platform/upstream/llvm.git
7 years agoFixed DoConnectRemote issues where ProcessKDP wasn't switched over to use the version...
Greg Clayton [Wed, 7 Dec 2016 23:51:49 +0000 (23:51 +0000)]
Fixed DoConnectRemote issues where ProcessKDP wasn't switched over to use the version that needed a StringRef as the URL, and also updated all virtual functions to say "override" to make sure this doesn't happen again.

llvm-svn: 288999

7 years agoRefactor how the MSVC toolchain searches for a compatibility version.
David L. Jones [Wed, 7 Dec 2016 23:41:58 +0000 (23:41 +0000)]
Refactor how the MSVC toolchain searches for a compatibility version.

Summary:
The MSVC toolchain and Clang driver combination currently uses a fairly complex
sequence of steps to determine the MS compatibility version to pass to cc1.
There is some oddness in this sequence currently, with some code which inspects
flags in the toolchain, and some code which inspects the triple and local
environment in the driver code.

This change is an attempt to consolidate most of this logic so that
Win32-specific code lives in MSVCToolChain.cpp. I'm not 100% happy with the
split, so any suggestions are welcome.

There are a few things you might want to watch for for specifically:

 - On all platforms, if MSVC compatibility flags are provided (and valid), use
   those.
 - The fallback sequence should be the same as before, but is now consolidated
   into MSVCToolChain::getMSVCVersion:
   - Otherwise, try to use the Triple.
   - Otherwise, on Windows, check the executable.
   - Otherwise, on Windows or with --fms-extensions, default to 18.
   - Otherwise, we can't determine the version.
 - MSVCToolChain::ComputeEffectiveTriple no longer calls the base
   ToolChain::ComputeEffectiveClangTriple. The only thing it would change for
   Windows the architecture, which we don't care about for the compatibility
   version.
    - I'm not sure whether this is philosophically correct (but it should
      be easy to add back to MSVCToolChain::getMSVCVersionFromTriple if not).
    - Previously, Tools.cpp just called getTriple() anyhow, so it doesn't look
      like the effective triple was always being used previously anyhow.

Reviewers: hans, compnerd, llvm-commits, rnk

Subscribers: amccarth

Differential Revision: https://reviews.llvm.org/D27477

llvm-svn: 288998

7 years agoAdd more tests for MSVC version handling.
David L. Jones [Wed, 7 Dec 2016 23:39:44 +0000 (23:39 +0000)]
Add more tests for MSVC version handling.

Summary:
This change adds more test cases for the default MSVC compatibility version:
 1. When -fms-extensions is supplied, but -fmsc-version and
    -fms-compatibility-version are not.
 2. With the target triple specifies an MSVC environment, but no other
    -fms* flags.

Reviewers: rnk, llvm-commits

Subscribers: hans, compnerd, amccarth

Differential Revision: https://reviews.llvm.org/D27498

llvm-svn: 288997

7 years agoRemove redundant call of std::unique_ptr::get.
Rui Ueyama [Wed, 7 Dec 2016 23:26:39 +0000 (23:26 +0000)]
Remove redundant call of std::unique_ptr::get.

Obj is an instance of std::unique_ptr, so *Obj.get() is the same as *Obj.

llvm-svn: 288996

7 years agoFix Windows buildbots.
Rui Ueyama [Wed, 7 Dec 2016 23:24:32 +0000 (23:24 +0000)]
Fix Windows buildbots.

clang-format-diff sorted these #include's in the asciibetical order,
but they need to be in this order.

llvm-svn: 288995

7 years ago[Driver] Add tests for default stack protector values on Darwin
Bruno Cardoso Lopes [Wed, 7 Dec 2016 23:20:30 +0000 (23:20 +0000)]
[Driver] Add tests for default stack protector values on Darwin

llvm-svn: 288994

7 years agoMake demangle() return None instead of "" if a given string is not a mangled symbol.
Rui Ueyama [Wed, 7 Dec 2016 23:17:05 +0000 (23:17 +0000)]
Make demangle() return None instead of "" if a given string is not a mangled symbol.

llvm-svn: 288993

7 years agoCOFF: Define overloaded toString functions.
Rui Ueyama [Wed, 7 Dec 2016 23:17:02 +0000 (23:17 +0000)]
COFF: Define overloaded toString functions.

Previously, we had different way to stringize SymbolBody and InputFile
to construct error messages. This patch defines overloaded function
toString() so that we don't need to memorize all these different
function names.

With that change, it is now easy to include demangled names in error
messages. Now, if there is a symbol name conflict, we'll print out
both mangled and demangled names.

llvm-svn: 288992

7 years ago[asan] Split load and store checks in test. NFCI
Filipe Cabecinhas [Wed, 7 Dec 2016 22:37:11 +0000 (22:37 +0000)]
[asan] Split load and store checks in test. NFCI

llvm-svn: 288991

7 years ago[yaml2obj] Refactor and abstract yaml2dwarf functions
Chris Bieneman [Wed, 7 Dec 2016 22:30:15 +0000 (22:30 +0000)]
[yaml2obj] Refactor and abstract yaml2dwarf functions

This abstracts the code for emitting DWARF binary from the DWARFYAML types into reusable interfaces that could be used by ELF and COFF.

