platform/upstream/llvm.git
7 years ago[sanitizer] allocator: split the local cache class into two, one for 32-bit allocator...
Kostya Serebryany [Tue, 23 Aug 2016 00:30:43 +0000 (00:30 +0000)]
[sanitizer] allocator: split the local cache class into two, one for 32-bit allocator and one for 64-bit one. NFC. The two imlementations will diverge in the following changes.

llvm-svn: 279495

7 years agoFix style in some Clang-tidy checks documentation.
Eugene Zelenko [Tue, 23 Aug 2016 00:19:43 +0000 (00:19 +0000)]
Fix style in some Clang-tidy checks documentation.

Differential revision: https://reviews.llvm.org/D23728

llvm-svn: 279494

7 years ago[MemorySSA] Remove unused field. NFC.
George Burgess IV [Mon, 22 Aug 2016 23:40:01 +0000 (23:40 +0000)]
[MemorySSA] Remove unused field. NFC.

Given that we're not currently using blocker info, and whether or not we
will end up using it it is unclear, don't waste 8 (or 4) bytes of memory
per path node.

llvm-svn: 279493

7 years ago[InstSimplify] add helper function for SimplifyICmpInst(); NFCI
Sanjay Patel [Mon, 22 Aug 2016 23:12:02 +0000 (23:12 +0000)]
[InstSimplify] add helper function for SimplifyICmpInst(); NFCI

And add a FIXME because the helper excludes folds for vectors. It's
not clear yet how many of these are actually testable (and therefore
necessary?) because later analysis uses computeKnownBits and other
methods to catch many of these cases.

llvm-svn: 279492

7 years agoTypo.
Adrian Prantl [Mon, 22 Aug 2016 22:38:18 +0000 (22:38 +0000)]
Typo.

llvm-svn: 279491

7 years agoAdd comments. NFC
Adrian Prantl [Mon, 22 Aug 2016 22:38:16 +0000 (22:38 +0000)]
Add comments. NFC

llvm-svn: 279490

7 years agoAdd the second half of the testcase I should have added in 279485.
Adrian Prantl [Mon, 22 Aug 2016 22:30:34 +0000 (22:30 +0000)]
Add the second half of the testcase I should have added in 279485.

llvm-svn: 279489

7 years agoMigrate from NodeType * to NodeRef.
Tim Shen [Mon, 22 Aug 2016 22:30:27 +0000 (22:30 +0000)]
Migrate from NodeType * to NodeRef.

llvm-svn: 279488

7 years agoFix crash from assert in r279466.
Pete Cooper [Mon, 22 Aug 2016 22:27:05 +0000 (22:27 +0000)]
Fix crash from assert in r279466.

The assert in r279466 checks that we call the correct version of
Intrinsic::getName.  The version which accepts only an ID should not
be used for intrinsics with overloaded types.  The global-isel
code was calling the wrong version.  The test CodeGen/AArch64/GlobalISel/arm64-irtranslator.ll
will ensure that we call the correct version from now on.

llvm-svn: 279487

7 years agoFix regression introduced by r279164: only pass definitions as the PatternDef
Richard Smith [Mon, 22 Aug 2016 22:25:03 +0000 (22:25 +0000)]
Fix regression introduced by r279164: only pass definitions as the PatternDef
to DiagnoseUninstantiableTemplate, teach hasVisibleDefinition to correctly
determine whether a function definition is visible, and mark both the function
and the template as visible when merging function template definitions to
provide hasVisibleDefinition with the relevant information.

The change to always pass the right declaration as the PatternDef to
DiagnoseUninstantiableTemplate also caused those checks to happen before other
diagnostics in InstantiateFunctionDefinition, giving worse diagnostics for the
same situations, so I sunk the relevant diagnostics into
DiagnoseUninstantiableTemplate. Those parts of this patch are based on changes
in reviews.llvm.org/D23492 by Vassil Vassilev.

llvm-svn: 279486

7 years agoModule debug info: Don't assert when encountering an incomplete definition
Adrian Prantl [Mon, 22 Aug 2016 22:23:58 +0000 (22:23 +0000)]
Module debug info: Don't assert when encountering an incomplete definition
in isDefinedInClangModule() and assume that the incomplete definition
is not defined in the module.

This broke the -gmodules self host recently.
rdar://problem/27894367

llvm-svn: 279485

7 years agoADT: Separate some list manipulation API into ilist_base, NFC
Duncan P. N. Exon Smith [Mon, 22 Aug 2016 22:21:07 +0000 (22:21 +0000)]
ADT: Separate some list manipulation API into ilist_base, NFC

Separate algorithms in iplist<T> that don't depend on T into ilist_base,
and unit test them.

While I was adding unit tests for these algorithms anyway, I also added
unit tests for ilist_node_base and ilist_sentinel<T>.

To make the algorithms and unit tests easier to write, I also did the
following minor changes as a drive-by:
- encapsulate Prev/Next in ilist_node_base to so that algorithms are
  easier to read, and
- update ilist_node_access API to take nodes by reference.

There should be no real functionality change here.

llvm-svn: 279484

7 years agoFix header comment for unittests/ADT/ilistTest.cpp
Duncan P. N. Exon Smith [Mon, 22 Aug 2016 22:04:16 +0000 (22:04 +0000)]
Fix header comment for unittests/ADT/ilistTest.cpp

llvm-svn: 279483

7 years ago[ADT] Actually mutate the iterator VisitStack.back().second, not its copy.
Tim Shen [Mon, 22 Aug 2016 21:59:26 +0000 (21:59 +0000)]
[ADT] Actually mutate the iterator VisitStack.back().second, not its copy.

