platform/upstream/llvm.git
8 years agoNFC: skip FenceInst up-front in AtomicExpandPass.
James Y Knight [Mon, 28 Mar 2016 15:05:30 +0000 (15:05 +0000)]
NFC: skip FenceInst up-front in AtomicExpandPass.

llvm-svn: 264583

8 years agoReduce size of DefinitionData from 120 to 96 bytes on Windows.
Nico Weber [Mon, 28 Mar 2016 14:55:24 +0000 (14:55 +0000)]
Reduce size of DefinitionData from 120 to 96 bytes on Windows.

In the Microsoft ABI, only bitfields with identical types get
packed together, so use unsigned consistently instead of a
bool / unsigned mix.

No intended behavior change.

llvm-svn: 264582

8 years ago[Hexagon] Only use restore functions for single register at -Oz
Krzysztof Parzyszek [Mon, 28 Mar 2016 14:52:21 +0000 (14:52 +0000)]
[Hexagon] Only use restore functions for single register at -Oz

llvm-svn: 264581

8 years ago[Hexagon] Speed up frame lowering when no optimizations are enabled
Krzysztof Parzyszek [Mon, 28 Mar 2016 14:42:03 +0000 (14:42 +0000)]
[Hexagon] Speed up frame lowering when no optimizations are enabled

- Do not optimize stack slots in optnone functions.
- Get aligned-base register from HexagonMachineFunctionInfo instead of
  looking for ALIGNA instruction in the function's body.

llvm-svn: 264580

8 years agoSparc: silently ignore .proc assembler directive
Douglas Katzman [Mon, 28 Mar 2016 14:00:11 +0000 (14:00 +0000)]
Sparc: silently ignore .proc assembler directive

Differential Revision: http://reviews.llvm.org/D18463

llvm-svn: 264579

8 years ago[lanai] Add Lanai backend.
Jacques Pienaar [Mon, 28 Mar 2016 13:09:54 +0000 (13:09 +0000)]
[lanai] Add Lanai backend.

Add the Lanai backend to lib/Target.

General Lanai backend discussion on llvm-dev thread "[RFC] Lanai backend" (http://lists.llvm.org/pipermail/llvm-dev/2016-February/095118.html).

Differential Revision: http://reviews.llvm.org/D17011

llvm-svn: 264578

8 years agoRevert "[OPENMP] Allow runtime insert its own code inside OpenMP regions."
Alexey Bataev [Mon, 28 Mar 2016 12:58:34 +0000 (12:58 +0000)]
Revert "[OPENMP] Allow runtime insert its own code inside OpenMP regions."

Reverting because of failed tests.

llvm-svn: 264577

8 years ago[OPENMP] Allow runtime insert its own code inside OpenMP regions.
Alexey Bataev [Mon, 28 Mar 2016 12:52:58 +0000 (12:52 +0000)]
[OPENMP] Allow runtime insert its own code inside OpenMP regions.

Solution unifies interface of RegionCodeGenTy type to allow insert
runtime-specific code before/after main codegen action defined in
CGStmtOpenMP.cpp file. Runtime should not define its own RegionCodeGenTy
for general OpenMP directives, but must be allowed to insert its own
 (required) code to support target specific codegen.

llvm-svn: 264576

8 years agoAdd fine-grain dependences analysis to release notes.
Hongbin Zheng [Mon, 28 Mar 2016 12:41:49 +0000 (12:41 +0000)]
Add fine-grain dependences analysis to release notes.

Differential Revision: http://reviews.llvm.org/D17905

llvm-svn: 264575

8 years ago[CLANG][avx512][BUILTIN] Adding fixupimm{pd|ps|sd|ss}
Michael Zuckerman [Mon, 28 Mar 2016 12:23:09 +0000 (12:23 +0000)]
[CLANG][avx512][BUILTIN] Adding fixupimm{pd|ps|sd|ss}
getexp{sd|ss} getmant{sd|ss} kunpck{di|si} loada{pd|ps} loaddqu{di|hi|qi|si} max{sd|ss} min{sd|ss} kmov16 builtins to clang

Differential Revision: http://reviews.llvm.org/D18215

llvm-svn: 264574

8 years ago[SROA] Fix typo in comment
Hal Finkel [Mon, 28 Mar 2016 11:23:21 +0000 (11:23 +0000)]
[SROA] Fix typo in comment

llvm-svn: 264573

8 years agoC++11 is required, remove some preprocessor checks for it
Hal Finkel [Mon, 28 Mar 2016 11:13:03 +0000 (11:13 +0000)]
C++11 is required, remove some preprocessor checks for it

We require C++11 to build, so remove a few remaining preprocessor checks for
'__cplusplus >= 201103L'. This should always be true.

llvm-svn: 264572

8 years agoFix-up for OS X test failure after r264451 ("Add memset, memmove, and memcpy to the...
Kuba Brecka [Mon, 28 Mar 2016 10:21:35 +0000 (10:21 +0000)]
Fix-up for OS X test failure after r264451 ("Add memset, memmove, and memcpy to the common interceptors")

llvm-svn: 264571

8 years agoRevert "[OPENMP] Allow runtime insert its own code inside OpenMP regions."
Alexey Bataev [Mon, 28 Mar 2016 10:12:03 +0000 (10:12 +0000)]
Revert "[OPENMP] Allow runtime insert its own code inside OpenMP regions."

This reverts commit 3ee791165100607178073f14531a0dc90c622b36.

llvm-svn: 264570

8 years ago[OPENMP] Allow runtime insert its own code inside OpenMP regions.
Alexey Bataev [Mon, 28 Mar 2016 09:53:43 +0000 (09:53 +0000)]
[OPENMP] Allow runtime insert its own code inside OpenMP regions.

Solution unifies interface of RegionCodeGenTy type to allow insert
runtime-specific code before/after main codegen action defined in
CGStmtOpenMP.cpp file. Runtime should not define its own RegionCodeGenTy
for general OpenMP directives, but must be allowed to insert its own
  (required) code to support target specific codegen.

llvm-svn: 264569

8 years ago[Power9] Implement new altivec instructions: bcd* series
Chuang-Yu Cheng [Mon, 28 Mar 2016 09:04:23 +0000 (09:04 +0000)]
[Power9] Implement new altivec instructions: bcd* series

This patch implements the following altivec instructions:

- Decimal Convert From/to National/Zoned/Signed-QWord:
    bcdcfn. bcdcfz. bcdctn. bcdctz. bcdcfsq. bcdctsq.

