platform/upstream/llvm.git
9 years agoAVX-512: Fixed encoding of VPBROADCASTM and added SKX forms of this instruction
Elena Demikhovsky [Sun, 26 Oct 2014 09:52:24 +0000 (09:52 +0000)]
AVX-512: Fixed encoding of VPBROADCASTM and added SKX forms of this instruction

llvm-svn: 220638

9 years agoChange StackDepot interface to use StackTrace more extensively
Alexey Samsonov [Sun, 26 Oct 2014 06:23:07 +0000 (06:23 +0000)]
Change StackDepot interface to use StackTrace more extensively

llvm-svn: 220637

9 years agoFixup deadlock sanitizer after r220635
Alexey Samsonov [Sun, 26 Oct 2014 05:43:17 +0000 (05:43 +0000)]
Fixup deadlock sanitizer after r220635

llvm-svn: 220636

9 years ago[Sanitizer] Make StackTrace a lightweight reference to array of PCs, and
Alexey Samsonov [Sun, 26 Oct 2014 03:35:14 +0000 (03:35 +0000)]
[Sanitizer] Make StackTrace a lightweight reference to array of PCs, and
introduce a BufferedStackTrace class, which owns this array.

Summary:
This change splits __sanitizer::StackTrace class into a lightweight
__sanitizer::StackTrace, which doesn't own array of PCs, and BufferedStackTrace,
which owns it. This would allow us to simplify the interface of StackDepot,
and eventually merge __sanitizer::StackTrace with __tsan::StackTrace.

Test Plan: regression test suite.

Reviewers: kcc, dvyukov

Reviewed By: dvyukov

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5985

llvm-svn: 220635

9 years ago[clang-tidy] check_clang_tidy_fix.sh -> check_clang_tidy.sh
Alexander Kornienko [Sun, 26 Oct 2014 02:58:07 +0000 (02:58 +0000)]
[clang-tidy] check_clang_tidy_fix.sh -> check_clang_tidy.sh

Summary: Make the script suitable for checking just messages. Move most of the tests to use it. Clean up the tests: shorten messages, insert line numbers, remove unnecessary RUN: lines, etc.

Reviewers: klimek

Reviewed By: klimek

Subscribers: curdeius, cfe-commits

Differential Revision: http://reviews.llvm.org/D5989

llvm-svn: 220634

9 years ago[clang-tidy] Clean up test.
Alexander Kornienko [Sun, 26 Oct 2014 02:21:32 +0000 (02:21 +0000)]
[clang-tidy] Clean up test.

Simplify RUN lines and make checks stricter, check messages instead of fixes.

llvm-svn: 220633

9 years ago[clang-tidy] Simplify RUN: lines in the test
Alexander Kornienko [Sun, 26 Oct 2014 02:18:06 +0000 (02:18 +0000)]
[clang-tidy] Simplify RUN: lines in the test

llvm-svn: 220632

9 years ago[clang-tidy] Bring order to check registration.
Alexander Kornienko [Sun, 26 Oct 2014 01:41:14 +0000 (01:41 +0000)]
[clang-tidy] Bring order to check registration.

Summary:
Register readability checks in a separate module. Renamed the checks
and test file names accordingly.

Reviewers: djasper, klimek

Reviewed By: klimek

Subscribers: curdeius, cfe-commits

Differential Revision: http://reviews.llvm.org/D5936

llvm-svn: 220631

9 years agoDriver: reduce search logic duplication
Saleem Abdulrasool [Sat, 25 Oct 2014 23:33:21 +0000 (23:33 +0000)]
Driver: reduce search logic duplication

Refactor the path search into a helper function to avoid duplicating the path
handling for the search.  NFC.

llvm-svn: 220628

9 years agobuiltins: avoid endian.h in favour of compiler builtins
Saleem Abdulrasool [Sat, 25 Oct 2014 20:54:51 +0000 (20:54 +0000)]
builtins: avoid endian.h in favour of compiler builtins

Use the __BYTE_ORDER__ macro which has been available in GCC since 4.6 and is
available in clang as well.

llvm-svn: 220626

9 years agotest: attempt to make test more hermetic
Saleem Abdulrasool [Sat, 25 Oct 2014 20:49:12 +0000 (20:49 +0000)]
test: attempt to make test more hermetic

Add a fake linker in to a sysroot to use for testing the driver's tool
invocation.  Should make the test behave similarly on all platforms.  Addresses
review comments from Reid Kleckner from SVN r220546.

llvm-svn: 220625

9 years agoDriver: assume that all architectures are supported for libstc++
Saleem Abdulrasool [Sat, 25 Oct 2014 20:48:35 +0000 (20:48 +0000)]
Driver: assume that all architectures are supported for libstc++

Rather than asserting that the target is unsupported, make a guess at what the
tree for a port would look like and use that for the search path.

Addresses review comments from Ried Kleckner for SVN r220547.

llvm-svn: 220624

9 years agoLSR: Minor cleanup after Daniel's patch.
Andrew Trick [Sat, 25 Oct 2014 19:59:30 +0000 (19:59 +0000)]
LSR: Minor cleanup after Daniel's patch.