Summary: Before the change, *Opt never actually gets updated by the end
of toNext(), so for every next time the loop has to start over from
child_begin(). This bug doesn't affect the correctness, since Visited prevents
it from re-entering the same node again; but it's slow.

Reviewers: dberris, dblaikie, dannyb

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D23649

llvm-svn: 279482

7 years ago[SemaObjC] Do not RebuildObjCMessageExpr without valid method decl
Bruno Cardoso Lopes [Mon, 22 Aug 2016 21:50:22 +0000 (21:50 +0000)]
[SemaObjC] Do not RebuildObjCMessageExpr without valid method decl

Fix crash-on-invalid in ObjC Sema by avoiding to rebuild a message
expression to a 'super' class in case the method to call does not exist
(i.e. comes from another missing identifier).

In this case, the typo transform is invoked upon the message expression
in an attempt to solve a typo in a 'super' call parameters, but it
crashes since it assumes the method to call has a valid declaration.

rdar://problem/27305403

llvm-svn: 279481

7 years agoAdd support for '|' in expressions.
Rafael Espindola [Mon, 22 Aug 2016 21:33:35 +0000 (21:33 +0000)]
Add support for '|' in expressions.

llvm-svn: 279480

7 years ago[AST] Remove unused function, to silence a GCC7 warning.
Davide Italiano [Mon, 22 Aug 2016 21:33:12 +0000 (21:33 +0000)]
[AST] Remove unused function, to silence a GCC7 warning.

llvm-svn: 279479

7 years ago[InstCombine] change param type from Instruction to BinaryOperator for icmp helpers...
Sanjay Patel [Mon, 22 Aug 2016 21:24:29 +0000 (21:24 +0000)]
[InstCombine] change param type from Instruction to BinaryOperator for icmp helpers; NFCI

This saves some casting in the helper functions and eases some further refactoring.

llvm-svn: 279478

7 years ago[GraphTraits] Replace all NodeType usage with NodeRef
Tim Shen [Mon, 22 Aug 2016 21:09:30 +0000 (21:09 +0000)]
[GraphTraits] Replace all NodeType usage with NodeRef

This should finish the GraphTraits migration.

Differential Revision: http://reviews.llvm.org/D23730

llvm-svn: 279475

7 years agoADT: Remove ilist_*sentinel_traits, NFC
Duncan P. N. Exon Smith [Mon, 22 Aug 2016 20:51:00 +0000 (20:51 +0000)]
ADT: Remove ilist_*sentinel_traits, NFC

Remove all the dead code around ilist_*sentinel_traits.  This is a
follow-up to gutting them as part of r279314 (originally r278974),
staged to prevent broken builds in sub-projects.

Uses were removed from clang in r279457 and lld in r279458.

llvm-svn: 279473

7 years ago[InstCombine] use m_APInt to allow icmp (shr exact X, Y), 0 folds for splat constant...
Sanjay Patel [Mon, 22 Aug 2016 20:45:06 +0000 (20:45 +0000)]
[InstCombine] use m_APInt to allow icmp (shr exact X, Y), 0 folds for splat constant vectors

llvm-svn: 279472

7 years agoAdd ADT headers to the cmake headers directory for LLVMSupport. NFC.
Pete Cooper [Mon, 22 Aug 2016 20:38:53 +0000 (20:38 +0000)]
Add ADT headers to the cmake headers directory for LLVMSupport.  NFC.

Xcode and MSVC list the headers and source files for each library.

LLVMSupport lists included the source files for ADT but not the headers.  This
add the ADT headers so that they are browsable by the UI.

llvm-svn: 279470

7 years ago[compiler-rt] Use flags found when configuring builtins during compilation
Francis Ricci [Mon, 22 Aug 2016 20:33:47 +0000 (20:33 +0000)]
[compiler-rt] Use flags found when configuring builtins during compilation

Summary: This fixes the omission of -fPIC when building the builtins.

Reviewers: compnerd, beanz

Subscribers: dberris, llvm-commits

Differential Revision: https://reviews.llvm.org/D23729

llvm-svn: 279469

7 years ago[compiler-rt] Don't build ubsan cxxabi sources when unused
Francis Ricci [Mon, 22 Aug 2016 20:27:21 +0000 (20:27 +0000)]
[compiler-rt] Don't build ubsan cxxabi sources when unused

Summary:
On apple targets, when SANITIZER_CAN_USE_CXXABI is false,
the ubsan cxxabi sources aren't built, since they're unused.
Do this on non-apple targets as well.

This fixes errors when linking sanitizers if c++ abi is
unavailable.

Reviewers: pcc, kubabrecka, beanz

Subscribers: rnk, llvm-commits, kubabrecka, compnerd, dberris

Differential Revision: https://reviews.llvm.org/D23638

llvm-svn: 279467

7 years agoAdd comments and an assert to follow-up on r279113. NFC.
Pete Cooper [Mon, 22 Aug 2016 20:18:28 +0000 (20:18 +0000)]
Add comments and an assert to follow-up on r279113.  NFC.

Philip commented on r279113 to ask for better comments as to
when to use the different versions of getName.  Its also possible
to assert in the simple case that we aren't an overloaded intrinsic
as those have to use the more capable version of getName.