llvm-svn: 288990

7 years ago[ADT, IR] Fix some Clang-tidy modernize-use-equals-delete and Include What You Use...
Eugene Zelenko [Wed, 7 Dec 2016 22:06:02 +0000 (22:06 +0000)]
[ADT, IR] Fix some Clang-tidy modernize-use-equals-delete and Include What You Use warnings; other minor fixes (NFC).

llvm-svn: 288989

7 years ago[XRay][AArch64] Disable the unstable test XRay-aarch64-linux::patching-unpatching.cc
Serge Rogatch [Wed, 7 Dec 2016 22:00:02 +0000 (22:00 +0000)]
[XRay][AArch64] Disable the unstable test XRay-aarch64-linux::patching-unpatching.cc

Summary:
The test `XRay-aarch64-linux::patching-unpatching.cc` sometimes passes, sometimes fails on buildbots.
This patch disables test `patching-unpatching.cc` for AArch64 targets.

Reviewers: rengolin, dberris

Subscribers: llvm-commits, iid_iunknown, aemerson

Differential Revision: https://reviews.llvm.org/D27528

llvm-svn: 288988

7 years ago[BDCE] Skip metadata while replacing uses.
Davide Italiano [Wed, 7 Dec 2016 21:47:32 +0000 (21:47 +0000)]
[BDCE] Skip metadata while replacing uses.

The fix committed in r288851 doesn't cover all the cases.
In particular, if we have an instruction with side effects
which has a no non-dbg use not depending on the bits, we still
perform RAUW destroying the dbg.value's first argument.
Prevent metadata from being replaced here to avoid the issue.

Differential Revision:  https://reviews.llvm.org/D27534

llvm-svn: 288987

7 years ago[obj2yaml] Refactor and abstract dwarf2yaml
Chris Bieneman [Wed, 7 Dec 2016 21:47:28 +0000 (21:47 +0000)]
[obj2yaml] Refactor and abstract dwarf2yaml

This makes the dwarf2yaml code separated and reusable allowing ELF and COFF to share implementations with MachO.

llvm-svn: 288986

7 years agoGlobalISel: use correct builder for ConstantExprs.
Tim Northover [Wed, 7 Dec 2016 21:29:15 +0000 (21:29 +0000)]
GlobalISel: use correct builder for ConstantExprs.

ConstantExpr instances were emitting code into the current block rather than
the entry block. This meant they didn't necessarily dominate all uses, which is
clearly wrong.

llvm-svn: 288985

7 years ago[ObjectYAML] Pull DWARF support into DWARFYAML namespace
Chris Bieneman [Wed, 7 Dec 2016 21:26:32 +0000 (21:26 +0000)]
[ObjectYAML] Pull DWARF support into DWARFYAML namespace

Since DWARF formatting is agnostic to the object file it is stored in, it doesn't make sense for this to be in the MachOYAML implementation. Pulling it into its own namespace means we could modify the ELF and COFF YAML tools to emit DWARF as well.

In a follow-up patch I will better abstract this in obj2yaml and yaml2obj so that the DWARF bits in the tools can be re-used too.

llvm-svn: 288984

7 years agoGlobalISel: store the current MachineFunction as direct state. NFC.
Tim Northover [Wed, 7 Dec 2016 21:17:47 +0000 (21:17 +0000)]
GlobalISel: store the current MachineFunction as direct state. NFC.

Having to ask the MIRBuilder for the current function is a little awkward, and
I'm intending to improve how that's threaded through anyway.

llvm-svn: 288983

7 years agoRename MaxPageSize to DefaultMaxPageSize to avoid confusion.
Rafael Espindola [Wed, 7 Dec 2016 21:13:27 +0000 (21:13 +0000)]
Rename MaxPageSize to DefaultMaxPageSize to avoid confusion.

Thanks to Rui for the suggestion.

llvm-svn: 288982

7 years ago[ObjectYAML] Rename DWARF entries to match section names
Chris Bieneman [Wed, 7 Dec 2016 21:09:37 +0000 (21:09 +0000)]
[ObjectYAML] Rename DWARF entries to match section names

This change makes the yaml tags for the members of the DWARF data match the names of the DWARF sections.

llvm-svn: 288981

7 years agoGlobalISel: simplify MachineIRBuilder interface.
Tim Northover [Wed, 7 Dec 2016 21:05:38 +0000 (21:05 +0000)]
GlobalISel: simplify MachineIRBuilder interface.

MachineIRBuilder had weird before/after and beginning/end flags for the insert
point. Unfortunately the non-default means that instructions will be inserted
in reverse order which is almost never what anyone wants.

Really, I think we just want (like IRBuilder has) the ability to insert at any
C++ iterator-style point (i.e. before any instruction or before MBB.end()). So
this fixes MIRBuilders to behave like IRBuilders in this respect.

llvm-svn: 288980

7 years ago[libFuzzer] include FuzzerIO.h and hopefully fix the Mac build. reported by Dejan...
Kostya Serebryany [Wed, 7 Dec 2016 21:02:48 +0000 (21:02 +0000)]
[libFuzzer] include FuzzerIO.h and hopefully fix the Mac build. reported by Dejan Mircevski

llvm-svn: 288979

7 years agoInstCombine: Fold bitcast of vector to FP scalar
Matt Arsenault [Wed, 7 Dec 2016 20:56:11 +0000 (20:56 +0000)]
InstCombine: Fold bitcast of vector to FP scalar

llvm-svn: 288978

7 years ago[CMake] Add check for HAVE_CRASHREPORTER_INFO
Chris Bieneman [Wed, 7 Dec 2016 20:55:38 +0000 (20:55 +0000)]
[CMake] Add check for HAVE_CRASHREPORTER_INFO

This was also explicitly undef in CMake for some unknown reason.

