platform/upstream/llvm.git
10 years agoRemove "virtual" and add "override".
Rui Ueyama [Thu, 3 Apr 2014 18:25:36 +0000 (18:25 +0000)]
Remove "virtual" and add "override".

Seems clang-modernize couldn't add "override" to nested classes, so
doing it by hand. Also removed unused virtual member function that
is not overriding anything, that seems to have been added by mistake.

llvm-svn: 205552

10 years agoAttempt to XFAIL this on mingw and cygwin hosts. The line table on
Eric Christopher [Thu, 3 Apr 2014 18:23:52 +0000 (18:23 +0000)]
Attempt to XFAIL this on mingw and cygwin hosts. The line table on
these is very much off and is more than just the branch
from this bug incorrect:

Address            Line   Column File   ISA Discriminator Flags
------------------ ------ ------ ------ --- ------------- -------------
0x30830a0100000002      3      0      1   0             0  is_stmt
0x30830a0100000008      3      0      1   0             0  is_stmt end_sequence

llvm-svn: 205551

10 years agoRemove "this->".
Rui Ueyama [Thu, 3 Apr 2014 18:13:14 +0000 (18:13 +0000)]
Remove "this->".

For most member function calls we do not use "this->" in this file.
Remove the rest for consistency.

llvm-svn: 205550

10 years agoAdd a paragraph describing how to configure the python interpreter.
Tobias Grosser [Thu, 3 Apr 2014 18:12:13 +0000 (18:12 +0000)]
Add a paragraph describing how to configure the python interpreter.

Contributed-by: cib123@googlemail.com
llvm-svn: 205549

10 years agoRevert r205436:
Roman Divacky [Thu, 3 Apr 2014 18:04:52 +0000 (18:04 +0000)]
Revert r205436:

        Extend the SSE2 comment lexing to AVX2. Only 16byte align when not on AVX2.

        This provides some 3% speedup when preprocessing gcc.c as a single file.

The patch is wrong, it always uses SSE2, and when I fix that there's no speedup
at all. I am not sure where the 3% came from previously.

--Thi lie, and those below, will be ignored--

M    Lex/Lexer.cpp

llvm-svn: 205548

10 years agoFix PR19270 - type mismatch caused by invalid optimization.
Eli Bendersky [Thu, 3 Apr 2014 17:51:58 +0000 (17:51 +0000)]
Fix PR19270 - type mismatch caused by invalid optimization.

Patch by Jingyue Wu.

llvm-svn: 205547

10 years agoLoosen up check so that we can pass on platforms that generate
Eric Christopher [Thu, 3 Apr 2014 17:40:08 +0000 (17:40 +0000)]
Loosen up check so that we can pass on platforms that generate
slightly more verbose than needed line tables, e.g.:

Address            Line   Column File   ISA Discriminator Flags
------------------ ------ ------ ------ --- ------------- -------------
0x0000000000000000      1      0      1   0             0  is_stmt
0x0000000000000000      1      0      1   0             0  is_stmt prologue_end
0x0000000000000010      2      0      1   0             0  is_stmt
0x0000000000000018      4      0      1   0             0  is_stmt

these should probably be looked at, but it isn't affecting the correctness
of the testcase.

llvm-svn: 205546

10 years agoARM: update even more tests
Saleem Abdulrasool [Thu, 3 Apr 2014 17:35:22 +0000 (17:35 +0000)]
ARM: update even more tests

More updating of tests to be explicit about the target triple rather than
relying on the default target triple supporting ARM mode.

Indicate to lit that object emission is not yet available for Windows on ARM.

llvm-svn: 205545

10 years agoGet "dis -c -s" working again.
Jim Ingham [Thu, 3 Apr 2014 17:16:17 +0000 (17:16 +0000)]
Get "dis -c -s" working again.

pr19324

llvm-svn: 205544

10 years agoTeach getTemplateInstantiationPattern to deal with generic lambdas.
Faisal Vali [Thu, 3 Apr 2014 16:32:21 +0000 (16:32 +0000)]
Teach getTemplateInstantiationPattern to deal with generic lambdas.

No functionality change.

When determining the pattern for instantiating a generic lambda call operator specialization - we must not go drilling down for the 'prototype' (i.e. as written) pattern - rather we must use our partially transformed  pattern (whose DeclRefExprs are wired correctly to any enclosing lambda's decls that should be mapped correctly in a local instantiation scope) that is the templated pattern of the specialization's primary template (even though the primary template might be instantiated from a 'prototype' member-template).  Previously, the drilling down was haltted by marking the instantiated-from primary template as a member-specialization (incorrectly).

This prompted Richard to remark (http://llvm-reviews.chandlerc.com/D1784?id=4687#inline-10272)
"It's a bit nasty to (essentially) set this bit incorrectly. Can you put the check into getTemplateInstantiationPattern instead?"