Thanks for the comments Philip.

llvm-svn: 279466

7 years agoIDFCalculator: Remove unused field.
Daniel Berlin [Mon, 22 Aug 2016 19:52:23 +0000 (19:52 +0000)]
IDFCalculator: Remove unused field.

llvm-svn: 279465

7 years agoAMDGPU: Split SILowerControlFlow into two pieces
Matt Arsenault [Mon, 22 Aug 2016 19:33:16 +0000 (19:33 +0000)]
AMDGPU: Split SILowerControlFlow into two pieces

Do most of the lowering in a pre-RA pass. Keep the skip jump
insertion late, plus a few other things that require more
work to move out.

One concern I have is now there may be COPY instructions
which do not have the necessary implicit exec uses
if they will be lowered to v_mov_b32.

This has a positive effect on SGPR usage in shader-db.

llvm-svn: 279464

7 years agoAMDGPU: Handle structs directly in AMDGPUABIInfo
Matt Arsenault [Mon, 22 Aug 2016 19:25:59 +0000 (19:25 +0000)]
AMDGPU: Handle structs directly in AMDGPUABIInfo

Structs are currently handled as pointer + byval, which makes AMDGPU
LLVM backend generate incorrect code when structs are used. This patch
changes struct argument to be handled directly and without flattening,
which Clover (Mesa 3D Gallium OpenCL state tracker) will be able to
handle. Flattening would expand the struct to individual elements and
pass each as a separate argument, which Clover can not
handle. Furthermore, such expansion does not fit the OpenCL
programming model which requires to explicitely specify each argument
index, size and memory location.

Patch by Vedran Miletić

llvm-svn: 279463

7 years agoMSSA: Factor out phi node placement
Daniel Berlin [Mon, 22 Aug 2016 19:14:30 +0000 (19:14 +0000)]
MSSA: Factor out phi node placement

llvm-svn: 279462

7 years agoMSSA: Only rename accesses whose defining access is nullptr
Daniel Berlin [Mon, 22 Aug 2016 19:14:16 +0000 (19:14 +0000)]
MSSA: Only rename accesses whose defining access is nullptr

llvm-svn: 279461

7 years ago[SimplifyCFG] Rewrite SinkThenElseCodeToEnd
James Molloy [Mon, 22 Aug 2016 19:07:15 +0000 (19:07 +0000)]
[SimplifyCFG] Rewrite SinkThenElseCodeToEnd

[Recommitting now an unrelated assertion in SROA is sorted out]

The new version has several advantages:
  1) IMSHO it's more readable and neater
  2) It handles loads and stores properly
  3) It can handle any number of incoming blocks rather than just two. I'll be taking advantage of this in a followup patch.

With this change we can now finally sink load-modify-store idioms such as:

    if (a)
      return *b += 3;
    else
      return *b += 4;

    =>

    %z = load i32, i32* %y
    %.sink = select i1 %a, i32 5, i32 7
    %b = add i32 %z, %.sink
    store i32 %b, i32* %y
    ret i32 %b

When this works for switches it'll be even more powerful.

Round 4. This time we should handle all instructions correctly, and not replace any operands that need to be constant with variables.

This was really hard to determine safely, so the helper function should be put into the Instruction API. I'll do that as a followup.

llvm-svn: 279460

7 years ago[ELF] Only print symbol name when it is available
Petr Hosek [Mon, 22 Aug 2016 19:01:53 +0000 (19:01 +0000)]
[ELF] Only print symbol name when it is available

Not only symbols (like sections) have names, in case where we
fail to create relocation against such symbol, we should not
print out an empty string, instead we should print a generic
message.

Differential Revision: https://reviews.llvm.org/D23731

llvm-svn: 279459

7 years agoADT: Remove uses of ilist_*sentinel_traits, NFC
Duncan P. N. Exon Smith [Mon, 22 Aug 2016 18:59:33 +0000 (18:59 +0000)]
ADT: Remove uses of ilist_*sentinel_traits, NFC

llvm-svn: 279458

7 years agoADT: Remove uses of ilist_*sentinel_traits, NFC
Duncan P. N. Exon Smith [Mon, 22 Aug 2016 18:57:44 +0000 (18:57 +0000)]
ADT: Remove uses of ilist_*sentinel_traits, NFC

llvm-svn: 279457

7 years ago[ELF] ICF should respect section alignment
Petr Hosek [Mon, 22 Aug 2016 18:53:09 +0000 (18:53 +0000)]
[ELF] ICF should respect section alignment

When performing ICF, we have to respect the alignment requirement
of each section within each group.

Differential Revision: https://reviews.llvm.org/D23732

llvm-svn: 279456

7 years ago[CUDA] Collapsed offload actions should not be top-level jobs.
Artem Belevich [Mon, 22 Aug 2016 18:50:34 +0000 (18:50 +0000)]
[CUDA] Collapsed offload actions should not be top-level jobs.

If they are, we end up with the last intermediary output preserved
in the current directory after compilation.

Added a test case to verify that we're using appropriate filenames
for outputs of different phases.

Differential Revision: https://reviews.llvm.org/D23526

llvm-svn: 279455

7 years ago[SROA] Remove incorrect assertion
James Molloy [Mon, 22 Aug 2016 18:49:42 +0000 (18:49 +0000)]
[SROA] Remove incorrect assertion

Confirmed with aprantl, this assertion is incorrect - code can get here (for example 80-bit FP types) and if it does it's benign. This is exposed by a completely unrelated patch of mine, so stop the compiler falling over.

