platform/upstream/llvm.git
7 years ago[CMake] darwin-debug is an hard dependency for tests on macOS.
Davide Italiano [Fri, 15 Dec 2017 23:27:10 +0000 (23:27 +0000)]
[CMake] darwin-debug is an hard dependency for tests on macOS.

Fixes a few failured on the testsuite with CMake.

llvm-svn: 320891

7 years ago[TableGen][GlobalISel] Make the different Matcher comparable
Quentin Colombet [Fri, 15 Dec 2017 23:24:39 +0000 (23:24 +0000)]
[TableGen][GlobalISel] Make the different Matcher comparable

This opens refactoring opportunities in the match table now that we can
check that two predicates are the same.

NFC.

llvm-svn: 320890

7 years ago[TableGen][GlobalISel] Fix unused variable warning in release mode
Quentin Colombet [Fri, 15 Dec 2017 23:24:36 +0000 (23:24 +0000)]
[TableGen][GlobalISel] Fix unused variable warning in release mode

Introduced in r320887.

NFC.

llvm-svn: 320889

7 years agoRevert "Recommit "[DWARFv5] Dump an MD5 checksum in the line-table header.""
Paul Robinson [Fri, 15 Dec 2017 23:21:52 +0000 (23:21 +0000)]
Revert "Recommit "[DWARFv5] Dump an MD5 checksum in the line-table header.""

This reverts commit 0afef672f63f0e4e91938656bc73424a8c058bfc.
Still failing at runtime on bots.

llvm-svn: 320888

7 years ago[TableGen][GlobalISel] Have the predicate directly know which data they are dealing...
Quentin Colombet [Fri, 15 Dec 2017 23:07:42 +0000 (23:07 +0000)]
[TableGen][GlobalISel] Have the predicate directly know which data they are dealing with

Prior to this patch, a predicate wouldn't make sense outside of its
rule. Indeed, it was only during emitting a rule that a predicate would
be made aware of the IDs of the data it is checking. Because of that,
predicates could not be moved around or compared between each other.

NFC.

llvm-svn: 320887

7 years agoRecommit "[DWARFv5] Dump an MD5 checksum in the line-table header."
Paul Robinson [Fri, 15 Dec 2017 22:57:17 +0000 (22:57 +0000)]
Recommit "[DWARFv5] Dump an MD5 checksum in the line-table header."
Adds missing support for DW_FORM_data16.

Update of r320852, fixing the unittest to use a hand-coded struct
instead of std::array to guarantee data layout.

Differential Revision: https://reviews.llvm.org/D41090

llvm-svn: 320886

7 years agoFix unused variable in non-assert builds
Matthias Braun [Fri, 15 Dec 2017 22:53:33 +0000 (22:53 +0000)]
Fix unused variable in non-assert builds

llvm-svn: 320885

7 years agoMachineFunction: Return reference from getFunction(); NFC
Matthias Braun [Fri, 15 Dec 2017 22:22:58 +0000 (22:22 +0000)]
MachineFunction: Return reference from getFunction(); NFC

The Function can never be nullptr so we can return a reference.

llvm-svn: 320884

7 years ago[MacOSX/Queues] Relax an overly aggressive assertion in a test.
Davide Italiano [Fri, 15 Dec 2017 22:22:51 +0000 (22:22 +0000)]
[MacOSX/Queues] Relax an overly aggressive assertion in a test.

"Default" is a valid QoS for a thread on older versions of macOS,
like the one installed in the bot.
Thanks to Jason Molenda for helping me figuring out the problem.

<rdar://problem/28346273>

llvm-svn: 320883

7 years agoMachineFunction: Slight refactoring; NFC
Matthias Braun [Fri, 15 Dec 2017 22:22:46 +0000 (22:22 +0000)]
MachineFunction: Slight refactoring; NFC

Slight cleanup/refactor in preparation for upcoming commit.

llvm-svn: 320882

7 years agoMachineModuleInfo: Remove unused function; NFC
Matthias Braun [Fri, 15 Dec 2017 22:22:42 +0000 (22:22 +0000)]
MachineModuleInfo: Remove unused function; NFC

Remove the unused setModule() function; it would be dangerous if someone
actually used it as it wouldn't reset/recompute various other module
related data.

llvm-svn: 320881

7 years ago[WebAssembly] Don't include lazy symbols in import table
Sam Clegg [Fri, 15 Dec 2017 22:17:15 +0000 (22:17 +0000)]
[WebAssembly] Don't include lazy symbols in import table

This bug was introduced in: https://reviews.llvm.org/D41304.
Add a test for this case.

Differential Revision: https://reviews.llvm.org/D41309

llvm-svn: 320872

7 years agoFixed the gcc 'enumeral and non-enumeral type in conditional expression [-Werror...
Galina Kistanova [Fri, 15 Dec 2017 22:15:29 +0000 (22:15 +0000)]
Fixed the gcc 'enumeral and non-enumeral type in conditional expression [-Werror=extra]' warning introduced by r320750

llvm-svn: 320868

7 years ago[Hexagon] Remove recursion in visitUsesOf, replace with use queue
Krzysztof Parzyszek [Fri, 15 Dec 2017 21:34:05 +0000 (21:34 +0000)]
[Hexagon] Remove recursion in visitUsesOf, replace with use queue

This is primarily to reduce stack usage, but ordering the use queue
according to the position in the code (earlier instructions visited
before later ones) reduces the number of unnecessary bottoms due to
visiting instructions out of order, e.g.
  %reg1 = copy %reg0
  %reg2 = copy %reg0
  %reg3 = and %reg1, %reg2
Here, reg3 should be known to be same as reg0-2, but if reg3 is
evaluated after reg1 is updated, but before reg2 is updated, the two
inputs to the and will appear different, causing reg3 to become
bottom.

llvm-svn: 320866

7 years ago[Hexagon] Handle concat_vectors of all allowed HVX types
Krzysztof Parzyszek [Fri, 15 Dec 2017 21:23:12 +0000 (21:23 +0000)]
[Hexagon] Handle concat_vectors of all allowed HVX types

llvm-svn: 320865

7 years ago[X86] Use AND32ri8 instead of AND64ri8 in Asan code in EmitCallAsanReport for 32...
Craig Topper [Fri, 15 Dec 2017 21:18:06 +0000 (21:18 +0000)]
[X86] Use AND32ri8 instead of AND64ri8 in Asan code in EmitCallAsanReport for 32-bit mode.

