platform/upstream/llvm.git
6 years ago[CGP] Avoid segmentation fault when doing PHI node simplifications
Bjorn Pettersson [Tue, 20 Mar 2018 09:06:37 +0000 (09:06 +0000)]
[CGP] Avoid segmentation fault when doing PHI node simplifications

Summary:
Made PHI node simplifiations more robust in several ways:

- Minor refactoring to let the SimplificationTracker own the
sets with new PHI/Select nodes that are introduced. This is
maybe not mapping to the original intention with the
SimplificationTracker, but IMHO it encapsulates the logic behind
those sets a little bit better.

- MatchPhiNode can sometimes populate the Matched set with
several entries, where it maps one PHI node to different candidates
for replacement. The Matched set is changed into a SmallSetVector
to make sure we get a deterministic iteration when doing
the replacements.

- As described above we may get several different replacements
for a single PHI node. The loop in MatchPhiSet that is doing
the replacements could end up calling eraseFromParent several
times for the same PHI node, resulting in segmentation faults.
This problem was supposed to be fixed in rL327250, but due to
the non-determinism(?) it only appeared to be fixed (I still
got crashes sometime when turning on/off -print-after-all etc
to get different iteration order in the DenseSets).
With this patch we follow the deterministic ordering in the
Matched set when replacing the PHI nodes. If we find a new
replacement for an already replaced PHI node we replace the
new replacement by the old replacement instead. This is quite
similar to what happened in the rl327250 patch, but here we
also recursively verify that the old replacement hasn't been
replaced already.

- It was really hard to track down the fault described above
(segementation fault due to doing eraseFromParent multiple
times for the same instruction). The fault was intermittent and
small changes in the code, or simply turning on -print-after-all
etc could make the problem go away. This was basically due to
the iteration over PhiNodesToMatch in MatchPhiSet no being
deterministic. Therefore I've changed the data structure for
the SimplificationTracker::AllPhiNodes into an SmallSetVector.
This gives a deterministic behavior.

Reviewers: skatkov, john.brawn

Reviewed By: skatkov

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44571

llvm-svn: 327961

6 years ago[LV] Let recordVectorLoopValueForInductionCast to check if IV was created from the...
Andrei Elovikov [Tue, 20 Mar 2018 09:04:39 +0000 (09:04 +0000)]
[LV] Let recordVectorLoopValueForInductionCast to check if IV was created from the cast.

Summary:
It turned out to be error-prone to expect the callers to handle that - better to
leave the decision to this routine and make the required data to be explicitly
passed to the function.

This handles the case that was missed in the r322473 and fixes the assert
mentioned in PR36524.

Reviewers: dorit, mssimpso, Ayal, dcaballe

Reviewed By: dcaballe

Subscribers: Ka-Ka, hiraditya, dneilson, hsaito, llvm-commits

Differential Revision: https://reviews.llvm.org/D43812

llvm-svn: 327960

6 years ago[ms] Parse #pragma optimize and ignore it behind its own flag
Hans Wennborg [Tue, 20 Mar 2018 08:53:11 +0000 (08:53 +0000)]
[ms] Parse #pragma optimize and ignore it behind its own flag

This allows users to turn off warnings about this pragma specifically,
while still receiving warnings about other ignored pragmas.

Differential Revision: https://reviews.llvm.org/D44630

llvm-svn: 327959

6 years ago[SystemZ] Add "REQUIRES: asserts" to test case to fix build bots.
Jonas Paulsson [Tue, 20 Mar 2018 08:29:19 +0000 (08:29 +0000)]
[SystemZ]  Add "REQUIRES: asserts" to test case to fix build bots.

llvm-svn: 327958

6 years ago[X86] Properly implement the calling convention for f80 for mingw/x86_64
Martin Storsjo [Tue, 20 Mar 2018 06:19:38 +0000 (06:19 +0000)]
[X86] Properly implement the calling convention for f80 for mingw/x86_64

In these cases, both parameters and return values are passed
as a pointer to a stack allocation.

MSVC doesn't use the f80 data type at all, while it is used
for long doubles on mingw.

Normally, this part of the calling convention is handled
within clang, but for intrinsics that are lowered to libcalls,
it may need to be handled within llvm as well.

Differential Revision: https://reviews.llvm.org/D44592

llvm-svn: 327957

6 years ago[ORC] Don't fully qualify explicit destructor call -- it confuses some compilers.
Lang Hames [Tue, 20 Mar 2018 05:56:58 +0000 (05:56 +0000)]
[ORC] Don't fully qualify explicit destructor call -- it confuses some compilers.

This should fix the builder failure at
http://lab.llvm.org:8011/builders/lld-x86_64-darwin13/builds/19224

llvm-svn: 327955

6 years ago[X86] Rename MOVSX32_NOREXrr8 to MOVSX32rr8_NOREX so that the scheduler model regular...
Craig Topper [Tue, 20 Mar 2018 05:00:20 +0000 (05:00 +0000)]
[X86] Rename MOVSX32_NOREXrr8 to MOVSX32rr8_NOREX so that the scheduler model regular expressions will pick it up with the regular version.

Do the same for MOVSX32_NOREXrm8, MOVZX32_NOREXrr8, and MOVZX32_NOREXrm8

llvm-svn: 327948

6 years ago[X86] Fix the SchedRW for memory forms of CMP and TEST.
Craig Topper [Tue, 20 Mar 2018 03:55:17 +0000 (03:55 +0000)]
[X86] Fix the SchedRW for memory forms of CMP and TEST.

They were incorrectly marked as RMW operations. Some of the CMP instrucions worked, but the ones that use a similar encoding as RMW form of ADD ended up marked as RMW.

TEST used the same tablegen class as some of the CMPs.

llvm-svn: 327947

6 years ago[ORC] Rename SymbolSource to MaterializationUnit, and make the materialization
Lang Hames [Tue, 20 Mar 2018 03:49:29 +0000 (03:49 +0000)]
[ORC] Rename SymbolSource to MaterializationUnit, and make the materialization
operation all-or-nothing, rather than allowing materialization on a per-symbol
basis.