Original differential: http://reviews.llvm.org/D16187
aprantl's advice to remove assertion: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160815/382129.html

llvm-svn: 279454

7 years agoAdd missing include that caused a test failure on Windows. Thanks to STL for the...
Marshall Clow [Mon, 22 Aug 2016 18:45:31 +0000 (18:45 +0000)]
Add missing include that caused a test failure on Windows. Thanks to STL for the patch. No functional change.

llvm-svn: 279453

7 years ago[ELF] Define __ehdr_start symbol
Petr Hosek [Mon, 22 Aug 2016 18:44:04 +0000 (18:44 +0000)]
[ELF] Define __ehdr_start symbol

This symbol can be used by the program to examine its own headers.

Differential Revision: https://reviews.llvm.org/D23750

llvm-svn: 279452

7 years ago[compiler-rt] Make cxxabi availability configurable in cmake
Francis Ricci [Mon, 22 Aug 2016 18:31:37 +0000 (18:31 +0000)]
[compiler-rt] Make cxxabi availability configurable in cmake

Summary:
Allow for use of cxxabi to be disabled from cmake.

This will make sanitizers usable when c++ abi is unavailable.

Reviewers: pcc, rnk, samsonov, beanz, compnerd

Subscribers: llvm-commits, compnerd, dberris

Differential Revision: https://reviews.llvm.org/D23639

llvm-svn: 279451

7 years ago[cmake] Fix append_rtti_flag macro and bad var name
Filipe Cabecinhas [Mon, 22 Aug 2016 18:30:37 +0000 (18:30 +0000)]
[cmake] Fix append_rtti_flag macro and bad var name

llvm-svn: 279450

7 years ago[SSP] Do not set __guard_local to hidden for OpenBSD SSP
Tim Shen [Mon, 22 Aug 2016 18:26:27 +0000 (18:26 +0000)]
[SSP] Do not set __guard_local to hidden for OpenBSD SSP

__guard_local is defined as long on OpenBSD. If the source file contains
a definition of __guard_local, it mismatches with the int8 pointer type
used in LLVM. In that case, Module::getOrInsertGlobal() returns a
cast operation instead of a GlobalVariable. Trying to set the
visibility on the cast operation leads to random segfaults (seen when
compiling the OpenBSD kernel, which also runs with stack protection).

In the kernel, the hidden attribute does not matter. For userspace code,
__guard_local is defined as hidden in the startup code. If a program
re-defines __guard_local, the definition from the startup code will
either win or the linker complains about multiple definitions
(depending on whether the re-defined __guard_local is placed in the
common segment or not).

It also matches what gcc on OpenBSD does.

Thanks Stefan Kempf <sisnkemp@gmail.com> for the patch!

Differential Revision: http://reviews.llvm.org/D23674

llvm-svn: 279449

7 years ago[InstCombine] Allow sinking from unique predecessor with multiple edges
Jun Bum Lim [Mon, 22 Aug 2016 18:21:56 +0000 (18:21 +0000)]
[InstCombine] Allow sinking from unique predecessor with multiple edges

Summary: We can allow sinking if the single user block has only one unique predecessor, regardless of the number of edges. Note that a switch statement with multiple cases can have the same destination.

Reviewers: mcrosier, majnemer, spatel, reames

Subscribers: reames, mcrosier, llvm-commits

Differential Revision: https://reviews.llvm.org/D23722

llvm-svn: 279448

7 years agoRevert "[SimplifyCFG] Rewrite SinkThenElseCodeToEnd"
James Molloy [Mon, 22 Aug 2016 18:13:12 +0000 (18:13 +0000)]
Revert "[SimplifyCFG] Rewrite SinkThenElseCodeToEnd"

This reverts commit r279443. It caused buildbot failures.

llvm-svn: 279447

7 years agoAdd a non-code-running data formatter for __NSCFBoolean
Enrico Granata [Mon, 22 Aug 2016 18:07:52 +0000 (18:07 +0000)]
Add a non-code-running data formatter for __NSCFBoolean

llvm-svn: 279446

7 years agoPR29086: DebugInfo: Improve support for fixed array dimensions in variable length...
David Blaikie [Mon, 22 Aug 2016 17:49:56 +0000 (17:49 +0000)]
PR29086: DebugInfo: Improve support for fixed array dimensions in variable length arrays

llvm-svn: 279445

7 years agoRemove redundant test
David Blaikie [Mon, 22 Aug 2016 17:49:50 +0000 (17:49 +0000)]
Remove redundant test

test/CodeGenCXX/debug-info-zero-length-arrays.cpp tests this
functionality more comprehensively

llvm-svn: 279444

7 years ago[SimplifyCFG] Rewrite SinkThenElseCodeToEnd
James Molloy [Mon, 22 Aug 2016 17:40:23 +0000 (17:40 +0000)]
[SimplifyCFG] Rewrite SinkThenElseCodeToEnd

The new version has several advantages:
  1) IMSHO it's more readable and neater
  2) It handles loads and stores properly
  3) It can handle any number of incoming blocks rather than just two. I'll be taking advantage of this in a followup patch.

With this change we can now finally sink load-modify-store idioms such as:

    if (a)
      return *b += 3;
    else
      return *b += 4;

    =>

    %z = load i32, i32* %y
    %.sink = select i1 %a, i32 5, i32 7
    %b = add i32 %z, %.sink
    store i32 %b, i32* %y
    ret i32 %b

When this works for switches it'll be even more powerful.

