platform/upstream/llvm.git
4 years agoUse diff -b on zlib tests so they pass on Windows
Amy Huang [Wed, 4 Dec 2019 22:47:05 +0000 (14:47 -0800)]
Use diff -b on zlib tests so they pass on Windows

Reviewers: hubert.reinterpretcast

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71036

4 years ago[clangd] Highlighting dependent types in more contexts
Nathan Ridge [Thu, 5 Dec 2019 19:28:56 +0000 (14:28 -0500)]
[clangd] Highlighting dependent types in more contexts

Fixes https://github.com/clangd/clangd/issues/214.

Reviewers: ilya-biryukov

Reviewed By: ilya-biryukov

Subscribers: MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D70746

4 years ago[clangd] Find reference to template parameter in 'sizeof...' expression
Nathan Ridge [Thu, 5 Dec 2019 19:27:23 +0000 (14:27 -0500)]
[clangd] Find reference to template parameter in 'sizeof...' expression

Fixes https://github.com/clangd/clangd/issues/213.

Reviewers: ilya-biryukov

Reviewed By: ilya-biryukov

Subscribers: MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D70740

4 years agoRevert "[DSE] Fix for a dangling point bug in DeadStoreElimination."
Florian Hahn [Thu, 5 Dec 2019 19:29:21 +0000 (19:29 +0000)]
Revert "[DSE] Fix for a dangling point bug in DeadStoreElimination."

The commit causes a failure:
http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-win/builds/20911

This reverts commit 1847fd9d85506ecee692230cb2500e3774ec628e.

4 years agoLowerDbgDeclare: look through bitcasts.
Evgenii Stepanov [Mon, 25 Nov 2019 23:45:17 +0000 (15:45 -0800)]
LowerDbgDeclare: look through bitcasts.

Summary:
Emit a value debug intrinsic (with OP_deref) when an alloca address is
passed to a function call after going through a bitcast.

This generates an FP or SP-relative location for the local variable in
the following case:
  int x;
  use((void *)&x;

Reviewers: aprantl, vsk, pcc

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70752

4 years agoscudo: Fix one of the C wrapper tests on Android.
Peter Collingbourne [Thu, 5 Dec 2019 18:41:37 +0000 (10:41 -0800)]
scudo: Fix one of the C wrapper tests on Android.

The test ScudoWrappersCTest.Realloc expects realloc of memalign to work on
Android, but this relies on dealloc_type_mismatch being set to false. Commit
0d3d4d3b0 caused us to start setting it to true in the C wrapper tests,
which broke the test. Set it to the correct value on Android.

Differential Revision: https://reviews.llvm.org/D71078

4 years agoscudo: Fix the build of wrappers_c_test.cpp on Android.
Peter Collingbourne [Tue, 19 Nov 2019 23:22:09 +0000 (15:22 -0800)]
scudo: Fix the build of wrappers_c_test.cpp on Android.

The Android headers don't provide a declaration of valloc or pvalloc, so we
need to declare them ourselves.

Differential Revision: https://reviews.llvm.org/D71077

4 years ago[GlobalISel] Localizer: Allow targets not to run the pass conditionally
Volkan Keles [Thu, 5 Dec 2019 19:09:50 +0000 (11:09 -0800)]
[GlobalISel] Localizer: Allow targets not to run the pass conditionally

Summary:
Previously, it was not possible to skip running the localizer pass
conditionally. This patch adds an input function to the pass which
decides if the pass should run on the given MachineFunction or not.

No test case as there is no upstream target needs this functionality.

Reviewers: qcolombet

Reviewed By: qcolombet

Subscribers: rovka, hiraditya, Petar.Avramovic, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71038

4 years ago[X86] Remove ProcIntelGLM/ProcIntelGLP/ProcIntelTRM and replace them with a single...
Craig Topper [Thu, 5 Dec 2019 18:24:10 +0000 (10:24 -0800)]
[X86] Remove ProcIntelGLM/ProcIntelGLP/ProcIntelTRM and replace them with a single feature flag covers the two places they were used.

Differential Revision: https://reviews.llvm.org/D71048

4 years agoRevert "[InstCombine] keep assumption before sinking calls"
Bob Haarman [Tue, 3 Dec 2019 01:23:54 +0000 (17:23 -0800)]
Revert "[InstCombine] keep assumption before sinking calls"

Summary:
This reverts commit c3b06d0c393e533eab712922911d14e5a079fa5d.

Reason for revert: Caused miscompiles when inserting assume for undef.

Also adds a test to prevent similar breakage in future.

Fixes PR44154.

Reviewers: rnk, jdoerfert, efriedma, xbolva00

Reviewed By: rnk

Subscribers: thakis, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70933

4 years ago[OPENMP50]Add support for if clause for simd part in distribute parallel for simd...
Alexey Bataev [Thu, 5 Dec 2019 18:22:15 +0000 (13:22 -0500)]
[OPENMP50]Add support for if clause for simd part in distribute parallel for simd directive.

According to OpenMP 5.0, the if clause can be applied to simd
subdirective in the combined directives.

4 years ago[InstCombine] Invert `add A, sext(B) --> sub A, zext(B)` canonicalization (to `sub...
Roman Lebedev [Thu, 5 Dec 2019 17:44:22 +0000 (20:44 +0300)]
[InstCombine] Invert `add A, sext(B) --> sub A, zext(B)` canonicalization (to `sub A, zext B -> add A, sext B`)

Summary:
D68408 proposes to greatly improve our negation sinking abilities.
But in current canonicalization, we produce `sub A, zext(B)`,
which we will consider non-canonical and try to sink that negation,
undoing the existing canonicalization.
So unless we explicitly stop producing previous canonicalization,
we will have two conflicting folds, and will end up endlessly looping.

This inverts canonicalization, and adds back the obvious fold
that we'd miss:
* `sub [nsw] Op0, sext/zext (bool Y) -> add [nsw] Op0, zext/sext (bool Y)`
  https://rise4fun.com/Alive/xx4
* `sext(bool) + C -> bool ? C - 1 : C`
  https://rise4fun.com/Alive/fBl

It is obvious that `@ossfuzz_9880()` / `@lshr_out_of_range()`/`@ashr_out_of_range()`
(oss-fuzz 4871) are no longer folded as much, though those aren't really worrying.

Reviewers: spatel, efriedma, t.p.northover, hfinkel

Reviewed By: spatel

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71064

4 years ago[ELF][PPC64] Support long branch thunks with addends
Fangrui Song [Mon, 2 Dec 2019 22:20:42 +0000 (14:20 -0800)]
[ELF][PPC64] Support long branch thunks with addends

Fixes PPC64 part of PR40438

  // clang -target ppc64le -c a.cc
  // .text.unlikely may be placed in a separate output section (via -z keep-text-section-prefix)
  // The distance between bar in .text.unlikely and foo in .text may be larger than 32MiB.
  static void foo() {}
  __attribute__((section(".text.unlikely"))) static int bar() { foo(); return 0; }
  __attribute__((used)) static int dummy = bar();

This patch makes such thunks with addends work for PPC64.