This addresses a shortcoming of per-symbol materialization: If a
MaterializationUnit (/SymbolSource) wants to materialize more symbols than
requested (which is likely: most materializers will want to materialize whole
modules) then it needs a way to notify the symbol table about the extra symbols
being materialized. This process (checking what has been requested against what
is being provided and notifying the symbol table about the difference) has to
be repeated at every level of the JIT stack. Making materialization
all-or-nothing eliminates this issue, simplifying both materializer
implementations and the symbol table (VSO class) API. The cost is that
per-symbol materialization (e.g. for individual symbols in a module) now
requires multiple MaterializationUnits.

llvm-svn: 327946

6 years agoProperly construct `inline` members without initializers
George Burgess IV [Tue, 20 Mar 2018 03:27:44 +0000 (03:27 +0000)]
Properly construct `inline` members without initializers

Digging through commit logs, it appears the checks in this block predate
`inline` class variables. With them, we fail to emit dynamic
initializers for members that don't have an explicit initializer, and we
won't go out of our way to instantiate the class denoted by
`Var->getType()`.

Fixes PR35599.

llvm-svn: 327945

6 years ago[X86] Add TEST16mi/TEST32mi/TEST64mi32 to the Sandybridge/Haswell/Broadwell/Skylake...
Craig Topper [Tue, 20 Mar 2018 03:02:03 +0000 (03:02 +0000)]
[X86] Add TEST16mi/TEST32mi/TEST64mi32 to the Sandybridge/Haswell/Broadwell/Skylake scheduler models.

Move it from a load+store group on SNB to a load only group, the same group as CMP.

llvm-svn: 327944

6 years ago[X86] Add ROR/ROL/SHR/SAR by 1 instructions to the Sandy Bridge scheduler model.
Craig Topper [Tue, 20 Mar 2018 03:01:59 +0000 (03:01 +0000)]
[X86] Add ROR/ROL/SHR/SAR by 1 instructions to the Sandy Bridge scheduler model.

I assume these match the generic immediate version like they do in the other models.

llvm-svn: 327943

6 years ago[ShrinkWrap] Take into account landing pad
Quentin Colombet [Tue, 20 Mar 2018 02:44:40 +0000 (02:44 +0000)]
[ShrinkWrap] Take into account landing pad

When scanning the function for CSRs uses and defs, also check if
the basic block are landing pads.
Consider that landing pads needs the CSRs to be properly set.
That way we force the prologue/epilogue to always be pushed out
of the problematic "throw" region. The "throw" region is
problematic because the jumps are not properly modeled.

Fixes PR36513

llvm-svn: 327942

6 years agoAdd a suggestion to convert dotest tests to use run_to_source_breakpoint.
Jim Ingham [Tue, 20 Mar 2018 02:15:23 +0000 (02:15 +0000)]
Add a suggestion to convert dotest tests to use run_to_source_breakpoint.

llvm-svn: 327941

6 years agoFix layering between llvm-c and Support by factoring out some typedefs into Support
David Blaikie [Tue, 20 Mar 2018 02:14:50 +0000 (02:14 +0000)]
Fix layering between llvm-c and Support by factoring out some typedefs into Support

llvm-c depends on Support, but Support (TargetRegistry) uses some of
llvm-c's typedefs. Move those into a Support header to be used from both
llvm-c and Support.

llvm-svn: 327940

6 years ago[CodeGen] Ignore OpaqueValueExprs that are unique references to their
Akira Hatanaka [Tue, 20 Mar 2018 01:47:58 +0000 (01:47 +0000)]
[CodeGen] Ignore OpaqueValueExprs that are unique references to their
source expressions when iterating over a PseudoObjectExpr's semantic
subexpression list.

Previously the loop in emitPseudoObjectExpr would emit the IR for each
OpaqueValueExpr that was in a PseudoObjectExpr's semantic-form
expression list and use the result when the OpaqueValueExpr later
appeared in other expressions. This caused an assertion failure when
AggExprEmitter tried to copy the result of an OpaqueValueExpr and the
copied type didn't have trivial copy/move constructors or assignment
operators.

This patch adds flag IsUnique to OpaqueValueExpr which indicates it is a
unique reference to its source expression (it is not used in multiple
places). The loop in emitPseudoObjectExpr ignores OpaqueValueExprs that
are unique and CodeGen visitors simply traverse the source expressions
of such OpaqueValueExprs.

rdar://problem/34363596

Differential Revision: https://reviews.llvm.org/D39562

llvm-svn: 327939

6 years ago[RISCV] Preserve stack space for outgoing arguments when the function contain variabl...
Shiva Chen [Tue, 20 Mar 2018 01:39:17 +0000 (01:39 +0000)]
[RISCV] Preserve stack space for outgoing arguments when the function contain variable size objects

E.g.

bar (int x)
{
  char p[x];

  push outgoing variables for foo.
  call foo
}

We need to generate stack adjustment instructions for outgoing arguments by
eliminateCallFramePseudoInstr when the function contains variable size
objects to avoid outgoing variables corrupt the variable size object.

Default hasReservedCallFrame will return !hasFP().
We don't want to generate extra sp adjustment instructions when hasFP()
return true, So We override hasReservedCallFrame as !hasVarSizedObjects().

Differential Revision: https://reviews.llvm.org/D43752

llvm-svn: 327938

6 years ago[X86] Fix the SNB scheduler for BLENDVB.
Craig Topper [Tue, 20 Mar 2018 01:30:21 +0000 (01:30 +0000)]
[X86] Fix the SNB scheduler for BLENDVB.

PBLENDVBrr0 was with the memory version of VBLENDVB and PBLENDVBrm0 was missing.

llvm-svn: 327937

6 years agoRevert "[compiler-rt] Change std::sort to llvm::sort in response to r327219"
Mandeep Singh Grang [Tue, 20 Mar 2018 01:17:18 +0000 (01:17 +0000)]
Revert "[compiler-rt] Change std::sort to llvm::sort in response to r327219"

This reverts commit 2ee210e1963e03aacc0f71c50e4994bb5c66586e.

llvm-svn: 327936

6 years ago[analyzer] Improve performance of NoStoreFuncVisitor
George Karpenkov [Tue, 20 Mar 2018 01:16:46 +0000 (01:16 +0000)]
[analyzer] Improve performance of NoStoreFuncVisitor

Compute modifying frames lazily on demand.