Round 4. This time we should handle all instructions correctly, and not replace any operands that need to be constant with variables.

This was really hard to determine safely, so the helper function should be put into the Instruction API. I'll do that as a followup.

llvm-svn: 279443

7 years ago[clang-tidy docs] Further cleanup of options
Alexander Kornienko [Mon, 22 Aug 2016 17:19:23 +0000 (17:19 +0000)]
[clang-tidy docs] Further cleanup of options

llvm-svn: 279442

7 years ago[X86][AVX] Don't use SubVectorBroadcast if there are additional users of the chain...
Simon Pilgrim [Mon, 22 Aug 2016 16:47:55 +0000 (16:47 +0000)]
[X86][AVX] Don't use SubVectorBroadcast if there are additional users of the chain (PR29088)

We could improve on this by making X86SubVBroadcast a full memory intrinsic similar to X86vzload

llvm-svn: 279441

7 years agoFix Gold Plugin after API change in the LTO API (constify callback type)
Mehdi Amini [Mon, 22 Aug 2016 16:41:58 +0000 (16:41 +0000)]
Fix Gold Plugin after API change in the LTO API (constify callback type)

llvm-svn: 279440

7 years ago[mips][ias] Support .dtprel[d]word and .tprel[d]word directives
Simon Atanasyan [Mon, 22 Aug 2016 16:18:42 +0000 (16:18 +0000)]
[mips][ias] Support .dtprel[d]word and .tprel[d]word directives

Assembler directives .dtprelword, .dtpreldword, .tprelword, and
.tpreldword generates relocations R_MIPS_TLS_DTPREL32, R_MIPS_TLS_DTPREL64,
R_MIPS_TLS_TPREL32, and R_MIPS_TLS_TPREL64 respectively.

The main motivation for this patch is to be able to write test cases
for checking correctness of the LLD linker's behaviour.

Differential Revision: https://reviews.llvm.org/D23669

llvm-svn: 279439

7 years ago[LTO] Constify the Module Hook function (NFC)
Mehdi Amini [Mon, 22 Aug 2016 16:17:40 +0000 (16:17 +0000)]
[LTO] Constify the Module Hook function (NFC)

It use to be non-const for the sole purpose of custom handling of
commons symbol. This is moved now in the regular LTO handling now
and such we can constify the callback.

llvm-svn: 279438

7 years agoReset isUndef when removing subreg from a def operand
Krzysztof Parzyszek [Mon, 22 Aug 2016 14:50:12 +0000 (14:50 +0000)]
Reset isUndef when removing subreg from a def operand

llvm-svn: 279437

7 years agoclang-format: [JS] supports casts to types starting with punctuation ("{[(").
Martin Probst [Mon, 22 Aug 2016 14:23:30 +0000 (14:23 +0000)]
clang-format: [JS] supports casts to types starting with punctuation ("{[(").

Before:

    x as{x: number}

After:

    x as {x: number}

Reviewers: djasper

Subscribers: klimek, cfe-commits

Differential Revision: https://reviews.llvm.org/D23761

llvm-svn: 279436

7 years ago[X86] Only accept SM_SentinelUndef (-1) as an undefined shuffle mask in range
Simon Pilgrim [Mon, 22 Aug 2016 13:18:56 +0000 (13:18 +0000)]
[X86] Only accept SM_SentinelUndef (-1) as an undefined shuffle mask in range

As discussed on D23027 we should be trying to be more strict on what is an undefined mask value.

llvm-svn: 279435

7 years agoRemove missing file from r279433 reversal
Artur Pilipenko [Mon, 22 Aug 2016 13:18:19 +0000 (13:18 +0000)]
Remove missing file from r279433 reversal

llvm-svn: 279434

7 years agoRevert -r278267 [ValueTracking] An improvement to IR ValueTracking on Non-negative...
Artur Pilipenko [Mon, 22 Aug 2016 13:14:07 +0000 (13:14 +0000)]
Revert -r278267 [ValueTracking] An improvement to IR ValueTracking on Non-negative Integers

This change cause performance regression on MultiSource/Benchmarks/TSVC/Symbolics-flt/Symbolics-flt from LNT and some other bechmarks.

See https://reviews.llvm.org/D18777 for details.

llvm-svn: 279433

7 years agoRevert -r278269 [IndVarSimplify] Eliminate zext of a signed IV when the IV is known...
Artur Pilipenko [Mon, 22 Aug 2016 13:12:07 +0000 (13:12 +0000)]
Revert -r278269 [IndVarSimplify] Eliminate zext of a signed IV when the IV is known to be non-negative

This change needs to be reverted in order to revert -r278267 which cause performance regression on MultiSource/Benchmarks/TSVC/Symbolics-flt/Symbolics-flt from LNT and some other bechmarks.

See comments on https://reviews.llvm.org/D18777 for details.

llvm-svn: 279432

7 years ago[PM] Port LoopDataPrefetch AArch64 tests to new pass manager
Balaram Makam [Mon, 22 Aug 2016 12:59:58 +0000 (12:59 +0000)]
[PM] Port LoopDataPrefetch AArch64 tests to new pass manager

Reviewers: mcrosier, tejohnson

Subscribers: aemerson, rengolin, mcrosier, llvm-commits

Differential Revision: https://reviews.llvm.org/D23724

llvm-svn: 279431

7 years ago[X86][SSE] Avoid specifying unused arguments in SHUFPD lowering
Simon Pilgrim [Mon, 22 Aug 2016 12:56:54 +0000 (12:56 +0000)]
[X86][SSE] Avoid specifying unused arguments in SHUFPD lowering

As discussed on PR26491, we are missing the opportunity to make use of the smaller MOVHLPS instruction because we set both arguments of a SHUFPD when using it to lower a single input shuffle.