Hopefully this one won't kill all the bots.

llvm-svn: 288977

7 years ago[RecursiveASTVisitor] Improve post-order traversal unit test
Malcolm Parsons [Wed, 7 Dec 2016 20:38:20 +0000 (20:38 +0000)]
[RecursiveASTVisitor] Improve post-order traversal unit test

llvm-svn: 288976

7 years agoFix Windows buildbots.
Rui Ueyama [Wed, 7 Dec 2016 20:31:46 +0000 (20:31 +0000)]
Fix Windows buildbots.

llvm-svn: 288975

7 years agoUse the correct MaxPageSize.
Rafael Espindola [Wed, 7 Dec 2016 20:29:46 +0000 (20:29 +0000)]
Use the correct MaxPageSize.

Now Target->MaxPageSize is only used as the default value of
Config->MaxPageSize.

llvm-svn: 288974

7 years agoDo not pass line number to convertToUnixPathSeparator.
Rui Ueyama [Wed, 7 Dec 2016 20:25:45 +0000 (20:25 +0000)]
Do not pass line number to convertToUnixPathSeparator.

Line number can never contain '/' or '\', so the previous code
was pointless at that point.

llvm-svn: 288973

7 years agoMake convertToUnixPathSeparator return a new string instead of mutating argument.
Rui Ueyama [Wed, 7 Dec 2016 20:22:27 +0000 (20:22 +0000)]
Make convertToUnixPathSeparator return a new string instead of mutating argument.

llvm-svn: 288972

7 years agoSimplify. NFC.
Rafael Espindola [Wed, 7 Dec 2016 20:20:39 +0000 (20:20 +0000)]
Simplify. NFC.

llvm-svn: 288971

7 years agoUse the correct MaxPageSize.
Rafael Espindola [Wed, 7 Dec 2016 20:10:43 +0000 (20:10 +0000)]
Use the correct MaxPageSize.

Found by inspection.

llvm-svn: 288970

7 years ago[change-namespace] always add a '::' prefix when a symbol reference needs to be fully...
Eric Liu [Wed, 7 Dec 2016 20:08:02 +0000 (20:08 +0000)]
[change-namespace] always add a '::' prefix when a symbol reference needs to be fully-qualified.

llvm-svn: 288969

7 years ago[GVNHoist] Invalidate MemDep when an instruction is moved.
Eli Friedman [Wed, 7 Dec 2016 19:55:59 +0000 (19:55 +0000)]
[GVNHoist] Invalidate MemDep when an instruction is moved.

See also r279907.

Fixes https://llvm.org/bugs/show_bug.cgi?id=30991 .

Differential Revision: https://reviews.llvm.org/D27493

llvm-svn: 288968

7 years agoFormat. NFC.
George Rimar [Wed, 7 Dec 2016 19:44:27 +0000 (19:44 +0000)]
Format. NFC.

llvm-svn: 288967

7 years ago[ELF] - Print absolute file name in errors when possible.
George Rimar [Wed, 7 Dec 2016 19:42:25 +0000 (19:42 +0000)]
[ELF] - Print absolute file name in errors when possible.

Currently LLD prints basename of source file name in error messages,
for example:
$ mkdir foo
$ echo 'void _start(void) { foobar(); }' > foo/bar.c
$ gcc -g -c foo/bar.c
$ bin/ld.lld -o out bar.o
bin/ld.lld: error: bar.c:1: undefined symbol 'foobar'
$
This should say:
bin/ld.lld: error: foo/bar.c:1: undefined symbol 'foobar'

This is PR31299

Differential revision: https://reviews.llvm.org/D27506

llvm-svn: 288966

7 years ago[X86] Skip over DEBUG_VALUE while looking for start of call sequence
Michael Kuperstein [Wed, 7 Dec 2016 19:31:08 +0000 (19:31 +0000)]
[X86] Skip over DEBUG_VALUE while looking for start of call sequence

If we don't skip over DEBUG_VALUEs, we get differences between -g and non-g
code.

This fixes PR31242.

Differential Revision: https://reviews.llvm.org/D27485

llvm-svn: 288965

7 years ago[X86] Do not assume "ri" instructions always have an immediate operand
Michael Kuperstein [Wed, 7 Dec 2016 19:29:18 +0000 (19:29 +0000)]
[X86] Do not assume "ri" instructions always have an immediate operand

The second operand of an "ri" instruction may be an immediate, but it may
also be a globalvariable, so we should make any assumptions.

This fixes PR31271.

Differential Revision: https://reviews.llvm.org/D27481

llvm-svn: 288964

7 years agoFix the apple build issue caused by r288956
Chris Bieneman [Wed, 7 Dec 2016 19:28:22 +0000 (19:28 +0000)]
Fix the apple build issue caused by r288956

Should be checking if HAVE_CRASHREPORTERCLIENT_H is defined not relying on it having a value.

llvm-svn: 288963

7 years ago[CUDA] Improve target attribute checking for function templates.
Artem Belevich [Wed, 7 Dec 2016 19:27:16 +0000 (19:27 +0000)]
[CUDA] Improve target attribute checking for function templates.