This seemed to work due to a quirk in the X86 MC encoder that didn't emit a REX byte that the AND64ri8 implies when in 32-bit mode. This made the encoding the same as AND32ri8. I tried to add an assert to catch the dropped REX prefix that caught this.

llvm-svn: 320864

7 years ago[X86] In LowerVectorCTPOP use ISD::ZERO_EXTEND/ISD::TRUNCATE instead of the target...
Craig Topper [Fri, 15 Dec 2017 21:18:05 +0000 (21:18 +0000)]
[X86] In LowerVectorCTPOP use ISD::ZERO_EXTEND/ISD::TRUNCATE instead of the target specific nodes.

The target independent nodes will get legalized to the target specific nodes by their own legalization process. Someday I'd like to stop using a target specific for zero extends and truncates of legal types so the less places we reference the target specific opcode the better.

llvm-svn: 320863

7 years ago[X86] Remove unnecessary TODO.
Craig Topper [Fri, 15 Dec 2017 20:57:18 +0000 (20:57 +0000)]
[X86] Remove unnecessary TODO.

When I wrote it I thought we were missing a potential optimization for KNL. But investigating further shows that for KNL we still do the optimal thing by widening to v4f32 and then using special isel patterns to widen again to zmm a register.

llvm-svn: 320862

7 years ago[MinGW] Ignore the --no-seh flag
Martin Storsjo [Fri, 15 Dec 2017 20:53:10 +0000 (20:53 +0000)]
[MinGW] Ignore the --no-seh flag

The COFF linker automatically sets the IMAGE_DLL_CHARACTERISTICS_NO_SEH
when suitable, similarly to link.exe.

Differential Revision: https://reviews.llvm.org/D41275

llvm-svn: 320861

7 years ago[COFF] Set the IMAGE_DLL_CHARACTERISTICS_NO_SEH flag automatically
Martin Storsjo [Fri, 15 Dec 2017 20:53:03 +0000 (20:53 +0000)]
[COFF] Set the IMAGE_DLL_CHARACTERISTICS_NO_SEH flag automatically

This seems to match how link.exe sets it.

Differential Revision: https://reviews.llvm.org/D41252

llvm-svn: 320860

7 years ago[LTO] Remove unused RegularLTOState::HasModule
Vitaly Buka [Fri, 15 Dec 2017 20:50:25 +0000 (20:50 +0000)]
[LTO] Remove unused RegularLTOState::HasModule

llvm-svn: 320859

7 years agoRe-commit : [LICM] Allow sinking when foldable in loop
Jun Bum Lim [Fri, 15 Dec 2017 20:33:24 +0000 (20:33 +0000)]
Re-commit : [LICM] Allow sinking when foldable in loop

This recommits r320823 reverted due to the test failure in sink-foldable.ll and
an unused variable. Added "REQUIRES: aarch64-registered-target" in the test
and removed unused variable.

Original commit message:

  Continue trying to sink an instruction if its users in the loop is foldable.
  This will allow the instruction to be folded in the loop by decoupling it from
  the user outside of the loop.

  Reviewers: hfinkel, majnemer, davidxl, efriedma, danielcdh, bmakam, mcrosier

  Reviewed By: hfinkel

  Subscribers: javed.absar, bmakam, mcrosier, llvm-commits

  Differential Revision: https://reviews.llvm.org/D37076

llvm-svn: 320858

7 years agoRevert "[DWARFv5] Dump an MD5 checksum in the line-table header."
Paul Robinson [Fri, 15 Dec 2017 20:29:25 +0000 (20:29 +0000)]
Revert "[DWARFv5] Dump an MD5 checksum in the line-table header."

Unit test fails on some bots.

llvm-svn: 320857

7 years ago[llvm-objcopy] Reformat everything using clang-format -i
Jake Ehrlich [Fri, 15 Dec 2017 20:17:55 +0000 (20:17 +0000)]
[llvm-objcopy] Reformat everything using clang-format -i

Overtime some non-clang formatted code has creeped into llvm-objcopy. This
patch fixes all of that.

Differential Revision: https://reviews.llvm.org/D41262

llvm-svn: 320856

7 years ago[Hexagon] Fix operand-swapping PatFrag for atomic stores
Krzysztof Parzyszek [Fri, 15 Dec 2017 20:13:57 +0000 (20:13 +0000)]
[Hexagon] Fix operand-swapping PatFrag for atomic stores

PatFrag now has the atomicity information stored as bit fields. They
need to be copied to the new PatFrag.

llvm-svn: 320855

7 years ago__is_target_environment: Check the environment after parsing it
Alex Lorenz [Fri, 15 Dec 2017 20:07:53 +0000 (20:07 +0000)]
__is_target_environment: Check the environment after parsing it

This ensures that target triples with environment versions can still work with
__is_target_environment.

llvm-svn: 320854

7 years ago__is_target_arch: Check the arch and subarch instead of the arch name
Alex Lorenz [Fri, 15 Dec 2017 19:58:38 +0000 (19:58 +0000)]
__is_target_arch: Check the arch and subarch instead of the arch name

This ensures that when compiling for "arm64" __is_target_arch will succeed for
both "arm64" and "aarch64".

Thanks to Bob Wilson who pointed this out!

llvm-svn: 320853

7 years ago[DWARFv5] Dump an MD5 checksum in the line-table header.
Paul Robinson [Fri, 15 Dec 2017 19:52:34 +0000 (19:52 +0000)]
[DWARFv5] Dump an MD5 checksum in the line-table header.
Adds missing support for DW_FORM_data16.

