platform/upstream/llvm.git
6 years agoAdd missing include (cstdlib) to Demangle.h
Konstantin Zhuravlyov [Tue, 22 Jan 2019 19:18:18 +0000 (19:18 +0000)]
Add missing include (cstdlib) to Demangle.h

Differential Revision: https://reviews.llvm.org/D57035

llvm-svn: 351861

6 years agoAMDGPU/GlobalISel: Remove vectors from legal constant types
Matt Arsenault [Tue, 22 Jan 2019 19:04:51 +0000 (19:04 +0000)]
AMDGPU/GlobalISel: Remove vectors from legal constant types

llvm-svn: 351859

6 years agoGlobalISel: Support narrowing zextload/sextload
Matt Arsenault [Tue, 22 Jan 2019 19:02:10 +0000 (19:02 +0000)]
GlobalISel: Support narrowing zextload/sextload

llvm-svn: 351856

6 years ago[libFuzzer][MSVC] Enable building libFuzzer with MSVC
Jonathan Metzman [Tue, 22 Jan 2019 18:59:25 +0000 (18:59 +0000)]
[libFuzzer][MSVC] Enable building libFuzzer with MSVC

Summary:
Enable building libFuzzer with MSVC.

* Don't try to include <endian.h> in FuzzerSHA1.cpp. MSVC
  doesn't have this header, and WINDOWS is always little
  endian (even on ARM)

Subscribers: srhines, mgorny, javed.absar, kristof.beyls

Differential Revision: https://reviews.llvm.org/D56510

llvm-svn: 351855

6 years ago[SelectionDAGBuilder] Defer C_Register Assignments to be in line with
Nirav Dave [Tue, 22 Jan 2019 18:57:49 +0000 (18:57 +0000)]
[SelectionDAGBuilder] Defer C_Register Assignments to be in line with
those of C_RegisterClass. NFCI.

llvm-svn: 351854

6 years agoGlobalISel: Disallow vectors for G_CONSTANT/G_FCONSTANT
Matt Arsenault [Tue, 22 Jan 2019 18:53:41 +0000 (18:53 +0000)]
GlobalISel: Disallow vectors for G_CONSTANT/G_FCONSTANT

llvm-svn: 351853

6 years agoFileOutputBuffer: Handle "-" as stdout.
Rui Ueyama [Tue, 22 Jan 2019 18:44:04 +0000 (18:44 +0000)]
FileOutputBuffer: Handle "-" as stdout.

I was honestly a bit surprised that we didn't do this before. This
patch is to handle "-" as the stdout so that if you pass `-o -` to
lld, for example, it writes an output to stdout instead of file `-`.

I thought that we might want to handle this at a higher level than
FileOutputBuffer, because if we land this patch, we can no longer
create a file whose name is `-` (there's a workaround though; you can
pass `./-` instead of `-`). However, because raw_fd_ostream already
handles `-` as a special file name, I think it's okay and actually
consistent to handle `-` as a special name in FileOutputBuffer.

Differential Revision: https://reviews.llvm.org/D56940

llvm-svn: 351852

6 years agoCodegen support for atomicrmw fadd/fsub
Matt Arsenault [Tue, 22 Jan 2019 18:36:06 +0000 (18:36 +0000)]
Codegen support for atomicrmw fadd/fsub

llvm-svn: 351851

6 years agoReapply "IR: Add fp operations to atomicrmw"
Matt Arsenault [Tue, 22 Jan 2019 18:18:02 +0000 (18:18 +0000)]
Reapply "IR: Add fp operations to atomicrmw"

This reapplies commits r351778 and r351782 with
RISCV test fixes.

llvm-svn: 351850

6 years ago[ASTImporter] Fix importing OperatorDelete from CXXConstructorDecl
Raphael Isemann [Tue, 22 Jan 2019 17:59:45 +0000 (17:59 +0000)]
[ASTImporter] Fix importing OperatorDelete from CXXConstructorDecl

Summary:
Shafik found out that importing a CXXConstructorDecl will create a translation unit that
causes Clang's CodeGen to crash. The reason for that is that we don't copy the OperatorDelete
from the CXXConstructorDecl when importing. This patch fixes it and adds a test case for that.

Reviewers: shafik, martong, a_sidorin, a.sidorin

Reviewed By: martong, a_sidorin

Subscribers: rnkovacs, cfe-commits

Differential Revision: https://reviews.llvm.org/D56651

llvm-svn: 351849

6 years ago[llvm-mca][X86] Tidyup avx512 placeholder tests
Simon Pilgrim [Tue, 22 Jan 2019 17:52:15 +0000 (17:52 +0000)]
[llvm-mca][X86] Tidyup avx512 placeholder tests

Ensure we keep avx512f/bw/dq + vl versions separate, add example broadcast tests - this should allow us to better the test coverage of test\CodeGen\X86\avx512-schedule.ll

llvm-svn: 351848

6 years ago[libcxx] Include <cstring> in tests that use strcmp
Louis Dionne [Tue, 22 Jan 2019 17:45:00 +0000 (17:45 +0000)]
[libcxx] Include <cstring> in tests that use strcmp

Reviewed as https://reviews.llvm.org/D56503.
Thanks to Andrey Maksimov for the patch.

llvm-svn: 351847

6 years ago[DEBUGINFO, NVPTX] Enable support for the debug info on NVPTX target.
Alexey Bataev [Tue, 22 Jan 2019 17:43:37 +0000 (17:43 +0000)]
[DEBUGINFO, NVPTX] Enable support for the debug info on NVPTX target.

Summary: Enable full support for the debug info.