Combine the Inserted an Done sets into a Visited set.

llvm-svn: 220623

9 years agoFix LSR compile time.
Andrew Trick [Sat, 25 Oct 2014 19:42:07 +0000 (19:42 +0000)]
Fix LSR compile time.

This is a simple fix that brings the compilation time from 5min to 5s
on a specific real-world example. It's a large chain of computation in
a crypto routine (always a problem for SCEV). A unit test is not
feasible and there would be no way to check it. The fix is just basic
good practice for dealing with SCEVs, there's no risk of regression.

Patch by Daniel Reynaud!

llvm-svn: 220622

9 years ago[OCaml] hexagon can't run MCJIT tests, XFAIL it.
Peter Zotov [Sat, 25 Oct 2014 19:01:14 +0000 (19:01 +0000)]
[OCaml] hexagon can't run MCJIT tests, XFAIL it.

llvm-svn: 220621

9 years ago[OCaml] Unbreak Llvm_executionengine.initialize_native_target.
Peter Zotov [Sat, 25 Oct 2014 18:50:02 +0000 (18:50 +0000)]
[OCaml] Unbreak Llvm_executionengine.initialize_native_target.

First, return true on success, as it is the OCaml convention.
Second, also initialize the native assembly printer, which is,
despite the name, required for MCJIT operation.

Since this function did not initialize the assembly printer earlier
and no function to initialize native assembly printer was available
elsewhere, it is safe to break its interface: it means that it
simply could not be used successfully before.

llvm-svn: 220620

9 years ago[OCaml] Expose Llvm_executionengine.ExecutionEngine.create_mcjit.
Peter Zotov [Sat, 25 Oct 2014 18:49:56 +0000 (18:49 +0000)]
[OCaml] Expose Llvm_executionengine.ExecutionEngine.create_mcjit.

llvm-svn: 220619

9 years ago[SeparateConstOffsetFromGEP] Fixed a bug related to unsigned modulo
Jingyue Wu [Sat, 25 Oct 2014 18:34:03 +0000 (18:34 +0000)]
[SeparateConstOffsetFromGEP] Fixed a bug related to unsigned modulo

The dividend in "signed % unsigned" is treated as unsigned instead of signed,
causing unexpected behavior such as -64 % (uint64_t)24 == 0.

Added a regression test in split-gep.ll

Patched by Hao Liu.

llvm-svn: 220618

9 years agoUnbreak the build.
Benjamin Kramer [Sat, 25 Oct 2014 18:20:17 +0000 (18:20 +0000)]
Unbreak the build.

llvm-svn: 220617

9 years agoClean up assume intrinsic pattern matching, no need to check that the argument is...
Benjamin Kramer [Sat, 25 Oct 2014 18:09:01 +0000 (18:09 +0000)]
Clean up assume intrinsic pattern matching, no need to check that the argument is a value.

Also make it const safe and remove superfluous casting. NFC.

llvm-svn: 220616

9 years ago[SeparateConstOffsetFromGEP] Fixed a bug in rebuilding OR expressions
Jingyue Wu [Sat, 25 Oct 2014 17:36:21 +0000 (17:36 +0000)]
[SeparateConstOffsetFromGEP] Fixed a bug in rebuilding OR expressions

The two operands of the new OR expression should be NextInChain and TheOther
instead of the two original operands.

Added a regression test in split-gep.ll.

Hao Liu reported this bug, and provded the test case and an initial patch.
Thanks!

llvm-svn: 220615

9 years agoLex: Fix an invalid access into a SmallString
David Majnemer [Sat, 25 Oct 2014 11:40:40 +0000 (11:40 +0000)]
Lex: Fix an invalid access into a SmallString

We would crash because we used operator[] to access past the end of a
SmallString.  This occured because our token had length zero.

Instead, form the pointer using .data() and arithmetic.  This is safe
because this forms a one-past-the-end pointer and it is only used to
compare with another one-past-the-end pointer.

This fixes PR21379.

llvm-svn: 220614

9 years ago[X86][SSE] Vector integer/float conversion memory folding
Simon Pilgrim [Sat, 25 Oct 2014 08:11:20 +0000 (08:11 +0000)]
[X86][SSE] Vector integer/float conversion memory folding

Tidied up some entries in the folding tables so that they are under the correct comment section (they were categorised as AVX2 instructions when they're AVX1).

Minor patch agreed with qcolombet.

llvm-svn: 220613

9 years agoInstCombine: Remove overzealous asserts
David Majnemer [Sat, 25 Oct 2014 07:13:13 +0000 (07:13 +0000)]
InstCombine: Remove overzealous asserts

These asserts can trigger if the worklist iteration order is
sufficiently unlucky.  Instead of adding special case logic to handle
these edge conditions, just bail out on trying to transform them:
InstSimplify will get them when it reaches them on the worklist.

This fixes PR21378.

N.B.  No test case is included because any test would rely on the
fragile worklist iteration order.

llvm-svn: 220612

9 years agoAllow the C API users to keep relying on the OutMessages parameter.
Rafael Espindola [Sat, 25 Oct 2014 04:31:08 +0000 (04:31 +0000)]
Allow the C API users to keep relying on the OutMessages parameter.