Differential Revision: https://reviews.llvm.org/D41090

llvm-svn: 320852

7 years agoDo not add .dynamic entries that don't have to be added early in the ctor. NFC.
Rui Ueyama [Fri, 15 Dec 2017 19:39:59 +0000 (19:39 +0000)]
Do not add .dynamic entries that don't have to be added early in the ctor. NFC.

We add dynamic section entries both in the ctor of the class and
DynamicSection::finalizeContents(). Some entries need to be added early
in the ctor because they add strings to .dynstr. Other entries were
intended to be added in finalizeContents(). However, some entries are
added in the ctor even though they don't add strings. This patch
fix the issue.

llvm-svn: 320851

7 years ago[X86] Remove assert in X86MCCodeEmitter.cpp that was added in r320830.
Craig Topper [Fri, 15 Dec 2017 19:38:14 +0000 (19:38 +0000)]
[X86] Remove assert in X86MCCodeEmitter.cpp that was added in r320830.

It seems to be failing real code which is concerning, but we were silently getting away with it. I'll investigate further.

llvm-svn: 320850

7 years ago[SelectionDAG][X86] Fix insert_vector_elt lowering for v32i1/v64i1 with non-constant...
Craig Topper [Fri, 15 Dec 2017 19:35:22 +0000 (19:35 +0000)]
[SelectionDAG][X86] Fix insert_vector_elt lowering for v32i1/v64i1 with non-constant index

Summary:
Currently we don't handle v32i1/v64i1 insert_vector_elt correctly as we fail to look at the number of elements closely and assume it can only be v16i1 or v8i1.

We also can't type legalize v64i1 insert_vector_elt correctly on KNL due to the type not being byte addressable as required by the legalizing through memory accesses path requires.

For the first issue, the patch now tries to pick a 512-bit register with the correct number of elements and promotes to that.

For the second issue, we now extend the vector to a byte addressable type, do the stores to memory, load the two halves, and then truncate the halves back to the original type. Technically since we changed the type, we may not need two loads, but actually checking that is more work and for the v64i1 case we do need them.

Reviewers: RKSimon, delena, spatel, zvi

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D40942

llvm-svn: 320849

7 years ago[Memcpy Loop Lowering] Insert loop BB inbetween the split BB.
Sean Fertile [Fri, 15 Dec 2017 19:29:12 +0000 (19:29 +0000)]
[Memcpy Loop Lowering] Insert loop BB inbetween the split BB.

The original memcpy expansion inserted the loop basic block inbetween
the 2 new basic blocks created by splitting the original block the memcpy
call was in. This commit makes the new memcpy expansion do the same to keep the
layout of the IR matching between the old and new implementations.

Differential Review: https://reviews.llvm.org/D41197

llvm-svn: 320848

7 years ago[WebAssembly] Base imports on Symtab. NFC.
Sam Clegg [Fri, 15 Dec 2017 19:23:49 +0000 (19:23 +0000)]
[WebAssembly] Base imports on Symtab. NFC.

Since imports are undefined symbols we know we can
find all of them my looking at the symbol table alone.
(i.e. imports cannot be have local binding).

This will be strictly faster and also allows us
to to remove a method from Symbol class

Differential Revision: https://reviews.llvm.org/D41304

llvm-svn: 320847

7 years ago[X86] Add 'Requires<[In64BitMode]>' to a bunch of instructions that only have memory...
Craig Topper [Fri, 15 Dec 2017 19:01:51 +0000 (19:01 +0000)]
[X86] Add 'Requires<[In64BitMode]>' to a bunch of instructions that only have memory and immediate operands.

The asm parser wasn't preventing these from being accepted in 32-bit mode. Instructions that use a GR64 register are protected by the parser rejecting the register in 32-bit mode.

llvm-svn: 320846

7 years ago[X86] Change BNDLDX to use anymem instead of i64mem for itsmemory operand.
Craig Topper [Fri, 15 Dec 2017 19:01:50 +0000 (19:01 +0000)]
[X86] Change BNDLDX to use anymem instead of i64mem for itsmemory operand.

This instruction doesn't access memory. It juse use a similar looking memory encoding. Don't require Intel syntax to put "qword ptr" in front of it.

llvm-svn: 320845

7 years ago[X86] Remove the 'Requires' In64BitMode/Not64BitMode from the LWP instructions.
Craig Topper [Fri, 15 Dec 2017 19:01:49 +0000 (19:01 +0000)]
[X86] Remove the 'Requires' In64BitMode/Not64BitMode from the LWP instructions.

These aren't doing anything due to a top level "let Predicates =". I think the GR32/GR64 register class protects these anyway.

llvm-svn: 320844

7 years ago[X86] Remove the 'Requires<[In64BitMode]>' from SHSTK instructions.
Craig Topper [Fri, 15 Dec 2017 19:01:48 +0000 (19:01 +0000)]
[X86] Remove the 'Requires<[In64BitMode]>' from SHSTK instructions.

This has no effect due to a top level "let Predicates =" around the instructions. But its also not required because the GR64 usage in the instruction guarantees it can never match.

llvm-svn: 320843

7 years ago[TargetLibraryInfo] fix documentation comment; NFC
Sanjay Patel [Fri, 15 Dec 2017 18:54:29 +0000 (18:54 +0000)]
[TargetLibraryInfo] fix documentation comment; NFC

llvm-svn: 320842

7 years ago[clang-tidy] Adding Fuchsia checker for virtual inheritance
Julie Hockett [Fri, 15 Dec 2017 18:54:28 +0000 (18:54 +0000)]
[clang-tidy] Adding Fuchsia checker for virtual inheritance

Adds a check to the Fuchsia module to warn if classes are defined
with virtual inheritance.

See https://fuchsia.googlesource.com/zircon/+/master/docs/cxx.md for
reference.