This patch sets the lowered argument to UNDEF if that shuffle element is undefined. This in turn makes it easier for target shuffle combining to decode UNDEF shuffle elements, allowing combines to MOVHLPS to occur.

A fix to match against MOVHPD stores was necessary as well.

This builds on the improved MOVLHPS/MOVHLPS lowering and memory folding support added in D16956

Adding similar support for SHUFPS will have to wait until have better support for target combining of binary shuffles.

Differential Revision: https://reviews.llvm.org/D23027

llvm-svn: 279430

7 years ago[mips][microMIPS] Implement BLTZC, BLEZC, BGEZC and BGTZC instructions, fix disassemb...
Hrvoje Varga [Mon, 22 Aug 2016 12:17:59 +0000 (12:17 +0000)]
[mips][microMIPS] Implement BLTZC, BLEZC, BGEZC and BGTZC instructions, fix disassembly and add operand checking to existing B<cond>C implementations
Differential Revision: https://reviews.llvm.org/D22667

llvm-svn: 279429

7 years ago[MC] Remove guard(s). NFCI.
Davide Italiano [Mon, 22 Aug 2016 11:55:22 +0000 (11:55 +0000)]
[MC] Remove guard(s). NFCI.

All the methods are already marked with
LLVM_DUMP_METHOD.

llvm-svn: 279428

7 years agoReapply "[analyzer] Added valist related checkers."
Gabor Horvath [Mon, 22 Aug 2016 11:21:30 +0000 (11:21 +0000)]
Reapply "[analyzer] Added valist related checkers."

Differential Revision: https://reviews.llvm.org/D15227

llvm-svn: 279427

7 years ago[ThinLTO][X86] Fix windows build
Simon Pilgrim [Mon, 22 Aug 2016 10:49:37 +0000 (10:49 +0000)]
[ThinLTO][X86] Fix windows build

Windows 'rm' complains about non-existent files if a wildcard is used. Be more explicit about the files deleted to avoid this.

llvm-svn: 279426

7 years ago[analyzer] Correctly add assumptions based on array bounds.
Gabor Horvath [Mon, 22 Aug 2016 10:07:32 +0000 (10:07 +0000)]
[analyzer] Correctly add assumptions based on array bounds.

Also simplify the constraints generated by the checker.

Differential Revision: https://reviews.llvm.org/D23112

llvm-svn: 279425

7 years ago[X86] Create a new instruction format to handle 4VOp3 encoding. This saves one bit...
Craig Topper [Mon, 22 Aug 2016 07:38:50 +0000 (07:38 +0000)]
[X86] Create a new instruction format to handle 4VOp3 encoding. This saves one bit in TSFlags and simplifies MRMSrcMem/MRMSrcReg format handling.

llvm-svn: 279424

7 years ago[X86] Create a new instruction format to handle MemOp4 encoding. This saves one bit...
Craig Topper [Mon, 22 Aug 2016 07:38:45 +0000 (07:38 +0000)]
[X86] Create a new instruction format to handle MemOp4 encoding. This saves one bit in TSFlags and simplifies MRMSrcMem/MRMSrcReg format handling.

llvm-svn: 279423

7 years ago[X86] Space out the encodings of X86 instruction formats. I plan to add some new...
Craig Topper [Mon, 22 Aug 2016 07:38:41 +0000 (07:38 +0000)]
[X86] Space out the encodings of X86 instruction formats. I plan to add some new encodings in future commits and this will reduce the size of those commits. NFC

This tries to keep all the ModRM memory and register forms in their own regions of the encodings. Hoping to make it simple on some of the switch statements that operate on these encodings.

llvm-svn: 279422

7 years ago[X86] Merge small helper function into the switch that calls it since they both opera...
Craig Topper [Mon, 22 Aug 2016 07:38:36 +0000 (07:38 +0000)]
[X86] Merge small helper function into the switch that calls it since they both operate on the same variable. NFC

llvm-svn: 279421

7 years ago[X86] Explicitly list all X86 instruction forms in switch statement so its easier...
Craig Topper [Mon, 22 Aug 2016 07:38:30 +0000 (07:38 +0000)]
[X86] Explicitly list all X86 instruction forms in switch statement so its easier to detect when one is missing. NFC

llvm-svn: 279420

7 years agoTest commit (Removing trailing whitespace).
Elad Cohen [Mon, 22 Aug 2016 07:34:21 +0000 (07:34 +0000)]
Test commit (Removing trailing whitespace).

llvm-svn: 279419

7 years agoAdd REQUIRES:X86 to test/tools/llvm-lto2/common.ll
Mehdi Amini [Mon, 22 Aug 2016 06:37:41 +0000 (06:37 +0000)]
Add REQUIRES:X86 to test/tools/llvm-lto2/common.ll

llvm-svn: 279418

7 years ago[LTO] Handles commons in monolithic LTO
Mehdi Amini [Mon, 22 Aug 2016 06:25:46 +0000 (06:25 +0000)]
[LTO] Handles commons in monolithic LTO

The gold-plugin was doing this internally, now the API is handling
commons correctly based on the given resolution.

