)]
DAG: Make mayBeEmittedAsTailCall parameter const

llvm-svn: 300603

7 years agoCleanup _LIBCPP_HAS_NO_<c++11-feature> macro uses in std::stack.
Eric Fiselier [Tue, 18 Apr 2017 21:16:26 +0000 (21:16 +0000)]
Cleanup _LIBCPP_HAS_NO_<c++11-feature> macro uses in std::stack.

llvm-svn: 300602

7 years ago[sanitizer] Define lsan-x86 in tests for both i386 and i686.
Evgeniy Stepanov [Tue, 18 Apr 2017 21:10:50 +0000 (21:10 +0000)]
[sanitizer] Define lsan-x86 in tests for both i386 and i686.

llvm-svn: 300601

7 years agoCleanup _LIBCPP_HAS_NO_<c++11-feature> macros in std::map and std::multimap
Eric Fiselier [Tue, 18 Apr 2017 21:08:06 +0000 (21:08 +0000)]
Cleanup _LIBCPP_HAS_NO_<c++11-feature> macros in std::map and std::multimap

llvm-svn: 300600

7 years agoImplement suspended thread register count for darwin
Francis Ricci [Tue, 18 Apr 2017 21:05:11 +0000 (21:05 +0000)]
Implement suspended thread register count for darwin

Reviewers: kubamracek, alekseyshl

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32165

llvm-svn: 300599

7 years agoRemove mips64 defines from darwin-specific file
Francis Ricci [Tue, 18 Apr 2017 21:05:09 +0000 (21:05 +0000)]
Remove mips64 defines from darwin-specific file

Reviewers: kubamracek, alekseyshl

Subscribers: llvm-commits, arichardson

Differential Revision: https://reviews.llvm.org/D32183

llvm-svn: 300598

7 years agoFix typo
Matt Arsenault [Tue, 18 Apr 2017 20:59:46 +0000 (20:59 +0000)]
Fix typo

llvm-svn: 300597

7 years agoAMDGPU: Make MFI fields private
Matt Arsenault [Tue, 18 Apr 2017 20:59:40 +0000 (20:59 +0000)]
AMDGPU: Make MFI fields private

llvm-svn: 300596

7 years agoCleanup _LIBCPP_HAS_NO_<c++11-feature> for std::set and std::multiset
Eric Fiselier [Tue, 18 Apr 2017 20:58:03 +0000 (20:58 +0000)]
Cleanup _LIBCPP_HAS_NO_<c++11-feature> for std::set and std::multiset

llvm-svn: 300595

7 years agoPR30508: Downgrade error to warning if the umbrella folder doesn't exist.
Vassil Vassilev [Tue, 18 Apr 2017 20:57:29 +0000 (20:57 +0000)]
PR30508: Downgrade error to warning if the umbrella folder doesn't exist.

Patch by Yuka Takahashi (D32119)!

llvm-svn: 300594

7 years agoDon't use abort_on_error for lsan darwin test suite
Francis Ricci [Tue, 18 Apr 2017 20:56:59 +0000 (20:56 +0000)]
Don't use abort_on_error for lsan darwin test suite

Summary:
This option is disabled by our other test suites, and will cause
failures when unit tests abort instead of failing with an error code.
Will also prevent the test suite from being too slow.

Reviewers: kubamracek, alekseyshl

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32129

llvm-svn: 300593

7 years agoAllow for setting of global platform-specific lsan options in the test suite
Francis Ricci [Tue, 18 Apr 2017 20:56:56 +0000 (20:56 +0000)]
Allow for setting of global platform-specific lsan options in the test suite

Reviewers: kubamracek, alekseyshl

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32131

llvm-svn: 300592

7 years agoThe SubjectMatchRule enum should not be used as a DenseMap key to avoid
Alex Lorenz [Tue, 18 Apr 2017 20:54:23 +0000 (20:54 +0000)]
The SubjectMatchRule enum should not be used as a DenseMap key to avoid
UBSAN 'invalid value' failures

The commit r300556 introduced a UBSAN issue that was caught by
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap. The DenseMap
failed to create an empty/tombstone value as the empty/tombstone values for the
SubjectMatchRule enum were not valid enum constants.

llvm-svn: 300591

7 years ago[CMake] Adding configure-time check for sigaction
Chris Bieneman [Tue, 18 Apr 2017 20:49:09 +0000 (20:49 +0000)]
[CMake] Adding configure-time check for sigaction

Hopefully this will fix the netbsd bot that I broke...

llvm-svn: 300590

7 years agoWriting multi-platform code is hard...
Chris Bieneman [Tue, 18 Apr 2017 20:49:05 +0000 (20:49 +0000)]
Writing multi-platform code is hard...

Fixing another error from r300579.

llvm-svn: 300589

7 years ago[clang-tidy] Address a few late comments.
Haojian Wu [Tue, 18 Apr 2017 20:47:34 +0000 (20:47 +0000)]
[clang-tidy] Address a few late comments.

llvm-svn: 300588

7 years agoFix broken windows build.
Chris Bieneman [Tue, 18 Apr 2017 20:37:05 +0000 (20:37 +0000)]
Fix broken windows build.

This is not ideal, but it should get the bot going again. I'll need to revisit this if we want to get signal handling working on Windows.

llvm-svn: 300587

7 years ago[MemoryBuiltins] Use ImmutableCallSite instead of CallSite to remove a const_cast...
Craig Topper [Tue, 18 Apr 2017 20:17:23 +0000 (20:17 +0000)]
[MemoryBuiltins] Use ImmutableCallSite instead of CallSite to remove a const_cast and const correct. NFCI

llvm-svn: 300585

7 years agoNewGVN: Fix memory congruence verification. The return true should be a return false...
Daniel Berlin [Tue, 18 Apr 2017 20:15:47 +0000 (20:15 +0000)]
NewGVN: Fix memory congruence verification. The return true should be a return false. Merge the appropriate if statements so it doesn't happen again.

llvm-svn: 300584

7 years ago[X86] Keep EXTRACT_VECTOR_ELT result type as f128 for Android x86_64.
Chih-Hung Hsieh [Tue, 18 Apr 2017 20:15:18 +0000 (20:15 +0000)]
[X86] Keep EXTRACT_VECTOR_ELT result type as f128 for Android x86_64.