- Decimal Copy-Sign/Set-Sign:
    bcdcpsgn. bcdsetsgn.

- Decimal Shift/Unsigned-Shift/Shift-and-Round:
    bcds. bcdus. bcdsr.

- Decimal (Unsigned) Truncate:
    bcdtrunc. bcdutrunc.

Total 13 instructions

Thanks Amehsan's advice! Thanks Kit's great help!
Reviewers: hal, nemanja, kbarton, tjablin, amehsan

http://reviews.llvm.org/D17838

llvm-svn: 264568

8 years ago[Power9] Implement new vsx instructions: insert, extract, test data class, min/max...
Chuang-Yu Cheng [Mon, 28 Mar 2016 08:34:28 +0000 (08:34 +0000)]
[Power9] Implement new vsx instructions: insert, extract, test data class, min/max, reverse, permute, splat

This change implements the following vsx instructions:

- Scalar Insert/Extract
    xsiexpdp xsiexpqp xsxexpdp xsxsigdp xsxexpqp xsxsigqp

- Vector Insert/Extract
    xviexpdp xviexpsp xvxexpdp xvxexpsp xvxsigdp xvxsigsp
    xxextractuw xxinsertw

- Scalar/Vector Test Data Class
    xststdcdp xststdcsp xststdcqp
    xvtstdcdp xvtstdcsp

- Maximum/Minimum
    xsmaxcdp xsmaxjdp
    xsmincdp xsminjdp

- Vector Byte-Reverse/Permute/Splat
    xxbrd xxbrh xxbrq xxbrw
    xxperm xxpermr
    xxspltib

30 instructions

Thanks Nemanja for invaluable discussion! Thanks Kit's great help!
Reviewers: hal, nemanja, kbarton, tjablin, amehsan

http://reviews.llvm.org/D16842

llvm-svn: 264567

8 years agoAVX-512: Fixed ICMP instruction selection for i1 operands
Elena Demikhovsky [Mon, 28 Mar 2016 07:47:58 +0000 (07:47 +0000)]
AVX-512: Fixed ICMP instruction selection for i1 operands

ICMP instruction selection fails on SKX and KNL for i1 operand.
I use XOR to resolve:
(A == B) is equivalent to (A xor B) == 0

Differential Revision: http://reviews.llvm.org/D18511

llvm-svn: 264566

8 years ago[Power9] Implement new vsx instructions: quad-precision move, fp-arithmetic
Chuang-Yu Cheng [Mon, 28 Mar 2016 07:38:01 +0000 (07:38 +0000)]
[Power9] Implement new vsx instructions: quad-precision move, fp-arithmetic

This change implements the following vsx instructions:

- quad-precision move
    xscpsgnqp, xsabsqp, xsnegqp, xsnabsqp

- quad-precision fp-arithmetic
    xsaddqp(o) xsdivqp(o) xsmulqp(o) xssqrtqp(o) xssubqp(o)
    xsmaddqp(o) xsmsubqp(o) xsnmaddqp(o) xsnmsubqp(o)

22 instructions

Thanks Nemanja and Kit for careful review and invaluable discussion!
Reviewers: hal, nemanja, kbarton, tjablin, amehsan

http://reviews.llvm.org/D16110

llvm-svn: 264565

8 years agoP0138R2: Allow direct-list-initialization of an enumeration from an integral
Richard Smith [Mon, 28 Mar 2016 06:08:37 +0000 (06:08 +0000)]
P0138R2: Allow direct-list-initialization of an enumeration from an integral
value that can convert to the enum's underlying type.

llvm-svn: 264564

8 years agoclang-tidy: Fix broken buildbot
Richard Thomson [Mon, 28 Mar 2016 04:15:41 +0000 (04:15 +0000)]
clang-tidy: Fix broken buildbot

VS 2013 does not support char16_t or char32_t

llvm-svn: 264563

8 years agoFix serialization/deserialization for __uuidof
David Majnemer [Mon, 28 Mar 2016 03:19:50 +0000 (03:19 +0000)]
Fix serialization/deserialization for __uuidof

I broke this back in r264529 because I forgot to serialize the UuidAttr
member.  Fix this by replacing the UuidAttr with a StringRef which is
properly serialized and deserialized.

llvm-svn: 264562

8 years agollvm/test/Transforms/FunctionImport/funcimport.ll: -stats REQUIRES +Asserts.
NAKAMURA Takumi [Mon, 28 Mar 2016 02:14:49 +0000 (02:14 +0000)]
llvm/test/Transforms/FunctionImport/funcimport.ll: -stats REQUIRES +Asserts.

llvm-svn: 264561

8 years agoR_AARCH64_PREL64 is relative.
Rafael Espindola [Mon, 28 Mar 2016 01:31:11 +0000 (01:31 +0000)]
R_AARCH64_PREL64 is relative.

llvm-svn: 264560

8 years ago[Coverage] Fix the way we load "<unknown>:func" records
Vedant Kumar [Mon, 28 Mar 2016 01:16:12 +0000 (01:16 +0000)]
[Coverage] Fix the way we load "<unknown>:func" records

When emitting coverage mappings for functions with local linkage and an
unknown filename, we use "<unknown>:func" for the PGO function name. The
problem is that we don't strip "<unknown>" from the name when loading
coverage data, like we do for other file names. Fix that and add a test.

llvm-svn: 264559

8 years agoBitcodeWriter: Replace dead code with an assertion, NFC
Duncan P. N. Exon Smith [Mon, 28 Mar 2016 00:03:12 +0000 (00:03 +0000)]
BitcodeWriter: Replace dead code with an assertion, NFC

The caller of ValueEnumerator::EnumerateOperandType never sends in
metadata.  Assert that, and remove the unnecessary logic.