Should fix the Ocaml tests.

llvm-svn: 220611

9 years agoRemove unused variable.
Rafael Espindola [Sat, 25 Oct 2014 04:07:53 +0000 (04:07 +0000)]
Remove unused variable.

llvm-svn: 220610

9 years agoUpdate for LLVM api change.
Rafael Espindola [Sat, 25 Oct 2014 04:06:14 +0000 (04:06 +0000)]
Update for LLVM api change.

llvm-svn: 220609

9 years agoUpdate the error handling of lib/Linker.
Rafael Espindola [Sat, 25 Oct 2014 04:06:10 +0000 (04:06 +0000)]
Update the error handling of lib/Linker.

Instead of passing a std::string&, use the new diagnostic infrastructure.

llvm-svn: 220608

9 years ago[NVPTX] aligned byte-buffers for vector return types
Jingyue Wu [Sat, 25 Oct 2014 03:46:16 +0000 (03:46 +0000)]
[NVPTX] aligned byte-buffers for vector return types

Summary:
Fixes PR21100 which is caused by inconsistency between the declared return type
and the expected return type at the call site. The new behavior is consistent
with nvcc and the NVPTXTargetLowering::getPrototype function.

Test Plan: test/Codegen/NVPTX/vector-return.ll

Reviewers: jholewinski

Reviewed By: jholewinski

Subscribers: llvm-commits, meheff, eliben, jholewinski

Differential Revision: http://reviews.llvm.org/D5612

llvm-svn: 220607

9 years agoMake this test a bit stricter by checking clang's output too.
Rafael Espindola [Sat, 25 Oct 2014 01:51:19 +0000 (01:51 +0000)]
Make this test a bit stricter by checking clang's output too.

llvm-svn: 220604

9 years agoAdd a test for the -suppress-warnings option.
Rafael Espindola [Sat, 25 Oct 2014 01:14:15 +0000 (01:14 +0000)]
Add a test for the -suppress-warnings option.

llvm-svn: 220603

9 years agoSetting breakpoints with name mask eFunctionNameTypeBase was broken for straight...
Jim Ingham [Sat, 25 Oct 2014 00:33:55 +0000 (00:33 +0000)]
Setting breakpoints with name mask eFunctionNameTypeBase was broken for straight C names by 220432.  Get
that working again.

llvm-svn: 220602

9 years ago[msan] Make -msan-check-constant-shadow a bit stronger.
Evgeniy Stepanov [Fri, 24 Oct 2014 23:34:15 +0000 (23:34 +0000)]
[msan] Make -msan-check-constant-shadow a bit stronger.

Allow (under the experimental flag) non-Instructions to participate in MSan checks.

llvm-svn: 220601

9 years agoModernize the error handling of the Materialize function.
Rafael Espindola [Fri, 24 Oct 2014 22:50:48 +0000 (22:50 +0000)]
Modernize the error handling of the Materialize function.

llvm-svn: 220600

9 years agoFix a Mach-O assembler segfault for a subtraction expression with an undefined symbol.
Kevin Enderby [Fri, 24 Oct 2014 22:39:40 +0000 (22:39 +0000)]
Fix a Mach-O assembler segfault for a subtraction expression with an undefined symbol.

In a Mach-O object file a relocatable expression of the form
SymbolA - SymbolB + constant is allowed when both symbols are
defined in a section.  But when either symbol is undefined it
is an error.

The code was crashing when it had an undefined symbol in this case.
And should have printed a error message using the location information
in the relocation entry.

rdar://18678402

llvm-svn: 220599

9 years ago[mach-o] Support old style objc class names in export lists
Nick Kledzik [Fri, 24 Oct 2014 22:28:54 +0000 (22:28 +0000)]
[mach-o] Support old style objc class names in export lists

Objective-C switched to a new ABI which uses a different mangling for class
names.  But to keep projects building that use export lists that use the old
class name mangling, the linker recognizes the old names and transforms them
to the new mangling.

llvm-svn: 220598

9 years ago[mach-o] Support data-only dylibs
Nick Kledzik [Fri, 24 Oct 2014 22:19:22 +0000 (22:19 +0000)]
[mach-o] Support data-only dylibs

In final linked shared images, the __TEXT segment contains both code and
the mach-o header/load-commands.  In the case of a data-only dylib, there is
no code, so we need to force the addition of the __TEXT segment.

llvm-svn: 220597

9 years agoImplement explicit thread stack size specification on Windows.
Zachary Turner [Fri, 24 Oct 2014 22:06:29 +0000 (22:06 +0000)]
Implement explicit thread stack size specification on Windows.

llvm-svn: 220596

9 years agoCodeGen: Don't form weak dllexport aliases
David Majnemer [Fri, 24 Oct 2014 22:05:27 +0000 (22:05 +0000)]
CodeGen: Don't form weak dllexport aliases

The MS linker cannot do anything interesting with these, it doesn't make
sense to emit them.

This fixes PR21373.

Differential Revision: http://reviews.llvm.org/D5986

llvm-svn: 220595

9 years agoSink DwarfUnit::constructImportedEntityDIE into DwarfCompileUnit.
Frederic Riss [Fri, 24 Oct 2014 21:31:09 +0000 (21:31 +0000)]
Sink DwarfUnit::constructImportedEntityDIE into DwarfCompileUnit.