Differential Revision: https://reviews.llvm.org/D23739

llvm-svn: 279417

7 years ago[LTO] Add a "CodeGenOnly" option. Allows the client to skip the optimizer.
Mehdi Amini [Mon, 22 Aug 2016 06:25:41 +0000 (06:25 +0000)]
[LTO] Add a "CodeGenOnly" option. Allows the client to skip the optimizer.

Summary: Slowly getting on par with libLTO

Reviewers: tejohnson

Subscribers: llvm-commits, mehdi_amini

Differential Revision: https://reviews.llvm.org/D23615

llvm-svn: 279416

7 years agoFormat.
Rui Ueyama [Mon, 22 Aug 2016 04:55:41 +0000 (04:55 +0000)]
Format.

llvm-svn: 279415

7 years agoDo not add .interp, .dynamic nor .eh_frame_hdr to segments just by type.
Rui Ueyama [Mon, 22 Aug 2016 04:55:20 +0000 (04:55 +0000)]
Do not add .interp, .dynamic nor .eh_frame_hdr to segments just by type.

Summary:
We previously added these output sections to segments just by type.
Therefore, if there's a PHDRS command like this

  PHDRS {
    headers PT_PHDR PHDRS;
    interp PT_INTERP;
  }

  SECTIONS {
    . = SIZEOF_HEADERS;
    .interp : { *(.interp) } :text
  }

then .interp was added to "interp" segment even though the linker
is not instructed to do so by SECTIONS command. This patch removes
the default behavior to simplify.

Differential Revision: https://reviews.llvm.org/D23702

llvm-svn: 279414

7 years ago[asan] Use 1 byte aligned stores to poison shadow memory
Vitaly Buka [Mon, 22 Aug 2016 04:16:14 +0000 (04:16 +0000)]
[asan] Use 1 byte aligned stores to poison shadow memory

Summary: r279379 introduced crash on arm 32bit bot. I suspect this is alignment issue.

Reviewers: eugenis

Subscribers: llvm-commits, aemerson

Differential Revision: https://reviews.llvm.org/D23762

llvm-svn: 279413

7 years ago[X86] Merge hasVEX_i8ImmReg into the ImmFormat type which had extra unused encodings...
Craig Topper [Mon, 22 Aug 2016 01:37:19 +0000 (01:37 +0000)]
[X86] Merge hasVEX_i8ImmReg into the ImmFormat type which had extra unused encodings. This saves one bit in TSFlags. NFC

llvm-svn: 279412

7 years ago[X86] Remove ignoreVEX_L from TSFlags. Only the disassembler needs it and the disasse...
Craig Topper [Mon, 22 Aug 2016 01:37:16 +0000 (01:37 +0000)]
[X86] Remove ignoreVEX_L from TSFlags. Only the disassembler needs it and the disassembler doesn't use TSFlags. NFC

llvm-svn: 279411

7 years agoReformat.
NAKAMURA Takumi [Mon, 22 Aug 2016 00:58:47 +0000 (00:58 +0000)]
Reformat.

llvm-svn: 279409

7 years agoUntabify.
NAKAMURA Takumi [Mon, 22 Aug 2016 00:58:04 +0000 (00:58 +0000)]
Untabify.

llvm-svn: 279408

7 years agoCOFF: hoist a local variable
Saleem Abdulrasool [Sun, 21 Aug 2016 23:05:43 +0000 (23:05 +0000)]
COFF: hoist a local variable

Create a local variable for the rdata section.  NFC.

llvm-svn: 279407

7 years ago[CostModel][X86] Removed shift tests
Simon Pilgrim [Sun, 21 Aug 2016 19:56:02 +0000 (19:56 +0000)]
[CostModel][X86] Removed shift tests

There are more thorough tests found in vshift-*-cost.ll

llvm-svn: 279406

7 years ago[CostModel][X86] Added costs for vXi16 and vXi8 vectors for add/sub/mul/and/or/xor...
Simon Pilgrim [Sun, 21 Aug 2016 19:44:44 +0000 (19:44 +0000)]
[CostModel][X86] Added costs for vXi16 and vXi8 vectors for add/sub/mul/and/or/xor tests

llvm-svn: 279405

7 years ago[CostModel][X86] Replaced SSSE3 with SSE2 costs to create a better baseline
Simon Pilgrim [Sun, 21 Aug 2016 19:14:48 +0000 (19:14 +0000)]
[CostModel][X86] Replaced SSSE3 with SSE2 costs to create a better baseline

llvm-svn: 279404

7 years ago[CostModel][X86] Added fsqrt and fma costs
Simon Pilgrim [Sun, 21 Aug 2016 19:06:25 +0000 (19:06 +0000)]
[CostModel][X86] Added fsqrt and fma costs

llvm-svn: 279403

7 years ago[CostModel][X86] Split off float arithmetic cost tests
Simon Pilgrim [Sun, 21 Aug 2016 18:34:47 +0000 (18:34 +0000)]
[CostModel][X86] Split off float arithmetic cost tests

llvm-svn: 279402

7 years ago[InstCombine] use m_APInt to allow icmp (shl X, Y), C folds for splat constant vector...
Sanjay Patel [Sun, 21 Aug 2016 17:10:07 +0000 (17:10 +0000)]
[InstCombine] use m_APInt to allow icmp (shl X, Y), C folds for splat constant vectors, part 4