In my reckless youth, I chose to ignore that comment.  With the passage of time, I have come to learn the value of bowing to the will of the angry Gods ;)

llvm-svn: 205543

10 years agoArrayRef: use std::vector::data() now that we are building in C++11 mode
Dmitri Gribenko [Thu, 3 Apr 2014 16:29:11 +0000 (16:29 +0000)]
ArrayRef: use std::vector::data() now that we are building in C++11 mode

llvm-svn: 205542

10 years agoARM: fixup more tests to specify the target more explicitly
Saleem Abdulrasool [Thu, 3 Apr 2014 16:01:44 +0000 (16:01 +0000)]
ARM: fixup more tests to specify the target more explicitly

This changes the tests that were targeting ARM EABI to explicitly specify the
environment rather than relying on the default.  This breaks with the new
Windows on ARM support when running the tests on Windows where the default
environment is no longer EABI.

Take the opportunity to avoid a pointless redirect (helps when trying to debug
with providing a command line invocation which can be copy and pasted) and
removing a few greps in favour of FileCheck.

llvm-svn: 205541

10 years agoARM: tell LLVM about zext properties of ldrexb/ldrexh
Tim Northover [Thu, 3 Apr 2014 15:10:35 +0000 (15:10 +0000)]
ARM: tell LLVM about zext properties of ldrexb/ldrexh

Implementing this via ComputeMaskedBits has two advantages:
  + It actually works. DAGISel doesn't deal with the chains properly
    in the previous pattern-based solution, so they never trigger.
  + The information can be used in other DAG combines, as well as the
    trivial "get rid of truncs". For example if the trunc is in a
    different basic block.

rdar://problem/16227836

llvm-svn: 205540

10 years agoAdd explict dependencies on swig .i files for cmake builds
Ed Maste [Thu, 3 Apr 2014 15:03:11 +0000 (15:03 +0000)]
Add explict dependencies on swig .i files for cmake builds

llvm.org/pr19316

llvm-svn: 205539

10 years ago[mips] Add negative tests confirm that supported ISA's don't allow instructions added...
Daniel Sanders [Thu, 3 Apr 2014 14:14:22 +0000 (14:14 +0000)]
[mips] Add negative tests confirm that supported ISA's don't allow instructions added in later ISA's

Summary:
test/MC/Mips/<isa1>/invalid-<isa2>.s
    Test that <isa1> does not support <isa2>'s instructions.
test/MC/Mips/<isa1>/invalid-<isa2>-xfail.s
    Things that should be invalid but currently aren't. Will XPASS if any
    become invalid.

Reviewers: matheusalmeida

Reviewed By: matheusalmeida

Differential Revision: http://llvm-reviews.chandlerc.com/D3262

llvm-svn: 205538

10 years ago[mips] Implement ehb, ssnop, and pause in assembler
Daniel Sanders [Thu, 3 Apr 2014 13:21:51 +0000 (13:21 +0000)]
[mips] Implement ehb, ssnop, and pause in assembler

Summary: Add negative tests for pause

Reviewers: matheusalmeida

Reviewed By: matheusalmeida

Differential Revision: http://llvm-reviews.chandlerc.com/D3246

llvm-svn: 205537

10 years agoCode cleanup (re-indent)
Logan Chien [Thu, 3 Apr 2014 13:12:44 +0000 (13:12 +0000)]
Code cleanup (re-indent)

llvm-svn: 205536

10 years agoARM: skip cmpxchg failure barrier if ordering is monotonic.
Tim Northover [Thu, 3 Apr 2014 13:06:54 +0000 (13:06 +0000)]
ARM: skip cmpxchg failure barrier if ordering is monotonic.

The terminal barrier of a cmpxchg expansion will be either Acquire or
SequentiallyConsistent. In either case it can be skipped if the
operation has Monotonic requirements on failure.

rdar://problem/15996804

llvm-svn: 205535

10 years ago[TSan] Fix a rare deadlock on multithreaded fork.
Alexey Samsonov [Thu, 3 Apr 2014 12:51:26 +0000 (12:51 +0000)]
[TSan] Fix a rare deadlock on multithreaded fork.

If a multi-threaded program calls fork(), TSan ignores all memory accesses
in the child to prevent deadlocks in TSan runtime. This is OK, as child is
probably going to call exec() as soon as possible. However, a rare deadlocks
could be caused by ThreadIgnoreBegin() function itself.

ThreadIgnoreBegin() remembers the current stack trace and puts it into the
StackDepot to report a warning later if a thread exited with ignores enabled.
Using StackDepotPut in a child process is dangerous: it locks a mutex on
a slow path, which could be already locked in a parent process.

The fix is simple: just don't put current stack traces to StackDepot in
ThreadIgnoreBegin() and ThreadIgnoreSyncBegin() functions if we're
running after a multithreaded fork. We will not report any
"thread exited with ignores enabled" errors in this case anyway.

Submitting this without a testcase, as I believe the standalone reproducer
is pretty hard to construct.

llvm-svn: 205534

10 years agoRevert "Add support for named values in the parser."
Samuel Benzaquen [Thu, 3 Apr 2014 12:50:47 +0000 (12:50 +0000)]
Revert "Add support for named values in the parser."
This was submitted before it was ready.