So that it has access to getOrCreateGlobalVariableDIE. If we ever support
decsribing using directive in C++ classes (thus requiring support in type
units), it will certainly use another mechanism anyway.

Differential Revision: http://reviews.llvm.org/D5975

llvm-svn: 220594

9 years agotest: correct an overzealous search-and-replace
Saleem Abdulrasool [Fri, 24 Oct 2014 21:23:49 +0000 (21:23 +0000)]
test: correct an overzealous search-and-replace

The temporary initialized is referenced as %0, not as the auto-release pool.
Fixes R+A tests.

llvm-svn: 220593

9 years ago[X86][SSE] Bitcast assertion in XFormVExtractWithShuffleIntoLoad
Simon Pilgrim [Fri, 24 Oct 2014 21:04:41 +0000 (21:04 +0000)]
[X86][SSE] Bitcast assertion in XFormVExtractWithShuffleIntoLoad

Minor patch to fix an issue in XFormVExtractWithShuffleIntoLoad where a load is unary shuffled, then bitcast (to a type with the same number of elements) before extracting an element.

An undef was created for the second shuffle operand using the original (post-bitcasted) vector type instead of the pre-bitcasted type like the rest of the shuffle node - this was then causing an assertion on the different types later on inside SelectionDAG::getVectorShuffle.

Differential Revision: http://reviews.llvm.org/D5917

llvm-svn: 220592

9 years agoRemove duplicated new file content
Ed Maste [Fri, 24 Oct 2014 20:49:50 +0000 (20:49 +0000)]
Remove duplicated new file content

llvm-svn: 220591

9 years agoCodeGen: correct materialize temporary aggregates in ARC mode
Saleem Abdulrasool [Fri, 24 Oct 2014 20:23:43 +0000 (20:23 +0000)]
CodeGen: correct materialize temporary aggregates in ARC mode

Avoid an assertion when materializing a lifetime type aggregate temporary.  When
performing CodeGen for ObjC++, we could generate a lifetime-only aggregate
temporary by using an initializer list (which is effectively an array).  We
would reach through the temporary expression, fishing out the inner expression.
If this expression was a lifetime expression, we would attempt to emit this as a
scalar.  This would eventually result in an assertion as the emission would
eventually assert that the expression being emitted has a scalar evaluation
kind.

Add a case to handle the aggregate expressions.  Use the EmitAggExpr to emit the
aggregate expression rather than the EmitScalarInit.

Addresses PR21347.

llvm-svn: 220590

9 years ago[modules] Support combining 'textual' with 'private'.
Richard Smith [Fri, 24 Oct 2014 20:23:01 +0000 (20:23 +0000)]
[modules] Support combining 'textual' with 'private'.

llvm-svn: 220589

9 years agoItanium ABI: Template template parameters are usable as substitutions
David Majnemer [Fri, 24 Oct 2014 20:22:57 +0000 (20:22 +0000)]
Itanium ABI: Template template parameters are usable as substitutions

Template template parameters weren't added to the list of substitutions.
This would make the substitution map contain inaccurate mappings,
leading to Clang violating the Itanium ABI and breaking compatibility
with GCC.

This fixes PR21351.

Differential Revision: http://reviews.llvm.org/D5959

llvm-svn: 220588

9 years agoCodeGen: trivial conversion to range based loop
Saleem Abdulrasool [Fri, 24 Oct 2014 19:54:32 +0000 (19:54 +0000)]
CodeGen: trivial conversion to range based loop

Switch to a range-based for loop.  NFC.

llvm-svn: 220587

9 years agotest: add -std=c++11
Saleem Abdulrasool [Fri, 24 Oct 2014 19:54:29 +0000 (19:54 +0000)]
test: add -std=c++11

This test uses C++11 features (r-value references) in ObjC++.  Add a -std=c++11
to silence a warning.  NFC.

llvm-svn: 220586

9 years agoCodeGen: GLValue exprs in template parameters should have reference type
David Majnemer [Fri, 24 Oct 2014 19:49:04 +0000 (19:49 +0000)]
CodeGen: GLValue exprs in template parameters should have reference type

This fixes a corner-case where __uuidof as a template argument would
result in us trying to emit a GLValue as an RValue.  This would lead to
a crash down the road.

llvm-svn: 220585

9 years ago[Hexagon] Resubmission of 220427
Colin LeMahieu [Fri, 24 Oct 2014 19:00:32 +0000 (19:00 +0000)]
[Hexagon] Resubmission of 220427
Modified library structure to deal with circular dependency between HexagonInstPrinter and HexagonMCInst.
Adding encoding bits for add opcode.
Adding llvm-mc tests.
Removing unit tests.

http://reviews.llvm.org/D5624

llvm-svn: 220584

9 years agoPatch from ovyalov@google.com:
Jim Ingham [Fri, 24 Oct 2014 18:51:57 +0000 (18:51 +0000)]
Patch from ovyalov@google.com:

Handle pexpect exceptions correctly so that processes spawned with
pexpect are always reaped.

llvm-svn: 220583

9 years agoAdd runtime flag 'symbolize_inline_frames' to disable symbolization of inlined frames...
Alexey Samsonov [Fri, 24 Oct 2014 18:34:43 +0000 (18:34 +0000)]
Add runtime flag 'symbolize_inline_frames' to disable symbolization of inlined frames done in llvm-symbolizer

llvm-svn: 220582

9 years agoFix copy paste comment
Matt Arsenault [Fri, 24 Oct 2014 18:13:10 +0000 (18:13 +0000)]
Fix copy paste comment

llvm-svn: 220581

9 years agoDon't ever call materializeAllPermanently during LTO.
Rafael Espindola [Fri, 24 Oct 2014 18:13:04 +0000 (18:13 +0000)]
Don't ever call materializeAllPermanently during LTO.