* __host__ __device__ functions are no longer considered to be
  redeclarations of __host__ or __device__ functions. This prevents
  unintentional merging of target attributes across them.
* Function target attributes are not considered (and must match) during
  explicit instantiation and specialization of function templates.

Differential Revision: https://reviews.llvm.org/D25809

llvm-svn: 288962

7 years agoRevert "[CMake] Use cmakedefine01 instead of cmakedefine"
Chris Bieneman [Wed, 7 Dec 2016 19:25:38 +0000 (19:25 +0000)]
Revert "[CMake] Use cmakedefine01 instead of cmakedefine"

This reverts commit r288959.

Apparently using cmakedefine01 explodes.

llvm-svn: 288961

7 years ago[builtin] Add Thumb1 implementation for idivsi3 and aeabi_idivmod
Weiming Zhao [Wed, 7 Dec 2016 19:14:29 +0000 (19:14 +0000)]
[builtin] Add Thumb1 implementation for idivsi3 and aeabi_idivmod

Summary:
For idivsi3, convert the Thumb2 only instruction to thumb1.
For aeabi_idivmod, using __divsi3.

Reviewers: rengolin, compnerd

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D27472

llvm-svn: 288960

7 years ago[CMake] Use cmakedefine01 instead of cmakedefine
Chris Bieneman [Wed, 7 Dec 2016 19:13:32 +0000 (19:13 +0000)]
[CMake] Use cmakedefine01 instead of cmakedefine

Looks like we need a 01 value for HAVE_CRASHREPORTERCLIENT_H.

llvm-svn: 288959

7 years ago[InstCombine] add tests for smin+icmp; NFC
Sanjay Patel [Wed, 7 Dec 2016 18:56:55 +0000 (18:56 +0000)]
[InstCombine] add tests for smin+icmp; NFC

The tests that already work are folded in InstSimplify, so those
tests should be redundant and we can remove them if they don't
seem worthwhile for completeness.

llvm-svn: 288957

7 years ago[CMake] Add a check for HAVE_CRASHREPORTERCLIENT_H
Chris Bieneman [Wed, 7 Dec 2016 18:53:04 +0000 (18:53 +0000)]
[CMake] Add a check for HAVE_CRASHREPORTERCLIENT_H

The CMake build has been hardcoding this to undef forever, we shouldn't have been doing that.

llvm-svn: 288956

7 years ago[ObjectYAML] Support for DWARF __debug_abbrev section
Chris Bieneman [Wed, 7 Dec 2016 18:52:59 +0000 (18:52 +0000)]
[ObjectYAML] Support for DWARF __debug_abbrev section

This patch adds support for round-tripping DWARF debug abbreviations through the obj<->yaml tools.

llvm-svn: 288955

7 years ago[builtin] for the condition for check __ARM_FEATURE_CLZ
Weiming Zhao [Wed, 7 Dec 2016 18:41:07 +0000 (18:41 +0000)]
[builtin] for the condition for check __ARM_FEATURE_CLZ

Summary: Since CLZ is not available for Thumb1, we use __ARM_ARCH_ISA_THUMB != 1 as one of the conditions.

Reviewers: rnk, compnerd, rengolin

Subscribers: aemerson, rengolin, llvm-commits

Differential Revision: https://reviews.llvm.org/D27530

llvm-svn: 288954

7 years ago[asan] Disable yet another unstable test under AArch64
Renato Golin [Wed, 7 Dec 2016 18:37:49 +0000 (18:37 +0000)]
[asan] Disable yet another unstable test under AArch64

llvm-svn: 288953

7 years ago[SelectionDAG] Add knownbits support for vector demandedelts in SMAX/SMIN/UMAX/UMIN...
Simon Pilgrim [Wed, 7 Dec 2016 17:54:00 +0000 (17:54 +0000)]
[SelectionDAG] Add knownbits support for vector demandedelts in SMAX/SMIN/UMAX/UMIN opcodes

llvm-svn: 288926

7 years agoAdd unittests for foreach(Elt|Piece). NFC.
Michael Kruse [Wed, 7 Dec 2016 17:48:02 +0000 (17:48 +0000)]
Add unittests for foreach(Elt|Piece). NFC.

llvm-svn: 288925

7 years agoAdd more ISL foreachElt functions. NFC.
Michael Kruse [Wed, 7 Dec 2016 17:47:57 +0000 (17:47 +0000)]
Add more ISL foreachElt functions. NFC.

Add and implement foreachElt for isl_map, isl_set and isl_union_set. These are
used by an out-of-tree patch which is in process of being upstreamed.

llvm-svn: 288924

7 years ago[RecursiveASTVisitor] Fix post-order traversal of UnaryOperator
Malcolm Parsons [Wed, 7 Dec 2016 17:39:04 +0000 (17:39 +0000)]
[RecursiveASTVisitor] Fix post-order traversal of UnaryOperator

Reviewers: aaron.ballman, klimek, doug.gregor, teemperor, rsmith

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D26742

llvm-svn: 288923

7 years ago[analyzer] Fix typo in nullability checker diagnostic
Devin Coughlin [Wed, 7 Dec 2016 17:36:27 +0000 (17:36 +0000)]
[analyzer] Fix typo in nullability checker diagnostic

'infered' --> 'inferred'

llvm-svn: 288922

7 years agoELF/AArch64: Refactor R_AARCH64_LDST{8,15,32,64,128}_ABS_LO12_NC Relocations
Adhemerval Zanella [Wed, 7 Dec 2016 17:31:48 +0000 (17:31 +0000)]
ELF/AArch64: Refactor R_AARCH64_LDST{8,15,32,64,128}_ABS_LO12_NC Relocations

This patch refactor how to apply the R_AARCH64_LDST{8,16,32,64,128}_ABS_NC
relocations by adding a new function to correct extract the bits expected
by each relocation.  This make is explicit which are the bits range expected
and simplify the code to mask and shift the deriable values.