Reviewers: echristo

Subscribers: jholewinski, aprantl, JDevlieghere, llvm-commits

Differential Revision: https://reviews.llvm.org/D46189

llvm-svn: 351846

6 years agoRevert r351520, "Re-enable terminator folding in LoopSimplifyCFG"
Jordan Rupprecht [Tue, 22 Jan 2019 17:39:02 +0000 (17:39 +0000)]
Revert r351520, "Re-enable terminator folding in LoopSimplifyCFG"

This is still causing compilation crashes in some targets. Will follow up shortly with a repro.

llvm-svn: 351845

6 years ago[DEBUG_INFO, NVPTX] Fix relocation info.
Alexey Bataev [Tue, 22 Jan 2019 17:24:16 +0000 (17:24 +0000)]
[DEBUG_INFO, NVPTX] Fix relocation info.

Summary: Initial function labels must follow the debug location for the correct relocation info generation.

Reviewers: tra, jlebar, echristo

Subscribers: jholewinski, llvm-commits

Differential Revision: https://reviews.llvm.org/D45784

llvm-svn: 351843

6 years ago[llvm-mca][X86] Add VPOPCNTDQ tests
Simon Pilgrim [Tue, 22 Jan 2019 17:19:44 +0000 (17:19 +0000)]
[llvm-mca][X86] Add VPOPCNTDQ tests

Matches test coverage of test\CodeGen\X86\avx512vpopcntdq-schedule.ll

llvm-svn: 351842

6 years agoReleaseNotes: remove openmp notes from r351580
Hans Wennborg [Tue, 22 Jan 2019 17:01:39 +0000 (17:01 +0000)]
ReleaseNotes: remove openmp notes from r351580

They were for the 8.0 branch, and have been committed there in r351839.

llvm-svn: 351841

6 years ago[x86] add partial undef 'and' test; NFC
Sanjay Patel [Tue, 22 Jan 2019 17:01:06 +0000 (17:01 +0000)]
[x86] add partial undef 'and' test; NFC

llvm-svn: 351840

6 years ago[docs] Scudo: document error messages & their potential cause
Kostya Kortchinsky [Tue, 22 Jan 2019 16:43:45 +0000 (16:43 +0000)]
[docs] Scudo: document error messages & their potential cause

Summary:
A couple of changes in the Scudo documentation:
- tag the shell code blocks as `console`;
- document error messages that are displayed in some termination conditions,
  the reason they triggered, and potential causes.

Reviewers: eugenis, enh

Reviewed By: eugenis

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D56857

llvm-svn: 351838

6 years agoAdd DIGlobalVariableExpression to LangRef
Adrian Prantl [Tue, 22 Jan 2019 16:40:18 +0000 (16:40 +0000)]
Add DIGlobalVariableExpression to LangRef

llvm-svn: 351837

6 years ago[llvm-mca][X86] Add missing CLWB/CLZERO/FSGSBASE/LWP/MWAITX/RDPID/SHA tests
Simon Pilgrim [Tue, 22 Jan 2019 16:39:28 +0000 (16:39 +0000)]
[llvm-mca][X86] Add missing CLWB/CLZERO/FSGSBASE/LWP/MWAITX/RDPID/SHA tests

We're getting pretty close to matching/exceeding test coverage of the test\CodeGen\X86\*-schedule.ll files, which should allow us to get rid of -print-schedule and fix PR37160

llvm-svn: 351836

6 years ago[llvm-mca][X86] Add missing enter/leave, invlpg/invlpga, rdmsr/wrmsr, rdpmc and rdtsc...
Simon Pilgrim [Tue, 22 Jan 2019 16:29:26 +0000 (16:29 +0000)]
[llvm-mca][X86] Add missing enter/leave, invlpg/invlpga, rdmsr/wrmsr, rdpmc and rdtsc/rdtscp tests

llvm-svn: 351835

6 years ago[x86] add another partial undef vector binop test; NFC
Sanjay Patel [Tue, 22 Jan 2019 16:26:09 +0000 (16:26 +0000)]
[x86] add another partial undef vector binop test; NFC

The existing test unintentionally shows that we have prematurely
optimized the shuffle into a vector concat and lost the undef info,
so it is not affected by a basic improvement to
SimplifyDemandedVectorElts.

llvm-svn: 351834

6 years agoUse response file when generating LLVM-C.dll
Serge Guelton [Tue, 22 Jan 2019 16:25:17 +0000 (16:25 +0000)]
Use response file when generating LLVM-C.dll

As discovered in D56774 the command line gets to long, so use a response file to give the script the libs. This change has been tested and is confirmed working for me.

Commited on behalf of Jakob Bornecrantz

Differential Revision: https://reviews.llvm.org/D56781

llvm-svn: 351833

6 years agoNote that we have a patch for LWG3101
Marshall Clow [Tue, 22 Jan 2019 16:22:53 +0000 (16:22 +0000)]
Note that we have a patch for LWG3101

llvm-svn: 351832

6 years ago[llvm-mca][X86] Add missing mfence/pinsrw tests
Simon Pilgrim [Tue, 22 Jan 2019 16:01:08 +0000 (16:01 +0000)]
[llvm-mca][X86] Add missing mfence/pinsrw tests

llvm-svn: 351831

6 years ago[CMake] Turn LLDB_FRAMEWORK_TOOLS into STRING to allow overrides from cache files
Stefan Granitz [Tue, 22 Jan 2019 15:59:47 +0000 (15:59 +0000)]
[CMake] Turn LLDB_FRAMEWORK_TOOLS into STRING to allow overrides from cache files

llvm-svn: 351830

6 years ago[llvm-mca][X86] Add missing monitor/mwait tests
Simon Pilgrim [Tue, 22 Jan 2019 15:48:16 +0000 (15:48 +0000)]
[llvm-mca][X86] Add missing monitor/mwait tests

These technically should be under a MONITOR cpuid bit, but we tag them as SSE3 so I've done that here as well.

llvm-svn: 351829

6 years ago[llvm-mca][X86] Add missing vperm2i128 tests
Simon Pilgrim [Tue, 22 Jan 2019 14:54:24 +0000 (14:54 +0000)]
[llvm-mca][X86] Add missing vperm2i128 tests

llvm-svn: 351828

6 years ago[llvm-mca][X86] Add missing tzcntw tests
Simon Pilgrim [Tue, 22 Jan 2019 14:53:52 +0000 (14:53 +0000)]
[llvm-mca][X86] Add missing tzcntw tests

llvm-svn: 351827

6 years ago[clangd] Followup fix of rL351818
Haojian Wu [Tue, 22 Jan 2019 14:48:04 +0000 (14:48 +0000)]
[clangd] Followup fix of rL351818