This reverts commit 62060a01e095cf35eb9ca42a333752d12714f35c.

llvm-svn: 205533

10 years agoImplementation of 16-bit microMIPS instructions MFHI and MFLO.
Zoran Jovanovic [Thu, 3 Apr 2014 12:47:34 +0000 (12:47 +0000)]
Implementation of 16-bit microMIPS instructions MFHI and MFLO.
Differential Revision: http://llvm-reviews.chandlerc.com/D3141

llvm-svn: 205532

10 years ago[TSan] Fix a typo ThreadIgnoreSyncEnd. Found by inspection
Alexey Samsonov [Thu, 3 Apr 2014 12:28:16 +0000 (12:28 +0000)]
[TSan] Fix a typo ThreadIgnoreSyncEnd. Found by inspection

llvm-svn: 205531

10 years ago[mips] Add initial (experimental) MIPS-IV support.
Daniel Sanders [Thu, 3 Apr 2014 12:13:36 +0000 (12:13 +0000)]
[mips] Add initial (experimental) MIPS-IV support.

Summary:
Adds the 'mips4' processor and a simple test of the ELF e_flags.

Patch by David Chisnall
His work was sponsored by: DARPA, AFRL

I made one small change to the testcase so that it uses
mips64-unknown-linux instead of mips4-unknown-linux.

This patch indirectly adds FeatureCondMov to FeatureMips64. This is ok
because it's supposed to be there anyway and it turns out that
FeatureCondMov is not a predicate of any instructions at the moment
(this is a bug that hasn't been noticed because there are no targets
without the conditional move instructions yet).

CC: theraven
Differential Revision: http://llvm-reviews.chandlerc.com/D3244

llvm-svn: 205530

10 years agoFix for PR 19261:
Eric Christopher [Thu, 3 Apr 2014 12:11:51 +0000 (12:11 +0000)]
Fix for PR 19261:

llc doesn't generate nodes for unconditional fall-through branches for targets
without FastISel implementation (X86 has it, but can be disabled by
"-fast-isel=false") in SelectionDAGBuilder::visitBr().

So for line 4 in the following testcase

1: void foo(int i){
2:   switch(i){
3:   default:
4:     break;
5:   }
6:   return;
7: }

there is no corresponding line in .debug_line section, and a debugger
cannot set a breakpoint at line 4.

Fix this by always emitting a branch when we're not optimizing and add a
testcase to ensure that there's code on every line we'd want to break.

Patch by Daniil Fukalov.

llvm-svn: 205529

10 years agoMicroMIPS specific little endian fixup data byte ordering.
Zoran Jovanovic [Thu, 3 Apr 2014 12:01:01 +0000 (12:01 +0000)]
MicroMIPS specific little endian fixup data byte ordering.
Differential Revision: http://llvm-reviews.chandlerc.com/D3245

llvm-svn: 205528

10 years agoclang-format: Prefer an additional line-break over hanging indent.
Daniel Jasper [Thu, 3 Apr 2014 12:00:33 +0000 (12:00 +0000)]
clang-format: Prefer an additional line-break over hanging indent.

Don't allow the RHS of an operator to be split over multiple
lines unless there is a line-break right after the operator.

Before:
  if (aaaa && bbbbb || // break
                  cccc) {
  }

After:
  if (aaaa &&
      bbbbb || // break
          cccc) {
  }

In most cases, this seems to increase readability.

llvm-svn: 205527

10 years agox
Daniel Jasper [Thu, 3 Apr 2014 12:00:27 +0000 (12:00 +0000)]
x

llvm-svn: 205526

10 years agoARM: expand atomic ldrex/strex loops in IR
Tim Northover [Thu, 3 Apr 2014 11:44:58 +0000 (11:44 +0000)]
ARM: expand atomic ldrex/strex loops in IR

The previous situation where ATOMIC_LOAD_WHATEVER nodes were expanded
at MachineInstr emission time had grown to be extremely large and
involved, to account for the subtly different code needed for the
various flavours (8/16/32/64 bit, cmpxchg/add/minmax).

Moving this transformation into the IR clears up the code
substantially, and makes future optimisations much easier:

1. an atomicrmw followed by using the *new* value can be more
   efficient. As an IR pass, simple CSE could handle this
   efficiently.
2. Making use of cmpxchg success/failure orderings only has to be done
   in one (simpler) place.
3. The common "cmpxchg; did we store?" idiom can be exposed to
   optimisation.

I intend to gradually improve this situation within the ARM backend
and make sure there are no hidden issues before moving the code out
into CodeGen to be shared with (at least ARM64/AArch64, though I think
PPC & Mips could benefit too).

llvm-svn: 205525

10 years agoPR19320:
Stepan Dyatkovskiy [Thu, 3 Apr 2014 11:29:15 +0000 (11:29 +0000)]
PR19320:
The trouble as in ARMAsmParser, in ParseInstruction method. It assumes that ARM::R12 + 1 == ARM::SP.
It is wrong, since ARM::<Register> codes are generated by tablegen and actually could be any random numbers.