Differential Revision: https://reviews.llvm.org/D40813

llvm-svn: 320841

7 years ago[CodeGen] fix documentation comments; NFC
Sanjay Patel [Fri, 15 Dec 2017 18:34:45 +0000 (18:34 +0000)]
[CodeGen] fix documentation comments; NFC

llvm-svn: 320840

7 years ago[AArch64] Fix typo in the ASIMD instruction optimization pass
Evandro Menezes [Fri, 15 Dec 2017 18:26:54 +0000 (18:26 +0000)]
[AArch64] Fix typo in the ASIMD instruction optimization pass

Fix typo in the representative instruction replacement.

Also, fix formatting and reword some comments.

llvm-svn: 320839

7 years agofix typo in comment and remove inaccurate comment; NFC
Sanjay Patel [Fri, 15 Dec 2017 18:25:13 +0000 (18:25 +0000)]
fix typo in comment and remove inaccurate comment; NFC

llvm-svn: 320838

7 years agoFix for bug PR35549 - Repeated schedule comments.
Andrew V. Tischenko [Fri, 15 Dec 2017 18:13:05 +0000 (18:13 +0000)]
Fix for bug PR35549 - Repeated schedule comments.
Differential Revision: https://reviews.llvm.org/D40960

llvm-svn: 320837

7 years agoRevert "Re-commit : [LICM] Allow sinking when foldable in loop"
Jun Bum Lim [Fri, 15 Dec 2017 18:12:49 +0000 (18:12 +0000)]
Revert "Re-commit : [LICM] Allow sinking when foldable in loop"

This reverts commit r320833.

llvm-svn: 320836

7 years ago[CodeGen] fix documentation comments; NFC
Sanjay Patel [Fri, 15 Dec 2017 18:09:33 +0000 (18:09 +0000)]
[CodeGen] fix documentation comments; NFC

llvm-svn: 320835

7 years agoRe-commit : [LICM] Allow sinking when foldable in loop
Jun Bum Lim [Fri, 15 Dec 2017 17:58:59 +0000 (17:58 +0000)]
Re-commit : [LICM] Allow sinking when foldable in loop

 This recommit r320823 after fixing a test failure.

 Original commit message:

    Continue trying to sink an instruction if its users in the loop is foldable.
    This will allow the instruction to be folded in the loop by decoupling it from
    the user outside of the loop.

    Reviewers: hfinkel, majnemer, davidxl, efriedma, danielcdh, bmakam, mcrosier

    Reviewed By: hfinkel

    Subscribers: javed.absar, bmakam, mcrosier, llvm-commits

    Differential Revision: https://reviews.llvm.org/D37076

llvm-svn: 320833

7 years agoUpdated llvm-objdump to display local relocations in Mach-O binaries
Michael Trent [Fri, 15 Dec 2017 17:57:40 +0000 (17:57 +0000)]
Updated llvm-objdump to display local relocations in Mach-O binaries

Summary:
llvm-objdump's Mach-O parser was updated in r306037 to display external
relocations for MH_KEXT_BUNDLE file types. This change extends the Macho-O
parser to display local relocations for MH_PRELOAD files. When used with
the -macho option relocations will be displayed in a historical format.

All tests are passing for llvm, clang, and lld. llvm-objdump builds without
compiler warnings.

rdar://35778019

Reviewers: enderby

Reviewed By: enderby

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D41199

llvm-svn: 320832

7 years ago[ubsan-minimal] Add a path for non-POSIX (and bare-metal) use of the library
Filipe Cabecinhas [Fri, 15 Dec 2017 17:30:50 +0000 (17:30 +0000)]
[ubsan-minimal] Add a path for non-POSIX (and bare-metal) use of the library

Summary:
Hook on -DKERNEL_USE (which is also used in lib/builtins) to not import
strlen and not rely on write() being implemented with the stderr on fd 2.

With this, the only requirements to use this library are:
  - "Good enough" std::atomic<void*> and std::atomic<int>
  - abort() being implemented
  - ubsan_message(const char*) being implemented

Reviewers: eugenis

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D39791

llvm-svn: 320831

7 years ago[X86] Fix XSAVE64 and similar instructions to not be allowed by the assembler in...
Craig Topper [Fri, 15 Dec 2017 17:22:58 +0000 (17:22 +0000)]
[X86] Fix XSAVE64 and similar instructions to not be allowed by the assembler in 32-bit mode.

There was a top level "let Predicates =" in the .td file that was overriding the Requires on each instruction.

I've added an assert to the code emitter to catch more cases like this. I'm sure this isn't the only place where the right predicates aren't being applied. This assert already found that we don't block btq/btsq/btrq in 32-bit mode.

llvm-svn: 320830

7 years agoRemove "FunctionName -" from docs on FunctionDecl(NFC)
Erich Keane [Fri, 15 Dec 2017 16:37:14 +0000 (16:37 +0000)]
Remove "FunctionName -" from docs on FunctionDecl(NFC)

Removed the repetative usage of the operator name on the
documentation for FunctionDecl.  Also reflowed some of the
comments since this changes the 80 character rule.

llvm-svn: 320829

7 years agoRevert "[LICM] Allow sinking when foldable in loop"
Jun Bum Lim [Fri, 15 Dec 2017 16:35:09 +0000 (16:35 +0000)]
Revert "[LICM] Allow sinking when foldable in loop"

This reverts commit r320823.

llvm-svn: 320828

7 years ago[CodeGen] Print stack object references as %(fixed-)stack.0 in both MIR and debug...
Francis Visoiu Mistrih [Fri, 15 Dec 2017 16:33:45 +0000 (16:33 +0000)]
[CodeGen] Print stack object references as %(fixed-)stack.0 in both MIR and debug output

Work towards the unification of MIR and debug output by printing
`%stack.0` instead of `<fi#0>`, and `%fixed-stack.0` instead of
`<fi#-4>` (supposing there are 4 fixed stack objects).

Only debug syntax is affected.