To do this, change the representation of lazy loaded functions.

The previous representation cannot differentiate between a function whose body
has been removed and one whose body hasn't been read from the .bc file. That
means that in order to drop a function, the entire body had to be read.

llvm-svn: 220580

9 years agoAllow AVX vrsqrtps generation.
Sanjay Patel [Fri, 24 Oct 2014 17:59:18 +0000 (17:59 +0000)]
Allow AVX vrsqrtps generation.

This is a follow-on to r220570 that allows a 256-bit (v8f32)
version of vrsqrtps to be generated.

llvm-svn: 220579

9 years agoDebugInfo: Sink DwarfDebug::ScopeVariables down into DwarfFile
David Blaikie [Fri, 24 Oct 2014 17:57:34 +0000 (17:57 +0000)]
DebugInfo: Sink DwarfDebug::ScopeVariables down into DwarfFile

(part of refactoring to allow subprogram emission in both the skeleton
and main units to enable -gmlt-like data to be included in the skeleton
for live inlined backtracing purposes)

llvm-svn: 220578

9 years agoremove class/function/variable names from comments; NFC
Sanjay Patel [Fri, 24 Oct 2014 17:55:59 +0000 (17:55 +0000)]
remove class/function/variable names from comments; NFC

llvm-svn: 220577

9 years agoFix windows-cross.c test on my machine
Reid Kleckner [Fri, 24 Oct 2014 17:55:29 +0000 (17:55 +0000)]
Fix windows-cross.c test on my machine

I suspect it will need a custom sysroot to pass reliably elsewhere.

llvm-svn: 220576

9 years agoRemove DwarfDebug::FirstCU as it has no use
David Blaikie [Fri, 24 Oct 2014 17:53:38 +0000 (17:53 +0000)]
Remove DwarfDebug::FirstCU as it has no use

It was only being used as a flag to identify the lack of debug info from
within endModule - use the section labels for that instead.

llvm-svn: 220575

9 years agoMake ProcessWindows just use Host::LaunchProcess.
Zachary Turner [Fri, 24 Oct 2014 17:51:56 +0000 (17:51 +0000)]
Make ProcessWindows just use Host::LaunchProcess.

llvm-svn: 220574

9 years agoAdd frontend support for __vectorcall
Reid Kleckner [Fri, 24 Oct 2014 17:42:17 +0000 (17:42 +0000)]
Add frontend support for __vectorcall

Wire it through everywhere we have support for fastcall, essentially.

This allows us to parse the MSVC "14" CTP headers, but we will
miscompile them because LLVM doesn't support __vectorcall yet.

Reviewed By: Aaron Ballman

Differential Revision: http://reviews.llvm.org/D5808

llvm-svn: 220573

9 years agoUse enumerators instead of hardcoded integers when processing macro names.
Serge Pavlov [Fri, 24 Oct 2014 17:31:32 +0000 (17:31 +0000)]
Use enumerators instead of hardcoded integers when processing macro names.

llvm-svn: 220572

9 years agotsan: support mmap(MAP_32BIT)
Dmitry Vyukov [Fri, 24 Oct 2014 17:07:29 +0000 (17:07 +0000)]
tsan: support mmap(MAP_32BIT)
Allow user memory in the first TB of address space.
This also enabled non-pie binaries and freebsd.
Fixes issue:
https://code.google.com/p/thread-sanitizer/issues/detail?id=5

llvm-svn: 220571

9 years agoUse rsqrt (X86) to speed up reciprocal square root calcs
Sanjay Patel [Fri, 24 Oct 2014 17:02:16 +0000 (17:02 +0000)]
Use rsqrt (X86) to speed up reciprocal square root calcs

This is a first step for generating SSE rsqrt instructions for
reciprocal square root calcs when fast-math is allowed.

For now, be conservative and only enable this for AMD btver2
where performance improves significantly - for example, 29%
on llvm/projects/test-suite/SingleSource/Benchmarks/BenchmarkGame/n-body.c
(if we convert the data type to single-precision float).

This patch adds a two constant version of the Newton-Raphson
refinement algorithm to DAGCombiner that can be selected by any target
via a parameter returned by getRsqrtEstimate()..

See PR20900 for more details:
http://llvm.org/bugs/show_bug.cgi?id=20900

Differential Revision: http://reviews.llvm.org/D5658

llvm-svn: 220570

9 years ago[Modules] Free modules that failed signature verification.
Benjamin Kramer [Fri, 24 Oct 2014 16:31:42 +0000 (16:31 +0000)]
[Modules] Free modules that failed signature verification.