It also fixes the R_AARCH64_LDST128_ABS_LO12_NC mask, although in pratice
the mask/shift always returns a 16 bytes aligned value.

Checked on AArch64 and with test-suite.

llvm-svn: 288921

7 years ago[X86] Add knownbits vector UMAX test
Simon Pilgrim [Wed, 7 Dec 2016 17:21:13 +0000 (17:21 +0000)]
[X86] Add knownbits vector UMAX test

In preparation for demandedelts support

llvm-svn: 288920

7 years ago[change-namespace] don't fix using shadow decls in classes.
Eric Liu [Wed, 7 Dec 2016 17:04:07 +0000 (17:04 +0000)]
[change-namespace] don't fix using shadow decls in classes.

Summary:
Using shadow declarations in classes always refers to base class, which does not
need to be fixed/qualified since it can be inferred from inheritance.

Reviewers: bkramer

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D27523

llvm-svn: 288919

7 years ago[X86][SSE] Remove AND -> VZEXT combine
Simon Pilgrim [Wed, 7 Dec 2016 17:02:41 +0000 (17:02 +0000)]
[X86][SSE] Remove AND -> VZEXT combine

This is now performed more generally by the target shuffle combine code.

Already covered by tests that were originally added in D7666/rL229480 to support combineVectorZext (or VectorZextCombine as it was known then....).

Differential Revision: https://reviews.llvm.org/D27510

llvm-svn: 288918

7 years ago[analyzer] Remove an unused enum value in RetainCountChecker.
Artem Dergachev [Wed, 7 Dec 2016 16:51:54 +0000 (16:51 +0000)]
[analyzer] Remove an unused enum value in RetainCountChecker.

No functional change intended.

Differential Revision: https://reviews.llvm.org/D27408

llvm-svn: 288917

7 years ago[SelectionDAG] Add knownbits support for EXTRACT_VECTOR_ELT opcodes
Simon Pilgrim [Wed, 7 Dec 2016 16:28:21 +0000 (16:28 +0000)]
[SelectionDAG] Add knownbits support for EXTRACT_VECTOR_ELT opcodes

llvm-svn: 288916

7 years agoAdd IslPtr type traits. NFC.
Michael Kruse [Wed, 7 Dec 2016 16:17:59 +0000 (16:17 +0000)]
Add IslPtr type traits. NFC.

Add traits for isl_id and isl_multi_aff, required by out-of-tree patches
currently in progress of upstreaming.

isl_union_pw_aff_dump has been added to ISL during one of the last ISL
updates, such that we can also enable its dump() trait.

llvm-svn: 288915

7 years ago[analyzer] pr31226: Disable CastSizeChecker in C++ because it's not quite ready.
Artem Dergachev [Wed, 7 Dec 2016 16:12:26 +0000 (16:12 +0000)]
[analyzer] pr31226: Disable CastSizeChecker in C++ because it's not quite ready.

Avoids a crash and a related false positive.

Investigation by Daniel Krupp!

llvm-svn: 288914

7 years ago[SelectionDAG] Removed old knownbits TODO comment. NFCI.
Simon Pilgrim [Wed, 7 Dec 2016 15:31:12 +0000 (15:31 +0000)]
[SelectionDAG] Removed old knownbits TODO comment. NFCI.

EXTRACT_VECTOR_ELT does support demanded elts if the element index is known and in range.

llvm-svn: 288913

7 years ago[X86] Add test to show missed opportunities to calculate knownbits in INSERT_VECTOR_ELT
Simon Pilgrim [Wed, 7 Dec 2016 15:27:18 +0000 (15:27 +0000)]
[X86] Add test to show missed opportunities to calculate knownbits in INSERT_VECTOR_ELT

llvm-svn: 288912

7 years ago[X86][SSE] Fix vpextrd/vpextrq checks
Simon Pilgrim [Wed, 7 Dec 2016 15:10:05 +0000 (15:10 +0000)]
[X86][SSE] Fix vpextrd/vpextrq checks

They were testing for the pre-vex versions

llvm-svn: 288911

7 years ago[X86][SSE] Force execution domain of 32-bit extractps/pextrd in the stack folding...
Simon Pilgrim [Wed, 7 Dec 2016 15:06:14 +0000 (15:06 +0000)]
[X86][SSE] Force execution domain of 32-bit extractps/pextrd in the stack folding tests

llvm-svn: 288910

7 years ago[LV] Scalarize operands of predicated instructions
Matthew Simpson [Wed, 7 Dec 2016 15:03:32 +0000 (15:03 +0000)]
[LV] Scalarize operands of predicated instructions

This patch attempts to scalarize the operand expressions of predicated
instructions if they were conditionally executed in the original loop. After
scalarization, the expressions will be sunk inside the blocks created for the
predicated instructions. The transformation essentially performs
un-if-conversion on the operands.