Differential Revision: https://reviews.llvm.org/D44503

llvm-svn: 327935

6 years agoRun dos2unix on a test. NFC.
Rafael Espindola [Tue, 20 Mar 2018 01:06:29 +0000 (01:06 +0000)]
Run dos2unix on a test. NFC.

llvm-svn: 327934

6 years ago[DebugInfoPDB] Add backward support flags to PDB_NameSearchFlags
Aaron Smith [Tue, 20 Mar 2018 01:04:21 +0000 (01:04 +0000)]
[DebugInfoPDB] Add backward support flags to PDB_NameSearchFlags

For NS_CaseInFileNameExt support.

llvm-svn: 327933

6 years agoAllow builtins for Cross-DSO CFI on Android
Petr Hosek [Tue, 20 Mar 2018 00:56:08 +0000 (00:56 +0000)]
Allow builtins for Cross-DSO CFI on Android

This is needed to avoid the test failure in case when compiler-rt
is set as the default runtime library for Clang.

Differential Revision: https://reviews.llvm.org/D44655

llvm-svn: 327932

6 years ago[ELF] Make SectionHeaderOffset less sensitive.
Fangrui Song [Tue, 20 Mar 2018 00:55:39 +0000 (00:55 +0000)]
[ELF] Make SectionHeaderOffset less sensitive.

llvm-svn: 327931

6 years agoObject: Fix handling of @@@ in .symver directive
Vitaly Buka [Tue, 20 Mar 2018 00:45:03 +0000 (00:45 +0000)]
Object: Fix handling of @@@ in .symver directive

Summary:
name@@@nodename is going to be replaced with name@@nodename if symbols is
defined in the assembled file, or name@nodename if undefined.
https://sourceware.org/binutils/docs/as/Symver.html

Fixes PR36623

Reviewers: pcc, espindola

Subscribers: mehdi_amini, hiraditya

Differential Revision: https://reviews.llvm.org/D44274

llvm-svn: 327930

6 years ago[compiler-rt] Change std::sort to llvm::sort in response to r327219
Mandeep Singh Grang [Tue, 20 Mar 2018 00:44:59 +0000 (00:44 +0000)]
[compiler-rt] Change std::sort to llvm::sort in response to r327219

Summary:
r327219 added wrappers to std::sort which randomly shuffle the container before sorting.
This will help in uncovering non-determinism caused due to undefined sorting
order of objects having the same key.

To make use of that infrastructure we need to invoke llvm::sort instead of std::sort.

Reviewers: kcc, rsmith, RKSimon, eugenis

Reviewed By: RKSimon

Subscribers: efriedma, kubamracek, dberris, #sanitizers, llvm-commits

Differential Revision: https://reviews.llvm.org/D44360

llvm-svn: 327929

6 years agoObject: Move attribute calculation into RecordStreamer. NFC
Vitaly Buka [Tue, 20 Mar 2018 00:38:33 +0000 (00:38 +0000)]
Object: Move attribute calculation into RecordStreamer. NFC

Summary: Preparation for D44274

Reviewers: pcc, espindola

Subscribers: hiraditya

Differential Revision: https://reviews.llvm.org/D44276

llvm-svn: 327928

6 years ago[SymbolFilePDB] Simplify code with newer methods
Aaron Smith [Tue, 20 Mar 2018 00:34:18 +0000 (00:34 +0000)]
[SymbolFilePDB] Simplify code with newer methods

llvm-svn: 327927

6 years ago[analyzer] Fix the assertion failure when static globals are used in lambda by reference
George Karpenkov [Tue, 20 Mar 2018 00:20:58 +0000 (00:20 +0000)]
[analyzer] Fix the assertion failure when static globals are used in lambda by reference

Also use the opportunity to clean up the code and remove unnecessary duplication.

rdar://37625895

Differential Revision: https://reviews.llvm.org/D44594

llvm-svn: 327926

6 years ago[SymbolFilePDB] Simplify getting the source file path
Aaron Smith [Tue, 20 Mar 2018 00:18:22 +0000 (00:18 +0000)]
[SymbolFilePDB] Simplify getting the source file path

Summary: Replace SymbolFilePDB::GetSourceFileNameForPDBCompiland() with PDBSymbolCompiland::getSourceFileFullPath().

Reviewers: zturner, rnk, lldb-commits

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44456

llvm-svn: 327925

6 years agoModernize a test.
Jim Ingham [Mon, 19 Mar 2018 23:15:06 +0000 (23:15 +0000)]
Modernize a test.

llvm-svn: 327924

6 years agoOpenBSD UBsan support missing bits
Vitaly Buka [Mon, 19 Mar 2018 23:12:14 +0000 (23:12 +0000)]
OpenBSD UBsan support missing bits

Summary:
Lost bits since the WIP ticket

Patch by David CARLIER

Reviewers: vitalybuka, vsk

Reviewed By: vitalybuka

Subscribers: srhines, kubamracek, krytarowski, fedor.sergeev, llvm-commits, #sanitizers

Differential Revision: https://reviews.llvm.org/D44599

llvm-svn: 327923

6 years agoRename remotectl_com.apple.internal.xpc.remote.debugserver.plist
Jason Molenda [Mon, 19 Mar 2018 23:07:31 +0000 (23:07 +0000)]
Rename remotectl_com.apple.internal.xpc.remote.debugserver.plist
to com.apple.internal.xpc.remote.debugserver.plist, not sure where
that remotectl_ prefix came from.

<rdar://problem/36751222>

llvm-svn: 327922

6 years agoRemove GnuHashTableSection::getShift2().
Rui Ueyama [Mon, 19 Mar 2018 23:04:13 +0000 (23:04 +0000)]
Remove GnuHashTableSection::getShift2().

Choosing a Shift2 value based on wordsize is cargo-culted from gold.
Assuming that djb hash is a good hash function, choosing bits [4,9]
shouldn't be any worse or better than choosing bits [5,10]. We shouldn't
have copied that behavior that we can't justify in the first place.