Android x86_64 target uses f128 type and stores f128 values in %xmm* registers.
SoftenFloatRes_EXTRACT_VECTOR_ELT should not convert result value
from f128 to i128.

Differential Revision: http://reviews.llvm.org/D32102

llvm-svn: 300583

7 years agoFixing bot failure caused by r300579
Chris Bieneman [Tue, 18 Apr 2017 20:08:29 +0000 (20:08 +0000)]
Fixing bot failure caused by r300579

llvm-svn: 300582

7 years ago[test] Silence another unused-typedef warning
Casey Carter [Tue, 18 Apr 2017 20:04:39 +0000 (20:04 +0000)]
[test] Silence another unused-typedef warning

llvm-svn: 300581

7 years agoUpdate DebugServer to support IPv6 over TCP
Chris Bieneman [Tue, 18 Apr 2017 20:01:59 +0000 (20:01 +0000)]
Update DebugServer to support IPv6 over TCP

Summary: This patch adds IPv6 support to debugserver. It follows a similar pattern to the changes proposed for LLDB/Host except that the listen implementation is only with kqueue(2) because debugserver is only supported on Darwin.

Reviewers: jingham, jasonmolenda, clayborg

Reviewed By: clayborg

Subscribers: mgorny, lldb-commits

Differential Revision: https://reviews.llvm.org/D31824

llvm-svn: 300580

7 years agoUpdate LLDB Host to support IPv6 over TCP
Chris Bieneman [Tue, 18 Apr 2017 20:01:52 +0000 (20:01 +0000)]
Update LLDB Host to support IPv6 over TCP

Summary:
This patch adds IPv6 support to LLDB/Host's TCP socket implementation. Supporting IPv6 involved a few significant changes to the implementation of the socket layers, and I have performed some significant code cleanup along the way.

This patch changes the Socket constructors for all types of sockets to not create sockets until first use. This is required for IPv6 support because the socket type will vary based on the address you are connecting to. This also has the benefit of removing code that could have errors from the Socket subclass constructors (which seems like a win to me).

The patch also slightly changes the API and behaviors of the Listen/Accept pattern. Previously both Listen and Accept calls took an address specified as a string. Now only listen does. This change was made because the Listen call can result in opening more than one socket. In order to support listening for both IPv4 and IPv6 connections we need to open one AF_INET socket and one AF_INET6 socket. During the listen call we construct a map of file descriptors to addrin structures which represent the allowable incoming connection address. This map removes the need for taking an address into the Accept call.

This does have a change in functionality. Previously you could Listen for connections based on one address, and Accept connections from a different address. This is no longer supported. I could not find anywhere in LLDB where we actually used the APIs in that way. The new API does still support AnyAddr for allowing incoming connections from any address.

The Listen implementation is implemented using kqueue on FreeBSD and Darwin, WSAPoll on Windows and poll(2) everywhere else.

Reviewers: zturner, clayborg

Subscribers: jasonmolenda, labath, lldb-commits, emaste

Differential Revision: https://reviews.llvm.org/D31823

llvm-svn: 300579

7 years ago[DOXYGEN] Minor improvements in doxygen comments.
Ekaterina Romanova [Tue, 18 Apr 2017 19:44:07 +0000 (19:44 +0000)]
[DOXYGEN] Minor improvements in doxygen comments.

- To be consistent with the rest of the intrinsics headers, I removed the tags <i> .. </i> for marking instruction names in italics in in smmintrin.h.

- Formatting changes to fit into 80 characters.

I got an OK from Eric Christopher to commit doxygen comments without prior code
review upstream.

llvm-svn: 300578

7 years ago[APInt] Inline the single word case of lshrInPlace similar to what we do for <<=.
Craig Topper [Tue, 18 Apr 2017 19:13:27 +0000 (19:13 +0000)]
[APInt] Inline the single word case of lshrInPlace similar to what we do for <<=.

llvm-svn: 300577

7 years ago[X86][SSE] Add scheduling latency/throughput tests for (most) SSE1 instructions
Simon Pilgrim [Tue, 18 Apr 2017 19:04:40 +0000 (19:04 +0000)]
[X86][SSE] Add scheduling latency/throughput tests for (most) SSE1 instructions

llvm-svn: 300576

7 years ago[test] Silence unused parameter/typedef warnings
Casey Carter [Tue, 18 Apr 2017 18:44:33 +0000 (18:44 +0000)]
[test] Silence unused parameter/typedef warnings

llvm-svn: 300575

7 years ago[SLP vectorizer] Allow phi node reordering in tryToVectorizeList.
Easwaran Raman [Tue, 18 Apr 2017 18:16:57 +0000 (18:16 +0000)]
[SLP vectorizer] Allow phi node reordering in tryToVectorizeList.

In tryToVectorizeList, under a very limited circumstance (when entered
from tryToVectorizePair), the values may be reordered (swapped) and the
SLP tree is built with the new order. This extends that to the case when
starting from phis in vectorizeChainsInBlock when there are exactly two
phis. The textual order of phi nodes shouldn't really matter. Without
this change, the loop body in the accompnaying test case is fully vectorized
when we swap the orde of the phis but not with this order. While this
doesn't solve the phi-ordering problem in a general way (for more than 2
phis), this is simple fix that piggybacks on an existing mechanism and
is useful in cases like multiplying two complex numbers.

Differential revision: https://reviews.llvm.org/D32065

llvm-svn: 300574

7 years agoRemove unused varible
Reid Kleckner [Tue, 18 Apr 2017 17:44:41 +0000 (17:44 +0000)]
Remove unused varible

The Result variable is unused both in Sema::CheckARMBuiltinFunctionCall
and Sema::CheckAArch64BuiltinFunctionCall, remove it.