llvm-svn: 264558

8 years agoBitcodeWriter: Reuse writeMetadataRecords, NFC
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:59:32 +0000 (23:59 +0000)]
BitcodeWriter: Reuse writeMetadataRecords, NFC

Change writeFunctionMetadata to call writeMetadataRecords.  For now
there's no functionality change, but makes it easy to serialize other
types of metadata in the function block in the future.

llvm-svn: 264557

8 years agoBitcodeWriter: Rename some functions for consistency, NFC
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:56:04 +0000 (23:56 +0000)]
BitcodeWriter: Rename some functions for consistency, NFC

To match writeMetadataRecords, writeNamedMetadata and
writeMetadataStrings, change:

    WriteModuleMetadata        => writeModuleMetadata
    WriteFunctionLocalMetadata => writeFunctionMetadata
    Write##CLASS               => write##CLASS

The only major change is "FunctionLocal" => "Function".  The point is to
be less specific, in preparation for emitting normal metadata records
inside function metadata blocks (currently we only emit
`LocalAsMetadata` there).

llvm-svn: 264556

8 years agoBitcodeWriter: Split out writeMetadataRecords, NFC
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:53:30 +0000 (23:53 +0000)]
BitcodeWriter: Split out writeMetadataRecords, NFC

Besides being a nice cleanup, this is preparation for reusing the code
in function metadata blocks.

llvm-svn: 264555

8 years agoBitcodeWriter: Restructure WriteFunctionLocalMetadata, NFC
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:38:36 +0000 (23:38 +0000)]
BitcodeWriter: Restructure WriteFunctionLocalMetadata, NFC

Use an early return to simplify logic.

llvm-svn: 264554

8 years agoBitcode: Fix MSVC bot failure from r264549
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:36:55 +0000 (23:36 +0000)]
Bitcode: Fix MSVC bot failure from r264549

make_unique => llvm::make_unique

llvm-svn: 264553

8 years agoBitcodeWriter: Simplify tracking of function-local metadata, NFC
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:22:31 +0000 (23:22 +0000)]
BitcodeWriter: Simplify tracking of function-local metadata, NFC

We don't really need a separate vector here; instead, point at a range
inside the main MDs array.  This matches how r264551 references the
ranges of strings and non-strings.

llvm-svn: 264552

8 years agoReapply ~"Bitcode: Collect all MDString records into a single blob"
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:17:54 +0000 (23:17 +0000)]
Reapply ~"Bitcode: Collect all MDString records into a single blob"

Spiritually reapply commit r264409 (reverted in r264410), albeit with a
bit of a redesign.

Firstly, avoid splitting the big blob into multiple chunks of strings.

r264409 imposed an arbitrary limit to avoid a massive allocation on the
shared 'Record' SmallVector.  The bug with that commit only reproduced
when there were more than "chunk-size" strings.  A test for this would
have been useless long-term, since we're liable to adjust the chunk-size
in the future.

Thus, eliminate the motivation for chunk-ing by storing the string sizes
in the blob.  Here's the layout:

    vbr6: # of strings
    vbr6: offset-to-blob
    blob:
       [vbr6]: string lengths
       [char]: concatenated strings

Secondly, make the output of llvm-bcanalyzer readable.

I noticed when debugging r264409 that llvm-bcanalyzer was outputting a
massive blob all in one line.  Past a small number, the strings were
impossible to split in my head, and the lines were way too long.  This
version adds support in llvm-bcanalyzer for pretty-printing.

    <STRINGS abbrevid=4 op0=3 op1=9/> num-strings = 3 {
      'abc'
      'def'
      'ghi'
    }

From the original commit:

Inspired by Mehdi's similar patch, http://reviews.llvm.org/D18342, this
should (a) slightly reduce bitcode size, since there is less record
overhead, and (b) greatly improve reading speed, since blobs are super
cheap to deserialize.

llvm-svn: 264551

8 years agoBitcodeWriter: Simplify and test writing blobs, NFC
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:04:04 +0000 (23:04 +0000)]
BitcodeWriter: Simplify and test writing blobs, NFC

Split helper out of EmitRecordWithAbbrevImpl called emitBlob to reduce
code duplication, and add a few tests for it.

No functionality change intended.

llvm-svn: 264550

8 years agoSupport: Implement StreamingMemoryObject::getPointer
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 23:00:59 +0000 (23:00 +0000)]
Support: Implement StreamingMemoryObject::getPointer

The implementation is fairly obvious.  This is preparation for using
some blobs in bitcode.

For clarity (and perhaps future-proofing?), I moved the call to
JumpToBit in BitstreamCursor::readRecord ahead of calling
MemoryObject::getPointer, since JumpToBit can theoretically (a) read
bytes, which (b) invalidates the blob pointer.

This isn't strictly necessary the two memory objects we have:

  - The return of RawMemoryObject::getPointer is valid until the memory
    object is destroyed.

  - StreamingMemoryObject::getPointer is valid until the next chunk is
    read from the stream.  Since the JumpToBit call is only going ahead
    to a word boundary, we'll never load another chunk.

However, reordering makes it clear by inspection that the blob returned
by BitstreamCursor::readRecord will be valid.

I added some tests for StreamingMemoryObject::getPointer and
BitstreamCursor::readRecord.

llvm-svn: 264549

8 years agoSupport: Move StreamingMemoryObject{,Test}.cpp, NFC
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 22:55:19 +0000 (22:55 +0000)]
Support: Move StreamingMemoryObject{,Test}.cpp, NFC

Change the filename to indicate this is a test, rename the tests, move
them into an anonymous namespace, and rename some variables.  All to
match our usual style before making further changes.

llvm-svn: 264548

8 years agoBitcode: Add SimpleBitstreamCursor::setArtificialByteLimit
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 22:49:32 +0000 (22:49 +0000)]
Bitcode: Add SimpleBitstreamCursor::setArtificialByteLimit

Allow users of SimpleBitstreamCursor to limit the number of bytes
available to the cursor.  This is preparation for instantiating a cursor
that isn't allowed to load more bytes from a StreamingMemoryObject (just
move around the ones already-loaded).

llvm-svn: 264547

8 years agoBitcode: Add SimpleBitstreamCursor::getPointerToByte, etc.
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 22:45:25 +0000 (22:45 +0000)]
Bitcode: Add SimpleBitstreamCursor::getPointerToByte, etc.