Differential Revision: https://reviews.llvm.org/D44547

llvm-svn: 327921

6 years agoallow-multiple-definitions should completely suppress errors instead of making them...
Rui Ueyama [Mon, 19 Mar 2018 23:04:04 +0000 (23:04 +0000)]
allow-multiple-definitions should completely suppress errors instead of making them warnings.

We found that when you pass --allow-multiple-definitions or `-z muldefs`
to GNU linkers, they don't complain about duplicate symbols at all. They
don't even print out warnings on it. We emit warnings in that case.
If you pass --fatal-warnings, that difference results in a link failure.

Differential Revision: https://reviews.llvm.org/D44549

llvm-svn: 327920

6 years ago[SelectionDAG] Transfer DbgValues when integer operations are promoted
Aaron Smith [Mon, 19 Mar 2018 22:58:50 +0000 (22:58 +0000)]
[SelectionDAG] Transfer DbgValues when integer operations are promoted

Summary:
DbgValue nodes were not transferred when integer DAG nodes were promoted. For example, if an i32 add node was promoted to an i64 add node by DAGTypeLegalizer::PromoteIntegerResult(), its DbgValue node was not transferred to the new node. The simple fix is to update SetPromotedInteger() to transfer DbgValues.

Add AArch64/dbg-value-i8.ll to test this change and fix ARM/debug-info-d16-reg.ll which had the wrong DILocalVariable nodes with arg numbers even though they are not for function parameters.

Patch by Se Jong Oh!

Reviewers: vsk, JDevlieghere, aprantl

Reviewed By: JDevlieghere

Subscribers: javed.absar, kristof.beyls, llvm-commits

Tags: #debug-info

Differential Revision: https://reviews.llvm.org/D44546

llvm-svn: 327919

6 years agoI didn't see that SocketAddress.cpp was already being pulled
Jason Molenda [Mon, 19 Mar 2018 22:55:19 +0000 (22:55 +0000)]
I didn't see that SocketAddress.cpp was already being pulled
in to debugserver; my re-addition resulted in duplicated
symbols.  Remove my added SocketAddress.cpp, and change the
original one to also be included for the debugserver-mini target.

llvm-svn: 327918

6 years ago[MachineOutliner] AArch64: Emit CFI instructions when outlining calls
Jessica Paquette [Mon, 19 Mar 2018 22:48:40 +0000 (22:48 +0000)]
[MachineOutliner] AArch64: Emit CFI instructions when outlining calls

When outlining calls, the outliner needs to update CFI to ensure that, say,
exception handling works. This commit adds that functionality and adds a test
just for call outlining.

Call outlining stuff in machine-outliner.mir should be moved into
machine-outliner-calls.mir in a later commit.

llvm-svn: 327917

6 years ago[DAGCombiner] Fix type in comment. NFC
Craig Topper [Mon, 19 Mar 2018 22:25:26 +0000 (22:25 +0000)]
[DAGCombiner] Fix type in comment. NFC

llvm-svn: 327916

6 years agoAdd the ios-mini target to the top-level xcodeproj, which builds
Jason Molenda [Mon, 19 Mar 2018 22:04:45 +0000 (22:04 +0000)]
Add the ios-mini target to the top-level xcodeproj, which builds
the debugserver-mini target in debugserver.  Add a new plist
which needs to be installed for debugserver-mini.
<rdar://problem/36751222>

llvm-svn: 327915

6 years ago[X86] Simplify the AVX512 code in LowerTruncate a little.
Craig Topper [Mon, 19 Mar 2018 21:58:02 +0000 (21:58 +0000)]
[X86] Simplify the AVX512 code in LowerTruncate a little.

We don't need to create an ISD::TRUNCATE node to return, we started with one and can return it. Also remove the call to getExtendInVec, the result is just going to be a getNode of that value passed in.

llvm-svn: 327914

6 years agoAdd DWARFUnit.cpp to the correct target.
Jason Molenda [Mon, 19 Mar 2018 21:55:48 +0000 (21:55 +0000)]
Add DWARFUnit.cpp to the correct target.

llvm-svn: 327913

6 years agoAccept any filepath in llvm_check_source_file_list
Serge Guelton [Mon, 19 Mar 2018 21:35:30 +0000 (21:35 +0000)]
Accept any filepath in llvm_check_source_file_list

Cmake function llvm_check_source_file_list currently only accepts paths
relative to current CMAKE_SOURCE_DIR or relative to argument SOURCE_DIR.

Extend it to accept any path, including absolute ones.

Differential revision: https://reviews.llvm.org/D44625

llvm-svn: 327912

6 years agoFixup test to explicitly use -fexceptions since exceptions are disabled by default...
Douglas Yung [Mon, 19 Mar 2018 21:22:58 +0000 (21:22 +0000)]
Fixup test to explicitly use -fexceptions since exceptions are disabled by default on the PS4 target.

llvm-svn: 327911

6 years ago[PDB] Add a method to get the full path of the source file for PDBSymbolCompiland
Aaron Smith [Mon, 19 Mar 2018 21:20:04 +0000 (21:20 +0000)]
[PDB] Add a method to get the full path of the source file for PDBSymbolCompiland

Summary:
Redefine PDBSymbolCompiland::getSourceFileName() to return the filename (w/o directory) of the source file that is used to compile the compiland. This is because the result returned previously is ambiguous. It could be the filename, relative path or full path of the source file.

Move the implementation of SymbolFilePDB::GetSourceFileNameForPDBCompiland() into a new method PDBSymbolCompiland::getSourceFileFullPath().

Reviewers: zturner, rnk, llvm-commits

Reviewed By: zturner

Differential Revision: https://reviews.llvm.org/D44458

llvm-svn: 327910

6 years ago[PDB] Add exclusive methods to derived symbol class
Aaron Smith [Mon, 19 Mar 2018 21:18:39 +0000 (21:18 +0000)]
[PDB] Add exclusive methods to derived symbol class

Summary: This commit adds two methods to the PDBSymboFunc class used in parsing symbols. getLineNumbers() is used to determine a Function symbol's declaration and getCompilandId() is used to initialize the SymbolContext field sc.comp_unit.