AArch64: .text -> `__AArch64ADRPThunk_ (adrp x16, ...; add x16, x16, ...; br x16)` -> target
PPC64: .text -> `__long_branch_ (addis 12, 2, ...; ld 12, ...(12); mtctr 12; bctr)` -> target

AArch64 can leverage ADRP to jump to the target directly, but PPC64
needs to load an address from .branch_lt . Before Power ISA v3.0, the
PC-relative ADDPCIS was not available. .branch_lt was invented to work
around the limitation.

Symbol::ppc64BranchltIndex is replaced by
PPC64LongBranchTargetSection::entry_index which take addends into
consideration.

The tests are rewritten: ppc64-long-branch.s tests -no-pie and
ppc64-long-branch-pi.s tests -pie and -shared.

Reviewed By: sfertile

Differential Revision: https://reviews.llvm.org/D70937

4 years ago[AArch64] Fix MUL/SUB fusing
Sanne Wouda [Thu, 5 Dec 2019 18:09:08 +0000 (18:09 +0000)]
[AArch64] Fix MUL/SUB fusing

Summary:
When MUL is the first operand to SUB, we can't use MLS because the accumulator
should be negated.  Emit a NEG of the accumulator and an MLA instead, similar to
what we do for FMUL / FSUB fusing.

Reviewers: dmgreen, SjoerdMeijer, fhahn, Gerolf, mstorsjo, asbirlea

Reviewed By: asbirlea

Subscribers: kristof.beyls, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71067

4 years ago[LLDB] Replacing use of ul suffix in GetMaxU64Bitfield since it not guarenteed to...
shafik [Thu, 5 Dec 2019 17:28:08 +0000 (09:28 -0800)]
[LLDB] Replacing use of ul suffix in GetMaxU64Bitfield since it not guarenteed to be 64 bit

GetMaxU64Bitfield(...) uses the ul suffix but we require a 64 bit unsigned integer and ul could be 32 bit. So this replacing it with a explicit cast and refactors the code around it to use an early exit.

Differential Revision: https://reviews.llvm.org/D70992

4 years ago[llvm] llvm-ifs: Support for handling empty IFS and merging weak+strong symbols.
Puyan Lotfi [Thu, 28 Nov 2019 21:43:56 +0000 (16:43 -0500)]
[llvm] llvm-ifs: Support for handling empty IFS and merging weak+strong symbols.

The following changes enable llvm-ifs to handle the following merge conflicts:

* Weak + Strong symbol merging for the same symbol
* empty vs non-empty triple field
* empty vs non-empty object file format

Differential Revision: https://reviews.llvm.org/D70834

4 years ago[DSE] Fix for a dangling point bug in DeadStoreElimination.
Ankit [Thu, 5 Dec 2019 17:43:04 +0000 (17:43 +0000)]
[DSE] Fix for a dangling point bug in DeadStoreElimination.

The patch makes sure that the LastThrowing pointer does not point to any instruction deleted by call to DeleteDeadInstruction.

While iterating through the instructions the pass maintains a pointer to the lastThrowing Instruction. A call to deleteDeadInstruction deletes a dead store and other instructions feeding the original dead instruction which also become dead. The instruction pointed by the lastThrowing pointer could also be deleted by the call to DeleteDeadInstruction and thus it becomes a dangling pointer. Because of this, we see an error in the next iteration.

In the patch, we maintain a list of throwing instructions encountered previously and use the last non deleted throwing instruction from the container.

Patch by Ankit <quic_aankit@quicinc.com>

Reviewers: fhahn, bcahoon, efriedma

Reviewed By: fhahn

Differential Revision: https://reviews.llvm.org/D65326

4 years ago[ELF][PPC64] Don't copy ppc64BranchltIndex in replaceWithDefined
Fangrui Song [Mon, 2 Dec 2019 23:31:11 +0000 (15:31 -0800)]
[ELF][PPC64] Don't copy ppc64BranchltIndex in replaceWithDefined

replaceWithDefined is used by canonical PLT and copy relocations, which
imply that the symbol is preemptable. ppc64BranchltIndex is only used by
non-preemptable cases, and it can only be the default value in
replaceWithDefined.

4 years ago[scan-build-py] Set of small fixes
Gabor Horvath [Thu, 5 Dec 2019 17:00:16 +0000 (09:00 -0800)]
[scan-build-py] Set of small fixes

* Unhandled exceptions
* Typos

Differential Revision: https://reviews.llvm.org/D70693

4 years ago[ASTImporter] Various source location and range import fixes.
Balázs Kéri [Thu, 5 Dec 2019 15:21:21 +0000 (16:21 +0100)]
[ASTImporter] Various source location and range import fixes.

Summary:
ASTImporter contained wrong or missing imports of SourceLocation
and SourceRange for some objects. At least a part of such errors
is fixed now.
Source location import fixes in namespace, enum, record,
class template specialization declarations and DeclRefExpr,
UnresolvedLookupExpr, UnresolvedMemberExpr, NestedNameSpecifierLoc.

Reviewers: martong, a.sidorin, shafik

Reviewed By: shafik

Subscribers: rnkovacs, dkrupp, Szelethus, gamesh411, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D60499

4 years ago[OPENMP50]Add support for if clause for simd part in parallel master taskloop simd...
Alexey Bataev [Thu, 5 Dec 2019 16:31:45 +0000 (11:31 -0500)]
[OPENMP50]Add support for if clause for simd part in parallel master taskloop simd directive.

According to OpenMP 5.0, the if clause can be applied to simd
subdirective in the combined directives.

4 years ago[AMDGPU][HIP] Improve opt-level handling
Scott Linder [Tue, 3 Dec 2019 22:18:21 +0000 (17:18 -0500)]
[AMDGPU][HIP] Improve opt-level handling

Summary:
The HIP toolchain invokes `llc` without an explicit opt-level, meaning
it always uses the default (-O2). This makes it impossible to use -O1,
for example. The HIP toolchain also coerces -Os/-Oz to -O2 even when
invoking opt, and it coerces -Og to -O2 rather than -O1.

Forward the opt-level to `llc` as well as `opt`, and only coerce levels
where it is required.

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D70987

4 years ago[InstCombine] narrow select with FP casts
Sanjay Patel [Thu, 5 Dec 2019 16:12:44 +0000 (11:12 -0500)]
[InstCombine] narrow select with FP casts

Select doesn't change values, so truncate of extended operand cancels out.

4 years ago[InstCombine] add FMF guard to builder in fptrunc transform; NFC
Sanjay Patel [Thu, 5 Dec 2019 15:50:43 +0000 (10:50 -0500)]
[InstCombine] add FMF guard to builder in fptrunc transform; NFC

This makes no difference currently because we don't apply FMF
to FP casts, but that may change.

This could also be a place to add a fold for select with fptrunc,
so it will make that patch easier/smaller.

4 years ago[DebugInfo] Don't create multiple DBG_VALUEs when sinking
Jeremy Morse [Thu, 5 Dec 2019 15:12:25 +0000 (15:12 +0000)]
[DebugInfo] Don't create multiple DBG_VALUEs when sinking

This patch addresses a performance problem reported in PR43855, and
present in the reapplication in in 001574938e5. It turns out that
MachineSink will (often) move instructions to the first block that
post-dominates the current block, and then try to sink further. This
means if we have a lot of conditionals, we can needlessly create large
numbers of DBG_VALUEs, one in each block the sunk instruction passes
through.