Differential Revision: https://reviews.llvm.org/D41027

llvm-svn: 320827

7 years ago[OPENMP] Codegen `declare simd` for function declarations.
Alexey Bataev [Fri, 15 Dec 2017 16:28:31 +0000 (16:28 +0000)]
[OPENMP] Codegen `declare simd` for function declarations.

Previously the attributes were emitted only for function definitions.
Patch adds emission of the attributes for function declarations.

llvm-svn: 320826

7 years ago[ThinLTO] Disallow multiple prevailing defs
Eugene Leviant [Fri, 15 Dec 2017 16:27:33 +0000 (16:27 +0000)]
[ThinLTO] Disallow multiple prevailing defs

https://reviews.llvm.org/D41291

llvm-svn: 320825

7 years ago[X86] Widen (v2i32 (fp_to_uint v2f64)) to (v8i32 (fp_to_uint v8f64)) during legalizat...
Craig Topper [Fri, 15 Dec 2017 16:22:20 +0000 (16:22 +0000)]
[X86] Widen (v2i32 (fp_to_uint v2f64)) to (v8i32 (fp_to_uint v8f64)) during legalization if we have AVX512F, but not VLX. NFC

Previously we widened it using isel patterns.

llvm-svn: 320824

7 years ago[LICM] Allow sinking when foldable in loop
Jun Bum Lim [Fri, 15 Dec 2017 16:09:54 +0000 (16:09 +0000)]
[LICM] Allow sinking when foldable in loop

Summary:
Continue trying to sink an instruction if its users in the loop is foldable.
This will allow the instruction to be folded in the loop by decoupling it from
the user outside of the loop.

Reviewers: hfinkel, majnemer, davidxl, efriedma, danielcdh, bmakam, mcrosier

Reviewed By: hfinkel

Subscribers: javed.absar, bmakam, mcrosier, llvm-commits

Differential Revision: https://reviews.llvm.org/D37076

llvm-svn: 320823

7 years ago[ARM] Some DAG combine tests
Sam Parker [Fri, 15 Dec 2017 15:30:39 +0000 (15:30 +0000)]
[ARM] Some DAG combine tests

Add some more and and shift load combine tests.

llvm-svn: 320822

7 years ago[ELF] Reduce size of ELF file produced by test
Peter Smith [Fri, 15 Dec 2017 15:30:00 +0000 (15:30 +0000)]
[ELF] Reduce size of ELF file produced by test

We only need to exceed 128 Megabytes to provoke the generation of a range
extension thunk. This brings the file size down to just over 128 Megabytes.

llvm-svn: 320821

7 years agollgs-tests: Add support for "exit" stop-reply packets
Pavel Labath [Fri, 15 Dec 2017 15:19:45 +0000 (15:19 +0000)]
llgs-tests: Add support for "exit" stop-reply packets

Summary:
This makes StopReply class abstract, so that we can represent different
types of stop replies such as StopReplyStop and StopReplyExit (there
should also be a StopReplySignal, but I don't need that right now so I
haven't implemented it yet).

This prepares the ground for a new test I'm writing.

Reviewers: eugene, zturner

Subscribers: lldb-commits

Differential Revision: https://reviews.llvm.org/D41067

llvm-svn: 320820

7 years ago[MIR] Add support for missing CFI directives
Francis Visoiu Mistrih [Fri, 15 Dec 2017 15:17:18 +0000 (15:17 +0000)]
[MIR] Add support for missing CFI directives

The following CFI directives are suported by MC but not by MIR:

* .cfi_rel_offset
* .cfi_adjust_cfa_offset
* .cfi_escape
* .cfi_remember_state
* .cfi_restore_state
* .cfi_undefined
* .cfi_register
* .cfi_window_save

Add support for printing, parsing and update tests.

Differential Revision: https://reviews.llvm.org/D41230

llvm-svn: 320819

7 years agoDelete a really large test output.
Rafael Espindola [Fri, 15 Dec 2017 14:56:01 +0000 (14:56 +0000)]
Delete a really large test output.

llvm-svn: 320818

7 years agoHandle a VersymIndex of 0 as an error.
Rafael Espindola [Fri, 15 Dec 2017 14:52:40 +0000 (14:52 +0000)]
Handle a VersymIndex of 0 as an error.

I noticed that the continue this patch deletes was not tested. Trying
to add a test I realized that we never put a VER_NDX_LOCAL symbol in
the dynamic symbol table. There doesn't seem to be any reason for a
linker to use VER_NDX_LOCAL for a defined shared symbol.

llvm-svn: 320817

7 years agoFix 32-bit builds broken by 320813
Pavel Labath [Fri, 15 Dec 2017 14:39:12 +0000 (14:39 +0000)]
Fix 32-bit builds broken by 320813

cast to size_t to avoid narrowing error.

llvm-svn: 320816

7 years ago[X86] Add RTM schedule tests
Simon Pilgrim [Fri, 15 Dec 2017 14:37:28 +0000 (14:37 +0000)]
[X86] Add RTM schedule tests

llvm-svn: 320815

7 years ago[InlineCost] Find repeated loads in the callee
Haicheng Wu [Fri, 15 Dec 2017 14:34:41 +0000 (14:34 +0000)]
[InlineCost] Find repeated loads in the callee

SROA analysis of InlineCost can figure out that some stores can be removed
after inlining and then the repeated loads clobbered by these stores are also
free.  This patch finds these clobbered loads and adjust the inline cost
accordingly.

Differential Revision: https://reviews.llvm.org/D33946

llvm-svn: 320814

7 years agoObjectFileELF: Add support for compressed sections
Pavel Labath [Fri, 15 Dec 2017 14:23:58 +0000 (14:23 +0000)]
ObjectFileELF: Add support for compressed sections

Summary:
We use the llvm decompressor to decompress SHF_COMPRESSED sections. This enables
us to read data from debug info sections, which are sometimes compressed,
particuarly in the split-dwarf case.  This functionality is only available if
llvm is compiled with zlib support.