The cost model has been updated to determine if scalarization is profitable. It
compares the cost of a vectorized instruction, assuming it will be
if-converted, to the cost of the scalarized instruction, assuming that the
instructions corresponding to each vector lane will be sunk inside a predicated
block, possibly avoiding execution. If it's more profitable to scalarize the
entire expression tree feeding the predicated instruction, the expression will
be scalarized; otherwise, it will be vectorized. We only consider the cost of
the entire expression to accurately estimate the cost of the required
insertelement and extractelement instructions.

Differential Revision: https://reviews.llvm.org/D26083

llvm-svn: 288909

7 years ago[change-namespace] move template class forward-declarations and don't move fwd-decls...
Eric Liu [Wed, 7 Dec 2016 14:20:52 +0000 (14:20 +0000)]
[change-namespace] move template class forward-declarations and don't move fwd-decls in classes.

Summary:
Forward declarations in moved namespaces should be moved back to the old
namespaces. We should also move template class forward declarations.

Also fix a bug that moves forward declarations of nested classes.

Reviewers: bkramer

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D27515

llvm-svn: 288908

7 years agoTry unbreaking the MSVC build.
Benjamin Kramer [Wed, 7 Dec 2016 13:35:11 +0000 (13:35 +0000)]
Try unbreaking the MSVC build.

llvm-svn: 288907

7 years ago[X86][SSE] Regenerate test.
Simon Pilgrim [Wed, 7 Dec 2016 13:05:04 +0000 (13:05 +0000)]
[X86][SSE] Regenerate test.

llvm-svn: 288906

7 years ago[AVR] Expand 'SELECT_CC' nodes whereever possible
Dylan McKay [Wed, 7 Dec 2016 12:34:47 +0000 (12:34 +0000)]
[AVR] Expand 'SELECT_CC' nodes whereever possible

llvm-svn: 288905

7 years ago[LowerTypeTests] Use the TrailingObjects infrastructure for trailing objects.
Benjamin Kramer [Wed, 7 Dec 2016 12:31:45 +0000 (12:31 +0000)]
[LowerTypeTests] Use the TrailingObjects infrastructure for trailing objects.

Also avoid allocating ~3x as much memory as needed.

llvm-svn: 288904

7 years agoWhen GVN removes a redundant load, it should not modify the debug location of the...
Andrea Di Biagio [Wed, 7 Dec 2016 12:31:36 +0000 (12:31 +0000)]
When GVN removes a redundant load, it should not modify the debug location of the dominating load.

In the case of a fully redundant load LI dominated by an equivalent load V, GVN
should always preserve the original debug location of V. Otherwise, we risk to
introduce an incorrect stepping.
If V has debug info, then clearly it should not be modified. If V has a null
debugloc, then it is still potentially incorrect to propagate LI's debugloc
because LI may not post-dominate V.

Differential Revision: https://reviews.llvm.org/D27468

llvm-svn: 288903

7 years ago[X86][SSE] Consistently set MOVD/MOVQ load/store/move instructions to integer domain
Simon Pilgrim [Wed, 7 Dec 2016 12:10:49 +0000 (12:10 +0000)]
[X86][SSE] Consistently set MOVD/MOVQ load/store/move instructions to integer domain

We are being inconsistent with these instructions (and all their variants.....) with a random mix of them using the default float domain.

Differential Revision: https://reviews.llvm.org/D27419

llvm-svn: 288902

7 years ago[InlineFunction] Refactor code in function `fixupLineNumbers' as suggested by David...
Andrea Di Biagio [Wed, 7 Dec 2016 12:01:45 +0000 (12:01 +0000)]
[InlineFunction] Refactor code in function `fixupLineNumbers' as suggested by David in D27462. NFC

llvm-svn: 288901

7 years ago[mips][rtdyld] Merge code to write relocated values to the section. NFC
Simon Dardis [Wed, 7 Dec 2016 11:41:23 +0000 (11:41 +0000)]
[mips][rtdyld] Merge code to write relocated values to the section. NFC

Preparation work for implementing N32 support.

Patch By: Daniel Sanders

Reviewers: vkalintiris, atanasyan

Differential Revision: https://reviews.llvm.org/D27460

llvm-svn: 288900

7 years ago[AVR] Move a pseudo expansion test into a folder
Dylan McKay [Wed, 7 Dec 2016 11:21:45 +0000 (11:21 +0000)]
[AVR] Move a pseudo expansion test into a folder

llvm-svn: 288899

7 years ago[X86][XOP] Fix VPERMIL2 non-constant pool shuffle decoding (PR31296)
Simon Pilgrim [Wed, 7 Dec 2016 11:19:00 +0000 (11:19 +0000)]
[X86][XOP] Fix VPERMIL2 non-constant pool shuffle decoding (PR31296)

The non-constant pool version of DecodeVPERMIL2PMask was not offsetting correctly for the second input. I've updated the code to match the implementation in the constant-pool version.