llvm-svn: 205524

10 years ago[ARM] When generating a vpaddl node the input lane type is not always the type of the
Silviu Baranga [Thu, 3 Apr 2014 10:44:27 +0000 (10:44 +0000)]
[ARM] When generating a vpaddl node the input lane type is not always the type of the
add operation since extract_vector_elt can perform an extend operation. Get the input lane
type from the vector on which we're performing the vpaddl operation on and extend or
truncate it to the output type of the original add node.

llvm-svn: 205523

10 years ago[mips] Extend MipsMCExpr class to handle %higher(sym1 - sym2 + const) and
Sasa Stankovic [Thu, 3 Apr 2014 10:37:45 +0000 (10:37 +0000)]
[mips] Extend MipsMCExpr class to handle %higher(sym1 - sym2 + const) and
%highest(sym1 - sym2 + const) relocations. Remove "ABS_" from VK_Mips_HI
and VK_Mips_LO enums in MipsMCExpr, to be consistent with VK_Mips_HIGHER
and VK_Mips_HIGHEST.

This change also deletes test file test/MC/Mips/higher_highest.ll and moves
its CHECK's to the new test file test/MC/Mips/higher-highest-addressing.s.
The deleted file tests that R_MIPS_HIGHER and R_MIPS_HIGHEST relocations are
emitted in the .o file. Since it uses -force-mips-long-branch option, it was
created when MipsLongBranch's implementation was emitting R_MIPS_HIGHER and
R_MIPS_HIGHEST relocations in the .o file. It was disabled when MipsLongBranch
started to directly calculate offsets.

Differential Revision: http://llvm-reviews.chandlerc.com/D3230

llvm-svn: 205522

10 years agoImproved semantics for implicit scalar -> extvector conversions.
Stephen Canon [Thu, 3 Apr 2014 10:33:25 +0000 (10:33 +0000)]
Improved semantics for implicit scalar -> extvector conversions.

llvm-svn: 205521

10 years agoARM64: add regression test for r205519.
Tim Northover [Thu, 3 Apr 2014 09:36:05 +0000 (09:36 +0000)]
ARM64: add regression test for r205519.

llvm-svn: 205520

10 years agoARM64: always use i64 for the RHS of shift operations
Tim Northover [Thu, 3 Apr 2014 09:26:16 +0000 (09:26 +0000)]
ARM64: always use i64 for the RHS of shift operations

Switching between i32 and i64 based on the LHS type is a good idea in
theory, but pre-legalisation uses i64 regardless of our choice,
leading to potential ISel errors.

Should fix PR19294.

llvm-svn: 205519

10 years agoRTTI Uniqueness: remove __name_for_load function.
Tim Northover [Thu, 3 Apr 2014 09:12:38 +0000 (09:12 +0000)]
RTTI Uniqueness: remove __name_for_load function.

It's identical to name() these days. (At one point it avoided masking
of the RTTI uniqueness bit because ARM64 ignored it architecturally,
but no longer).

llvm-svn: 205518

10 years agoclang-format: Understand that "auto" is a type.
Daniel Jasper [Thu, 3 Apr 2014 09:00:49 +0000 (09:00 +0000)]
clang-format: Understand that "auto" is a type.

Before:
  MACRO(auto * a);

After:
  MACRO(auto *a);

llvm-svn: 205517

10 years agoARM: Use __STACK_LIMIT symbol for segmented stacks
Oliver Stannard [Thu, 3 Apr 2014 08:45:16 +0000 (08:45 +0000)]
ARM: Use __STACK_LIMIT symbol for segmented stacks

We cannot use STACK_LIMIT, as it is not reserved for the compiler
by the C spec.

llvm-svn: 205516

10 years agoStack map docs. Remove some stray markup.
Andrew Trick [Thu, 3 Apr 2014 07:08:21 +0000 (07:08 +0000)]
Stack map docs. Remove some stray markup.

llvm-svn: 205515

10 years agoARM64: don't generate __sincos_stret calls unless on MachO
Tim Northover [Thu, 3 Apr 2014 07:06:13 +0000 (07:06 +0000)]
ARM64: don't generate __sincos_stret calls unless on MachO

This should fix PR19314.

llvm-svn: 205514

10 years agoMinor update to the stack map documentation.
Andrew Trick [Thu, 3 Apr 2014 07:03:28 +0000 (07:03 +0000)]
Minor update to the stack map documentation.

llvm-svn: 205513

10 years agoDebugInfo: Use a 64 bit type for the subrange
David Blaikie [Thu, 3 Apr 2014 06:28:20 +0000 (06:28 +0000)]
DebugInfo: Use a 64 bit type for the subrange

While we were encoding 64 bit values (data8) in the subrange itself,
using a 32 bit type for the subrange was still confusing the gdb. Oh,
and make it unsigned too.

As the comment points out, this could be pushed into the frontend so
that it would be 32 or 64 bit as appropriate, etc.

llvm-svn: 205512

10 years ago[CodeGen] Fix peephole optimizer bug introduced in r205481. Fixes PR19318.
Lang Hames [Thu, 3 Apr 2014 05:03:20 +0000 (05:03 +0000)]
[CodeGen] Fix peephole optimizer bug introduced in r205481. Fixes PR19318.