Reviewers: clayborg, zturner

Subscribers: emaste, mgorny, aprantl, lldb-commits

Differential Revision: https://reviews.llvm.org/D40616

llvm-svn: 320813

7 years ago[X86] Add MWAITX/MONITORX schedule tests
Simon Pilgrim [Fri, 15 Dec 2017 14:22:15 +0000 (14:22 +0000)]
[X86] Add MWAITX/MONITORX schedule tests

llvm-svn: 320812

7 years agoFix the second build bot break introduced by r320791.
Nemanja Ivanovic [Fri, 15 Dec 2017 14:17:45 +0000 (14:17 +0000)]
Fix the second build bot break introduced by r320791.

llvm-svn: 320811

7 years ago[X86] Add XOP schedule tests
Simon Pilgrim [Fri, 15 Dec 2017 14:02:35 +0000 (14:02 +0000)]
[X86] Add XOP schedule tests

llvm-svn: 320810

7 years agollgs-tests: Make addition of new tests easier
Pavel Labath [Fri, 15 Dec 2017 13:56:22 +0000 (13:56 +0000)]
llgs-tests: Make addition of new tests easier

Summary:
Adding a new test would require one to duplicate a significant part of
the existing test that we have. This attempts to reduce that by moving
some part of that code to the test fixture. The StandardStartupTest
fixture automatically starts up the server and connects it to the
client. I also add a more low-level TestBase fixture, which allows one
to start up the client and server in a custom way (I am going to need
this for the test I am writing).

Reviewers: eugene, zturner

Subscribers: lldb-commits, mgorny

Differential Revision: https://reviews.llvm.org/D41066

llvm-svn: 320809

7 years agoAdded a separate install target for compilert-rt-headers
Ilya Biryukov [Fri, 15 Dec 2017 13:20:13 +0000 (13:20 +0000)]
Added a separate install target for compilert-rt-headers

Summary:
This patch adds a new install target `install-compilert-rt-headers`,
that is similar to `install-clang-headers`.

It allows to install the headers without installing all of
compiler-rt.

Reviewers: alekseyshl, beanz

Reviewed By: beanz

Subscribers: smeenai, beanz, mgorny, #sanitizers, kcc, llvm-commits

Differential Revision: https://reviews.llvm.org/D41244

llvm-svn: 320808

7 years ago[clangd] Build in-memory index on symbols in files.
Eric Liu [Fri, 15 Dec 2017 12:25:02 +0000 (12:25 +0000)]
[clangd] Build in-memory index on symbols in files.

Reviewers: sammccall

Reviewed By: sammccall

Subscribers: klimek, mgorny, ilya-biryukov, cfe-commits

Differential Revision: https://reviews.llvm.org/D41276

llvm-svn: 320807

7 years agoFix code causing fallthrough warnings in the PPC back end.
Nemanja Ivanovic [Fri, 15 Dec 2017 11:47:48 +0000 (11:47 +0000)]
Fix code causing fallthrough warnings in the PPC back end.

llvm-svn: 320806

7 years ago[X86] Add AVX512 VPOPCNTDQ schedule tests
Simon Pilgrim [Fri, 15 Dec 2017 11:32:31 +0000 (11:32 +0000)]
[X86] Add AVX512 VPOPCNTDQ schedule tests

Demonstrates how to perform full coverage avx512 schedule tests

llvm-svn: 320805

7 years ago[clang] Add PPCallbacks list to preprocessor when building a preacompiled preamble.
Ilya Biryukov [Fri, 15 Dec 2017 11:27:51 +0000 (11:27 +0000)]
[clang] Add PPCallbacks list to preprocessor when building a preacompiled preamble.

Summary:
Revision D38639 needs this commit in order to properly make open
definition calls on include statements work.

Patch by William Enright.

Reviewers: malaperle, krasimir, bkramer, ilya-biryukov

Reviewed By: malaperle, ilya-biryukov

Subscribers: cfe-commits, arphaman, ilya-biryukov

Differential Revision: https://reviews.llvm.org/D39375

llvm-svn: 320804

7 years ago[ELF] Remove Duplicate .ARM.exidx sections
Peter Smith [Fri, 15 Dec 2017 11:09:41 +0000 (11:09 +0000)]
[ELF] Remove Duplicate .ARM.exidx sections

The ARM.exidx section contains a table of 8-byte entries with the first
word of each entry an offset to the function it describes and the second
word instructions for unwinding if an exception is thrown from that
function. The SHF_LINK_ORDER processing will order the table in ascending
order of the functions described by the exception table entries. As the
address range of an exception table entry is terminated by the next table
entry, it is possible to merge consecutive table entries that have
identical unwind instructions.

For this implementation we define a table entry to be identical if:
- Both entries are the special EXIDX_CANTUNWIND.
- Both entries have the same inline unwind instructions.
We do not attempt to establish if table entries that are references to
.ARM.extab sections are identical.

This implementation works at a granularity of a single .ARM.exidx
InputSection. If all entries in the InputSection are identical to the
previous table entry we can remove the InputSection. A more sophisticated
but more complex implementation would rewrite InputSection contents so that
duplicates within a .ARM.exidx InputSection can be merged.

Differential Revision: https://reviews.llvm.org/D40967

llvm-svn: 320803

7 years ago[ELF] Add missing test from r320800
Peter Smith [Fri, 15 Dec 2017 11:02:50 +0000 (11:02 +0000)]
[ELF] Add missing test from r320800

Forgot to svn add the additional test for the commit. Adding with this one.

llvm-svn: 320802

7 years ago[ELF] Fix buildbot warning error.
Peter Smith [Fri, 15 Dec 2017 10:47:44 +0000 (10:47 +0000)]
[ELF] Fix buildbot warning error.