To fix this, rather than immediately sinking DBG_VALUEs, record them in
a pass structure. When sinking is complete and instructions won't be
sunk any further, new DBG_VALUEs are added, avoiding lots of
intermediate DBG_VALUE $noregs being created.

Differential revision: https://reviews.llvm.org/D70676

4 years ago[DebugInfo] Don't reorder DBG_VALUEs when sunk
Jeremy Morse [Thu, 5 Dec 2019 15:03:33 +0000 (15:03 +0000)]
[DebugInfo] Don't reorder DBG_VALUEs when sunk

Fix part of PR43855, resolving a problem that comes from the reapplication
in 001574938e5. If we have two DBG_VALUE insts in a block that specify
the location of the same variable, for example:

   %0 = someinst
   DBG_VALUE %0, !123, !DIExpression()
   %1 = anotherinst
   DBG_VALUE %1, !123, !DIExpression()

if %0 were to sink, the corresponding DBG_VALUE would sink too, past the
next DBG_VALUE, effectively re-ordering assignments. To fix this, I've
added a SeenDbgVars set recording what variable locations have been seen in
a block already (working bottom up), and now flag DBG_VALUEs that would
pass a later DBG_VALUE for the same variable.

NB, this only works for repeated DBG_VALUEs in the same basic block, the
general case involving control flow is much harder, which I've written
up in PR44117.

Differential revision: https://reviews.llvm.org/D70672

4 years ago[DebugInfo] Re-apply two patches to MachineSink
Jeremy Morse [Thu, 5 Dec 2019 14:46:11 +0000 (14:46 +0000)]
[DebugInfo] Re-apply two patches to MachineSink

These were:
 * D58386 / f5e1b718a67 / reverted in d382a8a768b
 * D58238 / ee50590e168 / reverted in a8db456b53a

Of which the latter has a performance regression tracked in PR43855,
fixed by D70672 / D70676, which will be committed atomically with this
reapplication.

Contains a minor difference to account for a change in the IsCopyInstr
signature.

4 years ago[InstCombine] add tests for fpext+select+fptrunc; NFC
Sanjay Patel [Thu, 5 Dec 2019 14:55:56 +0000 (09:55 -0500)]
[InstCombine] add tests for fpext+select+fptrunc; NFC

4 years ago[OPENMP][DOCS]Update list of the supported features, NFC.
Alexey Bataev [Thu, 5 Dec 2019 15:44:07 +0000 (10:44 -0500)]
[OPENMP][DOCS]Update list of the supported features, NFC.

4 years ago[Syntax] Show input code on tests failures. NFC
Ilya Biryukov [Thu, 5 Dec 2019 15:12:28 +0000 (16:12 +0100)]
[Syntax] Show input code on tests failures. NFC

4 years ago[OPENMP50]Add support for if clause for simd part in master taskloop simd directive.
Alexey Bataev [Thu, 5 Dec 2019 14:50:18 +0000 (09:50 -0500)]
[OPENMP50]Add support for if clause for simd part in master taskloop simd directive.

According to OpenMP 5.0, the if clause can be applied to simd
subdirective in the combined directives.

4 years ago[AArch64][SVE] Integer reduction instructions pattern/intrinsics.
Danilo Carvalho Grael [Thu, 5 Dec 2019 13:07:02 +0000 (08:07 -0500)]
[AArch64][SVE] Integer reduction instructions pattern/intrinsics.

Added pattern matching/intrinsics for the following SVE instructions:

-- saddv, uaddv
-- smaxv, sminv, umaxv, uminv
-- orv, eorv, andv

4 years ago[clangd] Simplify the code, NFC.
Haojian Wu [Thu, 5 Dec 2019 14:30:04 +0000 (15:30 +0100)]
[clangd] Simplify the code, NFC.

AST.getASTContext().getSourceManager() => AST.getSourceManager().

4 years agogn build: Merge 7f93cb62280
LLVM GN Syncbot [Thu, 5 Dec 2019 14:23:58 +0000 (14:23 +0000)]
gn build: Merge 7f93cb62280

4 years ago[ARM] Enable TypePromotion by default
Sam Parker [Thu, 5 Dec 2019 14:19:44 +0000 (14:19 +0000)]
[ARM] Enable TypePromotion by default

ARMCodeGenPrepare has already been generalized and renamed to
TypePromotion. We've had it enabled and tested downstream for a
while, so enable it by default.

Differential Revision: https://reviews.llvm.org/D70998

4 years ago[AST] Traverse the class type loc inside the member type loc.
Haojian Wu [Fri, 29 Nov 2019 12:05:42 +0000 (13:05 +0100)]
[AST] Traverse the class type loc inside the member type loc.

Summary:
We are missing this currently.

This would fix https://github.com/clangd/clangd/issues/216.

Reviewers: ilya-biryukov

Subscribers: mgorny, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D70849

4 years ago[lldb][NFC] Move Address and AddressRange functions out of Stream and let them take...
Raphael Isemann [Thu, 5 Dec 2019 13:41:09 +0000 (14:41 +0100)]
[lldb][NFC] Move Address and AddressRange functions out of Stream and let them take raw_ostream

Summary:
Yet another step on the long road towards getting rid of lldb's Stream class.

We probably should just make this some kind of member of Address/AddressRange, but it seems quite often we just push
in random integers in there and this is just about getting rid of Stream and not improving arbitrary APIs.

I had to rename another `DumpAddress` function in FormatEntity that is dumping the content of an address to make Clang happy.

Reviewers: labath

Reviewed By: labath

Subscribers: JDevlieghere, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D71052

4 years agoAlso check /Fo when deciding on the .gcna / .gcda filename (PR44208)
Hans Wennborg [Wed, 4 Dec 2019 14:19:21 +0000 (15:19 +0100)]
Also check /Fo when deciding on the .gcna / .gcda filename (PR44208)

Differential revision: https://reviews.llvm.org/D71012

4 years ago[InstCombine] Extend `0 - (X sdiv C) -> (X sdiv -C)` fold to non-splat vectors
Roman Lebedev [Thu, 5 Dec 2019 12:10:25 +0000 (15:10 +0300)]
[InstCombine] Extend `0 - (X sdiv C) -> (X sdiv -C)` fold to non-splat vectors

Split off from https://reviews.llvm.org/D68408

4 years ago[AST] Enable expression of OpenCL language address spaces an attribute
Victor Lomuller [Tue, 25 Jun 2019 12:57:48 +0000 (13:57 +0100)]
[AST] Enable expression of OpenCL language address spaces an attribute

Summary:
Enable a way to set OpenCL language address space using attributes
in addition to existing keywords.