I should have read that comment a little more carefully. ;)

Regression test in the works, committing in the mean time to un-break people.

llvm-svn: 205511

10 years agoAdd comment for Linker Script node.
Rui Ueyama [Thu, 3 Apr 2014 04:10:52 +0000 (04:10 +0000)]
Add comment for Linker Script node.

llvm-svn: 205510

10 years agoHandle getting UnknownAddressOrSize or section_end().
Rafael Espindola [Thu, 3 Apr 2014 03:57:03 +0000 (03:57 +0000)]
Handle getting UnknownAddressOrSize or section_end().

These should probably be error conditions.

llvm-svn: 205509

10 years agoImplement get getSymbolFileOffset with getSymbolAddress.
Rafael Espindola [Thu, 3 Apr 2014 03:13:33 +0000 (03:13 +0000)]
Implement get getSymbolFileOffset with getSymbolAddress.

This has the following advantages:
* Less code.
* The old ELF implementation was wrong for non-relocatable objects.
* The old ELF implementation (and I think MachO) was wrong for thumb.

No current testcase since this is only used from MCJIT and it only uses
relocatable objects and I don't think it supports thumb yet.

llvm-svn: 205508

10 years agoAdd a section about reporting bugs and contributing patches
Marshall Clow [Thu, 3 Apr 2014 03:13:12 +0000 (03:13 +0000)]
Add a section about reporting bugs and contributing patches

llvm-svn: 205507

10 years ago[OPENMP] Small update for C++11
Alexey Bataev [Thu, 3 Apr 2014 02:55:46 +0000 (02:55 +0000)]
[OPENMP] Small update for C++11

llvm-svn: 205506

10 years agoReword C++14 status to match C++11 status
Marshall Clow [Thu, 3 Apr 2014 02:38:12 +0000 (02:38 +0000)]
Reword C++14 status to match C++11 status

llvm-svn: 205505

10 years agoMark C++14 status as 'complete'
Marshall Clow [Thu, 3 Apr 2014 02:35:29 +0000 (02:35 +0000)]
Mark C++14 status as 'complete'

llvm-svn: 205504

10 years agoRemove getSymbolValue.
Rafael Espindola [Thu, 3 Apr 2014 02:32:47 +0000 (02:32 +0000)]
Remove getSymbolValue.

All existing users explicitly ask for an address or a file offset.

llvm-svn: 205503

10 years agoAdd support for the R_ARM_ABS32 relocation.
Rafael Espindola [Thu, 3 Apr 2014 02:27:00 +0000 (02:27 +0000)]
Add support for the R_ARM_ABS32 relocation.

This should bring the arm buildbots back.

llvm-svn: 205502

10 years agoRemove ordinals from Input Graph elements.
Rui Ueyama [Thu, 3 Apr 2014 02:21:44 +0000 (02:21 +0000)]
Remove ordinals from Input Graph elements.

An ordinal is set to each child of Input Graph, but no one actually
uses it. The only piece of code that gets ordinaly values is
sortInputElements in InputGraph.cpp, but it does not actually do
anything -- we assign ordinals in increasing order just before
calling sort, so when sort is called it's already sorted. It's no-op.
We can simply remove it. No functionality change.

Differential Revision: http://llvm-reviews.chandlerc.com/D3270

llvm-svn: 205501

10 years agoOnly clear the thumb bit from function addresses.
Rafael Espindola [Thu, 3 Apr 2014 02:20:43 +0000 (02:20 +0000)]
Only clear the thumb bit from function addresses.

llvm-svn: 205500

10 years agoSimplify ELFObjectFile<ELFT>::getSymbolAddress.
Rafael Espindola [Thu, 3 Apr 2014 01:51:15 +0000 (01:51 +0000)]
Simplify ELFObjectFile<ELFT>::getSymbolAddress.

In particular, we only need to fetch the section if this is a relocatable
object.

No functionality change.

llvm-svn: 205499

10 years agoRevert "[Constant Hoisting] Lazily compute the idom and cache the result."
Juergen Ributzka [Thu, 3 Apr 2014 01:38:47 +0000 (01:38 +0000)]
Revert "[Constant Hoisting] Lazily compute the idom and cache the result."

This code is no longer usefull, because we only compute and use the
IDom once. There is no benefit in caching it anymore.

llvm-svn: 205498

10 years agoMake the fail messages
Jim Ingham [Thu, 3 Apr 2014 01:26:14 +0000 (01:26 +0000)]
Make the fail messages

llvm-svn: 205497

10 years agoPrint the new eStopReasonExec in stop_reason_to_str.
Jim Ingham [Thu, 3 Apr 2014 01:25:28 +0000 (01:25 +0000)]
Print the new eStopReasonExec in stop_reason_to_str.

llvm-svn: 205496

10 years agoAccount for scalarization costs in BasicTTI::getMemoryOpCost for extending vector...
Hal Finkel [Thu, 3 Apr 2014 00:53:59 +0000 (00:53 +0000)]
Account for scalarization costs in BasicTTI::getMemoryOpCost for extending vector loads

When a vector type legalizes to a larger vector type, and the target does not
support the associated extending load (or truncating store), then legalization
will scalarize the load (or store) resulting in an associated scalarization
cost.  BasicTTI::getMemoryOpCost needs to account for this.