Reviewers: zturner, rnk, llvm-commits

Reviewed By: zturner

Differential Revision: https://reviews.llvm.org/D44457

llvm-svn: 327909

6 years ago[SymbolFilePDB] Remove a few null pointer checks by passing ref
Aaron Smith [Mon, 19 Mar 2018 21:14:19 +0000 (21:14 +0000)]
[SymbolFilePDB] Remove a few null pointer checks by passing ref

Reviewers: zturner, rnk, lldb-commits

Reviewed By: zturner

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44455

llvm-svn: 327908

6 years ago[Hexagon] Add REQUIRES: asserts to test/CodeGen/Hexagon/v6vec_inc1.ll
Krzysztof Parzyszek [Mon, 19 Mar 2018 21:05:21 +0000 (21:05 +0000)]
[Hexagon] Add REQUIRES: asserts to test/CodeGen/Hexagon/v6vec_inc1.ll

llvm-svn: 327907

6 years ago[test] Skip flaky tests in TestMiSyntax on Darwin
Vedant Kumar [Mon, 19 Mar 2018 20:50:50 +0000 (20:50 +0000)]
[test] Skip flaky tests in TestMiSyntax on Darwin

These tests tend to time out locally and on our bots.

llvm-svn: 327906

6 years ago[test] Skip flaky TestThreadStates tests on Darwin
Vedant Kumar [Mon, 19 Mar 2018 20:50:49 +0000 (20:50 +0000)]
[test] Skip flaky TestThreadStates tests on Darwin

These tests do not pass/fail consistently, so just skip them.

This is llvm.org/pr15824 & rdar://problem/28557237.

llvm-svn: 327905

6 years agoCleanup of "extern char **environ" declaration
Vitaly Buka [Mon, 19 Mar 2018 20:44:06 +0000 (20:44 +0000)]
Cleanup of "extern char **environ" declaration

llvm-svn: 327904

6 years agoRevert "Support embedding natvis files in PDBs."
Zachary Turner [Mon, 19 Mar 2018 20:41:59 +0000 (20:41 +0000)]
Revert "Support embedding natvis files in PDBs."

This is causing a test failure on a certain bot, so I'm removing
this temporarily until we can figure out the source of the error.

llvm-svn: 327903

6 years ago[clangd] Fix undefined behavior due to misaligned type cast
Jan Korous [Mon, 19 Mar 2018 20:26:18 +0000 (20:26 +0000)]
[clangd] Fix undefined behavior due to misaligned type cast

The current code was casting pointer to a misaligned type which is undefined behavior.
Found by compiling with Undefined Behavior Sanitizer and running tests (check-clang-tools).

llvm-svn: 327902

6 years ago[clangd][nfc] Give name to a magic constant
Jan Korous [Mon, 19 Mar 2018 20:26:15 +0000 (20:26 +0000)]
[clangd][nfc] Give name to a magic constant

llvm-svn: 327901

6 years agoRemove an unused private variable.
Zachary Turner [Mon, 19 Mar 2018 20:22:48 +0000 (20:22 +0000)]
Remove an unused private variable.

llvm-svn: 327900

6 years ago[X86] Replace a couple calls to getExtendInVec with getNode and the appropriate targe...
Craig Topper [Mon, 19 Mar 2018 20:20:22 +0000 (20:20 +0000)]
[X86] Replace a couple calls to getExtendInVec with getNode and the appropriate target independent EXTEND_VECTOR_INREG opcode.

llvm-svn: 327899

6 years ago[DAG, X86] Revert r327197 "Revert r327170, r327171, r327172"
Nirav Dave [Mon, 19 Mar 2018 20:19:46 +0000 (20:19 +0000)]
[DAG, X86] Revert r327197 "Revert r327170, r327171, r327172"

Reland ISel cycle checking improvements after simplifying node id
invariant traversal and correcting typo.

llvm-svn: 327898

6 years ago[ARM, AArch64] Check the no-stack-arg-probe attribute for dynamic stack probes
Martin Storsjo [Mon, 19 Mar 2018 20:06:50 +0000 (20:06 +0000)]
[ARM, AArch64] Check the no-stack-arg-probe attribute for dynamic stack probes

This extends the use of this attribute on ARM and AArch64 from
SVN r325900 (where it was only checked for fixed stack
allocations on ARM/AArch64, but for all stack allocations on X86).

This also adds a testcase for the existing use of disabling the
fixed stack probe with the attribute on ARM and AArch64.

Differential Revision: https://reviews.llvm.org/D44291

llvm-svn: 327897

6 years agoAdd cast to Type*, fix failure from r327894.
Alina Sbirlea [Mon, 19 Mar 2018 20:05:01 +0000 (20:05 +0000)]
Add cast to Type*, fix failure from r327894.

llvm-svn: 327896

6 years agoSupport embedding natvis files in PDBs.
Zachary Turner [Mon, 19 Mar 2018 19:53:51 +0000 (19:53 +0000)]
Support embedding natvis files in PDBs.

Natvis is a debug language supported by Visual Studio for
specifying custom visualizers.  The /NATVIS option is an
undocumented link.exe flag which will take a .natvis file
and "inject" it into the PDB.  This way, you can ship the
debug visualizers for a program along with the PDB, which
is very useful for postmortem debugging.

This is implemented by adding a new "named stream" to the
PDB with a special name of /src/files/<natvis file name>
and simply copying the contents of the xml into this file.

Additionally, we need to emit a single stream named
/src/headerblock which contains a hash table of embedded
files to records describing them.

This patch adds this functionality, including the /NATVIS
option to lld-link.

Differential Revision: https://reviews.llvm.org/D44328

llvm-svn: 327895

6 years agoMake ConstantDataArray::get constructor templated. Will support signed integers.
Alina Sbirlea [Mon, 19 Mar 2018 19:49:28 +0000 (19:49 +0000)]
Make ConstantDataArray::get constructor templated. Will support signed integers.

Summary: Make ConstantDataArray::get() constructors a single templated one.