The control flow and ownership is weird enough so unique_ptr doesn't help here :(

llvm-svn: 220569

9 years ago[mips] Replace MipsABIEnum with a MipsABIInfo class.
Daniel Sanders [Fri, 24 Oct 2014 16:15:27 +0000 (16:15 +0000)]
[mips] Replace MipsABIEnum with a MipsABIInfo class.

Summary:
No functional change yet, it's just an object replacement for an enum.
It will allow us to gather ABI information in a single place so that we can
start testing for properties of the ABI's instead of the ABI itself.

For example we will eventually be able to use:
  ABI.MinStackAlignmentInBytes()
instead of:
  (isABI_N32() || isABI_N64()) ? 16 : 8
which is clearer and more maintainable.

Reviewers: matheusalmeida

Reviewed By: matheusalmeida

Differential Revision: http://reviews.llvm.org/D3341

llvm-svn: 220568

9 years ago[Object] Fix MachO's getUuid to return a pointer into the object instead of a danglin...
Benjamin Kramer [Fri, 24 Oct 2014 15:52:05 +0000 (15:52 +0000)]
[Object] Fix MachO's getUuid to return a pointer into the object instead of a dangling ArrayRef.

This works because uuid's are always little endian so it's not swapped.
Fixes use-after-return reported by asan.

llvm-svn: 220567

9 years ago[mips] Mark aggregate arguments passed in registers with the inreg attribute
Daniel Sanders [Fri, 24 Oct 2014 15:30:16 +0000 (15:30 +0000)]
[mips] Mark aggregate arguments passed in registers with the inreg attribute

Summary:
This allows us to easily identify them in the backend which in turn allows us
to handle them correctly for big-endian targets (where they must be shifted
into the upper bits of the register).

Depends on D5961

Reviewers: atanasyan

Reviewed By: atanasyan

Subscribers: cfe-commits, theraven

Differential Revision: http://reviews.llvm.org/D5962

llvm-svn: 220566

9 years agoThese functions are not actually defined for NDEBUG or !LLVM_DUMP_ENABLED, so guardin...
Aaron Ballman [Fri, 24 Oct 2014 15:16:39 +0000 (15:16 +0000)]
These functions are not actually defined for NDEBUG or !LLVM_DUMP_ENABLED, so guarding the declarations as well. NFC, silences MSVC warnings in release builds.

llvm-svn: 220565

9 years ago[mips] Fix >80-column line
Daniel Sanders [Fri, 24 Oct 2014 14:46:00 +0000 (14:46 +0000)]
[mips] Fix >80-column line

llvm-svn: 220564

9 years ago[mips] Promote all integral/enumeration types to the GPR width
Daniel Sanders [Fri, 24 Oct 2014 14:42:42 +0000 (14:42 +0000)]
[mips] Promote all integral/enumeration types to the GPR width

Summary:
Ensure all integral/enumeration types are appropriately annotated with
signext/zeroext. In particular, i32 now has these attributes when using the
N32/N64 ABI. This paves the way for accurately representing the way the
N32/N64 ABI's promotes integer arguments to i64.

Reviewers: atanasyan

Reviewed By: atanasyan

Subscribers: cfe-commits, theraven

Differential Revision: http://reviews.llvm.org/D5961

llvm-svn: 220563

9 years agoFix initializing TypeOfTypeLoc
Olivier Goffart [Fri, 24 Oct 2014 13:52:55 +0000 (13:52 +0000)]
Fix initializing TypeOfTypeLoc

This fixes a crash in the RecursiveASTVisitor on such code
 __typeof__(struct F*) var[invalid];

The UnderlyingTInfo of a TypeOfTypeLoc was left uninitialized when
created from ASTContext::getTrivialTypeSourceInfo
This lead to a crash in RecursiveASTVisitor when trying to access it.

llvm-svn: 220562

9 years ago[mips] Remove redundant code in RetCC_MipsN. NFC.
Daniel Sanders [Fri, 24 Oct 2014 13:49:54 +0000 (13:49 +0000)]
[mips] Remove redundant code in RetCC_MipsN. NFC.

Summary:
i32 is always promoted to i64 so it no longer makes sense to assign i32 to
registers.

Reviewers: vmedic

Reviewed By: vmedic

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5964

llvm-svn: 220561

9 years agoASTMatchers: Peel off a layer of indirection from true matcher. NFC.
Benjamin Kramer [Fri, 24 Oct 2014 13:29:21 +0000 (13:29 +0000)]
ASTMatchers: Peel off a layer of indirection from true matcher. NFC.

llvm-svn: 220560

9 years agoASTMatchers: for-rangify loops. No functionality change.
Benjamin Kramer [Fri, 24 Oct 2014 13:29:15 +0000 (13:29 +0000)]
ASTMatchers: for-rangify loops. No functionality change.

llvm-svn: 220559

9 years agoAdd arguments() to CXXConstructExpr so the same interface as with CallExpr can be...
Benjamin Kramer [Fri, 24 Oct 2014 13:29:07 +0000 (13:29 +0000)]
Add arguments() to CXXConstructExpr so the same interface as with CallExpr can be used.

llvm-svn: 220558

9 years agoReport when a function-try-block does not return a value on all control paths. Fixed...
Aaron Ballman [Fri, 24 Oct 2014 13:19:19 +0000 (13:19 +0000)]
Report when a function-try-block does not return a value on all control paths. Fixed PR14620.

llvm-svn: 220557

9 years ago[mips] For N32/N64, structs must be passed in the upper bits of a register.
Daniel Sanders [Fri, 24 Oct 2014 13:09:19 +0000 (13:09 +0000)]
[mips] For N32/N64, structs must be passed in the upper bits of a register.