Add API to SimpleBitstreamCursor to allow users to translate between
byte addresses and pointers.

  - jumpToPointer: move the bit position to a particular pointer.
  - getPointerToByte: get the pointer for a particular byte.
  - getPointerToBit: get the pointer for the byte of the current bit.
  - getCurrentByteNo: convenience function for assertions and tests.

Mainly adds unit tests (getPointerToBit/Byte already has a use), but
also preparation for eventually using jumpToPointer.

llvm-svn: 264546

8 years agoBitcode: Split out SimpleBitstreamCursor
Duncan P. N. Exon Smith [Sun, 27 Mar 2016 22:40:55 +0000 (22:40 +0000)]
Bitcode: Split out SimpleBitstreamCursor

Split out SimpleBitstreamCursor from BitstreamCursor, which is a
lower-level cursor with no knowledge of bitcode blocks, abbreviations,
or records.  It just knows how to read bits and navigate the stream.

This is mainly organizational, to separate the API for manipulating raw
bits from that for bitcode concepts like Record and Block.

llvm-svn: 264545

8 years agoRevert "isPodLike: more precise"
JF Bastien [Sun, 27 Mar 2016 20:50:05 +0000 (20:50 +0000)]
Revert "isPodLike: more precise"

This reverts commit c45f2afac5d6855a4804456a0f718563dc47ada0.

Looks like it may be causing a failure, I'll revert for now.

                 from
lib/CodeGen/AsmPrinter/DwarfDebug.cpp:14:
/usr/include/c++/4.9.2/bits/stl_pair.h: In instantiation of
                 'std::pair<_T1, _T2>& std::pair<_T1,
                 _T2>::operator=(const std::pair<_T1, _T2>&) [with _T1 =
                 std::unique_ptr<llvm::DwarfTypeUnit>; _T2 = const
                 llvm::DICompositeType*]':

/usr/include/c++/4.9.2/bits/stl_pair.h:160:8: error: use of deleted
function 'std::unique_ptr<_Tp, _Dp>& std::unique_ptr<_Tp,
_Dp>::operator=(const std::unique_ptr<_Tp, _Dp>&) [with _Tp =
llvm::DwarfTypeUnit; _Dp = std::default_delete<llvm::DwarfTypeUnit>]'
  first = __p.first;
        ^

llvm-svn: 264544

8 years agoworkaround for an IR variable named %.
Sanjay Patel [Sun, 27 Mar 2016 20:44:35 +0000 (20:44 +0000)]
workaround for an IR variable named %.

(which SimplifyCFG can produce...)

llvm-svn: 264543

8 years agoadd scrubber for excessive leading whitespace
Sanjay Patel [Sun, 27 Mar 2016 20:43:02 +0000 (20:43 +0000)]
add scrubber for excessive leading whitespace

llvm-svn: 264542

8 years agoisPodLike: more precise
JF Bastien [Sun, 27 Mar 2016 20:32:21 +0000 (20:32 +0000)]
isPodLike: more precise

I tried to use isPodLike in:
  http://reviews.llvm.org/D18483

That failed because !is_class is too strict on platforms which don't yet
have is_trivially_copyable. This update tries to make isPodLike smarter
for platforms which don't have is_trivially_copyable, and AFAICT it
Should Just Work on all of them. I'll revert if the bots disagree with
me.

I'll also rename isPodLike to isTriviallyCopyable if this all works out,
since that's what the standard calls it now and one day we'll be rid of
isPodLike.

llvm-svn: 264541

8 years ago[modules] When encoding SourceLocations in bitcode, rotate the 'is macro' flag
Richard Smith [Sun, 27 Mar 2016 20:13:24 +0000 (20:13 +0000)]
[modules] When encoding SourceLocations in bitcode, rotate the 'is macro' flag
bit from the top bit to the bottom bit, so that we don't need 6 VBR6 hunks for
each macro location. Reduces libstdc++ module size by about 1%.

llvm-svn: 264540

8 years agoclang-tidy: Add check modernize-raw-string-literal
Richard Thomson [Sun, 27 Mar 2016 16:43:44 +0000 (16:43 +0000)]
clang-tidy: Add check modernize-raw-string-literal

llvm-svn: 264539

8 years agoUse DAG check to try to appease bot
Teresa Johnson [Sun, 27 Mar 2016 15:36:43 +0000 (15:36 +0000)]
Use DAG check to try to appease bot

Try to appease
http://bb.pgr.jp/builders/cmake-llvm-x86_64-linux/builds/34772. This was
the only check that didn't use DAG and it wasn't found.

llvm-svn: 264538

8 years ago[ThinLTO] Add optional import message and statistics
Teresa Johnson [Sun, 27 Mar 2016 15:27:30 +0000 (15:27 +0000)]
[ThinLTO] Add optional import message and statistics

Summary:
Add a statistic to count the number of imported functions. Also, add a
new -print-imports option to emit a trace of imported functions, that
works even for an NDEBUG build.

Note that emitOptimizationRemark does not work for the above printing as
it expects a Function object and DebugLoc, neither of which we have
with summary-based importing.

This is part 2 of D18487, the first part was committed separately as
r264536.

Reviewers: joker.eph

Subscribers: llvm-commits, joker.eph

Differential Revision: http://reviews.llvm.org/D18487

llvm-svn: 264537

8 years ago[ThinLTO] Don't try to import alias unless aliasee can be imported
Teresa Johnson [Sun, 27 Mar 2016 15:01:11 +0000 (15:01 +0000)]
[ThinLTO] Don't try to import alias unless aliasee can be imported

With r264503, aliases are now being added to the GlobalsToImport set
even when their aliasees can't be imported due to their linkage type.
While the importing worked correctly (the aliases imported as
declarations) due to the logic in doImportAsDefinition, there is no
point to adding them to the GlobalsToImport set.

Additionally, with D18487 it was resulting in incorrectly printing a
message indicating that the alias was imported.

To avoid this, delay adding aliases to the GlobalsToImport set until
after the linkage type of the aliasee is checked.