Between this, and r205487, PowerPC on the P7 with VSX enabled shows:

MultiSource/Benchmarks/PAQ8p/paq8p: 43% speedup
SingleSource/Benchmarks/BenchmarkGame/puzzle: 51% speedup
SingleSource/UnitTests/Vectorizer/gcc-loops 28% speedup

(some of these are new; some of these, such as PAQ8p, just reverse regressions
that VSX support would trigger)

llvm-svn: 205495

10 years agoI removed SetCallback from the Python API's since it wasn't actually useful, and...
Jim Ingham [Thu, 3 Apr 2014 00:50:56 +0000 (00:50 +0000)]
I removed SetCallback from the Python API's since it wasn't actually useful, and added SetScriptCallbackFunction,
and SetScriptCallbackBody.  So add these to the default constructor test case.

llvm-svn: 205494

10 years agoRevert "Fix a nomenclature error in llvm-nm."
Rafael Espindola [Thu, 3 Apr 2014 00:19:35 +0000 (00:19 +0000)]
Revert "Fix a nomenclature error in llvm-nm."

This reverts commit r205479.

It turns out that nm does use addresses, it is just that every reasonable
relocatable ELF object has sections with address 0. I have no idea if those
exist in reality, but it at least it shows that llvm-nm should use the name
address.

The added test was includes an unusual .o file with non 0 section addresses. I
created it by hacking ELFObjectWriter.cpp.

Really sorry for the churn.

llvm-svn: 205493

10 years agoFix test case.
Richard Trieu [Thu, 3 Apr 2014 00:14:18 +0000 (00:14 +0000)]
Fix test case.

llvm-svn: 205492

10 years agoAllow ASTFile to be reset to null
Ben Langmuir [Thu, 3 Apr 2014 00:14:16 +0000 (00:14 +0000)]
Allow ASTFile to be reset to null

Which we do in removeModules().

llvm-svn: 205491

10 years agoFix style.
Rui Ueyama [Thu, 3 Apr 2014 00:01:57 +0000 (00:01 +0000)]
Fix style.

llvm-svn: 205490

10 years ago[X86] As per suggestion from Craig Topper and Hal Finkel, override
Lang Hames [Wed, 2 Apr 2014 23:57:49 +0000 (23:57 +0000)]
[X86] As per suggestion from Craig Topper and Hal Finkel, override
TargetInstrInfo::findCommutedOpIndices to enable VFMA*231 commutation, rather
than abusing commuteInstruction.

Thanks very much for the suggestion guys!

llvm-svn: 205489

10 years agoAdd a missing arm64 idef.
Jason Molenda [Wed, 2 Apr 2014 23:52:55 +0000 (23:52 +0000)]
Add a missing arm64 idef.

llvm-svn: 205488

10 years agoFix multi-register costs in BasicTTI::getCastInstrCost
Hal Finkel [Wed, 2 Apr 2014 23:18:54 +0000 (23:18 +0000)]
Fix multi-register costs in BasicTTI::getCastInstrCost

For an cast (extension, etc.), the currently logic predicts a low cost if the
associated operation (keyed on the destination type) is legal (or promoted).
This is not true when the number of values required to legalize the type is
changing. For example, <8 x i16> being sign extended by <8 x i32> is not
generically cheap on PPC with VSX, even though sign extension to v4i32 is
legal, because two output v4i32 values are required compared to the single
v8i16 input value, and without custom logic in the target, this conversion will
scalarize.

llvm-svn: 205487

10 years agoMinor style fix.
Rui Ueyama [Wed, 2 Apr 2014 23:17:39 +0000 (23:17 +0000)]
Minor style fix.

llvm-svn: 205486

10 years agoCodeGen: Emit some functions as weak_odr under -fms-compatibility
David Majnemer [Wed, 2 Apr 2014 23:17:29 +0000 (23:17 +0000)]
CodeGen: Emit some functions as weak_odr under -fms-compatibility

Summary:
MSVC always emits inline functions marked with the extern storage class
specifier.  The result is something similar to the opposite of
__attribute__((gnu_inline)).

This extension is also available in C.

This fixes PR19264.

Reviewers: rnk, rsmith

CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D3207

llvm-svn: 205485

10 years agoAdd test case for [Constant Hoisting] Erase dead cast instructions (r204538).
Juergen Ributzka [Wed, 2 Apr 2014 23:06:22 +0000 (23:06 +0000)]
Add test case for [Constant Hoisting] Erase dead cast instructions (r204538).

llvm-svn: 205484

10 years agoARM Linux support
Renato Golin [Wed, 2 Apr 2014 23:03:28 +0000 (23:03 +0000)]
ARM Linux support

llvm-svn: 205483

10 years agoAdd decorator for pr19311 on FreeBSD
Ed Maste [Wed, 2 Apr 2014 23:01:04 +0000 (23:01 +0000)]
Add decorator for pr19311 on FreeBSD

A breakpoint at the resume address for "continue" is skipped.  This is
presumably the root cause for the Linux failure as well.

llvm-svn: 205482

10 years ago[CodeGen] Teach the peephole optimizer to remember (and exploit) all folding
Lang Hames [Wed, 2 Apr 2014 22:59:58 +0000 (22:59 +0000)]
[CodeGen] Teach the peephole optimizer to remember (and exploit) all folding
opportunities in the current basic block, rather than just the last one seen.