Signed-off-by: Victor Lomuller victor@codeplay.com
Reviewers: aaron.ballman, Anastasia

Subscribers: yaxunl, ebevhan, cfe-commits, Naghasan

Tags: #clang

Differential Revision: https://reviews.llvm.org/D71005

Signed-off-by: Alexey Bader <alexey.bader@intel.com>
4 years ago[lldb/DWARF] Switch to llvm debug_rnglists parser
Pavel Labath [Wed, 4 Dec 2019 15:06:44 +0000 (16:06 +0100)]
[lldb/DWARF] Switch to llvm debug_rnglists parser

Summary:
Our rnglist support was working only for the trivial cases (one CU),
because we only ever parsed one contribution out of the debug_rnglists
section. This means we were never able to resolve range lists for the
second and subsequent units (DW_FORM_sec_offset references came out
blang, and DW_FORM_rnglistx references always used the ranges lists from
the first unit).

Since both llvm and lldb rnglist parsers are sufficiently
self-contained, and operate similarly, we can fix this problem by
switching to the llvm parser instead. Besides the changes which are due
to variations in the interface, the main thing is that now the range
list object is a member of the DWARFUnit, instead of the entire symbol
file. This ensures that each unit can get it's own private set of range
list indices, and is consistent with how llvm's DWARFUnit does it
(overall, I've tried to structure the code the same way as the llvm
version).

I've also added a test case for the two unit scenario.

Reviewers: JDevlieghere, aprantl, clayborg

Subscribers: dblaikie, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D71021

4 years ago Reapply af57dbf12e54 "Add support for options -frounding-math, ftrapping...
Melanie Blower [Wed, 4 Dec 2019 20:23:46 +0000 (12:23 -0800)]
    Reapply af57dbf12e54 "Add support for options -frounding-math, ftrapping-math, -ffp-model=, and -ffp-exception-behavior="

        Patch was reverted because https://bugs.llvm.org/show_bug.cgi?id=44048
        The original patch is modified to set the strictfp IR attribute
        explicitly in CodeGen instead of as a side effect of IRBuilder.
        In the 2nd attempt to reapply there was a windows lit test fail, the
        tests were fixed to use wildcard matching.

        Differential Revision: https://reviews.llvm.org/D62731

4 years ago[lldb/cpluspluslanguage] Add constructor substitutor
Pavel Labath [Tue, 26 Nov 2019 15:36:09 +0000 (16:36 +0100)]
[lldb/cpluspluslanguage] Add constructor substitutor

Summary:
This patch adds code which will substitute references to the full object
constructors/destructors with their base object versions.

Like all substitutions in this category, this operation is not really
sound, but doing this in a more precise way allows us to get rid of a
much larger hack -- matching function according to their demangled
names, which effectively does the same thing, but also much more.

This is a (very late) follow-up to D54074.

Background: clang has an optimization which can eliminate full object
structors completely, if they are found to be equivalent to their base
object versions. It does this because it assumes they can be regenerated
on demand in the compile unit that needs them (e.g., because they are
declared inline). However, this doesn't work for the debugging scenario,
where we don't have the structor bodies available -- we pretend all
constructors are defined out-of-line as far as clang is concerned. This
causes clang to emit references to the (nonexisting) full object
structors during expression evaluation.

Fun fact: This is not a problem on darwin, because the relevant
optimization is disabled to work around a linker bug.

Reviewers: teemperor, JDevlieghere

Subscribers: lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D70721

4 years ago[llvm/DWARF] Return section offset from DWARFUnit::get{Loc,Rng}listOffset
Pavel Labath [Wed, 4 Dec 2019 13:03:18 +0000 (14:03 +0100)]
[llvm/DWARF] Return section offset from DWARFUnit::get{Loc,Rng}listOffset

Summary:
Currently these function return the raw content of the appropriate table
header, which means they are relative to the DW_AT_{loc,rng}list_base,
and one has to relocate them in order to do anything.

This changes the functions to perform the relocation themselves, which
seems more clearer, particularly as they are sitting right next to the
find{Rng,Loc}listFromOffset functions, but one *cannot* simply take the
result of these functions and take pass them there.

The only effect of this patch is to change what value is dumped for the
DW_AT_ranges attribute, which I think is for the better, as previously
the values appeared to point into thin air.

(The main reason I am looking at this is because I was trying to
implement equivalent functionality in lldb's DWARFUnit, and was stumped
by this behavior.

Reviewers: dblaikie, JDevlieghere, aprantl

Subscribers: hiraditya, llvm-commits, SouraVX

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71006

4 years ago[LiveDebugValues] Silence the unused var warning; NFC
Djordje Todorovic [Thu, 5 Dec 2019 11:21:51 +0000 (12:21 +0100)]
[LiveDebugValues] Silence the unused var warning; NFC

4 years ago[clangd] More unittests for cross-file rename.
Haojian Wu [Thu, 5 Dec 2019 11:08:31 +0000 (12:08 +0100)]
[clangd] More unittests for cross-file rename.

Summary:
The previous unittests for cross-file rename was kind of weak. With this
patch, we should have more test coverage, and it is easy to add more tests in
the future.

Reviewers: ilya-biryukov, kbobyrev

Reviewed By: ilya-biryukov

Subscribers: merge_guards_bot, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D71050

4 years ago[DebugInfo] Handle call site values for instructions before call bundle
David Stenberg [Thu, 5 Dec 2019 09:49:41 +0000 (10:49 +0100)]
[DebugInfo] Handle call site values for instructions before call bundle

Summary:
If a call is bundled then the code that looks for instructions that
produce parameter values would break when reaching the call's bundle
header, due to the `ifCall(/*AnyInBundle*/)` invocation returning true.

It is not enough to simply ignore bundle headers in the `isCall()`
invocation, as the bundle header may have defines of parameter registers
due to the call, meaning that such registers would incorrectly be
removed from the worklist. Therefore, do not look at bundle headers at
all.

Reviewers: djtodoro, NikolaPrica, aprantl, vsk

Reviewed By: aprantl, vsk

Subscribers: hiraditya, llvm-commits

Tags: #debug-info, #llvm

Differential Revision: https://reviews.llvm.org/D71024

4 years agoFix the clang interpreter example which was broken by 4fc68b9b7f3e0.
Lang Hames [Thu, 5 Dec 2019 10:41:40 +0000 (02:41 -0800)]
Fix the clang interpreter example which was broken by 4fc68b9b7f3e0.

4 years ago[lldb] Don't put compile unit name into the support file list and support DWARF5...
Pavel Labath [Thu, 28 Nov 2019 09:19:23 +0000 (10:19 +0100)]
[lldb] Don't put compile unit name into the support file list and support DWARF5 line tables

Summary:
Lldb's "format-independent" debug info made use of the fact that DWARF
(<=4) did not use the file index zero, and reused the support file index
zero for storing the compile unit name.

While this provided some convenience for DWARF<=4, it meant that the PDB
plugin needed to artificially remap file indices in order to free up
index 0. Furthermore, DWARF v5 make file index 0 legal, which meant that
similar remapping would be needed in the dwarf plugin too.

What this patch does instead is remove the requirement of having the
compile unit name in the index 0. It is not that useful since the name
can always be fetched from the CompileUnit object. Remapping code in the
pdb plugin(s) has been removed or simplified.

DWARF plugin has started inserting an empty FileSpec at index 0 to
ensure the indices keep matching up (in case of DWARF<=4). For DWARF5,
we insert the file 0 from the line table.