This patch is part of D18487.

llvm-svn: 264536

8 years agoFixing the non-x86 build by removing dependence on kmp_cpuid_t
Hal Finkel [Sun, 27 Mar 2016 13:24:09 +0000 (13:24 +0000)]
Fixing the non-x86 build by removing dependence on kmp_cpuid_t

The problem is that the definition of kmp_cpuinfo_t contains:

  char       name [3*sizeof (kmp_cpuid_t)]; // CPUID(0x80000002,0x80000003,0x80000004)

and kmp_cpuid_t is only defined when compiling for x86.

Differential Revision: http://reviews.llvm.org/D18245

llvm-svn: 264535

8 years agoEncapsulate a couple of on-disk structures a little more.
Richard Smith [Sun, 27 Mar 2016 07:28:06 +0000 (07:28 +0000)]
Encapsulate a couple of on-disk structures a little more.

llvm-svn: 264534

8 years agoRemove unused support for replacing declarations from chained AST files.
Richard Smith [Sun, 27 Mar 2016 05:52:25 +0000 (05:52 +0000)]
Remove unused support for replacing declarations from chained AST files.

llvm-svn: 264533

8 years ago[PowerPC] Map max/minnum intrinsics and fmax/fmin to ISD nodes for CTR-based loop...
Hal Finkel [Sun, 27 Mar 2016 05:40:56 +0000 (05:40 +0000)]
[PowerPC] Map max/minnum intrinsics and fmax/fmin to ISD nodes for CTR-based loop legality

Intrinsic::maxnum and Intrinsic::minnum, along with the associated libc
function calls (fmax[f], etc.) generally map to function calls after lowering.
For some vector types with QPX at least, however, we can legally lower these,
and we don't need to prohibit CTR-based loops on their account.

It turned out, however, that the logic that checked the opcodes associated with
intrinsics was broken (it would set the Opcode variable, but that variable was
later checked only if set for some otherwise-external function call.

This fixes the latter problem and adds the FMAX/MINNUM mappings.

llvm-svn: 264532

8 years agoAdd clang-tools-extra release notes
Richard Thomson [Sun, 27 Mar 2016 05:06:57 +0000 (05:06 +0000)]
Add clang-tools-extra release notes

llvm-svn: 264531

8 years agoUse the correct alignment for uuid descriptors
David Majnemer [Sun, 27 Mar 2016 04:46:14 +0000 (04:46 +0000)]
Use the correct alignment for uuid descriptors

The _GUID_ descriptors emitted by MSVC have alignment 8 for 64-bit
builds: we should do the same if the linker picks the "wrong" COMDAT.

llvm-svn: 264530

8 years agoImprove the representation of CXXUuidofExpr
David Majnemer [Sun, 27 Mar 2016 04:46:07 +0000 (04:46 +0000)]
Improve the representation of CXXUuidofExpr

Keep a pointer to the UuidAttr that the CXXUuidofExpr corresponds to.
This makes translating from __uuidof to the underlying constant a lot
more straightforward.

llvm-svn: 264529

8 years ago[Verifier] Reject PHIs using defs from own block.
Michael Kruse [Sat, 26 Mar 2016 23:32:57 +0000 (23:32 +0000)]
[Verifier] Reject PHIs using defs from own block.

Reject the following IR as malformed (assuming that %entry, %next are
not in a loop):

    next:
      %y = phi i32 [ 0, %entry ]
      %x = phi i32 [ %y, %entry ]

Such PHI nodes came up in PR26718. While there was no consensus on
whether or not this is valid IR, most opinions on that bug and in a
discussion on the llvm-dev mailing list tended towards a
"strict interpretation" (term by Joseph Tremoulet) of PHI node uses.
Also, the language reference explicitly states that "the use of each
incoming value is deemed to occur on the edge from the corresponding
predecessor block to the current block" and
`DominatorTree::dominates(Instruction*, Use&)` uses this definition as
well.

For the code mentioned in PR15384, clang does not compile to such PHIs
(anymore?). The test case still hangs when replacing `%tmp6` with `%tmp`
in revisions before r176366 (where PR15384 has been fixed). The
occurrence of %tmp6 therefore was probably unintentional. Its value is
not used except in other PHIs.

Reviewers: majnemer, reames, JosephTremoulet, bkramer, grosser, jdoerfert, kparzysz, sanjoy

Differential Revision: http://reviews.llvm.org/D18443

llvm-svn: 264528

8 years ago[SimplifyCFG] propagate branch metadata when creating select (PR26636)
Sanjay Patel [Sat, 26 Mar 2016 23:30:50 +0000 (23:30 +0000)]
[SimplifyCFG] propagate branch metadata when creating select (PR26636)

llvm-svn: 264527

8 years agominimize test cases
Sanjay Patel [Sat, 26 Mar 2016 23:09:25 +0000 (23:09 +0000)]
minimize test cases

These are tests for store transforms.
The loads, adds, and geps were irrelevant.

llvm-svn: 264526

8 years ago[NFC] Tweak diagnostic for template template arguments, to include template aliases.
Faisal Vali [Sat, 26 Mar 2016 20:46:45 +0000 (20:46 +0000)]
[NFC] Tweak diagnostic for template template arguments, to include template aliases.

The prior diagnostic (err_template_arg_not_class_template) would state that the template argument to a template template parameter can only be a class template, when it can also be a template alias.  The newly renamed diagnostic (err_template_arg_not_valid_template) mentions template aliases.

llvm-svn: 264522

8 years agoMake vp merge test more robust
Xinliang David Li [Sat, 26 Mar 2016 20:44:03 +0000 (20:44 +0000)]
Make vp merge test more robust

llvm-svn: 264521

8 years agollvm-dwp: Include the dwo name (if available) when diagnosing duplicate CU IDs from...
David Blaikie [Sat, 26 Mar 2016 20:32:14 +0000 (20:32 +0000)]
llvm-dwp: Include the dwo name (if available) when diagnosing duplicate CU IDs from dwp input files

If you're building dwps from other dwps, it can be hard to track down a
duplicate CU ID if it comes from two compilations of the same file in
different modes, etc. By including the .dwo path (which is hopefully
more unique than the file path) it can help track down where the
duplicates came from.

llvm-svn: 264520

8 years agoCheck if a path is already absolute before trying to make it so.
Bob Wilson [Sat, 26 Mar 2016 18:55:13 +0000 (18:55 +0000)]
Check if a path is already absolute before trying to make it so.