<rdar://problem/16478629>

llvm-svn: 205481

10 years agoWorkaround for collision between enum members in LLVM's MachO.h and system headers
Jim Ingham [Wed, 2 Apr 2014 22:53:21 +0000 (22:53 +0000)]
Workaround for collision between enum members in LLVM's MachO.h and system headers
on Mac OS X (in particular mach/machine.h).

<rdar://problem/16494607>

llvm-svn: 205480

10 years agoFix a nomenclature error in llvm-nm.
Rafael Espindola [Wed, 2 Apr 2014 22:52:46 +0000 (22:52 +0000)]
Fix a nomenclature error in llvm-nm.

What llvm-nm prints depends on the file format. On ELF for example, if the
file is relocatable, it prints offsets. If it is not, it prints addresses.
Since it doesn't really need to care what it is that it is printing, use the
generic term value.

Fix or implement getSymbolValue to keep llvm-nm working.

llvm-svn: 205479

10 years agoAdd ability to disable building LLVM utils
Pete Cooper [Wed, 2 Apr 2014 22:49:58 +0000 (22:49 +0000)]
Add ability to disable building LLVM utils

Patch by Chris Bieneman

llvm-svn: 205478

10 years agoObjective-C [IRGen]. Add encoding for 'weak' attribute of a
Fariborz Jahanian [Wed, 2 Apr 2014 22:49:42 +0000 (22:49 +0000)]
Objective-C [IRGen]. Add encoding for 'weak' attribute of a
'readonly' property. // rdar://16136439

llvm-svn: 205477

10 years ago[PowerPC] Make PPCTTI::getMemoryOpCost call BasicTTI::getMemoryOpCost
Hal Finkel [Wed, 2 Apr 2014 22:43:49 +0000 (22:43 +0000)]
[PowerPC] Make PPCTTI::getMemoryOpCost call BasicTTI::getMemoryOpCost

PPCTTI::getMemoryOpCost will now make use of BasicTTI::getMemoryOpCost to
calculate the base cost of the memory access, and then adjust on top of that.
There is no functionality change from this modification, but it will become
important so that PPCTTI can take advantage of scalarization information for which
BasicTTI::getMemoryOpCost will account in the near future.

llvm-svn: 205476

10 years agoConcatenate strings that are unnecessarily separated.
Rui Ueyama [Wed, 2 Apr 2014 22:32:51 +0000 (22:32 +0000)]
Concatenate strings that are unnecessarily separated.

llvm-svn: 205475

10 years agoAdd comments and test case for [DAG] Keep the opaque constant flag when performing...
Juergen Ributzka [Wed, 2 Apr 2014 22:21:01 +0000 (22:21 +0000)]
Add comments and test case for [DAG] Keep the opaque constant flag when performing unary constant folding operations (r204737).

llvm-svn: 205474

10 years agotypo
Adrian Prantl [Wed, 2 Apr 2014 22:17:30 +0000 (22:17 +0000)]
typo

llvm-svn: 205473

10 years ago[X86] Make the VFMA*231 variants commutable and relax the alignment restrictions
Lang Hames [Wed, 2 Apr 2014 22:06:16 +0000 (22:06 +0000)]
[X86] Make the VFMA*231 variants commutable and relax the alignment restrictions
on FMA3 memory operands. FMA3 instructions are VEX encoded, so they can load
from unaligned memory.

Testcase to follow, along with related patch.

<rdar://problem/16478629>

llvm-svn: 205472

10 years agoRevert "Reapply "LTO: add API to set strategy for -internalize""
Duncan P. N. Exon Smith [Wed, 2 Apr 2014 22:05:57 +0000 (22:05 +0000)]
Revert "Reapply "LTO: add API to set strategy for -internalize""

This reverts commit r199244.

Conflicts:
include/llvm-c/lto.h
include/llvm/LTO/LTOCodeGenerator.h
lib/LTO/LTOCodeGenerator.cpp

llvm-svn: 205471

10 years agoMinor cleanup.
Rui Ueyama [Wed, 2 Apr 2014 21:57:29 +0000 (21:57 +0000)]
Minor cleanup.

llvm-svn: 205470

10 years agoRemove unused fields/member functions from unit test.
Rui Ueyama [Wed, 2 Apr 2014 21:49:27 +0000 (21:49 +0000)]
Remove unused fields/member functions from unit test.

llvm-svn: 205469

10 years agoAdd comments and test case for [X86TTI] Make constant base pointers for GetElementPtr...
Juergen Ributzka [Wed, 2 Apr 2014 21:45:36 +0000 (21:45 +0000)]
Add comments and test case for [X86TTI] Make constant base pointers for GetElementPtr opaque (r204739).