Patch by Wei-Ren Chen!

Reviewers: craig.topper, rnk

Reviewed By: rnk

Subscribers: aemerson, cfe-commits, rengolin

Differential Revision: https://reviews.llvm.org/D32014

llvm-svn: 300572

7 years ago[AArch64][clang] Pass cpu/arch information to assembler for AArch64.
Manoj Gupta [Tue, 18 Apr 2017 17:36:10 +0000 (17:36 +0000)]
[AArch64][clang] Pass cpu/arch information to assembler for AArch64.

Summary:
Pass Cpu/Arch options to assembler for AArch64 with no-integrated-as.
This fixes PR20019.

Reviewers: richard.barton.arm, kristof.beyls, rengolin

Reviewed By: rengolin

Subscribers: srhines, pirama, aemerson, rengolin, cfe-commits

Differential Revision: https://reviews.llvm.org/D32132

llvm-svn: 300571

7 years agoTesting commit access.
Manoj Gupta [Tue, 18 Apr 2017 17:34:46 +0000 (17:34 +0000)]
Testing commit access.

Summary: Test commit access.

Reviewers: gbiv, george.burgess.iv

Reviewed By: george.burgess.iv

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D32161

llvm-svn: 300570

7 years ago[clang-tidy] Fix google-explicit-constructor issue with out-of-line conversions
Alexander Kornienko [Tue, 18 Apr 2017 17:26:00 +0000 (17:26 +0000)]
[clang-tidy] Fix google-explicit-constructor issue with out-of-line conversions

llvm-svn: 300569

7 years agoMark LWG#2788 as complete - we already do this
Marshall Clow [Tue, 18 Apr 2017 17:22:49 +0000 (17:22 +0000)]
Mark LWG#2788 as complete - we already do this

llvm-svn: 300568

7 years ago[X86] Use for-range loop. NFCI.
Simon Pilgrim [Tue, 18 Apr 2017 17:18:54 +0000 (17:18 +0000)]
[X86] Use for-range loop. NFCI.

llvm-svn: 300567

7 years ago[APInt] Use lshrInPlace to replace lshr where possible
Craig Topper [Tue, 18 Apr 2017 17:14:21 +0000 (17:14 +0000)]
[APInt] Use lshrInPlace to replace lshr where possible

This patch uses lshrInPlace to replace code where the object that lshr is called on is being overwritten with the result.

This adds an lshrInPlace(const APInt &) version as well.

Differential Revision: https://reviews.llvm.org/D32155

llvm-svn: 300566

7 years agoNewGVN: Don't waste time value numbering unreachable blocks
Daniel Berlin [Tue, 18 Apr 2017 17:06:11 +0000 (17:06 +0000)]
NewGVN: Don't waste time value numbering unreachable blocks

llvm-svn: 300565

7 years agoAdd back code to implement "frame var -a,-l,-g" filters.
Jim Ingham [Tue, 18 Apr 2017 16:52:16 +0000 (16:52 +0000)]
Add back code to implement "frame var -a,-l,-g" filters.

r285226 dropped the code that did these checks.  I am pretty
sure that was inadvertent, so I added that back in and added
a test for it.

<rdar://problem/31661252>

llvm-svn: 300564

7 years agoMake the test pass on x86_64-win32 target.
Haojian Wu [Tue, 18 Apr 2017 16:25:03 +0000 (16:25 +0000)]
Make the test pass on x86_64-win32 target.

llvm-svn: 300563

7 years agomingw-w64: enable support for __declspec(selectany)
Martell Malone [Tue, 18 Apr 2017 15:56:24 +0000 (15:56 +0000)]
mingw-w64: enable support for __declspec(selectany)

Add selectany as a GCC spelling for mingw-w64

Reviewers: rnk

Differential revision: https://reviews.llvm.org/D32083

llvm-svn: 300562

7 years ago[DAG] Improve store merge candidate pruning.
Nirav Dave [Tue, 18 Apr 2017 15:36:34 +0000 (15:36 +0000)]
[DAG] Improve store merge candidate pruning.

Remove non-consecutive stores from store merge candidate search as
they cannot be merged and will prevent us from finding subsequent
mergeable store cases.

Reviewers: jyknight, bogner, javed.absar, spatel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32086

llvm-svn: 300561

7 years ago[ASTPrinter] Print template parameter lists for out-of-line functions
Alex Lorenz [Tue, 18 Apr 2017 15:12:34 +0000 (15:12 +0000)]
[ASTPrinter] Print template parameter lists for out-of-line functions

llvm-svn: 300560

7 years agoAdd base-index-based store merge test
Nirav Dave [Tue, 18 Apr 2017 15:12:13 +0000 (15:12 +0000)]
Add base-index-based store merge test

llvm-svn: 300559

7 years agoLoopRerollPass: Prefer Value::hasOneUse() over Value::getNumUses(). NFC.
Zvi Rackover [Tue, 18 Apr 2017 14:55:43 +0000 (14:55 +0000)]
LoopRerollPass: Prefer Value::hasOneUse() over Value::getNumUses(). NFC.

getNumUses() can be more expensive as it iterates over all list's elements.

llvm-svn: 300558

7 years ago[LV] Cache block mask values
Gil Rapaport [Tue, 18 Apr 2017 14:43:43 +0000 (14:43 +0000)]
[LV] Cache block mask values

This patch is part of D28975's breakdown.

Add caching for block masks similar to the cache already used for edge masks,
replacing generation per user with reusing the first generated value which
dominates all uses.

Differential Revision: https://reviews.llvm.org/D32054

llvm-svn: 300557

7 years agoAdd #pragma clang attribute
Alex Lorenz [Tue, 18 Apr 2017 14:33:39 +0000 (14:33 +0000)]
Add #pragma clang attribute

This is a recommit of r300539 that was reverted in r300543 due to test failures.
The original commit message is displayed below:

The new '#pragma clang attribute' directive can be used to apply attributes to
multiple declarations. An attribute must satisfy the following conditions to
be supported by the pragma:
- It must have a subject list that's defined in the TableGen file.
- It must be documented.
- It must not be late parsed.
- It must have a GNU/C++11 spelling.