Summary:
Most structs were fixed by r218451 but those of between >32-bits and
<64-bits remained broken since they were not marked with [ASZ]ExtUpper.
This patch fixes the remaining cases by using
CCPromoteToUpperBitsInType<i64> on i64's in addition to i32 and smaller.

Reviewers: vmedic

Reviewed By: vmedic

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5963

llvm-svn: 220556

9 years ago[Thumb] Clang thinks "char" is signed when using a thumb triple
Oliver Stannard [Fri, 24 Oct 2014 11:28:47 +0000 (11:28 +0000)]
[Thumb] Clang thinks "char" is signed when using a thumb triple

'char' is unsigned on all ARM and Thumb architectures. Clang gets this
right for ARM, and for thumb when using and arm triple and the -mthumb
option, but gets it wrong for thumb triples. This fixes that.

llvm-svn: 220555

9 years ago[Tsan] Do not intercept non-FreeBSD functions on FreeBSD
Viktor Kutuzov [Fri, 24 Oct 2014 10:55:03 +0000 (10:55 +0000)]
[Tsan] Do not intercept non-FreeBSD functions on FreeBSD
Differential Revision: http://reviews.llvm.org/D5858

llvm-svn: 220554

9 years ago[AArch64] Fix fast-isel of cbz of i1, i8, i16
Oliver Stannard [Fri, 24 Oct 2014 09:54:41 +0000 (09:54 +0000)]
[AArch64] Fix fast-isel of cbz of i1, i8, i16

This fixes a miscompilation in the AArch64 fast-isel which was
triggered when a branch is based on an icmp with condition eq or ne,
and type i1, i8 or i16. The cbz instruction compares the whole 32-bit
register, so values with the bottom 1, 8 or 16 bits clear would cause
the wrong branch to be taken.

llvm-svn: 220553

9 years ago[Tsan] Fix the thread_name tests to build on FreeBSD
Viktor Kutuzov [Fri, 24 Oct 2014 09:27:34 +0000 (09:27 +0000)]
[Tsan] Fix the thread_name tests to build on FreeBSD
Differential Revision: http://reviews.llvm.org/D5855

llvm-svn: 220552

9 years ago[Tsan] Intercept pthread_set_name_np on FreeBSD
Viktor Kutuzov [Fri, 24 Oct 2014 09:23:06 +0000 (09:23 +0000)]
[Tsan] Intercept pthread_set_name_np on FreeBSD
Differential Revision: http://reviews.llvm.org/D5932

llvm-svn: 220551

9 years ago[Tsan] Fix the signal_recursive.cc test to build on FreeBSD
Viktor Kutuzov [Fri, 24 Oct 2014 09:20:20 +0000 (09:20 +0000)]
[Tsan] Fix the signal_recursive.cc test to build on FreeBSD
Differential Revision: http://reviews.llvm.org/D5841

llvm-svn: 220550

9 years ago[CMake] clangRewrite: Roll back clangAST in libdeps, due to clang::Stmt::printPretty().
NAKAMURA Takumi [Fri, 24 Oct 2014 08:44:01 +0000 (08:44 +0000)]
[CMake] clangRewrite: Roll back clangAST in libdeps, due to clang::Stmt::printPretty().

llvm-svn: 220549

9 years agoUpdate test/MC/ARM/coff-debugging-secrel.ll expectations to fix breakage caused by...
Timur Iskhodzhanov [Fri, 24 Oct 2014 06:24:07 +0000 (06:24 +0000)]
Update test/MC/ARM/coff-debugging-secrel.ll expectations to fix breakage caused by r220544

llvm-svn: 220548

9 years agoDriver: add missed file from previous commit
Saleem Abdulrasool [Fri, 24 Oct 2014 03:24:33 +0000 (03:24 +0000)]
Driver: add missed file from previous commit

llvm-svn: 220547

9 years agoDriver: add CrossWindowsToolChain
Saleem Abdulrasool [Fri, 24 Oct 2014 03:13:37 +0000 (03:13 +0000)]
Driver: add CrossWindowsToolChain

This is a very basic toolchain.  It supports cross-compiling Windows (primarily
inspired by the WoA target).  It is meant to use clang with the LLVM IAS and a
binutils ld-compatible interface for the linker (eventually to be lld).  It does
not perform any "standard" GCC lookup, nor does it perform any special
adjustments given that it is expected to be used in an environment where the
user is using MSVCRT (and as such Visual Studio headers) and the Windows SDK.
The primary runtime library is expected to be compiler-rt and the C++
implementation to be libc++.

It also expects that a sysroot has been setup given the usual Unix semantics
(standard C headers in /usr/include, all the import libraries available in
/usr/lib).  It also expects that an entry point stub is present in /usr/lib
(crtbegin.obj for executables, crtbeginS.obj for shared libraries).