The FileSystem::makeAbsolute function has been calculating the current
working directory unconditionally, even when it is not needed. This calls
down to llvm::sys::fs::current_path, which is relatively expensive
because it stats two directories, regardless of whether those paths are
already in the stat cache. The net effect is that when using the
VFS, every stat during header search turns into three stats. With this
change, we get back to a single stat for absolute directory paths.

llvm-svn: 264519

8 years ago[ELF/LTO] Refactor to reduce indentation.
Davide Italiano [Sat, 26 Mar 2016 18:33:09 +0000 (18:33 +0000)]
[ELF/LTO] Refactor to reduce indentation.

Suggested by: Rui Ueyama.

llvm-svn: 264518

8 years ago[X86][AVX] Enabled SMUL_LOHI/UMUL_LOHI v8i32 vectors on AVX1 targets
Simon Pilgrim [Sat, 26 Mar 2016 18:32:13 +0000 (18:32 +0000)]
[X86][AVX] Enabled SMUL_LOHI/UMUL_LOHI v8i32 vectors on AVX1 targets

Correct splitting of v8i32 vectors into v4i32 vectors to prevent scalarization

llvm-svn: 264517

8 years agoRevert "NFC: static_assert instead of comment"
JF Bastien [Sat, 26 Mar 2016 18:20:02 +0000 (18:20 +0000)]
Revert "NFC: static_assert instead of comment"

This reverts commit fa36fcff16c7d4f78204d6296bf96c3558a4a672.

Causes the following Windows failure:

  C:\Buildbot\Slave\llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast\llvm.src\lib\CodeGen\MachineInstr.cpp(762):
  error C2338: must be trivially copyable to memmove

llvm-svn: 264516

8 years agoNFC: static_assert instead of comment
JF Bastien [Sat, 26 Mar 2016 18:14:27 +0000 (18:14 +0000)]
NFC: static_assert instead of comment

Summary: isPodLike is as close as we have for is_trivially_copyable.

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D18483

llvm-svn: 264515

8 years agoGeneralize the domain complexity restrictions
Johannes Doerfert [Sat, 26 Mar 2016 16:17:00 +0000 (16:17 +0000)]
Generalize the domain complexity restrictions

  This patch applies the restrictions on the number of domain conjuncts
  also to the domain parts of piecewise affine expressions we generate.
  To this end the wording is change slightly. It was needed to support
  complex additions featuring zext-instructions but it also fixes PR27045.

  lnt profitable runs reports only little changes that might be noise:
  Compile Time:
    Polybench/[...]/2mm                     +4.34%
    SingleSource/[...]/stepanov_container   -2.43%
  Execution Time:
    External/[...]/186_crafty               -2.32%
    External/[...]/188_ammp                 -1.89%
    External/[...]/473_astar                -1.87%

llvm-svn: 264514

8 years ago[Cxx1z-constexpr-lambda-P0170R1] Support parsing of constexpr specifier (and its...
Faisal Vali [Sat, 26 Mar 2016 16:11:37 +0000 (16:11 +0000)]
[Cxx1z-constexpr-lambda-P0170R1]  Support parsing of constexpr specifier (and its inference) on lambda expressions

Support the constexpr specifier on lambda expressions - and support its inference from the lambda call operator's body.

i.e.
  auto L = [] () constexpr { return 5; };
  static_assert(L() == 5); // OK
  auto Implicit = [] (auto a) { return a; };
  static_assert(Implicit(5) == 5);

We do not support evaluation of lambda's within constant expressions just yet.

Implementation Strategy:
  - teach ParseLambdaExpressionAfterIntroducer to expect a constexpr specifier and mark the invented function call operator's declarator's decl-specifier with it; Have it emit fixits for multiple decl-specifiers (mutable or constexpr) in this location.
  - for cases where constexpr is not explicitly specified, have buildLambdaExpr check whether the invented function call operator satisfies the requirements of a constexpr function, by calling CheckConstexprFunctionDecl/Body.

Much obliged to Richard Smith for his patience and his care, in ensuring the code is clang-worthy.

llvm-svn: 264513

8 years ago[X86][AVX] Enabled MULHS/MULHU v16i16 vectors on AVX1 targets
Simon Pilgrim [Sat, 26 Mar 2016 15:44:55 +0000 (15:44 +0000)]
[X86][AVX] Enabled MULHS/MULHU v16i16 vectors on AVX1 targets

Correct splitting of v16i16 vectors into v8i16 vectors to prevent scalarization

Differential Revision: http://reviews.llvm.org/D18307

llvm-svn: 264512

8 years ago[X86][SSE] Add MULHS/MULHU custom lowering for i8 vectors
Simon Pilgrim [Sat, 26 Mar 2016 15:27:20 +0000 (15:27 +0000)]
[X86][SSE] Add MULHS/MULHU custom lowering for i8 vectors

Currently this is to mainly to prevent scalarization of integer division by constants.

Differential Revision: http://reviews.llvm.org/D18307

llvm-svn: 264511

8 years ago[X86][SSE] Added v64i8 vector integer multiply tests
Simon Pilgrim [Sat, 26 Mar 2016 09:50:06 +0000 (09:50 +0000)]
[X86][SSE] Added v64i8 vector integer multiply tests

llvm-svn: 264510

8 years ago[X86][AVX512BW] AVX512BW can sign-extend v32i8 to v32i16 for simpler v32i8 multiplies.
Simon Pilgrim [Sat, 26 Mar 2016 09:44:27 +0000 (09:44 +0000)]
[X86][AVX512BW] AVX512BW can sign-extend v32i8 to v32i16 for simpler v32i8 multiplies.

Only pre-AVX512BW targets need to split v32i8 vectors.

llvm-svn: 264509

8 years ago[PowerPC] Disable the CTR optimization in the presence of {min,max}num
David Majnemer [Sat, 26 Mar 2016 09:42:31 +0000 (09:42 +0000)]
[PowerPC] Disable the CTR optimization in the presence of {min,max}num

The minnum and maxnum intrinsics get lowered to libcalls which
invalidates the CTR optimization.