Forward declare InputSectionDescription as struct not class.

llvm-svn: 320801

7 years ago[ELF] Complete implementation of --fix-cortex-a53-843419
Peter Smith [Fri, 15 Dec 2017 10:32:34 +0000 (10:32 +0000)]
[ELF] Complete implementation of --fix-cortex-a53-843419

This patch provides the mechanism to fix instances of the instruction
sequence that may trigger the cortex-a53 843419 erratum. The fix is
provided by an alternative instruction sequence to remove one of the
erratum conditions. To reach this alternative instruction sequence we
replace the original instruction with a branch to the alternative
sequence. The alternative sequence is responsible for branching back to
the original.

As there is only erratum to fix the implementation is specific to
AArch64 and the specific erratum conditions. It should be generalizable
to other targets and erratum if needed.

Differential Revision: https://reviews.llvm.org/D36749

llvm-svn: 320800

7 years ago[RISCV] Change shift amount operand of RVC shift instructions to uimmlog2xlennonzero
Alex Bradbury [Fri, 15 Dec 2017 10:20:51 +0000 (10:20 +0000)]
[RISCV] Change shift amount operand of RVC shift instructions to uimmlog2xlennonzero

c.slli/c.srli/c.srai allow a 5-bit shift in RV32C and a 6-bit shift in RV64C.
This patch adds uimmlog2xlennonzero to reflect this constraint as well as
tests.

Differential Revision: https://reviews.llvm.org/D41216

Patch by Shiva Chen.

llvm-svn: 320799

7 years agoFix the build bot break introduced by r320791.
Nemanja Ivanovic [Fri, 15 Dec 2017 09:51:34 +0000 (09:51 +0000)]
Fix the build bot break introduced by r320791.

llvm-svn: 320798

7 years ago[RISCV] Enable emission of alias instructions by default
Alex Bradbury [Fri, 15 Dec 2017 09:47:01 +0000 (09:47 +0000)]
[RISCV] Enable emission of alias instructions by default

This patch switches the default for -riscv-no-aliases to false
and updates all affected MC and CodeGen tests. As recommended in
D41071, MC tests use the canonical instructions and the CodeGen
tests use the aliases.

Additionally, for the f and d instructions with rounding mode,
the tests for the aliased versions are moved and tightened such
that they can actually detect if alias emission is enabled.
(see D40902 for context)

Differential Revision: https://reviews.llvm.org/D41225

Patch by Mario Werner.

llvm-svn: 320797

7 years ago[PM] port Rewrite Statepoints For GC to the new pass manager.
Fedor Sergeev [Fri, 15 Dec 2017 09:32:11 +0000 (09:32 +0000)]
[PM] port Rewrite Statepoints For GC to the new pass manager.

Summary:
The port is nearly straightforward.
The only complication is related to the analyses handling,
since one of the analyses used in this module pass is domtree,
which is a function analysis. That requires asking for the results
of each function and disallows a single interface for run-on-module
pass action.

Decided to copy-paste the main body of this pass.
Most of its code is requesting analyses anyway, so not that much
of a copy-paste.

The rest of the code movement is to transform all the implementation
helper functions like stripNonValidData into non-member statics.

Extended all the related LLVM tests with new-pass-manager use.
No failures.

Reviewers: sanjoy, anna, reames

Reviewed By: anna

Subscribers: skatkov, llvm-commits

Differential Revision: https://reviews.llvm.org/D41162

llvm-svn: 320796

7 years ago[ARM] Add tests for D34515
Roger Ferrer Ibanez [Fri, 15 Dec 2017 09:24:46 +0000 (09:24 +0000)]
[ARM] Add tests for D34515

This is NFC and a preparatory step for D34515.

Differential Revision: https://reviews.llvm.org/D41122

llvm-svn: 320795

7 years ago[LLVMgold] Don't set undefined symbol as prevailing
Eugene Leviant [Fri, 15 Dec 2017 09:18:21 +0000 (09:18 +0000)]
[LLVMgold] Don't set undefined symbol as prevailing

Differential revision: https://reviews.llvm.org/D41113

llvm-svn: 320794

7 years agoRepair Windows buildbots after r320792
Shoaib Meenai [Fri, 15 Dec 2017 08:08:26 +0000 (08:08 +0000)]
Repair Windows buildbots after r320792

Windows paths have colons in them, so the regex will fail there. Just
match for any character; the rest of the message will restrict the match
to the path anyway.

llvm-svn: 320793

7 years ago[COFF] Warn for locally imported symbols
Shoaib Meenai [Fri, 15 Dec 2017 07:49:21 +0000 (07:49 +0000)]
[COFF] Warn for locally imported symbols

Locally imported symbols are a very surprising linker feature. link.exe
warns for them, and we should warn too.

Differential Revision: https://reviews.llvm.org/D41269

llvm-svn: 320792

7 years ago[PowerPC] Convert r+r instructions to r+i (pre and post RA)
Nemanja Ivanovic [Fri, 15 Dec 2017 07:27:53 +0000 (07:27 +0000)]
[PowerPC] Convert r+r instructions to r+i (pre and post RA)

This patch adds the necessary infrastructure to convert instructions that
take two register operands to those that take a register and immediate if
the necessary operand is produced by a load-immediate. Furthermore, it uses
this infrastructure to perform such conversions twice - first at MachineSSA
and then pre-emit.

There are a number of reasons we may end up with opportunities for this
transformation, including but not limited to:
- X-Form instructions chosen since the exact offset isn't available at ISEL time
- Atomic instructions with constant operands (we will add patterns for this
  in the future)
- Tail duplication may duplicate code where one block contains this redundancy
- When emitting compare-free code in PPCDAGToDAGISel, we don't handle constant
  comparands specially

Furthermore, this patch moves the initialization of PPCMIPeepholePass so that
it can be used for MIR tests.

llvm-svn: 320791

7 years ago[X86] Fix a couple bugs in my recent changes to vXi1 insert_subvector lowering.
Craig Topper [Fri, 15 Dec 2017 07:16:41 +0000 (07:16 +0000)]
[X86] Fix a couple bugs in my recent changes to vXi1 insert_subvector lowering.