Annoyingly this bug was hidden for so long as it's tricky to combine to useful variable shuffle masks that don't become constant-pool entries.

llvm-svn: 288898

7 years ago[AVR] Allow loading from stack slots where src and dest registers are identical
Dylan McKay [Wed, 7 Dec 2016 11:08:56 +0000 (11:08 +0000)]
[AVR] Allow loading from stack slots where src and dest registers are identical

Fixes PR 31256

llvm-svn: 288897

7 years agoImplement the -Wstrict-prototypes warning
Alex Lorenz [Wed, 7 Dec 2016 10:52:18 +0000 (10:52 +0000)]
Implement the -Wstrict-prototypes warning

This commit fixes PR20796. It implements the C only -Wstrict-prototypes warning.
Clang now emits a warning for function declarations which have no parameters
specified and for K&R function definitions with more than 0 parameters that are
not preceded by a previous prototype declaration.

The patch was originally submitted by Paul Titei!

rdar://15060615

Differential Revision: https://reviews.llvm.org/D16533

llvm-svn: 288896

7 years ago[InlineFunction] Do not propagate the callsite debug location to instructions inlined...
Andrea Di Biagio [Wed, 7 Dec 2016 10:37:26 +0000 (10:37 +0000)]
[InlineFunction] Do not propagate the callsite debug location to instructions inlined from functions with debug info.

When a function F is inlined, InlineFunction extends the debug location of every
instruction inlined from F by adding an InlinedAt.

However, if an instruction has a 'null' debug location, InlineFunction would
propagate the callsite debug location to it. This behavior existed since
revision 210459.

Revision 210459 was originally committed specifically to workaround the lack of
debug information for instructions inlined from intrinsic functions (which are
usually declared with attributes `__always_inline__, __nodebug__`).

The problem with revision 210459 is that it doesn't make any sort of distinction
between instructions inlined from a 'nodebug' function and instructions which
are inlined from a function built with debug info. This issue may lead to
incorrect stepping in the debugger.

This patch works under the assumption that a nodebug function does not have a
DISubprogram. When a function F is inlined into another function G,
InlineFunction checks if F has debug info associated with it.

For nodebug functions, the InlineFunction logic is unchanged (i.e. it would
still propagate the callsite debugloc to the inlined instructions). Otherwise,
InlineFunction no longer propagates the callsite debug location.

Differential Revision: https://reviews.llvm.org/D27462

llvm-svn: 288895

7 years ago[PM] Add some more logging to make it more clear when the CGSCC
Chandler Carruth [Wed, 7 Dec 2016 10:33:15 +0000 (10:33 +0000)]
[PM] Add some more logging to make it more clear when the CGSCC
infrastrucutre is skipping SCCs and RefSCCs.

llvm-svn: 288894

7 years ago[ObjC++] Don't enter a C++ declarator scope when the current context is
Alex Lorenz [Wed, 7 Dec 2016 10:24:44 +0000 (10:24 +0000)]
[ObjC++] Don't enter a C++ declarator scope when the current context is
an Objective-C declaration

This commit ensures that Sema won't enter a C++ declarator scope when the
current context is an Objective-C declaration. This prevents an assertion
failure in EnterDeclaratorContext that's used to ensure that current context
will be restored correctly after exiting the declarator context.

rdar://20560175

Differential Revision: https://reviews.llvm.org/D26922

llvm-svn: 288893

7 years agoCompilation database test: don't try to output to CWD
Sam McCall [Wed, 7 Dec 2016 09:19:07 +0000 (09:19 +0000)]
Compilation database test: don't try to output to CWD

Summary:
Write output from compilation database test to %T rather than the working dir.
Sometimes CWD isn't writable!
Also specify no-canonical-prefixes so that clang has 'clang' in the name.

Reviewers: bkramer

Subscribers: joerg, cfe-commits

Differential Revision: https://reviews.llvm.org/D27504

llvm-svn: 288892

7 years ago[OpenCL] Added a LIT test for ensuring address space mangling is done the same both...
Alexey Bader [Wed, 7 Dec 2016 08:43:49 +0000 (08:43 +0000)]
[OpenCL] Added a LIT test for ensuring address space mangling is done the same both in OpenCL1.2 and OpenCL2.0.

Patch by Egor Churaev (echuraev).

Reviewers: Anastasia

Subscribers: yaxunl, cfe-commits, bader

Differential Revision: https://reviews.llvm.org/D27403

llvm-svn: 288891

7 years ago[OpenCL] Fix SPIR version generation.
Alexey Bader [Wed, 7 Dec 2016 08:38:24 +0000 (08:38 +0000)]
[OpenCL] Fix SPIR version generation.

Patch by Egor Churaev (echuraev).

Reviewers: Anastasia

Subscribers: bader, yaxunl, cfe-commits

Differential Revision: https://reviews.llvm.org/D27300

llvm-svn: 288890

7 years agoReintroduce a check accidentally removed in 288873 to fix clang bots
Philip Reames [Wed, 7 Dec 2016 04:48:50 +0000 (04:48 +0000)]
Reintroduce a check accidentally removed in 288873 to fix clang bots

I believe this is the cause of the failure, but have not been able to confirm.  Note that this is a speculative fix; I'm still waiting for a full build to finish as I synced and ended up doing a clean build which takes 20+ minutes on my machine.

llvm-svn: 288886

7 years agoAdd comments and reorder code a bit to clarify the intention. NFC.
Rui Ueyama [Wed, 7 Dec 2016 04:45:34 +0000 (04:45 +0000)]
Add comments and reorder code a bit to clarify the intention. NFC.

llvm-svn: 288885

7 years agoFix a warning introduced in r288874
Philip Reames [Wed, 7 Dec 2016 04:11:22 +0000 (04:11 +0000)]
Fix a warning introduced in r288874

llvm-svn: 288884

7 years agoMake a decision about whether we should warn on missing entry or not early.
Rui Ueyama [Wed, 7 Dec 2016 04:06:21 +0000 (04:06 +0000)]
Make a decision about whether we should warn on missing entry or not early.