ClangTidyOptions::getDefaults is not free, it will initialize all
clang-tidy modules to get check-specific options, and we don't use this
information in CodeComplete, so using an empty one (constructed by
default constructor) is sufficient.

llvm-svn: 351826

6 years ago[DAGCombiner] narrow vector binop with 2 insert subvector operands
Sanjay Patel [Tue, 22 Jan 2019 14:24:13 +0000 (14:24 +0000)]
[DAGCombiner] narrow vector binop with 2 insert subvector operands

vecbo (insertsubv undef, X, Z), (insertsubv undef, Y, Z) --> insertsubv VecC, (vecbo X, Y), Z

This is another step in generic vector narrowing. It's also a step towards more horizontal op
formation specifically for x86 (although we still failed to match those in the affected tests).

The scalarization cases are also not optimal (we should be scalarizing those), but it's still
an improvement to use a narrower vector op when we know part of the result must be constant
because both inputs are undef in some vector lanes.

I think a similar match but checking for a constant operand might help some of the cases in
D51553.

Differential Revision: https://reviews.llvm.org/D56875

llvm-svn: 351825

6 years ago[llvm-objdump] - Introduce getRelocsMap() helper. NFCI.
George Rimar [Tue, 22 Jan 2019 14:09:37 +0000 (14:09 +0000)]
[llvm-objdump] - Introduce getRelocsMap() helper. NFCI.

Currently disassembleObject() is a ~550 lines length function.
This patch extracts the code that creates a section->their relocation
mapping into a new helper function to simplify/reduce it a bit.

Differential revision: https://reviews.llvm.org/D57019

llvm-svn: 351824

6 years ago[RISCV][NFC] Change naming scheme for RISC-V specific DAG nodes
Alex Bradbury [Tue, 22 Jan 2019 14:05:11 +0000 (14:05 +0000)]
[RISCV][NFC] Change naming scheme for RISC-V specific DAG nodes

Previously we had names like 'Call' or 'Tail'. This potentially clashes with
the naming scheme used elsewhere in RISCVInstrInfo.td. Many other backends
would use names like AArch64call or PPCtail. I prefer the SystemZ approach,
which uses prefixed all-lowercase names. This matches the naming scheme used
for target-independent SelectionDAG nodes.

llvm-svn: 351823

6 years ago[MCA] Add tests for int-to-fpu transfer delays. NFC
Andrea Di Biagio [Tue, 22 Jan 2019 13:59:08 +0000 (13:59 +0000)]
[MCA] Add tests for int-to-fpu transfer delays. NFC

llvm-svn: 351822

6 years agoSlight fix for r351820
Serge Guelton [Tue, 22 Jan 2019 13:57:29 +0000 (13:57 +0000)]
Slight fix for r351820

llvm-svn: 351821

6 years agoFix llvm::is_trivially_copyable portability issues
Serge Guelton [Tue, 22 Jan 2019 13:48:55 +0000 (13:48 +0000)]
Fix llvm::is_trivially_copyable portability issues

llvm::is_trivially_copyable portability is verified at compile time using
std::is_trivially_copyable as the reference implementation.

Unfortunately, the latter is not available on all platforms, so introduce
a proper configure check to detect if it is available on the target platform.

In a similar manner, std::is_copy_assignable is not fully supported for gcc4.9.
Provide a portable (?) implementation instead.

Differential Revision: https://reviews.llvm.org/D57018

llvm-svn: 351820

6 years ago[X86][SSE] Canonicalize OR(AND(X,C),AND(Y,~C)) -> OR(AND(X,C),ANDNP(C,Y))
Simon Pilgrim [Tue, 22 Jan 2019 13:44:49 +0000 (13:44 +0000)]
[X86][SSE] Canonicalize OR(AND(X,C),AND(Y,~C)) -> OR(AND(X,C),ANDNP(C,Y))

For constant bit select patterns, replace one AND with a ANDNP, allowing us to reuse the constant mask. Only do this if the mask has multiple uses (to avoid losing load folding) or if we have XOP as its VPCMOV can handle most folding commutations.

This also requires computeKnownBitsForTargetNode support for X86ISD::ANDNP and X86ISD::FOR to prevent regressions in fabs/fcopysign patterns.

Differential Revision: https://reviews.llvm.org/D55935

llvm-svn: 351819

6 years agoFix "missing field 'ClangTidyOpts' initializer" warning. NFCI.
Simon Pilgrim [Tue, 22 Jan 2019 13:35:16 +0000 (13:35 +0000)]
Fix "missing field 'ClangTidyOpts' initializer" warning. NFCI.

llvm-svn: 351818

6 years ago[X86][BtVer2] SSE2 vector shifts has local forwarding disabled
Simon Pilgrim [Tue, 22 Jan 2019 13:27:18 +0000 (13:27 +0000)]
[X86][BtVer2] SSE2 vector shifts has local forwarding disabled

Similar to horizontal ops on D56777, the sse2 (but not mmx) bit shift ops has local forwarding disabled, adding +1cy to the use latency for the result.

Differential Revision: https://reviews.llvm.org/D57026

llvm-svn: 351817

6 years agoFix "comparison of unsigned expression >= 0 is always true" warning. NFCI.
Simon Pilgrim [Tue, 22 Jan 2019 13:18:26 +0000 (13:18 +0000)]
Fix "comparison of unsigned expression >= 0 is always true" warning. NFCI.

llvm-svn: 351816

6 years ago[X86][BtVer2] X86ISD::VPERMILPV has local forwarding disabled
Simon Pilgrim [Tue, 22 Jan 2019 13:13:57 +0000 (13:13 +0000)]
[X86][BtVer2] X86ISD::VPERMILPV has local forwarding disabled

Similar to horizontal ops on D56777, the vpermilpd/vpermilps variable mask ops has local forwarding disabled, adding +1cy to the use latency for the result.