This concludes the fixes for icmp+shl in this series:
https://reviews.llvm.org/rL279339
https://reviews.llvm.org/rL279398
https://reviews.llvm.org/rL279399

llvm-svn: 279401

7 years agoremove FIXME comment; fixed by previous commit
Sanjay Patel [Sun, 21 Aug 2016 16:40:42 +0000 (16:40 +0000)]
remove FIXME comment; fixed by previous commit

llvm-svn: 279400

7 years ago[InstCombine] use m_APInt to allow icmp (shl X, Y), C folds for splat constant vector...
Sanjay Patel [Sun, 21 Aug 2016 16:35:34 +0000 (16:35 +0000)]
[InstCombine] use m_APInt to allow icmp (shl X, Y), C folds for splat constant vectors, part 3

This is a partial enablement (move the ConstantInt guard down).

llvm-svn: 279399

7 years ago[InstCombine] use m_APInt to allow icmp (shl X, Y), C folds for splat constant vector...
Sanjay Patel [Sun, 21 Aug 2016 16:28:22 +0000 (16:28 +0000)]
[InstCombine] use m_APInt to allow icmp (shl X, Y), C folds for splat constant vectors, part 2

This is a partial enablement (move the ConstantInt guard down).

llvm-svn: 279398

7 years ago[X86][AVX] Dropped combineShuffle256 - this can now be performed by EltsFromConsecuti...
Simon Pilgrim [Sun, 21 Aug 2016 15:39:45 +0000 (15:39 +0000)]
[X86][AVX] Dropped combineShuffle256 - this can now be performed by EltsFromConsecutiveLoads

llvm-svn: 279397

7 years ago[InstCombine] use APInt instead of ConstantInt in isSignBitCheck(); NFCI
Sanjay Patel [Sun, 21 Aug 2016 15:07:45 +0000 (15:07 +0000)]
[InstCombine] use APInt instead of ConstantInt in isSignBitCheck(); NFCI

The callers still have ConstantInt guards, so there is no functional change
intended from this change. But relaxing the callers will allow more folds
for vector types.

llvm-svn: 279396

7 years agoAdd a flag to dump SCoP optimized with the IslScheduleOptimizer pass
Roman Gareev [Sun, 21 Aug 2016 11:20:39 +0000 (11:20 +0000)]
Add a flag to dump SCoP optimized with the IslScheduleOptimizer pass

Dump polyhedral descriptions of Scops optimized with the isl scheduling
optimizer and the set of post-scheduling transformations applied
on the schedule tree to be able to check the work of the IslScheduleOptimizer
pass at the polyhedral level.

Reviewed-by: Tobias Grosser <tobias@grosser.es>
Differential Revision: https://reviews.llvm.org/D23740

llvm-svn: 279395

7 years agoSimplify AccFuncMap to vector<> AccessFunctions
Roman Gareev [Sun, 21 Aug 2016 11:09:19 +0000 (11:09 +0000)]
Simplify AccFuncMap to vector<> AccessFunctions

getAccessFunctions() is dead code and the 'BB' argument
of getOrCreateAccessFunctions() is not used. This patch deletes
getAccessFunctions and transforms AccFuncMap into
a std::vector<std::unique_ptr<MemoryAccess>> AccessFunctions.

Reviewed-by: Tobias Grosser <tobias@grosser.es>
Differential Revision: https://reviews.llvm.org/D23759

llvm-svn: 279394

7 years ago[AVX512][FastISel] Do not use K registers in TEST instructions
Guy Blank [Sun, 21 Aug 2016 08:02:27 +0000 (08:02 +0000)]
[AVX512][FastISel] Do not use K registers in TEST instructions

In some cases, FastIsel was emitting TEST instruction with K reg input, which is illegal.
Changed to using KORTEST when dealing with K regs.

Differential Revision: https://reviews.llvm.org/D23163

llvm-svn: 279393

7 years ago[X86][AVX512F] minor fix of the parameter names
Asaf Badouh [Sun, 21 Aug 2016 07:56:47 +0000 (07:56 +0000)]
[X86][AVX512F] minor fix of the parameter names
add "__" prefix

Bug 28842 https://llvm.org/bugs/show_bug.cgi?id=29040

Differential Revision: https://reviews.llvm.org/D23753

llvm-svn: 279392

7 years agoARM: Avoid dereferencing end() in ARMFrameLowering::emitEpilogue
Duncan P. N. Exon Smith [Sun, 21 Aug 2016 00:08:10 +0000 (00:08 +0000)]
ARM: Avoid dereferencing end() in ARMFrameLowering::emitEpilogue

This fixes the crash from PR29072, where the MachineBasicBlock::iterator
wasn't being properly checked against MachineBasicBlock::end() before
iterating.  This was another bug exposed by the new
ilist::iterator::operator*() assertion from r279314.

This testcase is poor quality.  bugpoint couldn't reduce any further,
and I haven't had time to dig into what's going on so I can't invent a
better one.  I didn't even get good CHECK lines in: this is just a
crasher.

I'm committing anyway since this is a real crash with an obvious fix,
but I'll leave PR29072 open and ask an ARM maintainer to help improve
the testcase.

llvm-svn: 279391

7 years agoFix broken macOS LLDB Xcode build from r279314
Todd Fiala [Sat, 20 Aug 2016 23:24:02 +0000 (23:24 +0000)]
Fix broken macOS LLDB Xcode build from r279314

llvm-svn: 279390