The entry point stub is responsible for running any GNU constructors.

llvm-svn: 220546

9 years agoAdded reset of LexicalScope in LiveDebugVariables reset function.
Marcello Maggioni [Fri, 24 Oct 2014 02:46:50 +0000 (02:46 +0000)]
Added reset of LexicalScope in LiveDebugVariables reset function.

llvm-svn: 220545

9 years agoFix PR21189 -- Emit symbol subsection required to debug LLVM-built binaries with...
Timur Iskhodzhanov [Fri, 24 Oct 2014 01:27:45 +0000 (01:27 +0000)]
Fix PR21189 -- Emit symbol subsection required to debug LLVM-built binaries with VS2012+

Reviewed at http://reviews.llvm.org/D5772

llvm-svn: 220544

9 years agoAdd a new -fmerge-functions -cc1 flag that enables function merging.
Nick Lewycky [Fri, 24 Oct 2014 00:49:29 +0000 (00:49 +0000)]
Add a new -fmerge-functions -cc1 flag that enables function merging.

llvm-svn: 220543

9 years agoDebugInfo: Remove DwarfDebug::addScopeVariable now that it's just a trivial wrapper
David Blaikie [Fri, 24 Oct 2014 00:43:47 +0000 (00:43 +0000)]
DebugInfo: Remove DwarfDebug::addScopeVariable now that it's just a trivial wrapper

llvm-svn: 220542

9 years agoMake test for r220533 more robust by using GPR pattern.
Ahmed Bougacha [Fri, 24 Oct 2014 00:03:46 +0000 (00:03 +0000)]
Make test for r220533 more robust by using GPR pattern.

llvm-svn: 220541

9 years ago[AVX512] FMA support for the 231 variants
Adam Nemet [Fri, 24 Oct 2014 00:03:00 +0000 (00:03 +0000)]
[AVX512] FMA support for the 231 variants

This is asm/diasm-only support, similar to AVX.

For ISeling the register variant, they are no different from 213 other than
whether the multiplication or the addition operand is destructed.

For ISeling the memory variant, i.e. to fold a load, they are no different
than the 132 variant.  The addition operand (op3) in both cases can come from
memory.  Again the ony difference is which operand is destructed.

There could be a post-RA pass that would convert a 213 or 132 into a 231.

Part of <rdar://problem/17082571>

llvm-svn: 220540

9 years ago[AVX512] Introduce fma3p_forms from AVX
Adam Nemet [Fri, 24 Oct 2014 00:02:55 +0000 (00:02 +0000)]
[AVX512] Introduce fma3p_forms from AVX

This multiclass generates the different forms: 213, 231, 132 in AVX.

132 in AVX512 is a separate class but I am planning to use this same
multiclass to generate 231 relying on the nice the null_frag trick from AVX to
disable codegen pattern for 231.

No functionality change, no change in X86.td.expanded except for the different
instruction definition names.

llvm-svn: 220539

9 years agoIf requested, apply function merging at -O0 too. It's useful there to reduce the...
Nick Lewycky [Thu, 23 Oct 2014 23:49:31 +0000 (23:49 +0000)]
If requested, apply function merging at -O0 too. It's useful there to reduce the time to compile.

llvm-svn: 220537

9 years agoMake getDISubprogram(const Function *F) available in LLVM
Timur Iskhodzhanov [Thu, 23 Oct 2014 23:46:28 +0000 (23:46 +0000)]
Make getDISubprogram(const Function *F) available in LLVM

Reviewed at http://reviews.llvm.org/D5950

llvm-svn: 220536

9 years agoChange uses of `sys.platform == 'linux2' to `sys.platform.startswith('linux')
Eric Fiselier [Thu, 23 Oct 2014 23:05:46 +0000 (23:05 +0000)]
Change uses of `sys.platform == 'linux2' to `sys.platform.startswith('linux')

Although the current method is valid up till python 3.3 (which is not supported)
this seems to be a clearer way of checking for linux and moves the tests towards
python 3 compatibility.

llvm-svn: 220535

9 years agoChange uses of `sys.platform == 'linux2' to `sys.platform.startswith('linux')
Eric Fiselier [Thu, 23 Oct 2014 22:57:56 +0000 (22:57 +0000)]
Change uses of `sys.platform == 'linux2' to `sys.platform.startswith('linux')

Although the current method is valid up till python 3.3 (which is not supported)
this seems to be a clearer way of checking for linux and moves the tests towards
python 3 compatibility.

llvm-svn: 220534

9 years ago[SelectionDAG] Teach the vector scalarizer about FP conversions.
Ahmed Bougacha [Thu, 23 Oct 2014 22:49:25 +0000 (22:49 +0000)]
[SelectionDAG] Teach the vector scalarizer about FP conversions.

This adds support for legalization of instructions of the form:

  [fp_conv] <1 x i1> %op to <1 x double>

where fp_conv is one of fpto[us]i, [us]itofp.  This used to assert
because they were simply missing from the vector operand scalarizer.

A similar problem arose in r190830, with trunc instead.

Fixes PR20778.

Differential Revision: http://reviews.llvm.org/D5810

llvm-svn: 220533