Differential Revision: https://reviews.llvm.org/D57022

llvm-svn: 351815

6 years ago[clang-tidy] Fix whitespace in docs. NFC
Alexander Kornienko [Tue, 22 Jan 2019 12:59:34 +0000 (12:59 +0000)]
[clang-tidy] Fix whitespace in docs. NFC

Actually, just testing commits via monorepo ;)

llvm-svn: 351814

6 years ago[clangd] NFC: reduce log noise from Diagnostics.
Eric Liu [Tue, 22 Jan 2019 12:55:27 +0000 (12:55 +0000)]
[clangd] NFC: reduce log noise from Diagnostics.

Reviewers: sammccall

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, cfe-commits

Differential Revision: https://reviews.llvm.org/D57042

llvm-svn: 351813

6 years ago[clangd] Fix the broken buildbot.
Haojian Wu [Tue, 22 Jan 2019 12:55:15 +0000 (12:55 +0000)]
[clangd] Fix the broken buildbot.

llvm-svn: 351812

6 years agoRevert "[llvm-objcopy] [COFF] Implement --add-gnu-debuglink"
Martin Storsjo [Tue, 22 Jan 2019 12:35:34 +0000 (12:35 +0000)]
Revert "[llvm-objcopy] [COFF] Implement --add-gnu-debuglink"

This reverts commit r351801, as it caused errors on (so far)
ppc64be and aarch64 buildbots - the reason is yet unknown.

llvm-svn: 351811

6 years ago[CostModel][X86] Add ICMP Predicate specific costs
Simon Pilgrim [Tue, 22 Jan 2019 12:29:38 +0000 (12:29 +0000)]
[CostModel][X86] Add ICMP Predicate specific costs

First step towards PR40376, this patch adds support for getCmpSelInstrCost to use the (optional) Instruction CmpInst predicate to indicate the type of integer comparison we're performing and alter the costs accordingly.

Differential Revision: https://reviews.llvm.org/D57013

llvm-svn: 351810

6 years ago[clangd] Fix the `-Wtype-limits` warning, NFC
Haojian Wu [Tue, 22 Jan 2019 12:21:25 +0000 (12:21 +0000)]
[clangd] Fix the `-Wtype-limits` warning, NFC

The assertion is always true, and triggers a compiler warning, so remove it.

llvm-svn: 351809

6 years agoFix/unify top comment in lib/Analysis/PolyhedralInfo.cpp
Michal Gorny [Tue, 22 Jan 2019 12:18:38 +0000 (12:18 +0000)]
Fix/unify top comment in lib/Analysis/PolyhedralInfo.cpp

Change the top comment in PolyhedralInfo.cpp to use // instead of ///,
similarly to headers in other files.  This fixes the issue of copyright
line exceeding textwidth and triggering polly-check-format45 failure,
e.g. seen here:

http://lab.llvm.org:8011/builders/lldb-amd64-ninja-netbsd8/builds/18293/steps/run%20unit%20tests/logs/stdio

llvm-svn: 351808

6 years ago[X86][SSE] Add selective commutation support for insertps (PR40340)
Simon Pilgrim [Tue, 22 Jan 2019 12:17:48 +0000 (12:17 +0000)]
[X86][SSE] Add selective commutation support for insertps (PR40340)

When we are inserting 1 "inline" element, and zeroing 2 of the other elements then we can safely commute the insertps source inputs to improve memory folding.

Differential Revision: https://reviews.llvm.org/D56843

llvm-svn: 351807

6 years ago[RISCV] Quick fix for PR40333
Alex Bradbury [Tue, 22 Jan 2019 12:11:53 +0000 (12:11 +0000)]
[RISCV] Quick fix for PR40333

Avoid the infinite loop caused by the target DAG combine converting ANYEXT to
SIGNEXT and the target-independent DAG combine logic converting back to
ANYEXT. Do this by not adding the new node to the worklist.

Committing directly as this definitely doesn't make the problem any worse, and
I intend to follow-up with a patch that avoids this custom combiner logic
altogether and just lowers the i32 operations to a target-specific
SelectionDAG node. This should be easier to reason about and improve codegen
quality in some cases (though may miss out on some later DAG combines).

llvm-svn: 351806

6 years ago[LoopPredication] Support guards expressed as branches by widenable condition
Max Kazantsev [Tue, 22 Jan 2019 11:49:06 +0000 (11:49 +0000)]
[LoopPredication] Support guards expressed as branches by widenable condition

This patch adds support of guards expressed as branches by widenable
conditions in Loop Predication.

Differential Revision: https://reviews.llvm.org/D56081
Reviewed By: reames

llvm-svn: 351805

6 years ago[X86] Add test for matchAddressRecursively's MUL handling
Simon Pilgrim [Tue, 22 Jan 2019 11:39:21 +0000 (11:39 +0000)]
[X86] Add test for matchAddressRecursively's MUL handling

Noticed in code coverage tests that this isn't tested.

llvm-svn: 351804

6 years ago[NFC] Add function to parse widenable conditional branches
Max Kazantsev [Tue, 22 Jan 2019 11:21:32 +0000 (11:21 +0000)]
[NFC] Add function to parse widenable conditional branches

llvm-svn: 351803

6 years ago[llvm-objcopy] [COFF] Implement --add-gnu-debuglink
Martin Storsjo [Tue, 22 Jan 2019 10:58:18 +0000 (10:58 +0000)]
[llvm-objcopy] [COFF] Implement --add-gnu-debuglink

Differential Revision: https://reviews.llvm.org/D57007

llvm-svn: 351801

6 years ago[llvm-objcopy] [COFF] Update symbol indices in weak externals
Martin Storsjo [Tue, 22 Jan 2019 10:58:09 +0000 (10:58 +0000)]
[llvm-objcopy] [COFF] Update symbol indices in weak externals

Differential Revision: https://reviews.llvm.org/D57006

llvm-svn: 351800

6 years ago[llvm-objcopy] Consistently use createStringError instead of make_error<StringError>
Martin Storsjo [Tue, 22 Jan 2019 10:57:59 +0000 (10:57 +0000)]
[llvm-objcopy] Consistently use createStringError instead of make_error<StringError>

This was requested in the review of D57006.