I add a test to ensure we can correctly lookup line table entries
referencing file 0, and in particular the case where the file 0 is also
duplicated in another file entry, as this is how clang produces line
tables in some circumstances (see pr44170). Though this is probably a
bug in clang, this is not forbidden by DWARF, and lldb already has
support for that in some (but not all) cases -- this adds a test for the
code path which was not fixed in this patch.

Reviewers: clayborg, JDevlieghere, jdoerfert

Subscribers: aprantl, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D70954

4 years ago[AArch64][SVE] Implement element count intrinsics
Cullen Rhodes [Thu, 5 Dec 2019 09:56:00 +0000 (09:56 +0000)]
[AArch64][SVE] Implement element count intrinsics

Summary:
Adds intrinsics for the following:

    * cntb
    * cnth
    * cntw
    * cntd
    * cntp

Reviewers: sdesmalen, huntergr, dancgr, rengolin, efriedma, rovka

Reviewed By: efriedma

Subscribers: tschuett, kristof.beyls, hiraditya, rkruppe, psnobl, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70967

4 years ago[LLD][ELF][AArch64] .note.gnu.property sections should have alignment 8
Peter Smith [Fri, 29 Nov 2019 18:44:21 +0000 (18:44 +0000)]
[LLD][ELF][AArch64] .note.gnu.property sections should have alignment 8

The .note.gnu.property SHT_NOTE sections on AArch64 (a 64-bit target)
should have alignment 8 to more closely match the binutils implementation
where alignment is 4-bytes on 32-bit machines and 8-bytes on 64-bit
machines.

Previously LLD was using 4 for both 32-bit and 64-bit machines.

Differential Revision: https://reviews.llvm.org/D70962

4 years agoReland "[LiveDebugValues] Introduce entry values of unmodified params"
Djordje Todorovic [Tue, 3 Dec 2019 13:18:02 +0000 (14:18 +0100)]
Reland "[LiveDebugValues] Introduce entry values of unmodified params"

Relanding this after resolving the cause of the test failure.

4 years ago[Checkers] Added support for freopen to StreamChecker.
Balázs Kéri [Wed, 4 Dec 2019 16:15:03 +0000 (17:15 +0100)]
[Checkers] Added support for freopen to StreamChecker.

Summary: Extend StreamChecker with a new evaluation function for API call 'freopen'.

Reviewers: NoQ, baloghadamsoftware, Szelethus, martong

Reviewed By: baloghadamsoftware, martong

Subscribers: martong, rnkovacs, dkrupp, Szelethus, gamesh411, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D69948

4 years ago[LLD][ELF] Add support for PT_GNU_PROPERTY
Peter Smith [Mon, 2 Dec 2019 18:34:56 +0000 (18:34 +0000)]
[LLD][ELF] Add support for PT_GNU_PROPERTY

The PT_GNU_PROPERTY program header describes the location of the
.note.gnu.property SHT_NOTES section. The linux kernel uses this program
header to find the .note.gnu.property section rather than parsing.
Executables that have properties that the kernel needs to act on that don't
have the PT_GNU_PROPERTY program header will not boot.

Differential Revision: https://reviews.llvm.org/D70961

4 years ago[clangd] Fix an incorrect comment, NFC.
Haojian Wu [Thu, 5 Dec 2019 09:43:29 +0000 (10:43 +0100)]
[clangd] Fix an incorrect comment, NFC.

4 years ago[ORC] Remove the automagic Main JITDylib fram ExecutionSession.
Lang Hames [Thu, 5 Dec 2019 06:45:38 +0000 (22:45 -0800)]
[ORC] Remove the automagic Main JITDylib fram ExecutionSession.

This patch removes the magic "main" JITDylib from ExecutionEngine. The main
JITDylib was created automatically at ExecutionSession construction time, and
all subsequently created JITDylibs were added to the main JITDylib's
links-against list by default. This saves a couple of lines of boilerplate for
simple JIT setups, but this isn't worth introducing magical behavior for.

ORCv2 clients should now construct their own main JITDylib using
ExecutionSession::createJITDylib and set up its linkages manually using
JITDylib::setSearchOrder (or related methods in JITDylib).

4 years ago[MCRegInfo] Add forward sub and super register iterators. (NFC)
Florian Hahn [Thu, 5 Dec 2019 09:16:08 +0000 (09:16 +0000)]
[MCRegInfo] Add forward sub and super register iterators. (NFC)

This patch adds forward iterators mc_difflist_iterator,
mc_subreg_iterator and mc_superreg_iterator, based on the existing
DiffListIterator. Those are used to provide iterator ranges over
sub- and super-register from TRI, which are slightly more convenient
than the existing MCSubRegIterator/MCSuperRegIterator. Unfortunately,
it duplicates a bit of functionality, but the new iterators are a bit
more convenient (and can be used with various existing iterator
utilities)  and should probably replace the old iterators in the future.

This patch updates some existing users.

Reviewers: evandro, qcolombet, paquette, MatzeB, arsenm

Reviewed By: qcolombet

Differential Revision: https://reviews.llvm.org/D70565

4 years ago[MIBundle] Turn MachineOperandIteratorBase into a forward iterator.
Florian Hahn [Thu, 5 Dec 2019 08:52:24 +0000 (08:52 +0000)]
[MIBundle] Turn MachineOperandIteratorBase into a forward iterator.

This patch turns MachineOperandIteratorBase into a regular forward
iterator, which can be used with iterator_range.

It also adds mi_bundle_ops and const_mi_bundle_ops that return iterator
ranges over all operands in a bundle and updates a use of the old
iterator.

Reviewers: evandro, t.p.northover, paquette, MatzeB, arsenm

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D70561

4 years agoAvoid triple corruption while merging core info
Muhammad Omair Javaid [Thu, 5 Dec 2019 08:04:04 +0000 (13:04 +0500)]
Avoid triple corruption while merging core info

Summary:
This patch fixes a bug where when target triple created from elf information
is arm-*-linux-eabihf and platform triple is armv8l-*-linux-gnueabihf. Merging
both triple results in armv8l--unknown-unknown.

This happens because we order a triple update while calling CoreUpdated and
CoreUpdated creates a new triple with no vendor or environment information.

Making sure we do not update triple and just update to more specific core
fixes the issue.

Reviewers: labath, jasonmolenda, clayborg

Reviewed By: jasonmolenda

Subscribers: jankratochvil, kristof.beyls, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D70155

4 years ago[llvm-readelf/llvm-readobj] - Remove getSecTypeName() helper.
Georgii Rymar [Wed, 4 Dec 2019 15:29:01 +0000 (18:29 +0300)]
[llvm-readelf/llvm-readobj] - Remove getSecTypeName() helper.

We do not need it, we have
`object::getELFSectionTypeName` that can be used instead.

Differential revision: https://reviews.llvm.org/D71017

4 years agoFix the macro fusion table for X86 according to Intel optimization
Shengchen Kan [Thu, 5 Dec 2019 01:40:11 +0000 (09:40 +0800)]
Fix the macro fusion table for X86 according to Intel optimization
manual and add function isMacroFused

Differential Revision: https://reviews.llvm.org/D70999

4 years agoReland [MachineCopyPropagation] Extend MCP to do trivial copy backward propagation.
Kai Luo [Thu, 5 Dec 2019 06:01:00 +0000 (14:01 +0800)]
Reland [MachineCopyPropagation] Extend MCP to do trivial copy backward propagation.