A couple places didn't use the same SDValue variables to connect everything all the way through.

I don't have a test case for a bug in insert into the lower bits of a non-zero, non-undef vector. Not sure the best way to create that. We don't create the case when lowering concat_vectors which is the main way to get insert_subvectors.

llvm-svn: 320790

7 years ago[SCEV] Fix the movement of insertion point in expander. PR35406.
Serguei Katkov [Fri, 15 Dec 2017 05:24:42 +0000 (05:24 +0000)]
[SCEV] Fix the movement of insertion point in expander. PR35406.

We cannot move the insertion point to header if SCEV contains div/rem
operations due to they may go over check for zero denominator.

Reviewers: sanjoy, mkazantsev, sebpop
Reviewed By: sebpop
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D41229

llvm-svn: 320789

7 years agoRecommit CodeGen: Fix assertion in machine inst sheduler due to llvm.dbg.value
Yaxun Liu [Fri, 15 Dec 2017 03:56:57 +0000 (03:56 +0000)]
Recommit CodeGen: Fix assertion in machine inst sheduler due to llvm.dbg.value

The regression on ppc64 was not due to this commit.

llvm-svn: 320788

7 years ago[libFuzzer] make the tests depend on ubsan
Kostya Serebryany [Fri, 15 Dec 2017 01:39:52 +0000 (01:39 +0000)]
[libFuzzer] make the tests depend on ubsan

llvm-svn: 320787

7 years agoDisabling r312514 as it causes miscompiles that show up on bootstrap
Nemanja Ivanovic [Fri, 15 Dec 2017 01:38:03 +0000 (01:38 +0000)]
Disabling r312514 as it causes miscompiles that show up on bootstrap

The compare elimination peephole introduced in https://reviews.llvm.org/rL312514
causes a miscompile in AMDGPUInstrInfo.cpp which in turn causes some AMDGPU
test case failures in stage2 bootstrap testing. This miscompile didn't cause any
test case failures until https://reviews.llvm.org/rL320614, so it appeared as if
that patch caused these failures.
Disabling this transformation for now to bring the build bots back to green and
the author of the patch will investigate the miscompile.

llvm-svn: 320786

7 years ago[cmake] Fix clang-cl cross-compilation on macOS
Shoaib Meenai [Fri, 15 Dec 2017 01:05:48 +0000 (01:05 +0000)]
[cmake] Fix clang-cl cross-compilation on macOS

macOS paths usually start with /Users, which clang-cl interprets as a
macro undefine, leading to pretty much everything failing to compile.

CMake should be taught to put a -- in its compilation rules for clang-cl
(and I've been meaning to submit that upstream for a while). In the
meantime, however, and to support older CMake versions, we can just
create a custom make rules override to fix the compilation rules.

Differential Revision: https://reviews.llvm.org/D41219

llvm-svn: 320785

7 years ago[X86] Add a TODO about v8i1 CONCAT_VECTORS.
Craig Topper [Fri, 15 Dec 2017 01:03:46 +0000 (01:03 +0000)]
[X86] Add a TODO about v8i1 CONCAT_VECTORS.

llvm-svn: 320784

7 years ago[SelectionDAG] Make getNode calls that take an ArrayRef of SDValue for operands call...
Craig Topper [Fri, 15 Dec 2017 01:03:45 +0000 (01:03 +0000)]
[SelectionDAG] Make getNode calls that take an ArrayRef of SDValue for operands call NewSDValueDbgMsg.

This makes it work better with some build_vector and concat_vectors creations.

Adjust the NewSDValueDbgMsg in getConstant to avoid duplicating the print when it calls getSplatBuildVector since getSplatBuildVector didn't trigger a print before.

llvm-svn: 320783

7 years ago[X86] Further rearrange the setOperationAction calls to separate the ones that requir...
Craig Topper [Fri, 15 Dec 2017 01:03:43 +0000 (01:03 +0000)]
[X86] Further rearrange the setOperationAction calls to separate the ones that require 512-bit registers OR VLX into separate sections. NFCI

We have several instructions that were introduced in AVX512F that are only available in 512-bit form on KNL. We still make use of them for 128/256 by artificially widening and extracting during isel.

This commit separates these operations from the true 512-bit operations. This way we can qualify the normal 512-bit operations with needing 512-bit register support. And these special operations will get qualified with needing 512-bit registers OR VLX.

The 512-bit register qualification will be introduced in a future patch this just gets everything grouped to minimize deltas on that patch.

llvm-svn: 320782

7 years ago[X86] Group setOperationActions related to vXi1 masks together. NFCI
Craig Topper [Fri, 15 Dec 2017 01:03:42 +0000 (01:03 +0000)]
[X86] Group setOperationActions related to vXi1 masks together. NFCI

Previously they were sort of interleaved in with XMM/YMM/ZMM action related code.

Trying to separate things so its easier to split 512-bit vectors later.

llvm-svn: 320781

7 years ago[X86] Make ISD::INSERT_SUBVECTOR v8i1 legal with AVX512F because we should be custom...
Craig Topper [Fri, 15 Dec 2017 01:03:40 +0000 (01:03 +0000)]
[X86] Make ISD::INSERT_SUBVECTOR v8i1 legal with AVX512F because we should be custom lowering inserting v1i1 into v8i1 under this.

I don't have a test case at the moment. Just noticed while auditing things.

llvm-svn: 320780

7 years ago[X86] Move some of the hasVLX qualified code out of the main hasAVX512 block in the...
Craig Topper [Fri, 15 Dec 2017 01:03:38 +0000 (01:03 +0000)]
[X86] Move some of the hasVLX qualified code out of the main hasAVX512 block in the X86ISelLowering constructor. NFCI

Move it into the separate hasVLX block later in the constructor.

I'm trying to separate 128/256 and 512-bit related code so we can eventually qualify the hasAVX512 block with support for 512-bit vectors required by the prefer-vector-width feature support being talked about in D41096.

llvm-svn: 320779