Also add missing quotes around symbol names in error messages.

Differential Revision: https://reviews.llvm.org/D57014

llvm-svn: 351799

6 years ago[NFC][llvm-readobj]Normalise --/- inconsistency in test options
James Henderson [Tue, 22 Jan 2019 10:57:21 +0000 (10:57 +0000)]
[NFC][llvm-readobj]Normalise --/- inconsistency in test options

llvm-svn: 351798

6 years ago[X86] HADDPS/HADDPD scalar lowering was added at rL350421
Simon Pilgrim [Tue, 22 Jan 2019 10:49:41 +0000 (10:49 +0000)]
[X86] HADDPS/HADDPD scalar lowering was added at rL350421

llvm-svn: 351797

6 years agoRevert r351778: IR: Add fp operations to atomicrmw
Chandler Carruth [Tue, 22 Jan 2019 10:29:58 +0000 (10:29 +0000)]
Revert r351778: IR: Add fp operations to atomicrmw

This broke the RISCV build, and even with that fixed, one of the RISCV
tests behaves surprisingly differently with asserts than without,
leaving there no clear test pattern to use. Generally it seems bad for
hte IR to differ substantially due to asserts (as in, an alloca is used
with asserts that isn't needed without!) and nothing I did simply would
fix it so I'm reverting back to green.

This also required reverting the RISCV build fix in r351782.

llvm-svn: 351796

6 years ago[llvm-symbolizer] Add support for --basenames/-s
James Henderson [Tue, 22 Jan 2019 10:24:32 +0000 (10:24 +0000)]
[llvm-symbolizer] Add support for --basenames/-s

This fixes https://bugs.llvm.org/show_bug.cgi?id=40068.

--basenames is a GNU addr2line switch which strips the directory names
from the file path in the output.

Reviewed by: ruiu

Differential Revision: https://reviews.llvm.org/D56919

llvm-svn: 351795

6 years ago[NFC] Factor out some reusable logic
Max Kazantsev [Tue, 22 Jan 2019 10:13:36 +0000 (10:13 +0000)]
[NFC] Factor out some reusable logic

llvm-svn: 351794

6 years ago[clangd] NFC: Use buildCompilerInvocation in CodeComplete
Kadir Cetinkaya [Tue, 22 Jan 2019 09:58:53 +0000 (09:58 +0000)]
[clangd] NFC: Use buildCompilerInvocation in CodeComplete

Reviewers: ilya-biryukov, sammccall

Reviewed By: sammccall

Subscribers: ioeric, MaskRay, jkorous, arphaman, cfe-commits

Differential Revision: https://reviews.llvm.org/D56860

llvm-svn: 351793

6 years ago[clangd] Support clang-tidy configuration in clangd.
Haojian Wu [Tue, 22 Jan 2019 09:39:05 +0000 (09:39 +0000)]
[clangd] Support clang-tidy configuration in clangd.

Summary:
This patch adds some basic supports for clang-tidy configurations in clangd:
      - clangd will respect .clang-tidy configurations for each file
      - we don't aim to support all clang-tidy options in clangd, only a
        small subset of condfigurations (options related to which checks will be
        enabled) are supported.
      - add a `clang-tidy-checks` CLI option that can override options from
        .clang-tidy file

Reviewers: ilya-biryukov, sammccall

Reviewed By: sammccall

Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits

Differential Revision: https://reviews.llvm.org/D55256

llvm-svn: 351792

6 years ago[NFC] Add detector for guards expressed as branch by widenable conditions
Max Kazantsev [Tue, 22 Jan 2019 09:36:22 +0000 (09:36 +0000)]
[NFC] Add detector for guards expressed as branch by widenable conditions

This patch adds a function to detect guards expressed in explicit control
flow form as branch by `and` with widenable condition intrinsic call:

    %wc = call i1 @llvm.experimental.widenable.condition()
    %guard_cond = and i1, %some_cond, %wc
    br i1 %guard_cond, label %guarded, label %deopt

  deopt:
    <maybe some non-side-effecting instructions>
    deoptimize()

This form can be used as alternative to implicit control flow guard
representation expressed by `experimental_guard` intrinsic.

Differential Revision: https://reviews.llvm.org/D56074
Reviewed By: reames

llvm-svn: 351791

6 years ago[LLD][ELF]Fix tests for D56910
James Henderson [Tue, 22 Jan 2019 09:35:47 +0000 (09:35 +0000)]
[LLD][ELF]Fix tests for D56910

r351789 changes the output of llvm-readelf --dyn-symbols. This causes 3
LLD tests to break. This patch fixes them.

Reviewed by: ruiu

Differential Revision: https://reviews.llvm.org/D56911

llvm-svn: 351790

6 years ago[llvm-readelf]Revert --dyn-symbols behaviour to make it GNU compatible, and add new...
James Henderson [Tue, 22 Jan 2019 09:35:35 +0000 (09:35 +0000)]
[llvm-readelf]Revert --dyn-symbols behaviour to make it GNU compatible, and add new --hash-symbols switch for old behaviour

In r287786, the behaviour of --dyn-symbols in llvm-readelf (but not
llvm-readobj) was changed to print the dynamic symbols as derived from
the hash table, rather than to print the dynamic symbol table contents
directly. The original change was initially submitted without review,
and some comments were made on the commit mailing list implying that the
new behavious is GNU compatible. I argue that it is not:

  1) It does not include a null symbol.
  2) It prints the symbols based on an order derived from the hash
     table.
  3) It prints an extra column indicating which bucket it came from.
     This could break parsers that expect a fixed number of columns,
     with the first column being the symbol index.
  4) If the input happens to have both .hash and .gnu.hash section, it
     prints interpretations of them both, resulting in most symbols
     being printed twice.
  5) There is no way of just printing the raw dynamic symbol table,
     because --symbols also prints the static symbol table.