Differential Revision: https://reviews.llvm.org/D30009

llvm-svn: 300556

7 years agoDriver: Better detection of mingw-gcc
Martell Malone [Tue, 18 Apr 2017 14:27:36 +0000 (14:27 +0000)]
Driver: Better detection of mingw-gcc

Stop blindly searching for "gcc.exe" on windows.
Stop assuming "/usr" on unix, fixes cross compiling.

Reviewers: mati865, yaron.keren

Subscribers: ismail, rnk

Differential revision: https://reviews.llvm.org/D15005

llvm-svn: 300555

7 years ago[ConstantRange] fix doxygen comment formatting; NFC
Sanjay Patel [Tue, 18 Apr 2017 14:27:24 +0000 (14:27 +0000)]
[ConstantRange] fix doxygen comment formatting; NFC

llvm-svn: 300554

7 years agoMake globalaa-retained.ll test catching more cases.
Nikolai Bozhenov [Tue, 18 Apr 2017 13:29:26 +0000 (13:29 +0000)]
Make globalaa-retained.ll test catching more cases.

Summary:
* Add checks for store. That is needed because GlobalsAA is called
  twice in the current pipeline with different sets of Function passes
  following it. However, the loads are eliminated using instcombine
  which happens everywhere. On the other hand, DeadStoreElimination is
  performed only once so by checking for store we'll be able to catch
  more cases when GlobalsAA is invalidated unintentionally.
* Add empty function above/below the test so that we don't depend on
  the relative order of instcombine/dead-store-elimination and the
  pass that invalidates the analysis (inside the same
  FunctionPassManager).

Reviewers: kristof.beyls

Reviewed By: kristof.beyls

Subscribers: llvm-commits, n.bozhenov

Differential Revision: https://reviews.llvm.org/D32015
Patch by Andrei Elovikov <andrei.elovikov@intel.com>

llvm-svn: 300553

7 years ago[GVNHoist] Mark GlobalsAA as preserved by GVNHoist.
Nikolai Bozhenov [Tue, 18 Apr 2017 13:25:49 +0000 (13:25 +0000)]
[GVNHoist] Mark GlobalsAA as preserved by GVNHoist.

Reviewers: sebpop, hiraditya

Reviewed By: sebpop

Subscribers: n.bozhenov, llvm-commits

Differential Revision: https://reviews.llvm.org/D32158
Patch by Andrei Elovikov <andrei.elovikov@intel.com>

llvm-svn: 300552

7 years agoAdd store Merge test.
Nirav Dave [Tue, 18 Apr 2017 13:25:19 +0000 (13:25 +0000)]
Add store Merge test.

llvm-svn: 300551

7 years ago[ARM] Add hardware build attributes in assembler
Oliver Stannard [Tue, 18 Apr 2017 13:21:05 +0000 (13:21 +0000)]
[ARM] Add hardware build attributes in assembler

This passes an option to the ARM assembly parser to emit build
attributes for the hardware selected by command line options, when
assembling an assembly file.

This is not enabled for C/C++, as this would result in duplicate build
attribute directives being emitted in each inline assembly block, when
emitting assembly.

This also adds an option to allow disabling this behaviour for assembly
files, for users who were relying on the old behaviour.

Differential revision: https://reviews.llvm.org/D31813

llvm-svn: 300550

7 years ago[ARM,AArch64] Define __ELF__ for arm-none-eabihf and AArch64
Oliver Stannard [Tue, 18 Apr 2017 13:12:36 +0000 (13:12 +0000)]
[ARM,AArch64] Define __ELF__ for arm-none-eabihf and AArch64

This macro is defined for arm-none-eabi as of r266625, but it should also be
defined for eabihf and aarch64.

llvm-svn: 300549

7 years agoIntroduce FPR and Debug Registers/NetBSD/amd64 support
Kamil Rytarowski [Tue, 18 Apr 2017 12:53:35 +0000 (12:53 +0000)]
Introduce FPR and Debug Registers/NetBSD/amd64 support

Summary:
This code offers Debug Registers (80386) model in LLDB/amd64.

This is initial support and has one issue that will be addressed later,
Debug Register trap (TRAP_DBREG) is registered as (TRAP_TRACE)
for unknown reason.  On the other hand this works good enough to
move on and leave this bug to be squashed later.

Improve the NativeProcessNetBSD::ReinitializeThreads() function,
stop setting inside it SetStoppedByExec(). This fixes incorrect
stop reason on attaching (SetStoppedBySignal(SIGSTOP)).

This commits also has no functional style improvements from
clang-format.

This code also ships with FXSAVE support on NetBSD.

Demo:

```
$ lldb ./watch
(lldb) target create "./watch"
Current executable set to './watch' (x86_64).
(lldb) b main
Breakpoint 1: where = watch`main + 15 at watch.c:8, address = 0x000000000040087f
(lldb) r
Process 1573 launched: './watch' (x86_64)
Process 1573 stopped
* thread #1, stop reason = breakpoint 1.1
    frame #0: 0x000000000040087f watch`main(argc=1, argv=0x00007f7fffa12b88) at watch.c:8
   5    {
   6            int i, j, k;
   7
-> 8            for (i = 0; i < 3; i++)
   9                    for (j = 0; j < 3; j++)
   10                           for (k = 0; k < 3; k++)
   11                                   printf("Hello world! i=%d j=%d k=%d\n", i, j, k);
(lldb) watch set var i
Watchpoint created: Watchpoint 1: addr = 0x7f7fffa12b4c size = 4 state = enabled type = w
    declare @ '/public/lldb_devel/watch.c:6'
    watchpoint spec = 'i'
    new value: 0
(lldb) c
Process 1573 resuming
Hello world! i=0 j=0 k=0
Hello world! i=0 j=0 k=1
Hello world! i=0 j=0 k=2
Hello world! i=0 j=1 k=0
Hello world! i=0 j=1 k=1
Hello world! i=0 j=1 k=2
Hello world! i=0 j=2 k=0
Hello world! i=0 j=2 k=1
Hello world! i=0 j=2 k=2
Process 1573 stopped
* thread #1, stop reason = trace
    frame #0: 0x00000000004008cc watch`main(argc=1, argv=0x00007f7fffa12b88) at watch.c:8
   5    {
   6            int i, j, k;
   7
-> 8            for (i = 0; i < 3; i++)
   9                    for (j = 0; j < 3; j++)
   10                           for (k = 0; k < 3; k++)
   11                                   printf("Hello world! i=%d j=%d k=%d\n", i, j, k)
```