Fix assertion error
```
bool llvm::MachineOperand::isRenamable() const: Assertion `Register::isPhysicalRegister(getReg()) && "isRenamable should only be checked on physical registers"' failed.
```
by checking if the register is 0 before invoking `isRenamable`.

4 years agoRevert "Properly convert all declaration non-type template arguments when"
David L. Jones [Thu, 5 Dec 2019 06:08:38 +0000 (22:08 -0800)]
Revert "Properly convert all declaration non-type template arguments when"

This reverts commit 11d10527852b4d3ed738aa90d8bec0f398160593.

This change is problematic with function pointer template parameters. For
example, building libcxxabi with futexes (-D_LIBCXXABI_USE_FUTEX) produces this
diagnostic:

    In file included from .../llvm-project/libcxxabi/src/cxa_guard.cpp:15:
    .../llvm-project/libcxxabi/src/cxa_guard_impl.h:416:54: error: address of function 'PlatformThreadID' will always evaluate to 'true' [-Werror,-Wpointer-bool-conversion]
        has_thread_id_support(this->thread_id_address && GetThreadIDArg),
                                                      ~~ ^~~~~~~~~~~~~~
    .../llvm-project/libcxxabi/src/cxa_guard.cpp:38:26: note: in instantiation of member function '__cxxabiv1::(anonymous namespace)::InitByteFutex<&__cxxabiv1::(anonymous namespace)::PlatformFutexWait, &__cxxabiv1::(anonymous namespace)::PlatformFutexWake, &__cxxabiv1::(anonymous namespace)::PlatformThreadID>::InitByteFutex' requested here
      SelectedImplementation imp(raw_guard_object);
                             ^
    .../llvm-project/libcxxabi/src/cxa_guard_impl.h:416:54: note: prefix with the address-of operator to silence this warning
        has_thread_id_support(this->thread_id_address && GetThreadIDArg),
                                                         ^
                                                         &
    1 error generated.

The diagnostic is incorrect: adding the address-of operator also fails ("cannot
take the address of an rvalue of type 'uint32_t (*)()' (aka 'unsigned int
(*)()')").

4 years agoRevert "[MachineCopyPropagation] Extend MCP to do trivial copy backward propagation"
Kai Luo [Thu, 5 Dec 2019 04:48:37 +0000 (12:48 +0800)]
Revert "[MachineCopyPropagation] Extend MCP to do trivial copy backward propagation"

This reverts commit 75b3a1c318ccad0f96c38689279bc5db63e2ad05, since it
breaks bootstrap build.

4 years agoAdd a default copy-assignment or copy-constructor for -Wdeprecated-copy warnings.
Eric Christopher [Thu, 5 Dec 2019 04:34:56 +0000 (20:34 -0800)]
Add a default copy-assignment or copy-constructor for -Wdeprecated-copy warnings.

4 years ago[AArch64][SVE] Add intrinsics and patterns for logical predicate instructions
Danilo Carvalho Grael [Thu, 5 Dec 2019 02:45:59 +0000 (21:45 -0500)]
[AArch64][SVE] Add intrinsics and patterns for logical predicate instructions

Add instrinics and patters for the following logical predicate instructions:
-- and, ands, bic, bics, eor, eors
-- sel
-- orr, orrs, orn, orns, nor, nors, nand, nads

4 years ago[Signal] Allow one-shot SIGPIPE handler to be reached
Vedant Kumar [Thu, 5 Dec 2019 03:21:14 +0000 (19:21 -0800)]
[Signal] Allow one-shot SIGPIPE handler to be reached

As SIGPIPE is no longer in the IntSigs array, handle SIGPIPE before
handling any interrupt signals.

Thanks to Alexandre Ganea for pointing out the issue here.

4 years ago[fix][unittests][llvm] Fix running unit tests without assertions. [NFCI]
Francesco Petrogalli [Thu, 5 Dec 2019 03:23:31 +0000 (03:23 +0000)]
[fix][unittests][llvm] Fix running unit tests without assertions. [NFCI]

4 years ago[clangd] Add xref for macro to static index.
Utkarsh Saxena [Wed, 20 Nov 2019 14:18:54 +0000 (15:18 +0100)]
[clangd] Add xref for macro to static index.

Summary:
This adds the references for macros to the SymbolCollector (used for static index).
Enabled if `CollectMacro` option is set.

Reviewers: hokein

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

Tags: #clang

Differential Revision: https://reviews.llvm.org/D70489

4 years ago[MachineCopyPropagation] Extend MCP to do trivial copy backward propagation
Kai Luo [Wed, 4 Dec 2019 03:19:50 +0000 (11:19 +0800)]
[MachineCopyPropagation] Extend MCP to do trivial copy backward propagation

Summary:
This patch mainly do such transformation
```
$R0 = OP ...
... // No read/clobber of $R0 and $R1
$R1 = COPY $R0 // $R0 is killed
```
Replace $R0 with $R1 and remove the COPY, we have
```
$R1 = OP ...
```
This transformation can also expose more opportunities for existing
copy elimination in MCP.

Differential Revision: https://reviews.llvm.org/D67794

4 years agoProperly convert all declaration non-type template arguments when
Richard Smith [Thu, 5 Dec 2019 02:53:34 +0000 (18:53 -0800)]
Properly convert all declaration non-type template arguments when
forming non-type template parameter values.

4 years ago[c++17] Fix assert / wrong code when passing a noexcept pointer to
Richard Smith [Thu, 5 Dec 2019 01:31:58 +0000 (17:31 -0800)]
[c++17] Fix assert / wrong code when passing a noexcept pointer to
member function to a non-noexcept pointer to member non-type template
parameter.

4 years ago[lldb/Reproducers] Don't instrument SBFileSpec::GetPath
Jonas Devlieghere [Thu, 5 Dec 2019 01:51:23 +0000 (17:51 -0800)]
[lldb/Reproducers] Don't instrument SBFileSpec::GetPath

This method uses a char* and length as output arguments and the
reproducer instrumentation doesn't know how to deal with that (yet).

4 years ago[lldb/Reproducers] Add missing instrumentation for SBFile (2/2)
Jonas Devlieghere [Thu, 5 Dec 2019 01:49:34 +0000 (17:49 -0800)]
[lldb/Reproducers] Add missing instrumentation for SBFile (2/2)

Found another issue while running TestDefaultConstructorForAPIObjects.

4 years ago[X86] Remove override of shouldUseStrictFP_TO_INT for fp80. NFC
Craig Topper [Thu, 5 Dec 2019 01:44:32 +0000 (17:44 -0800)]
[X86] Remove override of shouldUseStrictFP_TO_INT for fp80. NFC

I suspect this became unnecessary after r354161. Prior to that
we may have been going through the default expansion of FP_TO_UINT
on 64-bit targets and then ending up back in Custom X86 handling
to handle the FP_TO_SINT for it. Now we just Custom handle the
FP_TO_UINT directly. We already need to handle it for 32-bit mode
during type legalization so we wouldn't save any code by using
the default expansion on 64-bit.