This patch reverts the --dyn-symbols behaviour back to its old behaviour
of just printing the contents of the dynamic symbol table, similar to
what is printed by --symbols. As the hashed interpretation is still
desirable to validate the hash table, it puts it under a new switch
"--hash-symbols". This is a no-op on all output forms except for GNU
output style for ELF. If there is no hash table, it does nothing,
unlike the previous behaviour which printed the raw dynamic symbol
table, since the raw dynsym is available under --dyn-symbols.

The yaml input for the test is based on that in
test/tools/llvm-readobj/demangle.test, but stripped down to the bare
minimum to provide a valid dynamic symbol.

Note: some LLD tests needed updating. I will commit a separate patch for
those.

Reviewed by: grimar, rupprecht

Differential Revision: https://reviews.llvm.org/D56910

llvm-svn: 351789

6 years ago[clangd] Filter out plugin related flags and move all commandline manipulations into...
Kadir Cetinkaya [Tue, 22 Jan 2019 09:10:20 +0000 (09:10 +0000)]
[clangd] Filter out plugin related flags and move all commandline manipulations into OverlayCDB.

Summary:
Some projects make use of clang plugins when building, but clangd is
not aware of those plugins therefore can't work with the same compile command
arguments.

There were multiple places clangd performed commandline manipulations,
 this one also moves them all into OverlayCDB.

Reviewers: ilya-biryukov

Subscribers: klimek, sammccall, ioeric, MaskRay, jkorous, arphaman, cfe-commits

Differential Revision: https://reviews.llvm.org/D56841

llvm-svn: 351788

6 years agoRevert "Remove static_assert(value == std::is_trivially_copyable<T>::value)"
Vitaly Buka [Tue, 22 Jan 2019 07:22:45 +0000 (07:22 +0000)]
Revert "Remove static_assert(value == std::is_trivially_copyable<T>::value)"

Upgraded the bot as workaround.

This reverts commit r351784.

llvm-svn: 351786

6 years ago[RISCV][NFC] Add break to case statement in RISCVDAGToDAGISel::Select
Alex Bradbury [Tue, 22 Jan 2019 07:22:00 +0000 (07:22 +0000)]
[RISCV][NFC] Add break to case statement in RISCVDAGToDAGISel::Select

The break isn't strictly needed yet as there is no subsequent entry in the
case. But adding to prevent mistakes further down the road.

llvm-svn: 351785

6 years agoRemove static_assert(value == std::is_trivially_copyable<T>::value)
Vitaly Buka [Tue, 22 Jan 2019 06:26:50 +0000 (06:26 +0000)]
Remove static_assert(value == std::is_trivially_copyable<T>::value)

This fails to compile with clang ang libstdc++ 4.6

llvm-svn: 351784

6 years ago[safestack] Return syscalls for mmap, munmap and mprotect
Vitaly Buka [Tue, 22 Jan 2019 05:23:48 +0000 (05:23 +0000)]
[safestack] Return syscalls for mmap, munmap and mprotect

This function can be already intercepted by instrumented code.

llvm-svn: 351783

6 years ago[RISCV] Fix build after r351778
Alex Bradbury [Tue, 22 Jan 2019 05:06:57 +0000 (05:06 +0000)]
[RISCV] Fix build after r351778

Also add a comment to explain the expansion strategy for atomicrmw
{fadd,fsub}.

llvm-svn: 351782

6 years agobreakpad: Add FUNC records to the symtab
Pavel Labath [Tue, 22 Jan 2019 04:56:31 +0000 (04:56 +0000)]
breakpad: Add FUNC records to the symtab

This patch extends SymbolFileBreakpad::AddSymbols to include the symbols
from the FUNC records too. These symbols come from the debug info and
have a size associated with them, so they are given preference in case
there is a PUBLIC record for the same address.

To achieve this, I first pre-process the symbols into a temporary
DenseMap, and then insert the uniqued symbols into the module's symtab.

Reviewers: clayborg, lemo, zturner

Reviewed By: clayborg

Subscribers: lldb-commits

Differential Revision: https://reviews.llvm.org/D56590

llvm-svn: 351781

6 years ago[Test] Fix up tests affected by the new LLVM header.
Jonas Devlieghere [Tue, 22 Jan 2019 03:50:44 +0000 (03:50 +0000)]
[Test] Fix up tests affected by the new LLVM header.

The new LLVM header is one line shorter than the old one, which lead to
some test failures. Ideally tests should rely on line numbers for
breakpoints or output, but that's a different discussion. Hopefully this
turns the bots green again.

llvm-svn: 351779

6 years agoIR: Add fp operations to atomicrmw
Matt Arsenault [Tue, 22 Jan 2019 03:32:36 +0000 (03:32 +0000)]
IR: Add fp operations to atomicrmw

Add just fadd/fsub for now.

llvm-svn: 351778

6 years ago[ARM] Combine ands+lsls to lsls+lsrs for Thumb1.
Eli Friedman [Tue, 22 Jan 2019 01:51:37 +0000 (01:51 +0000)]
[ARM] Combine ands+lsls to lsls+lsrs for Thumb1.

This patch may seem familiar... but my previous patch handled the
equivalent lsls+and, not this case.  Usually instcombine puts the
"and" after the shift, so this case doesn't come up. However, if the
shift comes out of a GEP, it won't get canonicalized by instcombine,
and DAGCombine doesn't have an equivalent transform.

This also modifies isDesirableToCommuteWithShift to suppress DAGCombine
transforms which would make the overall code worse.

I'm not really happy adding a bunch of code to handle this, but it would
probably be tricky to substantially improve the behavior of DAGCombine
here.