FPR (in another program using libm)

```
(lldb) register read --all
General Purpose Registers:
       rax = 0x000000000000001c
       rbx = 0x00007f7fff1d4fe0
       rcx = 0x000000000000000c
       rdx = 0x0000000000000002
       rdi = 0x0000746711d5b018  __sF + 152
       rsi = 0x0000000000000001
       rbp = 0x00007f7fff1d3d80
       rsp = 0x00007f7fff1d3d60
        r8 = 0x00007f7fff1d3470
        r9 = 0x0000000000000000
       r10 = 0x0000000000000001
       r11 = 0x0000000000000202
       r12 = 0x00007f7fff1d3da0
       r13 = 0x00007d8ad2d88500
       r14 = 0x0000000000000002
       r15 = 0x00007f7fffa627e0
       rip = 0x00000000004009e9  fpr`main + 217 at fpr.c:15
    rflags = 0x0000000000000202
        cs = 0x0000000000000047
        fs = 0x0000000000000000
        gs = 0x0000000000000000
        ss = 0x000000000000003f
        ds = 0x000000000000003f
        es = 0x000000000000003f
       eax = 0x0000001c
       ebx = 0xff1d4fe0
       ecx = 0x0000000c
       edx = 0x00000002
       edi = 0x11d5b018
       esi = 0x00000001
       ebp = 0xff1d3d80
       esp = 0xff1d3d60
       r8d = 0xff1d3470
       r9d = 0x00000000
      r10d = 0x00000001
      r11d = 0x00000202
      r12d = 0xff1d3da0
      r13d = 0xd2d88500
      r14d = 0x00000002
      r15d = 0xffa627e0
        ax = 0x001c
        bx = 0x4fe0
        cx = 0x000c
        dx = 0x0002
        di = 0xb018
        si = 0x0001
        bp = 0x3d80
        sp = 0x3d60
       r8w = 0x3470
       r9w = 0x0000
      r10w = 0x0001
      r11w = 0x0202
      r12w = 0x3da0
      r13w = 0x8500
      r14w = 0x0002
      r15w = 0x27e0
        ah = 0x00
        bh = 0x4f
        ch = 0x00
        dh = 0x00
        al = 0x1c
        bl = 0xe0
        cl = 0x0c
        dl = 0x02
       dil = 0x18
       sil = 0x01
       bpl = 0x80
       spl = 0x60
       r8l = 0x70
       r9l = 0x00
      r10l = 0x01
      r11l = 0x02
      r12l = 0xa0
      r13l = 0x00
      r14l = 0x02
      r15l = 0xe0

unknown:
     fctrl = 0x037f
     fstat = 0x0220
      ftag = 0x00
       fop = 0x0000
     fiseg = 0x11e1a52c
     fioff = 0x11e1a52c
     foseg = 0xff1d3d54
     fooff = 0xff1d3d54
     mxcsr = 0x00001fa0
  mxcsrmask = 0x0000ffff
       st0 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       st1 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       st2 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       st3 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       st4 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       st5 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       st6 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       st7 = {0xa5 0xdb 0x2d 0xbd 0x93 0xae 0xb9 0xfe 0xfe 0x3f}
       mm0 = 0x3fe9d13800000000
       mm1 = 0x3e0485fcce89c000
       mm2 = 0x3fefd735e0000000
       mm3 = 0x0000000000000000
       mm4 = 0x3fe0000000000000
       mm5 = 0x3fe00000005217f3
       mm6 = 0x0000000000000000
       mm7 = 0x3fefd735e0000000
      xmm0 = {0x00 0x00 0x00 0x00 0x38 0xd1 0xe9 0x3f 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm1 = {0x00 0xc0 0x89 0xce 0xfc 0x85 0x04 0x3e 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm2 = {0x00 0x00 0x00 0xe0 0x35 0xd7 0xef 0x3f 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm3 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm4 = {0x00 0x00 0x00 0x00 0x00 0x00 0xe0 0x3f 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm5 = {0xf3 0x17 0x52 0x00 0x00 0x00 0xe0 0x3f 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm6 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm7 = {0x00 0x00 0x00 0xe0 0x35 0xd7 0xef 0x3f 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm8 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
      xmm9 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
     xmm10 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
     xmm11 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
     xmm12 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
     xmm13 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
     xmm14 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
     xmm15 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
       dr0 = 0x0000000000000000
       dr1 = 0x0000000000000000
       dr2 = 0x0000000000000000
       dr3 = 0x0000000000000000
       dr4 = 0x0000000000000000
       dr5 = 0x0000000000000000
       dr6 = 0x00000000ffff0ff0
       dr7 = 0x0000000000000400
22 registers were unavailable.
```

Sponsored by <The NetBSD Foundation>

Reviewers: labath, emaste, joerg, kettenis

Reviewed By: labath

Subscribers: #lldb

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D32080

llvm-svn: 300548

7 years ago[ARM] Add hardware build attributes in assembler
Oliver Stannard [Tue, 18 Apr 2017 12:52:35 +0000 (12:52 +0000)]
[ARM] Add hardware build attributes in assembler

In the assembler, we should emit build attributes based on the target
selected with command-line options. This matches the GNU assembler's
behaviour. We only do this for build attributes which describe the
hardware that is expected to be available, not the ones that describe
ABI compatibility.