Reviewers: timshen, rsmith

Subscribers: sanjoy, llvm-commits, jlebar

Differential Revision: https://reviews.llvm.org/D44337

llvm-svn: 327894

6 years ago[ClangASTContext] Remove dead code. NFCI.
Davide Italiano [Mon, 19 Mar 2018 19:35:20 +0000 (19:35 +0000)]
[ClangASTContext] Remove dead code. NFCI.

llvm-svn: 327893

6 years ago[CodeGen] Add funclet token to ARC marker
Shoaib Meenai [Mon, 19 Mar 2018 19:34:39 +0000 (19:34 +0000)]
[CodeGen] Add funclet token to ARC marker

The inline assembly generated for the ARC autorelease elision marker
must have a funclet token if it's emitted inside a funclet, otherwise
the inline assembly (and all subsequent code in the funclet) will be
marked unreachable. r324689 fixed this issue for regular inline assembly
blocks.

Note that clang only emits the marker at -O0, so this only fixes that
case. The optimizations case (where the marker is emitted by the
backend) will be fixed in a separate change.

Differential Revision: https://reviews.llvm.org/D44640

llvm-svn: 327892

6 years ago[AMDGPU] change test to avoid NaN math
Sanjay Patel [Mon, 19 Mar 2018 19:26:22 +0000 (19:26 +0000)]
[AMDGPU] change test to avoid NaN math

llvm-svn: 327891

6 years ago[AMDGPU] adjust tests to be nan-free
Sanjay Patel [Mon, 19 Mar 2018 19:23:53 +0000 (19:23 +0000)]
[AMDGPU] adjust tests to be nan-free

As suggested in D44521 - bitcast to integer for the math,
so we preserve the intent of these tests when NaN math
gets folded away.

llvm-svn: 327890

6 years ago[Power9]Legalize and emit code for quad-precision copySign/abs/nabs/neg/sqrt
Lei Huang [Mon, 19 Mar 2018 19:22:52 +0000 (19:22 +0000)]
[Power9]Legalize and emit code for quad-precision copySign/abs/nabs/neg/sqrt

Legalize and emit code for quad-precision floating point operations:

  * xscpsgnqp
  * xsabsqp
  * xsnabsqp
  * xsnegqp
  * xssqrtqp

Differential Revision: https://reviews.llvm.org/D44530

llvm-svn: 327889

6 years ago[llvm-mca] Remove unused method from ResourceManager. NFC
Andrea Di Biagio [Mon, 19 Mar 2018 19:14:06 +0000 (19:14 +0000)]
[llvm-mca] Remove unused method from ResourceManager. NFC

llvm-svn: 327888

6 years ago[clang-move] Fix the failing test caused by changes in clang-format.
Haojian Wu [Mon, 19 Mar 2018 19:13:03 +0000 (19:13 +0000)]
[clang-move] Fix the failing test caused by changes in clang-format.

llvm-svn: 327887

6 years ago[llvm-mca] Simplify code. NFC
Andrea Di Biagio [Mon, 19 Mar 2018 19:09:38 +0000 (19:09 +0000)]
[llvm-mca] Simplify code. NFC

llvm-svn: 327886

6 years ago[clang-move] Fix move-used-helper-decls.cpp test.
Fangrui Song [Mon, 19 Mar 2018 19:05:53 +0000 (19:05 +0000)]
[clang-move] Fix move-used-helper-decls.cpp test.

llvm-svn: 327885

6 years ago[Hexagon] Add a few more lit tests
Krzysztof Parzyszek [Mon, 19 Mar 2018 19:03:18 +0000 (19:03 +0000)]
[Hexagon] Add a few more lit tests

llvm-svn: 327884

6 years ago[X86] Add JMP16r and JMP32r to Sandybridge scheduler model.
Craig Topper [Mon, 19 Mar 2018 19:00:37 +0000 (19:00 +0000)]
[X86] Add JMP16r and JMP32r to Sandybridge scheduler model.

Fixes PR36010

llvm-svn: 327883

6 years ago[X86] Remove OUT32rr/OUT8rr/OUT32ri/OUT8ri from Sandybridge scheduler model.
Craig Topper [Mon, 19 Mar 2018 19:00:35 +0000 (19:00 +0000)]
[X86] Remove OUT32rr/OUT8rr/OUT32ri/OUT8ri from Sandybridge scheduler model.

PR35590 was already filed for this information being wrong. It's probably better to default to WriteSystem behavior instead of using something completely wrong.

llvm-svn: 327882

6 years ago[X86] Add JCXZ/JECXZ to Sandybridge/Haswell/Broadwell/Skylake scheduler models.
Craig Topper [Mon, 19 Mar 2018 19:00:32 +0000 (19:00 +0000)]
[X86] Add JCXZ/JECXZ to Sandybridge/Haswell/Broadwell/Skylake scheduler models.

JRCXZ was already present, but not the others.

We never codegen this instruction so this doesn't affect much just trying to get them all into a single generated scheduler class in the output.

llvm-svn: 327881

6 years ago[X86] Correct regular expression in Zen scheduler model that was excluding JECXZ...
Craig Topper [Mon, 19 Mar 2018 19:00:29 +0000 (19:00 +0000)]
[X86] Correct regular expression in Zen scheduler model that was excluding JECXZ instruction.

The regex was looking for JECXZ_32 or JECXZ_64, but their is just one instruction called JECXZ. They used to exist as separate instructions, but were merged over 3 years ago.

llvm-svn: 327880

6 years ago[X86] Correct the SchedRW on (V)MOVAPSrr_REV and similar to match their non _REV...
Craig Topper [Mon, 19 Mar 2018 19:00:26 +0000 (19:00 +0000)]
[X86] Correct the SchedRW on (V)MOVAPSrr_REV and similar to match their non _REV counterparts.

llvm-svn: 327879

6 years ago[PowerPC][Power9]Legalize and emit code for quad-precision add/div/mul/sub
Lei Huang [Mon, 19 Mar 2018 18:52:20 +0000 (18:52 +0000)]
[PowerPC][Power9]Legalize and emit code for quad-precision add/div/mul/sub

Legalize and emit code for quad-precision floating point operations:

  * xsaddqp
  * xssubqp
  * xsdivqp
  * xsmulqp

Differential Revision: https://reviews.llvm.org/D44506

llvm-svn: 327878

6 years ago[PowerPC] Make AddrSpaceCast noop
Nemanja Ivanovic [Mon, 19 Mar 2018 18:50:02 +0000 (18:50 +0000)]
[PowerPC] Make AddrSpaceCast noop

PowerPC targets do not use address spaces. As a result, we can get selection
failures with address space casts. This patch makes those casts noops.