Config->WarnMissingEntry is a single-purpose boolean variable, and
I think it's easier to understand than Config->HasEntry.

llvm-svn: 288883

7 years agoSimplify -e <number> option handling.
Rui Ueyama [Wed, 7 Dec 2016 03:23:06 +0000 (03:23 +0000)]
Simplify -e <number> option handling.

This patch is to parse the entry symbol name lazily.

llvm-svn: 288882

7 years agoLowerTypeTests: Add a test that covers "unsatisfiable" type metadata.
Peter Collingbourne [Wed, 7 Dec 2016 03:04:34 +0000 (03:04 +0000)]
LowerTypeTests: Add a test that covers "unsatisfiable" type metadata.

llvm-svn: 288881

7 years ago[ELF] Fix the broken PPC test
Petr Hosek [Wed, 7 Dec 2016 03:04:02 +0000 (03:04 +0000)]
[ELF] Fix the broken PPC test

This is related to the change in handling of entry point symbols.

Differential Revision: https://reviews.llvm.org/D27500

llvm-svn: 288880

7 years agoAMDGPU : Add S_SETREG instructions to fix fdiv precision issues.
Tom Stellard [Wed, 7 Dec 2016 02:42:15 +0000 (02:42 +0000)]
AMDGPU : Add S_SETREG instructions to fix fdiv precision issues.

Patch By: Wei Ding

Summary: This patch fixes the fdiv precision issues.

Reviewers: b-sumner, cfang, wdng, arsenm

Subscribers: kzhuravl, nhaehnle, yaxunl, tony-tye

Differential Revision: https://reviews.llvm.org/D26424

llvm-svn: 288879

7 years ago[ELF] Shared libraries should have entry point
Petr Hosek [Wed, 7 Dec 2016 02:26:16 +0000 (02:26 +0000)]
[ELF] Shared libraries should have entry point

Shared libraries should have entry set following the same rules as for
regular binaries. The only difference is that in case the default entry
point (_start or __start) isn't found (unless it was set explicitly), we
shouldn't give a warning as in case of regular binaries.

Differential Revision: https://reviews.llvm.org/D27497

llvm-svn: 288878

7 years ago[ELF] Only binaries should have DT_DEBUG entry
Petr Hosek [Wed, 7 Dec 2016 02:05:42 +0000 (02:05 +0000)]
[ELF] Only binaries should have DT_DEBUG entry

The presence of DT_DEBUG entry is unrelated to the existence of entry point.

Differential Revision: https://reviews.llvm.org/D27496

llvm-svn: 288877

7 years ago[AArch64] Correct the check of signed 9-bit imm in isLegalAddressingMode()
Haicheng Wu [Wed, 7 Dec 2016 01:45:04 +0000 (01:45 +0000)]
[AArch64] Correct the check of signed 9-bit imm in isLegalAddressingMode()

In the addressing mode, signed 9-bit imm is [-256, 255], not [-512, 511].

Differential Revision: https://reviews.llvm.org/D27480

llvm-svn: 288876

7 years ago[LCG] Add basic verification of the parent set and fix bugs it uncovers.
Chandler Carruth [Wed, 7 Dec 2016 01:42:40 +0000 (01:42 +0000)]
[LCG] Add basic verification of the parent set and fix bugs it uncovers.

The existing unittests actually cover this now that we verify things.

llvm-svn: 288875

7 years ago[LVI] Remove used return value from markX functions
Philip Reames [Wed, 7 Dec 2016 01:03:56 +0000 (01:03 +0000)]
[LVI] Remove used return value from markX functions

llvm-svn: 288874

7 years ago[LVI] Simplify mergeIn code
Philip Reames [Wed, 7 Dec 2016 00:54:21 +0000 (00:54 +0000)]
[LVI] Simplify mergeIn code

Remove the unused return type, use early return, use assignment operator.

llvm-svn: 288873

7 years agoDriver: Remove support for -fobjc-gc*
Duncan P. N. Exon Smith [Wed, 7 Dec 2016 00:31:10 +0000 (00:31 +0000)]
Driver: Remove support for -fobjc-gc*

As a first step toward removing Objective-C garbage collection from
Clang, remove support from the driver.  I'm hoping this will flush out
any expected bots/configurations/whatever that might rely on it.

I've left the options behind temporarily in -cc1 to keep tests passing.
I'll kill them off entirely in a follow up when I've had a chance to
update/delete the rest of Clang.

llvm-svn: 288872

7 years ago[LVI] Simplify obfuscated code
Philip Reames [Wed, 7 Dec 2016 00:28:28 +0000 (00:28 +0000)]
[LVI] Simplify obfuscated code

It doesn't matter why something is overdefined if it is...

llvm-svn: 288871

7 years agoFix test.
Richard Smith [Wed, 7 Dec 2016 00:24:40 +0000 (00:24 +0000)]
Fix test.

llvm-svn: 288870