This fixes PR27083.

llvm-svn: 264508

8 years ago[X86][SSE] Refreshed vector integer multiply tests
Simon Pilgrim [Sat, 26 Mar 2016 09:35:48 +0000 (09:35 +0000)]
[X86][SSE] Refreshed vector integer multiply tests

Add all 256-bit vector tests.
Added AVX512F/AVX512BW test targets.
Renamed tests something more meaningful.

llvm-svn: 264507

8 years ago[X86][SSE] Don't duplicate Lower256IntArith functionality in LowerMul. NFC.
Simon Pilgrim [Sat, 26 Mar 2016 09:29:04 +0000 (09:29 +0000)]
[X86][SSE] Don't duplicate Lower256IntArith functionality in LowerMul. NFC.

LowerMul v32i8 on AVX2 needs to split the 256-bit sources to allow sign-extension back to v16i16 to occur. Since this is basically the same as Lower256IntArith we simplify by using that here instead.

llvm-svn: 264506

8 years agoMinor code cleanup. NFC.
Junmo Park [Sat, 26 Mar 2016 06:04:55 +0000 (06:04 +0000)]
Minor code cleanup. NFC.

llvm-svn: 264505

8 years ago[Power9] Implement new altivec instructions: permute, count zero, extend sign, negate...
Chuang-Yu Cheng [Sat, 26 Mar 2016 05:46:11 +0000 (05:46 +0000)]
[Power9] Implement new altivec instructions: permute, count zero, extend sign, negate, parity, shift/rotate, mul10

This change implements the following vector operations:
- vclzlsbb vctzlsbb vctzb vctzd vctzh vctzw
- vextsb2w vextsh2w vextsb2d vextsh2d vextsw2d
- vnegd vnegw
- vprtybd vprtybq vprtybw
- vbpermd vpermr
- vrlwnm vrlwmi vrldnm vrldmi vslv vsrv
- vmul10cuq vmul10uq vmul10ecuq vmul10euq

28 instructions

Thanks Nemanja, Kit for invaluable hints and discussion!
Reviewers: hal, nemanja, kbarton, tjablin, amehsan

Phabricator: http://reviews.llvm.org/D15887
llvm-svn: 264504

8 years agoThinLTO: use the callgraph from the combined index to drive the FunctionImporter
Mehdi Amini [Sat, 26 Mar 2016 05:40:34 +0000 (05:40 +0000)]
ThinLTO: use the callgraph from the combined index to drive the FunctionImporter

Summary:
Now that the summary contains the full reference/call graph, we can
replace the existing function importer that loads and inspect the IR
to iteratively walk the call graph by a traversal based purely on the
summary information. Decouple the actual importing decision from any
IR manipulation.

Reviewers: tejohnson

Subscribers: llvm-commits, joker.eph

Differential Revision: http://reviews.llvm.org/D18343

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 264503

8 years agoabtest: Cleanup, improve comments
Matthias Braun [Sat, 26 Mar 2016 04:07:55 +0000 (04:07 +0000)]
abtest: Cleanup, improve comments

llvm-svn: 264502

8 years ago[libprofile] Adjust dummy condition to be more forgiving.
Sean Silva [Sat, 26 Mar 2016 04:01:57 +0000 (04:01 +0000)]
[libprofile] Adjust dummy condition to be more forgiving.

On PS4, we have to fake environment variables by passing extra command
line arguments, so the dummy test `argc > 1` was failing.
The condition is just a dummy condition that the compiler can't fold
away, so the number is arbitrary as long as the condition is false.

Increase the number it compares against.

llvm-svn: 264491

8 years ago[libprofile] Add some missing `env` prefixes on commands.
Sean Silva [Sat, 26 Mar 2016 03:37:45 +0000 (03:37 +0000)]
[libprofile] Add some missing `env` prefixes on commands.

llvm-svn: 264490

8 years agoRename ModuleSummaryIndex::modPathStringEntries() into modulePaths()
Mehdi Amini [Sat, 26 Mar 2016 03:35:38 +0000 (03:35 +0000)]
Rename ModuleSummaryIndex::modPathStringEntries() into modulePaths()

It now return the map instead of an iterator.

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 264489

8 years agoStop depending on the keys in a hash table being in a particular order.
Richard Smith [Sat, 26 Mar 2016 03:06:42 +0000 (03:06 +0000)]
Stop depending on the keys in a hash table being in a particular order.

llvm-svn: 264488

8 years agoStop testing the unspecified order in which the OnDiskHashTable stores entries.
Richard Smith [Sat, 26 Mar 2016 02:02:59 +0000 (02:02 +0000)]
Stop testing the unspecified order in which the OnDiskHashTable stores entries.

llvm-svn: 264487

8 years agoDon't force OnDiskHashTables to have a minimum of 64 buckets. That's
Richard Smith [Sat, 26 Mar 2016 01:49:50 +0000 (01:49 +0000)]
Don't force OnDiskHashTables to have a minimum of 64 buckets. That's
preposterously large for some lookup tables -- in C++ classes generated by
template instantiation, it's very common for the number of lookup results to be
either 1 or 2.

This reduces size of a libstdc++ module by 7-15%.

llvm-svn: 264486

8 years agoWhen we import the definition for a Tagdecl, propagate its completeness too.
Sean Callanan [Sat, 26 Mar 2016 00:37:55 +0000 (00:37 +0000)]
When we import the definition for a Tagdecl, propagate its completeness too.

The ASTImporter completes the full definiton for a TagDecl in several places,
including the type-deport logic.  When this happens, we should also propagate
the bit that says that this is a complete definition.  This makes (for example)
lambdas callable.

<rdar://problem/22864976>

llvm-svn: 264485

8 years agoabtest: Check all files, improve announcements
Matthias Braun [Sat, 26 Mar 2016 00:36:58 +0000 (00:36 +0000)]
abtest: Check all files, improve announcements

Simply searching for gooddir/* and baddir/* instead of *.s the script
should also work with .o files.

llvm-svn: 264484

8 years agoRecord all translation units with more than one function in them (e.g., blocks).
Sean Callanan [Sat, 26 Mar 2016 00:30:40 +0000 (00:30 +0000)]
Record all translation units with more than one function in them (e.g., blocks).