Patch by Valentin Churavy.

Differential revision: https://reviews.llvm.org/D43781

llvm-svn: 327877

6 years agoFix CMake/MSVC when compiler-rt and llvm are built separately
Vitaly Buka [Mon, 19 Mar 2018 18:22:35 +0000 (18:22 +0000)]
Fix CMake/MSVC when compiler-rt and llvm are built separately

Summary:
For some reason CMake can't find the `append` macro if LLVM is built separately and imported via `LLVM_CONFIG_PATH`.

Patch by Loo Rong Jie

Reviewers: rnk, vitalybuka

Reviewed By: rnk, vitalybuka

Subscribers: dberris, mgorny, llvm-commits, #sanitizers

Differential Revision: https://reviews.llvm.org/D43458

llvm-svn: 327876

6 years agoFix for Fix for https://bugs.llvm.org/show_bug.cgi?id=36705.
Andrey Churbanov [Mon, 19 Mar 2018 18:05:15 +0000 (18:05 +0000)]
Fix for Fix for https://bugs.llvm.org/show_bug.cgi?id=36705.

Differential Revision: https://reviews.llvm.org/D44637

llvm-svn: 327875

6 years ago[X86] Add the rest of the TEST with immediate instructions to the scheduler models...
Craig Topper [Mon, 19 Mar 2018 17:58:41 +0000 (17:58 +0000)]
[X86] Add the rest of the TEST with immediate instructions to the scheduler models to match their 8-bit counterpart.

llvm-svn: 327874

6 years ago[OPENMP, NVPTX] Reworked castToType() function, NFC.
Alexey Bataev [Mon, 19 Mar 2018 17:53:56 +0000 (17:53 +0000)]
[OPENMP, NVPTX] Reworked castToType() function, NFC.

Reworked function castToType to use more frontend functionality rather
than the backend.

llvm-svn: 327873

6 years ago[X86] Add MOV16ri*/MOV32ri*/MOV64ri* to scheduler models to match MOV8ri. Correct...
Craig Topper [Mon, 19 Mar 2018 17:46:59 +0000 (17:46 +0000)]
[X86] Add MOV16ri*/MOV32ri*/MOV64ri* to scheduler models to match MOV8ri. Correct SchedRW and itinerary for MOV32ri64.

llvm-svn: 327872

6 years ago[ELF] Add basic support for PPC LE
Zaara Syeda [Mon, 19 Mar 2018 17:40:14 +0000 (17:40 +0000)]
[ELF] Add basic support for PPC LE

This patch adds changes to start supporting the Power 64-Bit ELF V2 ABI.
This includes:
- Changing the ElfSym::GlobalOffsetTable to be named .TOC.
- Creating a GotHeader so the first entry in the .got is .TOC.
- Setting the e_flags to be 1 for ELF V1 and 2 for ELF V2

Differential Revision: https://reviews.llvm.org/D44483

llvm-svn: 327871

6 years ago[ObjC] Allow declaring __weak pointer fields in C structs in ARC.
Akira Hatanaka [Mon, 19 Mar 2018 17:38:40 +0000 (17:38 +0000)]
[ObjC] Allow declaring __weak pointer fields in C structs in ARC.

This patch uses the infrastructure added in r326307 for enabling
non-trivial fields to be declared in C structs to allow __weak fields in
C structs in ARC.

This recommits r327206, which was reverted because it caused
module-enabled builders to fail. I discovered that the
CXXRecordDecl::CanPassInRegisters flag wasn't being set correctly in
some cases after I moved it to RecordDecl.

Thanks to Eric Liu for helping me investigate the bug.

rdar://problem/33599681

https://reviews.llvm.org/D44095

llvm-svn: 327870

6 years ago[X86] Remove sse41 specific code from lowering v16i8 multiply
Craig Topper [Mon, 19 Mar 2018 17:31:41 +0000 (17:31 +0000)]
[X86] Remove sse41 specific code from lowering v16i8 multiply

With the SRAs removed from the SSE2 code in D44267, then there doesn't appear to be any advantage to the sse41 code. The punpcklbw instruction and pmovsx seem to have the same latency and throughput on most CPUs. And the SSE41 code requires moving the upper 64-bits into the lower 64-bit before the sign extend can be done. The unpckhbw in sse2 code can do better than that.

llvm-svn: 327869

6 years ago[OPENMP] Fix build with MSVC, NFC.
Alexey Bataev [Mon, 19 Mar 2018 17:18:13 +0000 (17:18 +0000)]
[OPENMP] Fix build with MSVC, NFC.

llvm-svn: 327868

6 years ago[OPENMP, NVPTX] Emit correct thread id.
Alexey Bataev [Mon, 19 Mar 2018 17:04:07 +0000 (17:04 +0000)]
[OPENMP, NVPTX] Emit correct thread id.

We emitted fake thread id for the outined function in NVPTX codegen.
Patch adds emission of the real thread id.

llvm-svn: 327867

6 years ago[X86] Make the multiply and divide itineraries more consistent.
Craig Topper [Mon, 19 Mar 2018 16:38:33 +0000 (16:38 +0000)]
[X86] Make the multiply and divide itineraries more consistent.

Sometimes we used the same itinerary for MEM and REG forms, but that seems inconsistent with our usual usage.

We also used the MUL8 itinerary for MULX32/64 which was also weird.