Differential Revision: https://reviews.llvm.org/D56032

llvm-svn: 351776

6 years ago[CVP] Use LVI to constant fold deopt operands
Philip Reames [Tue, 22 Jan 2019 01:34:33 +0000 (01:34 +0000)]
[CVP] Use LVI to constant fold deopt operands

Deopt operands are generally intended to record information about a site in code with minimal perturbation of the surrounding code. Idiomatically, they also tend to appear down rare paths. Putting these together, we have an obvious case for extending CVP w/deopt operand constant folding. Arguably, we should be doing this for all operands on all instructions, but that's definitely a much larger and risky change.

Differential Revision: https://reviews.llvm.org/D55678

llvm-svn: 351774

6 years agoUpdated issue 3144
Marshall Clow [Tue, 22 Jan 2019 01:05:58 +0000 (01:05 +0000)]
Updated issue 3144

llvm-svn: 351773

6 years ago[LangRef] Clarify semantics of volatile operations.
Eli Friedman [Tue, 22 Jan 2019 00:42:20 +0000 (00:42 +0000)]
[LangRef] Clarify semantics of volatile operations.

Specifically, clarify the following:

1. Volatile load and store may access addresses that are not memory.
2. Volatile load and store do not modify arbitrary memory.
3. Volatile load and store do not trap.

Prompted by recent volatile discussion on llvmdev.

Currently, there's sort of a split in the source code about whether
volatile operations are allowed to trap; this resolves that dispute in
favor of not allowing them to trap.

Differential Revision: https://reviews.llvm.org/D53184

llvm-svn: 351772

6 years ago[safestack] Fix NetBSD build
Vitaly Buka [Tue, 22 Jan 2019 00:39:59 +0000 (00:39 +0000)]
[safestack] Fix NetBSD build

llvm-svn: 351771

6 years agoUpdate with issues to be moved in San Diego
Marshall Clow [Tue, 22 Jan 2019 00:31:09 +0000 (00:31 +0000)]
Update with issues to be moved in San Diego

llvm-svn: 351770

6 years agoGlobalISel: Fix out of bounds crashes in verifier
Matt Arsenault [Tue, 22 Jan 2019 00:29:37 +0000 (00:29 +0000)]
GlobalISel: Fix out of bounds crashes in verifier

llvm-svn: 351769

6 years ago[AArch64] Add patterns for zext/sext of shift amount.
Eli Friedman [Tue, 22 Jan 2019 00:21:35 +0000 (00:21 +0000)]
[AArch64] Add patterns for zext/sext of shift amount.

Not sure this is the best fix, but it saves an instruction for certain
constructs involving variable shifts.

Differential Revision: https://reviews.llvm.org/D55572

llvm-svn: 351768

6 years agoAMDGPU/GlobalISel: Legalize more fp<->int conversions
Matt Arsenault [Tue, 22 Jan 2019 00:20:17 +0000 (00:20 +0000)]
AMDGPU/GlobalISel: Legalize more fp<->int conversions

llvm-svn: 351767

6 years ago[CodeGen] Always use string computed in Sema for PredefinedExpr
Eli Friedman [Tue, 22 Jan 2019 00:11:17 +0000 (00:11 +0000)]
[CodeGen] Always use string computed in Sema for PredefinedExpr

We can't use any other string, anyway, because its type wouldn't
match the type of the PredefinedExpr.

With this change, we don't compute a "nice" name for the __func__ global
when it's used in the initializer for a constant. This doesn't seem like
a great loss, and I'm not sure how to fix it without either storing more
information in the AST, or somehow threading through the information
from ExprConstant.cpp.

This could break some situations involving BlockDecl; currently,
CodeGenFunction::EmitPredefinedLValue has some logic to intentionally
emit a string different from what Sema computed.  This code skips that
logic... but that logic can't work correctly in general anyway.  (For
example, sizeof(__func__) returns the wrong result.) Hopefully this
doesn't affect practical code.

Fixes https://bugs.llvm.org/show_bug.cgi?id=40313 .

Differential Revision: https://reviews.llvm.org/D56821

llvm-svn: 351766

6 years agoDocument toolchain update policy
JF Bastien [Mon, 21 Jan 2019 23:53:52 +0000 (23:53 +0000)]
Document toolchain update policy

Summary:
Capture the current agreed-upon toolchain update policy based on the following
discussions:

  - LLVM dev meeting 2018 BoF "Migrating to C++14, and beyond!"
    llvm.org/devmtg/2018-10/talk-abstracts.html#bof3
  - A Short Policy Proposal Regarding Host Compilers
    lists.llvm.org/pipermail/llvm-dev/2018-May/123238.html
  - Using C++14 code in LLVM (2018)
    lists.llvm.org/pipermail/llvm-dev/2018-May/123182.html
  - Using C++14 code in LLVM (2017)
    lists.llvm.org/pipermail/llvm-dev/2017-October/118673.html
  - Using C++14 code in LLVM (2016)
    lists.llvm.org/pipermail/llvm-dev/2016-October/105483.html
  - Document and Enforce new Host Compiler Policy
    llvm.org/D47073
  - Require GCC 5.1 and LLVM 3.5 at a minimum
    llvm.org/D46723

Subscribers: jkorous, dexonsmith, llvm-commits

Differential Revision: https://reviews.llvm.org/D56819

llvm-svn: 351765

6 years ago[x86] add another test for xor with undefs; NFC
Sanjay Patel [Mon, 21 Jan 2019 22:12:35 +0000 (22:12 +0000)]
[x86] add another test for xor with undefs; NFC

llvm-svn: 351764

6 years ago[x86] add tests for vector ops with undef lanes; NFC
Sanjay Patel [Mon, 21 Jan 2019 21:52:27 +0000 (21:52 +0000)]
[x86] add tests for vector ops with undef lanes; NFC

llvm-svn: 351763

6 years ago[X86] Use X86ISD::VFPROUND instead of ISD::FP_ROUND for 256 and 512 bit cvtpd2ps...
Craig Topper [Mon, 21 Jan 2019 20:14:09 +0000 (20:14 +0000)]
[X86] Use X86ISD::VFPROUND instead of ISD::FP_ROUND for 256 and 512 bit cvtpd2ps intrinsics.