This is done by moving some of the attribute emission code to
ARMTargetStreamer, so that it can be shared between the assembly and
code-generation code paths. Since the assembler only creates a
MCSubtargetInfo, not an ARMSubtarget, the code had to be changed to
check raw features, and not use the convenience functions in
ARMSubtarget.

If different attributes are later specified using the .eabi_attribute
directive, then they will take precedence, as happens when the same
.eabi_attribute is specified twice.

This must be enabled by an option, because we don't want to do this when
parsing inline assembly. The attributes would match the ones emitted at
the start of the file, so wouldn't actually change the emitted object
file, but the extra directives would be added to every inline assembly
block when emitting assembly, which we'd like to avoid.

The majority of the changes in the build-attributes.ll test are just
re-ordering the directives, because the hardware attributes are now
emitted before the ABI ones. However, I did fix one bug which I spotted:
Tag_CPU_arch_profile was not being emitted for v6M.

Differential revision: https://reviews.llvm.org/D31812

llvm-svn: 300547

7 years ago[ARM] GlobalISel: Add support for G_SUB
Diana Picus [Tue, 18 Apr 2017 12:35:28 +0000 (12:35 +0000)]
[ARM] GlobalISel: Add support for G_SUB

Support G_SUB throughout the GlobalISel pipeline. It is exactly the same
as G_ADD, nothing fancy.

llvm-svn: 300546

7 years agoclang-tools-extra/test/clang-tidy/performance-inefficient-vector-operation.cpp: Appea...
NAKAMURA Takumi [Tue, 18 Apr 2017 12:13:30 +0000 (12:13 +0000)]
clang-tools-extra/test/clang-tidy/performance-inefficient-vector-operation.cpp: Appease targeting msvc with -fno-ms-extensions.

FIXME: This may work with -target x86_64-win32.
llvm-svn: 300545

7 years ago[SampleProfile] Don't assert when printing the DebugLoc of a branch. NFC.
Andrea Di Biagio [Tue, 18 Apr 2017 11:27:58 +0000 (11:27 +0000)]
[SampleProfile] Don't assert when printing the DebugLoc of a branch. NFC.

llvm-svn: 300544

7 years agoRevert r300539 - Add #pragma clang attribute
Alex Lorenz [Tue, 18 Apr 2017 10:46:41 +0000 (10:46 +0000)]
Revert r300539 - Add #pragma clang attribute

Some tests fail on the Windows buildbots. I will have to investigate more.
This commit reverts r300539, r300540 and r300542.

llvm-svn: 300543

7 years agoFix one more 'not all control paths return a value' MSVC warning
Alex Lorenz [Tue, 18 Apr 2017 10:17:41 +0000 (10:17 +0000)]
Fix one more 'not all control paths return a value' MSVC warning

The warning was caused by r300539.

llvm-svn: 300542

7 years ago[SampleProfile] Skip intrinsic calls when visiting callsites in InlineHotFunctions.
Andrea Di Biagio [Tue, 18 Apr 2017 10:08:53 +0000 (10:08 +0000)]
[SampleProfile] Skip intrinsic calls when visiting callsites in InlineHotFunctions.

Before this patch, we always called method 'findCalleeFunctionSamples()' on
intrinsic calls. However, intrinsic calls like llvm.dbg.value() are not viable
candidates for obvious reasons.

No functional change intended.

Differential Revision: https://reviews.llvm.org/D32008

llvm-svn: 300541

7 years ago[TableGen] Fix MSVC warning that occurs in TableGen generated function
Alex Lorenz [Tue, 18 Apr 2017 09:59:27 +0000 (09:59 +0000)]
[TableGen] Fix MSVC warning that occurs in TableGen generated function

This should fix the Windows buildbot failure that happened after r300539.

llvm-svn: 300540

7 years agoAdd #pragma clang attribute
Alex Lorenz [Tue, 18 Apr 2017 09:41:47 +0000 (09:41 +0000)]
Add #pragma clang attribute

The new '#pragma clang attribute' directive can be used to apply attributes to
multiple declarations. An attribute must satisfy the following conditions to
be supported by the pragma:
- It must have a subject list that's defined in the TableGen file.
- It must be documented.
- It must not be late parsed.
- It must have a GNU/C++11 spelling.

Differential Revision: https://reviews.llvm.org/D30009

llvm-svn: 300539

7 years agoRevert "[GlobalISel] Support vector-of-pointers in LLT"
Kristof Beyls [Tue, 18 Apr 2017 09:26:36 +0000 (09:26 +0000)]
Revert "[GlobalISel] Support vector-of-pointers in LLT"

This reverts r300535 and r300537.
The newly added tests in test/CodeGen/AArch64/GlobalISel/arm64-fallback.ll
produces slightly different code between LLVM versions being built with different compilers.
E.g., dependent on the compiler LLVM is built with, either one of the following
can be produced:

remark: <unknown>:0:0: unable to legalize instruction: %vreg0<def>(p0) = G_EXTRACT_VECTOR_ELT %vreg1, %vreg2; (in function: vector_of_pointers_extractelement)
remark: <unknown>:0:0: unable to legalize instruction: %vreg2<def>(p0) = G_EXTRACT_VECTOR_ELT %vreg1, %vreg0; (in function: vector_of_pointers_extractelement)

Non-determinism like this is clearly a bad thing, so reverting this until
I can find and fix the root cause of the non-determinism.

llvm-svn: 300538

7 years agoFix gcc build after r300535.
Kristof Beyls [Tue, 18 Apr 2017 08:47:55 +0000 (08:47 +0000)]
Fix gcc build after r300535.

llvm-svn: 300537

7 years ago[ARM] Check for correct HW div when lowering divmod
Diana Picus [Tue, 18 Apr 2017 08:32:27 +0000 (08:32 +0000)]
[ARM] Check for correct HW div when lowering divmod

For subtargets that use the custom lowering for divmod, e.g. gnueabi,
we used to check if the subtarget has hardware divide and then lower to
a div-mul-sub sequence if true, or to a libcall if false.