4 years agoClear out the python class name in OptionParsingStarted for the OptionGroupPythonClas...
Jim Ingham [Thu, 5 Dec 2019 01:40:57 +0000 (17:40 -0800)]
Clear out the python class name in OptionParsingStarted for the OptionGroupPythonClassWithDict
options class.  This value was hanging around so for instance if you made a scripted breakpoint
resolver, then went to set another breakpoint, it would still think you had passed in a class
name and the breakpoint wouldn't do what you expected.

4 years ago[lldb/Reproducers] Add missing instrumentation for SBFile
Jonas Devlieghere [Thu, 5 Dec 2019 01:37:08 +0000 (17:37 -0800)]
[lldb/Reproducers] Add missing instrumentation for SBFile

This was properly captured by the instrumentation framework when running
TestRunCommandInterpreterAPI.py in capture-mode.

4 years agoReland [AArch64][MachineOutliner] Return address signing for outlined functions
David Tellenbach [Thu, 5 Dec 2019 01:20:59 +0000 (02:20 +0100)]
Reland [AArch64][MachineOutliner] Return address signing for outlined functions

Summary:
Reland after fixing an ASan failure by stopping outlining early if the
constraints for return address signing removed too many outlining candidates.

During AArch64 frame lowering instructions to enable return address
signing are inserted into functions if needed. Functions generated during
machine outlining don't run through target frame lowering and hence are
missing such instructions.

This patch introduces the following changes:

1. If not all functions that potentially participate in function outlining agree
   on their return address signing scope and their return address signing key,
   outlining is disabled for these functions.
2. If not all functions that potentially participate in function outlining agree
   on their support for v8.3A features, outlining is disabled for these
   functions.
3. If an outlining candidate would outline instructions that modify sp in a way
   that invalidates return address signing, outlining is disabled for that
   particular candidate.
4. If all candidate functions agree on the signing scope, signing key and their
   support for v8.3 features, the outlined function behaves as if it had the
   same scope and key attributes and as if it would provide the same v8.3A
   support as the original functions.

Reviewers: ostannard, paquette

Reviewed By: ostannard

Subscribers: kristof.beyls, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70635

4 years ago[GlobalISel] Fix compiler crash lowering G_LOAD in AArch64.
Amara Emerson [Thu, 5 Dec 2019 01:01:07 +0000 (17:01 -0800)]
[GlobalISel] Fix compiler crash lowering G_LOAD in AArch64.

Patch by Daniel Rodríguez Troitiño.

Differential Revision: https://reviews.llvm.org/D70794

4 years ago[clang][IFS] Ignoring -Xlinker/-Xclang arguments in InterfaceStubs pass for now.
Puyan Lotfi [Fri, 29 Nov 2019 06:38:25 +0000 (01:38 -0500)]
[clang][IFS] Ignoring -Xlinker/-Xclang arguments in InterfaceStubs pass for now.

Many of the inputs to clang driver toolchain tools can be things other
than files such as -Xclang and -Xlinker arguments. For now we don't take
such tool pass-through type arguments (although having an -Xifs would be nice to
have to replace things like -emit-merged-ifs). So because for the moment
we are not doing any sort of argument pass-through, I am going to have
InterfaceStubs Merger ignore any non-file type input arguments.

Differential Revision: https://reviews.llvm.org/D70838

4 years agogn build: Merge 27f12444575
LLVM GN Syncbot [Thu, 5 Dec 2019 00:59:20 +0000 (00:59 +0000)]
gn build: Merge 27f12444575

4 years ago[clangd] Add a tweak refactoring to wrap Objective-C string literals in `NSLocalized...
Alex Lorenz [Thu, 5 Dec 2019 00:58:12 +0000 (16:58 -0800)]
[clangd]  Add a tweak refactoring to wrap Objective-C string literals in `NSLocalizedString` macros

The commit adds a refactoring to Clangd that mimics the existing refactoring action in Xcode that wraps around an Objective-C string literal in an NSLocalizedString macro.

Differential Revision: https://reviews.llvm.org/D69543

4 years ago[clangd] NFC, add getLangOpts helper to ParsedAST
Alex Lorenz [Wed, 4 Dec 2019 23:09:35 +0000 (15:09 -0800)]
[clangd] NFC, add getLangOpts helper to ParsedAST

The addition of the helper is split out from https://reviews.llvm.org/D69543
as suggested by Kadir. I also updated the existing uses to use the new API.

4 years ago[lldb/Reproducers] Propagate LLDB_CAPTURE_REPRODUCER to the test suite
Jonas Devlieghere [Thu, 5 Dec 2019 00:43:19 +0000 (16:43 -0800)]
[lldb/Reproducers] Propagate LLDB_CAPTURE_REPRODUCER to the test suite

4 years ago[lldb/Reproducers] Override capture with LLDB_CAPTURE_REPRODUCER env var
Jonas Devlieghere [Thu, 5 Dec 2019 00:06:48 +0000 (16:06 -0800)]
[lldb/Reproducers] Override capture with LLDB_CAPTURE_REPRODUCER env var

Make it possible to override reproducer capture with the
LLDB_CAPTURE_REPRODUCER environment variable.

The goal of this change is twofold.

(1) I want to be able to enable capturing reproducers during regular
    test runs, both locally and on the bots. To do so I need a way to
    force capture. I cannot do this through the Python API, because
    reproducer capture must be enabled *before* the debugger
    initialized, which happens automatically when doing `import lldb`.

(2) I want to provide an escape hatch for when reproducers are enabled
    by default. Downstream we have reproducer capture enabled by default
    in the driver.

This patch solves both problems by overriding the reproducer mode based
on the environment variable. Acceptable values are 0/1 and ON/OFF.

4 years agoRevert "Reland [AArch64][MachineOutliner] Return address signing for outlined functions"
Sterling Augustine [Thu, 5 Dec 2019 00:27:26 +0000 (16:27 -0800)]
Revert "Reland [AArch64][MachineOutliner] Return address signing for outlined functions"

This reverts commit 02760b750b2ffcc0e2f5d78ecb137c80930c42c3.

The original commit is not asan clean.
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/37147/steps/check-llvm%20asan/logs/stdio

4 years ago[analyzer] Fix more ObjC accessor body farms after 2073dd2d.
Artem Dergachev [Thu, 5 Dec 2019 00:26:16 +0000 (16:26 -0800)]
[analyzer] Fix more ObjC accessor body farms after 2073dd2d.

Fix a crash when constructing a body farm for accessors of a property
that is declared and @synthesize'd in different (but related) interfaces
with the explicit ivar syntax.

This is a follow-up for 0b58b80e.

4 years ago[clang][IFS] Adding support for new clang interface stubs decl types.
Puyan Lotfi [Tue, 26 Nov 2019 18:02:18 +0000 (13:02 -0500)]
[clang][IFS] Adding support for new clang interface stubs decl types.