Blocks and lambdas have their implementation functions stored in the IR for an
expression.  If we put the block/lambda into a result variable it needs to stay
around.  As a heuristic, remember any execution unit that has more than one
function in it.

<rdar://problem/22864976>

llvm-svn: 264483

8 years agoPut my abtest scripts into the util directory
Matthias Braun [Sat, 26 Mar 2016 00:23:59 +0000 (00:23 +0000)]
Put my abtest scripts into the util directory

See http://lists.llvm.org/pipermail/llvm-dev/2016-March/097640.html
There is also a description/tutorial in the comments of the abtest.py
file.

llvm-svn: 264482

8 years ago[asan] bump the scariness score of read-after-frees (based on feedback from the Chrom...
Kostya Serebryany [Sat, 26 Mar 2016 00:00:19 +0000 (00:00 +0000)]
[asan] bump the scariness score of read-after-frees (based on feedback from the Chrome security team)

llvm-svn: 264481

8 years ago[profile] Add integration test to validate PGO function names
Vedant Kumar [Fri, 25 Mar 2016 23:56:23 +0000 (23:56 +0000)]
[profile] Add integration test to validate PGO function names

Differential Revision: http://reviews.llvm.org/D18485

llvm-svn: 264480

8 years ago[Support] Switch to RAII helper for error-as-out-parameter idiom.
Lang Hames [Fri, 25 Mar 2016 23:54:32 +0000 (23:54 +0000)]
[Support] Switch to RAII helper for error-as-out-parameter idiom.

As discussed on the llvm-commits thread for r264467.

llvm-svn: 264479

8 years agoRemoved LoggingDiagnosticConsumer, an unused class.
Sean Callanan [Fri, 25 Mar 2016 23:51:25 +0000 (23:51 +0000)]
Removed LoggingDiagnosticConsumer, an unused class.

llvm-svn: 264478

8 years agoImprove the reliability of file renaming in Windows by having the compiler retry
Sunil Srivastava [Fri, 25 Mar 2016 23:41:28 +0000 (23:41 +0000)]
Improve the reliability of file renaming in Windows by having the compiler retry
the rename operation on 3 error conditions of ReplaceFileW() that it was
previously bailing out on.

Patch by Douglas Yung!

Differential Revision: http://reviews.llvm.org/D17903

llvm-svn: 264477

8 years agoFix FILE * leak in Python API
Stephane Sezer [Fri, 25 Mar 2016 23:40:32 +0000 (23:40 +0000)]
Fix FILE * leak in Python API

Summary:
This fixes a leak introduced by some of these changes:
r257644
r250530
r250525

The changes made in these patches result in leaking the FILE* passed
to SetImmediateOutputFile. GetStream() will dup() the fd held by the
python caller and create a new FILE*. It will then pass this FILE*
to SetImmediateOutputFile, which always uses the flag
transfer_ownership=false when it creates a File from the FILE*.

Since transfer_ownership is false, the lldb File destructor will not
close the underlying FILE*. Because this FILE* came from a dup-ed fd,
it will also not be closed when the python caller closes its file.

Leaking the FILE* causes issues if the same file is used multiple times
by different python callers during the same lldb run, even if these
callers open and close the python file properly, as you can end up
with issues due to multiple buffered writes to the same file.

Reviewers: granata.enrico, zturner, clayborg

Subscribers: zturner, lldb-commits, sas

Differential Revision: http://reviews.llvm.org/D18459

Change by Francis Ricci <fjricci@fb.com>

llvm-svn: 264476

8 years agoStringMap/DenseMap unittests: use piecewise_construct and ensure no copy occurs.
Mehdi Amini [Fri, 25 Mar 2016 23:25:06 +0000 (23:25 +0000)]
StringMap/DenseMap unittests: use piecewise_construct and ensure no copy occurs.

This makes us no longer relying on move-construction elision by the compiler.
Suggested by D. Blaikie.

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 264475

8 years agoAdd a 'language cplusplus demangle' command. This can be useful to provide a low...
Enrico Granata [Fri, 25 Mar 2016 23:14:24 +0000 (23:14 +0000)]
Add a 'language cplusplus demangle' command. This can be useful to provide a low-friction reproduction for issues with the LLDB demangling of C++ symbols

llvm-svn: 264474

8 years ago[Object] Make createMachOObjectFile return Expected<...> rather than
Lang Hames [Fri, 25 Mar 2016 23:11:52 +0000 (23:11 +0000)]
[Object] Make createMachOObjectFile return Expected<...> rather than
ErrorOr<...>.

llvm-svn: 264473

8 years agoAllow value forwarding past release fences in GVN
Philip Reames [Fri, 25 Mar 2016 22:40:35 +0000 (22:40 +0000)]
Allow value forwarding past release fences in GVN

A release fence acts as a publication barrier for stores within the current thread to become visible to other threads which might observe the release fence. It does not require the current thread to observe stores performed on other threads. As a result, we can allow store-load and load-load forwarding across a release fence.

We choose to be much more conservative about stores.  In theory, nothing prevents us from shifting a store from after a release fence to before it, and then eliminating the preceeding (previously fenced) store.  Doing this without actually moving the second store is likely also legal, but we chose to be conservative at this time.

The LangRef indicates only atomic loads and stores are effected by fences. This patch chooses to be far more conservative then that.

This is the GVN companion to http://reviews.llvm.org/D11434 which applied the same logic in EarlyCSE and has been baking in tree for a while now.

Differential Revision: http://reviews.llvm.org/D11436

llvm-svn: 264472

8 years agoDon't warn on "use" of undefined inline function that isn't actually an ODR
Richard Smith [Fri, 25 Mar 2016 22:29:27 +0000 (22:29 +0000)]
Don't warn on "use" of undefined inline function that isn't actually an ODR
use. In order for this to fire, the function needed to be a templated function
marked 'constexpr' and declared but not defined. This weird pattern appears in
libstdc++'s alloc_traits.h.

llvm-svn: 264471