The test changes are because we were using IIC_IMUL32_RR and IIC_IMUL64_RR instead of IIC_IMUL32_REG/IIC_IMUL64_REG for the 32 and 64 bit multiplies that produce double width result.

llvm-svn: 327866

6 years agoXcode: Include DWARFUnit.{h,cpp} in the build
Vedant Kumar [Mon, 19 Mar 2018 16:24:58 +0000 (16:24 +0000)]
Xcode: Include DWARFUnit.{h,cpp} in the build

This should address a bot failure due to r327809.

llvm-svn: 327865

6 years agoRevert [MachineLICM] This reverts commit rL327856
Zaara Syeda [Mon, 19 Mar 2018 16:19:44 +0000 (16:19 +0000)]
Revert [MachineLICM] This reverts commit rL327856

Failing build bots. Revert the commit now.

llvm-svn: 327864

6 years ago[Driver] Avoid invalidated iterator in insertTargetAndModeArgs
Serge Pavlov [Mon, 19 Mar 2018 16:13:43 +0000 (16:13 +0000)]
[Driver] Avoid invalidated iterator in insertTargetAndModeArgs

Doing an .insert() can potentially invalidate iterators by reallocating the
vector's storage. When all the stars align just right, this causes segfaults
or glibc aborts.

Gentoo Linux bug (crashes while building Chromium): https://bugs.gentoo.org/650082.

Patch by Hector Martin!

Differential Revision: https://reviews.llvm.org/D44607

llvm-svn: 327863

6 years ago[CodeGen] Avoid handling DBG_VALUE in the LivePhysRegs (addUses,removeDefs,stepForward)
Matt Davis [Mon, 19 Mar 2018 16:06:40 +0000 (16:06 +0000)]
[CodeGen] Avoid handling DBG_VALUE in the LivePhysRegs (addUses,removeDefs,stepForward)

Summary:
This patch prevents DBG_VALUE instructions from influencing
LivePhysRegs::stepBackwards and stepForwards.  In at least one case,
specifically branch folding, the stepBackwards logic was having an
influence on code generation.  The result was that certain code
compiled with '-g -O2' would differ from that compiled with '-O2'
alone. It seems that the original logic, accounting for DBG_VALUE,
was influencing the placement of an IMPLICIT_DEF which had a later
impact on how blocks were processed in branch folding.

Reviewers: kparzysz, MatzeB

Reviewed By: kparzysz

Subscribers: bjope, llvm-commits

Tags: #debug-info

Differential Revision: https://reviews.llvm.org/D43850

llvm-svn: 327862

6 years ago[clang-format] Remove empty lines before }[;] // comment
Krasimir Georgiev [Mon, 19 Mar 2018 15:33:40 +0000 (15:33 +0000)]
[clang-format] Remove empty lines before }[;] // comment

Summary:
This addresses bug 36766 and a FIXME in tests about empty lines before
`}[;] // comment` lines.

Subscribers: klimek, cfe-commits

Differential Revision: https://reviews.llvm.org/D44631

llvm-svn: 327861

6 years ago[CMake] Use libc++ and compiler-rt as default libraries in Fuchsia toolchain
Petr Hosek [Mon, 19 Mar 2018 15:19:19 +0000 (15:19 +0000)]
[CMake] Use libc++ and compiler-rt as default libraries in Fuchsia toolchain

Fuchsia already defaults to libc++ and compiler-rt, but we want to use
these as default runtimes even on the host platform.

Differential Revision: https://reviews.llvm.org/D39930

llvm-svn: 327860

6 years ago[demangler] Recopy the demangler from libcxxabi.
Erik Pilkington [Mon, 19 Mar 2018 15:18:23 +0000 (15:18 +0000)]
[demangler] Recopy the demangler from libcxxabi.

Some significant work has gone into libcxxabi's copy of this file:
  - Uses an AST to represent mangled names.
  - Support/bugfixes for many C++ features.
  - Uses LLVM coding style.

llvm-svn: 327859

6 years ago[InstCombine] canonicalize fcmp+select to fabs
Sanjay Patel [Mon, 19 Mar 2018 15:14:30 +0000 (15:14 +0000)]
[InstCombine] canonicalize fcmp+select to fabs

This is complicated by -0.0 and nan. This is based on the DAG patterns
as shown in D44091. I'm hoping that we can just remove those DAG folds
and always rely on IR canonicalization to handle the matching to fabs.

We would still need to delete the broken code from DAGCombiner to fix
PR36600:
https://bugs.llvm.org/show_bug.cgi?id=36600

Differential Revision: https://reviews.llvm.org/D44550

llvm-svn: 327858

6 years agoMove code generation test case to test/CodeGen/
Tobias Grosser [Mon, 19 Mar 2018 15:05:30 +0000 (15:05 +0000)]
Move code generation test case to test/CodeGen/

llvm-svn: 327857

6 years ago[MachineLICM] Add functions to MachineLICM to hoist invariant stores
Zaara Syeda [Mon, 19 Mar 2018 14:52:25 +0000 (14:52 +0000)]
[MachineLICM] Add functions to MachineLICM to hoist invariant stores

This patch adds functions to allow MachineLICM to hoist invariant stores.
Currently, MachineLICM does not hoist any store instructions, however
when storing the same value to a constant spot on the stack, the store
instruction should be considered invariant and be hoisted. The function
isInvariantStore iterates each operand of the store instruction and checks
that each register operand satisfies isCallerPreservedPhysReg. The store
may be fed by a copy, which is hoisted by isCopyFeedingInvariantStore.
This patch also adds the PowerPC changes needed to consider the stack
register as caller preserved.

Differential Revision: https://reviews.llvm.org/D40196

llvm-svn: 327856

6 years ago[X86] Generalize schedule classes to support multiple stages
Simon Pilgrim [Mon, 19 Mar 2018 14:46:07 +0000 (14:46 +0000)]
[X86] Generalize schedule classes to support multiple stages

Currently the WriteResPair style multi-classes take a single pipeline stage and latency, this patch generalizes this to make it easier to create complex schedules with ResourceCycles and NumMicroOps be overriden from their defaults.

This has already been done for the Jaguar scheduler to remove a number of custom schedule classes and adding it to the other x86 targets will make it much tidier as we add additional classes in the future to try and replace so many custom cases.

I've converted some instructions but a lot of the models need a bit of cleanup after the patch has been committed - memory latencies not being consistent, the class not actually being used when we could remove some/all customs, etc. I'd prefer to keep this as NFC as possible so later patches can be smaller and target specific.

Differential Revision: https://reviews.llvm.org/D44612

llvm-svn: 327855