NamespaceAliasDecl UnresolvedUsingTypenameDecl CXXDeductionGuideDecl
ConstructorUsingShadowDecl

Differential Revision: https://reviews.llvm.org/D70728

4 years agoAdd parray example for lldb, vrs. *ptr@count gdb cmd.
Jason Molenda [Wed, 4 Dec 2019 23:43:38 +0000 (15:43 -0800)]
Add parray example for lldb, vrs. *ptr@count gdb cmd.

4 years ago[llvm] Fixing MIRVRegNamerUtils to properly handle 2+ MachineBasicBlocks.
Puyan Lotfi [Sun, 1 Dec 2019 00:00:10 +0000 (19:00 -0500)]
[llvm] Fixing MIRVRegNamerUtils to properly handle 2+ MachineBasicBlocks.

An interplay of code from D70210, along with code from the
Value-Numbering-esque hash-based namer from D70210, as well as some
crusty code from the original MIR-Canon code lead to multiple causes of
failure when canonicalizing or renaming vregs for MIR with multiple
basic blocks. This patch fixes those issues while deleting some no
longer needed code and adding a nice diamond test case to boot.

Differential Revision: https://reviews.llvm.org/D70478

4 years agoAdd help text for parray and poarray aliases.
Jason Molenda [Wed, 4 Dec 2019 23:33:54 +0000 (15:33 -0800)]
Add help text for parray and poarray aliases.

4 years agoUpstream debugserver arm64e support.
Jason Molenda [Wed, 4 Dec 2019 23:20:56 +0000 (15:20 -0800)]
Upstream debugserver arm64e support.

The changes are minor; primarily debugserver needs to go through
accessor functions/macros when changing pc/fp/sp/lr, and debugserver
needs to clear any existing pointer auth bits from values in two
cases.  debugserver can fetch the number of bits used for addressing
from a sysctl, and will include that in the qHostInfo reply.  Update
qHostInfo documentation to document it.

4 years ago[llvm-symbolizer] Support debug file lookup using build ID
Petr Hosek [Wed, 27 Nov 2019 01:18:42 +0000 (17:18 -0800)]
[llvm-symbolizer] Support debug file lookup using build ID

Build ID is a protocol for looking up debug files that's already
supported by various tools including debuggers. For example, when
locating debug files, gdb would check the following directories:

- /usr/lib/debug/.build-id/ab/cdef1234.debug
- /usr/bin/ls.debug
- /usr/bin/.debug/ls.debug
- /usr/lib/debug/usr/bin/ls.debug

llvm-symbolizer currently consults all of these except for build ID
based one. This patch implements support for build ID lookup. The
set of debug directories to search is specified by the new option:
--debug-file-directory, whose name matches the debug-file-directory
variable used by gdb for the same purpose.

Differential Revision: https://reviews.llvm.org/D70759

4 years ago[ThinLTO] Fix importing of writeonly variables in distributed ThinLTO
Teresa Johnson [Tue, 3 Dec 2019 20:22:51 +0000 (12:22 -0800)]
[ThinLTO] Fix importing of writeonly variables in distributed ThinLTO

Summary:
D69561/dde5893 enabled importing of readonly variables with references,
however, it introduced a bug relating to importing/internalization of
writeonly variables with references.

A fix for this was added in D70006/7f92d66. But this didn't work in
distributed ThinLTO mode. The reason is that the fix (importing the
writeonly var with a zeroinitializer) was only applied when there were
references on the writeonly var summary. In distributed ThinLTO mode,
where we only have a small slice of the index, we will not have the
references on the importing side if we are not importing those
referenced values. Rather than changing this handshaking (which will
require a lot of other changes, since that's how we know what to import
in the distributed backend clang invocation), we can simply always give
the writeonly variable a zero initializer.

Reviewers: evgeny777, steven_wu

Subscribers: mehdi_amini, inglorion, hiraditya, dexonsmith, arphaman, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70977

4 years agoRevert "[OpenMP50] Add parallel master construct, by Chi Chun Chen."
Reid Kleckner [Wed, 4 Dec 2019 22:50:06 +0000 (14:50 -0800)]
Revert "[OpenMP50] Add parallel master construct, by Chi Chun Chen."

This reverts commit 713dab21e27c987b9114547ce7136bac2e775de9.

Tests do not pass on Windows.

4 years agoBug 43965 - Value of _MSVC_LANG doesn't match MSVC++ VS2019 /std:c++latest mode
Soumi Manna [Wed, 4 Dec 2019 22:38:00 +0000 (14:38 -0800)]
Bug 43965 - Value of _MSVC_LANG doesn't match MSVC++ VS2019 /std:c++latest mode

Summary:
The patch adds correct value of _MSVC_LANG to match with MSVC++ VS2019 /std:c++latest mode.

    Bugzilla - Bug 43965
    https://bugs.llvm.org/show_bug.cgi?id=43965

    The value for a MS specific macro differs from Microsoft starting with VS2019 in /std:c++latest mode.

    -bash-4.2$ cat msvclang.cpp
    _MSVC_LANG

    -bash-4.2$ cl /std:c++latest -E msvclang.cpp
    Microsoft (R) C/C++ Optimizing Compiler Version 19.20.27508.1 for x64
    Copyright (C) Microsoft Corporation.  All rights reserved.

    /std:c++latest is provided as a preview of language features from the latest C++
    working draft, and we're eager to hear about bugs and suggestions for improvements.
    However, note that these features are provided as-is without support, and subject
    to changes or removal as the working draft evolves. See
    https://go.microsoft.com/fwlink/?linkid=2045807 for details.

    msvclang.cpp

    201705L

    -bash-4.2$ clang-cl /std:c++latest -E  msvclang.cpp
    201704L

Reviewers: rnk

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D70996

4 years agoRevert "Add --strip-trailing-cr to compression tests so they pass on Windows."
Xiangling Liao [Wed, 4 Dec 2019 22:23:57 +0000 (17:23 -0500)]
Revert "Add --strip-trailing-cr to compression tests so they pass on Windows."

This reverts commit d6cbc9528d46d30416a6f9cd6c8570b704a0bd33.

It causes the tests to fail on AIX.

4 years ago[lit] Document the undocumented pre-defined substitutions
Daniel Sanders [Wed, 4 Dec 2019 22:24:51 +0000 (14:24 -0800)]
[lit] Document the undocumented pre-defined substitutions

4 years ago[compiler-rt] FuzzedDataProvider: do not call memcpy on empty vector.
Max Moroz [Wed, 4 Dec 2019 21:24:59 +0000 (13:24 -0800)]
[compiler-rt] FuzzedDataProvider: do not call memcpy on empty vector.

Summary:
Some versions of memcpy mark pointer arguments as __nonnull, that triggers UBSan
errors even when the length passed is 0.

Reviewers: manojgupta, metzman

Subscribers: dberris, #sanitizers, llvm-commits

Tags: #sanitizers, #llvm

Differential Revision: https://reviews.llvm.org/D71031

[compiler-rt] FDP: assert that num_bytes_to_consume == 0 when size == 0.

4 years ago[NFC][InstCombine] Autogenerate check lines in a few tests
Roman Lebedev [Wed, 4 Dec 2019 22:07:23 +0000 (01:07 +0300)]
[NFC][InstCombine] Autogenerate check lines in a few tests

These files are potentially affected by Negator (D68408) patch.