However, judging by the usage of hasDivide vs hasDivideInARMMode, it
seems that hasDivide only refers to Thumb. For instance, in the
ARMTargetLowering constructor, the code that specifies whether to use
libcalls for (S|U)DIV looks like this:

bool hasDivide = Subtarget->isThumb() ? Subtarget->hasDivide()
                                      : Subtarget->hasDivideInARMMode();

In the case of divmod for arm-gnueabi, using only hasDivide() to
determine what to do means that instead of lowering to __aeabi_idivmod
to get the remainder, we lower to div-mul-sub and then further lower the
div to __aeabi_idiv. Even worse, if we have hardware divide in ARM but
not in Thumb, we generate a libcall instead of using it (this is not an
issue in practice since AFAICT none of the cores that we support have
hardware divide in ARM but not Thumb).

This patch fixes the code dealing with custom lowering to take into
account the mode (Thumb or ARM) when deciding whether or not hardware
division is available.

Differential Revision: https://reviews.llvm.org/D32005

llvm-svn: 300536

7 years ago[GlobalISel] Support vector-of-pointers in LLT
Kristof Beyls [Tue, 18 Apr 2017 08:12:45 +0000 (08:12 +0000)]
[GlobalISel] Support vector-of-pointers in LLT

This fixes PR32471.

As comment 10 on that bug report highlights
(https://bugs.llvm.org//show_bug.cgi?id=32471#c10), there are quite a
few different defendable design tradeoffs that could be made, including
not representing pointers at all in LLT.

I decided to go for representing vector-of-pointer as a concept in LLT,
while keeping the size of the LLT type 64 bits (this is an increase from
48 bits before). My rationale for keeping pointers explicit is that on
some targets probably it's very handy to have the distinction between
pointer and non-pointer (e.g. 68K has a different register bank for
pointers IIRC). If we keep a scalar pointer, it probably is easiest to
also have a vector-of-pointers to keep LLT relatively conceptually clean
and orthogonal, while we don't have a very strong reason to break that
orthogonality. Once we gain more experience on the use of LLT, we can
of course reconsider this direction.

Rejecting vector-of-pointer types in the IRTranslator is also an option
to avoid the crash reported in PR32471, but that is only a very
short-term solution; also needs quite a bit of code tweaks in places,
and is probably fragile. Therefore I didn't consider this the best
option.

llvm-svn: 300535

7 years ago[clang-tidy] Add a clang-tidy check for possible inefficient vector operations
Haojian Wu [Tue, 18 Apr 2017 07:46:39 +0000 (07:46 +0000)]
[clang-tidy] Add a clang-tidy check for possible inefficient vector operations

Summary:
The "performance-inefficient-vector-operation" check finds vector oprations in
for-loop statements which may cause multiple memory reallocations.

This is the first version, only detects typical for-loop:

```
std::vector<int> v;
for (int i = 0; i < n; ++i) {
  v.push_back(i);
}

// or

for (int i = 0; i < v2.size(); ++i) {
  v.push_back(v2[i]);
}
```

We can extend it to handle more cases like for-range loop in the future.

Reviewers: alexfh, aaron.ballman

Reviewed By: aaron.ballman

Subscribers: zaks.anna, Eugene.Zelenko, mgorny, cfe-commits, djasper

Differential Revision: https://reviews.llvm.org/D31757

llvm-svn: 300534

7 years agoEmit benchmark tests into the benchmarks build directory, not lib/
Eric Fiselier [Tue, 18 Apr 2017 07:40:22 +0000 (07:40 +0000)]
Emit benchmark tests into the benchmarks build directory, not lib/

llvm-svn: 300533

7 years agotest commit
Leslie Zhai [Tue, 18 Apr 2017 07:28:54 +0000 (07:28 +0000)]
test commit

llvm-svn: 300532

7 years ago[sanitizer] Don't include <linux/user.h> in sanitizer_stoptheworld_linux_libcdep...
Maxim Ostapenko [Tue, 18 Apr 2017 07:22:26 +0000 (07:22 +0000)]
[sanitizer] Don't include <linux/user.h> in sanitizer_stoptheworld_linux_libcdep.cc on ARM Android

Turned out that adding defined(_arm_) in sanitizer_stoptheworld_linux_libcdep.cc breaks android arm with some toolchains.

.../llvm/projects/compiler-rt/lib/sanitizer_common/sanitizer_stoptheworld_linux_libcdep.cc:36:11: fatal error:
      'linux/user.h' file not found
# include <linux/user.h>  // for pt_regs
          ^
1 error generated.

Context:
#if SANITIZER_ANDROID && defined(__arm__)
# include <linux/user.h>  // for pt_regs
#else

This patch removes corresponding #if SANITIZER_ANDROID && defined(__arm__) and a bit rearranges adjacent сode.

Differential Revision: https://reviews.llvm.org/D32128

llvm-svn: 300531

7 years agoUpdate google benchmark
Eric Fiselier [Tue, 18 Apr 2017 07:17:20 +0000 (07:17 +0000)]
Update google benchmark

llvm-svn: 300530

7 years agoAssert that a valid operator new/delete signature is always found by the coroutine...
Eric Fiselier [Tue, 18 Apr 2017 05:30:39 +0000 (05:30 +0000)]
Assert that a valid operator new/delete signature is always found by the coroutine body

llvm-svn: 300529

7 years agoSpeculatively attempt to fix bot failures caused by recent coroutine changes.
Eric Fiselier [Tue, 18 Apr 2017 05:08:08 +0000 (05:08 +0000)]
Speculatively attempt to fix bot failures caused by recent coroutine changes.

llvm-svn: 300528

7 years ago[APInt] Cleanup the reverseBits slow case a little.
Craig Topper [Tue, 18 Apr 2017 05:02:21 +0000 (05:02 +0000)]
[APInt] Cleanup the reverseBits slow case a little.