Summary:
Use X86ISD::VFPROUND in the instruction isel patterns. Add new patterns for ISD::FP_ROUND to maintain support for fptrunc in IR.

In the process I found a couple duplicate isel patterns which I also deleted in this patch.

Reviewers: RKSimon, spatel

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D56991

llvm-svn: 351762

6 years ago[X86] Change avx512 COMPRESS and EXPAND lowering to use a single masked node instead...
Craig Topper [Mon, 21 Jan 2019 20:02:28 +0000 (20:02 +0000)]
[X86] Change avx512 COMPRESS and EXPAND lowering to use a single masked node instead of expand/compress+select.

Summary:
For compress, a select node doesn't semantically reflect the behavior of the instruction. The mask would have holes in it, but the resulting write is to contiguous elements at the bottom of the vector.

Furthermore, as far as the compressing and expanding is concerned the behavior is depended on the mask. You can't just have an expand/compress node that only reads the input vector. That node would have no meaning by itself.

This all only works because we pattern match the compress/expand+select back to the instruction. But conceivably an optimization of the select could break the pattern and leave something meaningless.

This patch modifies the expand and compress node to take the mask and passthru as additional inputs and gets rid of the select all together.

Reviewers: RKSimon, spatel

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D57002

llvm-svn: 351761

6 years agoFix test after AST dump output change
Stephen Kelly [Mon, 21 Jan 2019 19:18:10 +0000 (19:18 +0000)]
Fix test after AST dump output change

llvm-svn: 351760

6 years ago[AMDGPU] Fixed hazard recognizer to walk predecessors
Stanislav Mekhanoshin [Mon, 21 Jan 2019 19:11:26 +0000 (19:11 +0000)]
[AMDGPU] Fixed hazard recognizer to walk predecessors

Fixes two problems with GCNHazardRecognizer:
1. It only scans up to 5 instructions emitted earlier.
2. It does not take control flow into account. An earlier instruction
from the previous basic block is not necessarily a predecessor.
At the same time a real predecessor block is not scanned.

The patch provides a way to distinguish between scheduler and
hazard recognizer mode. It is OK to work with emitted instructions
in the scheduler because we do not really know what will be emitted
later and its order. However, when pass works as a hazard recognizer
the schedule is already finalized, and we have full access to the
instructions for the whole function, so we can properly traverse
predecessors and their instructions.

Differential Revision: https://reviews.llvm.org/D56923

llvm-svn: 351759

6 years agogn build: Stop passing -DLLVM_LIBXML2_ENABLED to some targets
Nico Weber [Mon, 21 Jan 2019 18:59:11 +0000 (18:59 +0000)]
gn build: Stop passing -DLLVM_LIBXML2_ENABLED to some targets

This is a remnant from before the gn build had a working config.h.

Defining LLVM_LIBXML2_ENABLED only for targets that depend on build/libs/xml is
nice in that only some of the codebase needs to be rebuilt when
llvm_enable_libxml2 changes -- but config.h already defines it and defining it
there and then redundantly a second time for some targets is worse than having
it just in config.h.

No behavior change.

Differential Revision: https://reviews.llvm.org/D56908

llvm-svn: 351758

6 years agogn build: Merge r351627, r351548, r351701
Nico Weber [Mon, 21 Jan 2019 18:56:39 +0000 (18:56 +0000)]
gn build: Merge r351627, r351548, r351701

llvm-svn: 351757

6 years agoFix compilation error with gcc 4.8
Pavel Labath [Mon, 21 Jan 2019 18:21:03 +0000 (18:21 +0000)]
Fix compilation error with gcc 4.8

This version of gcc seems to be having issues with raw literals inside macro
arguments. I change the string to use regular string literals instead.

llvm-svn: 351756

6 years ago[X86][BtVer2] Update latency of mmx horizontal operations
Simon Pilgrim [Mon, 21 Jan 2019 18:04:25 +0000 (18:04 +0000)]
[X86][BtVer2] Update latency of mmx horizontal operations

D56777 added +1cy local forwarding penalty for horizontal operations, but this penalty only affects sse2/xmm variants, the mmx variants don't suffer the penalty.

Confirmed with @andreadb

llvm-svn: 351755

6 years ago[AArch64] add more tests for buildvec to shuffle transform; NFC
Sanjay Patel [Mon, 21 Jan 2019 17:46:35 +0000 (17:46 +0000)]
[AArch64] add more tests for buildvec to shuffle transform; NFC

These are copied from the sibling x86 file. I'm not sure which
of the current outputs (if any) is considered optimal, but
someone more familiar with AArch may want to take a look.

llvm-svn: 351754

6 years ago[DAGCombiner] fix crash when converting build vector to shuffle
Sanjay Patel [Mon, 21 Jan 2019 17:30:14 +0000 (17:30 +0000)]
[DAGCombiner] fix crash when converting build vector to shuffle

The regression test is reduced from the example shown in D56281.
This does raise a question as noted in the test file: do we want
to handle this pattern? I don't have a motivating example for
that on x86 yet, but it seems like we could have that pattern
there too, so we could avoid the back-and-forth using a shuffle.

llvm-svn: 351753

6 years ago[test] Pass -ccc-install-dir in mac compilation db test
Michal Gorny [Mon, 21 Jan 2019 17:05:43 +0000 (17:05 +0000)]
[test] Pass -ccc-install-dir in mac compilation db test

Pass -ccc-install-dir explicitly as the compilation database code does
not pass argv[0] to getMainExecutable(), while some systems require it
to return the correct path.  Since the relevant code is apparently only
applicable to Darwin, just pass correct -ccc-install-dir to make
the tests pass on *BSD systems.

Differential Revision: https://reviews.llvm.org/D56976

llvm-svn: 351752