Use lshrInPlace. Use single bit extract and operator|=(uint64_t) to avoid a few temporary APInts.

llvm-svn: 300527

7 years ago[APInt] Make operator<<= shift in place. Improve the implementation of tcShiftLeft...
Craig Topper [Tue, 18 Apr 2017 04:39:48 +0000 (04:39 +0000)]
[APInt] Make operator<<= shift in place. Improve the implementation of tcShiftLeft and use it to implement operator<<=.

llvm-svn: 300526

7 years ago[XRay][compiler-rt] Use emulated TSC when CPU supports rdtscp, but cannot determine...
Douglas Yung [Tue, 18 Apr 2017 03:25:11 +0000 (03:25 +0000)]
[XRay][compiler-rt] Use emulated TSC when CPU supports rdtscp, but cannot determine the CPU frequency

A problem arises if a machine supports the rdtscp instruction, but the processor
frequency cannot be determined by the function getTSCFrequency(). In this case,
we want to use the emulated TSC instead. This patch implements that by adding a
call to getTSCFrequency() from probeRequiredCPUFeatures(), and the function only
returns true if both the processor supports rdtscp and the CPU frequency can be
determined.

This should fix PR32620.

Reviewers: dberris

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32067

llvm-svn: 300525

7 years ago[coroutines] Fix building of new/delete expressions when get_return_object_on_allocat...
Eric Fiselier [Tue, 18 Apr 2017 03:12:48 +0000 (03:12 +0000)]
[coroutines] Fix building of new/delete expressions when get_return_object_on_allocation_failure() is present.

Summary:
This patch implements [dcl.fct.def.coroutine]p8:
> The unqualified-id get_return_object_on_allocation_failure is looked up in the scope of
> class P by class member access lookup (3.4.5). If a declaration is found, ..., and if a
> global allocation function is selected, the ::operator new(size_t, nothrow_t) form shall be used.
> [...]
> The allocation function used in this case must have a non-throwing noexcept-specification.

Reviewers: GorNishanov, rsmith, majnemer, aaron.ballman

Reviewed By: GorNishanov

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D31562

llvm-svn: 300524

7 years agoDebug Info: Remove special-casing of indirect function argument handling.
Adrian Prantl [Tue, 18 Apr 2017 01:22:01 +0000 (01:22 +0000)]
Debug Info: Remove special-casing of indirect function argument handling.

LLVM has changed the semantics of dbg.declare for describing function
arguments. After this patch a dbg.declare always takes the *address*
of a variable as the first argument, even if the argument is not an
alloca.

https://bugs.llvm.org/show_bug.cgi?id=32382
rdar://problem/31205000

llvm-svn: 300523

7 years agoPR32382: Fix emitting complex DWARF expressions.
Adrian Prantl [Tue, 18 Apr 2017 01:21:53 +0000 (01:21 +0000)]
PR32382: Fix emitting complex DWARF expressions.

The DWARF specification knows 3 kinds of non-empty simple location
descriptions:
1. Register location descriptions
  - describe a variable in a register
  - consist of only a DW_OP_reg
2. Memory location descriptions
  - describe the address of a variable
3. Implicit location descriptions
  - describe the value of a variable
  - end with DW_OP_stack_value & friends

The existing DwarfExpression code is pretty much ignorant of these
restrictions. This used to not matter because we only emitted very
short expressions that we happened to get right by accident.  This
patch makes DwarfExpression aware of the rules defined by the DWARF
standard and now chooses the right kind of location description for
each expression being emitted.

This would have been an NFC commit (for the existing testsuite) if not
for the way that clang describes captured block variables. Based on
how the previous code in LLVM emitted locations, DW_OP_deref
operations that should have come at the end of the expression are put
at its beginning. Fixing this means changing the semantics of
DIExpression, so this patch bumps the version number of DIExpression
and implements a bitcode upgrade.

There are two major changes in this patch:

I had to fix the semantics of dbg.declare for describing function
arguments. After this patch a dbg.declare always takes the *address*
of a variable as the first argument, even if the argument is not an
alloca.

When lowering a DBG_VALUE, the decision of whether to emit a register
location description or a memory location description depends on the
MachineLocation — register machine locations may get promoted to
memory locations based on their DIExpression. (Future) optimization
passes that want to salvage implicit debug location for variables may
do so by appending a DW_OP_stack_value. For example:
  DBG_VALUE, [RBP-8]                        --> DW_OP_fbreg -8
  DBG_VALUE, RAX                            --> DW_OP_reg0 +0
  DBG_VALUE, RAX, DIExpression(DW_OP_deref) --> DW_OP_reg0 +0

All testcases that were modified were regenerated from clang. I also
added source-based testcases for each of these to the debuginfo-tests
repository over the last week to make sure that no synchronized bugs
slip in. The debuginfo-tests compile from source and run the debugger.

https://bugs.llvm.org/show_bug.cgi?id=32382
<rdar://problem/31205000>

Differential Revision: https://reviews.llvm.org/D31439

llvm-svn: 300522

7 years ago[asan] Fixup for r300483 (which is a fixup for r300473).
Evgeniy Stepanov [Tue, 18 Apr 2017 01:08:00 +0000 (01:08 +0000)]
[asan] Fixup for r300483 (which is a fixup for r300473).

Sanitizer Printf() does not know about %lu.

llvm-svn: 300521

7 years agoAdd const to a const method. NFC
George Burgess IV [Tue, 18 Apr 2017 01:04:05 +0000 (01:04 +0000)]
Add const to a const method. NFC

llvm-svn: 300520

7 years agoTestStaticVariables still fails on Linux.
Jim Ingham [Tue, 18 Apr 2017 00:44:14 +0000 (00:44 +0000)]
TestStaticVariables still fails on Linux.

llvm-svn: 300519