llvm-svn: 205468

10 years agoIf a using-declaration names a class member, but appears outside a class, try
Richard Smith [Wed, 2 Apr 2014 21:44:35 +0000 (21:44 +0000)]
If a using-declaration names a class member, but appears outside a class, try
to suggest a different syntax to get the same effect.

llvm-svn: 205467

10 years agoRemove unused include.
Rui Ueyama [Wed, 2 Apr 2014 21:39:44 +0000 (21:39 +0000)]
Remove unused include.

llvm-svn: 205466

10 years agoARM: fixup tests to specify the target more explicitly
Saleem Abdulrasool [Wed, 2 Apr 2014 21:22:03 +0000 (21:22 +0000)]
ARM: fixup tests to specify the target more explicitly

This changes the tests that were targeting ARM EABI to explicitly specify the
environment rather than relying on the default.  This breaks with the new
Windows on ARM support when running the tests on Windows where the default
environment is no longer EABI.

Take the opportunity to avoid a pointless redirect (helps when trying to debug
with providing a command line invocation which can be copy and pasted) and
removing a few greps in favour of FileCheck.

llvm-svn: 205465

10 years agoAdd test case for [Stackmaps][X86TTI] Fix think-o in getIntImmCost calculation (r204738).
Juergen Ributzka [Wed, 2 Apr 2014 21:15:36 +0000 (21:15 +0000)]
Add test case for [Stackmaps][X86TTI] Fix think-o in getIntImmCost calculation (r204738).

llvm-svn: 205464

10 years agoSimplify communication between Resolver and Input Graph.
Rui Ueyama [Wed, 2 Apr 2014 21:02:44 +0000 (21:02 +0000)]
Simplify communication between Resolver and Input Graph.

Resolver is sending too much information to Input Graph than Input
Graph actually needs. In order to collect the detailed information,
which wouldn't be consumed by anyone, we have a good amount of code
in Resolver, Input Graph and Input Elements. This patch is to
simplify it. No functionality change.

Specifically, this patch replaces ResolverState enum with a boolean
value. The enum defines many bits to notify the progress about
linking to Input Graph using bit masks, however, what Input Graph
actually does is to compare a given value with 0. The details of
the bit mask is simply being ignored, so the efforts to collect
such data is wasted.

This patch also changes the name of the notification interface from
setResolverState to notifyProgress, to make it sounds more like
message passing style. It's not a setter but something to notify of
an update, so the new name should be more appropriate than before.

Differential Revision: http://llvm-reviews.chandlerc.com/D3267

llvm-svn: 205463

10 years agoDon’t #include "llvm/Support/MachO.h” as it isn’t needed here.
Greg Clayton [Wed, 2 Apr 2014 20:39:20 +0000 (20:39 +0000)]
Don’t #include "llvm/Support/MachO.h” as it isn’t needed here.

llvm-svn: 205462

10 years agoDon’t include “llvm/Support/MachO.h” as it isn’t needed here.
Greg Clayton [Wed, 2 Apr 2014 20:38:03 +0000 (20:38 +0000)]
Don’t include “llvm/Support/MachO.h” as it isn’t needed here.

llvm-svn: 205461

10 years agoDon’t #include “llvm/Support/MachO.h” in a header file if we can avoid it.
Greg Clayton [Wed, 2 Apr 2014 20:36:22 +0000 (20:36 +0000)]
Don’t #include “llvm/Support/MachO.h” in a header file if we can avoid it.

llvm-svn: 205460

10 years agoARM: update subtarget information for Windows on ARM
Saleem Abdulrasool [Wed, 2 Apr 2014 20:32:05 +0000 (20:32 +0000)]
ARM: update subtarget information for Windows on ARM

Update the subtarget information for Windows on ARM.  This enables using the MC
layer to target Windows on ARM.

llvm-svn: 205459

10 years agoMake a few more range-based loops use explicit types.
Jim Grosbach [Wed, 2 Apr 2014 20:21:22 +0000 (20:21 +0000)]
Make a few more range-based loops use explicit types.

No functional change.

llvm-svn: 205458

10 years agoUse explit types in foreach iterators
Tobias Grosser [Wed, 2 Apr 2014 20:18:19 +0000 (20:18 +0000)]
Use explit types in foreach iterators

llvm-svn: 205457

10 years agoAdd back an assert that was lost in the ELFObjectFile.h split.
Rafael Espindola [Wed, 2 Apr 2014 20:00:33 +0000 (20:00 +0000)]
Add back an assert that was lost in the ELFObjectFile.h split.

llvm-svn: 205456

10 years agoTargetLibraryInfo: Disable memcpy and memset on R600
Tom Stellard [Wed, 2 Apr 2014 19:53:29 +0000 (19:53 +0000)]
TargetLibraryInfo: Disable memcpy and memset on R600

There are no implementations of these for R600.

llvm-svn: 205455

10 years agoRemove dead code.
Rui Ueyama [Wed, 2 Apr 2014 19:42:35 +0000 (19:42 +0000)]
Remove dead code.

llvm-svn: 205454

10 years agoSimplify resolveFrameIndex() signature.
Jim Grosbach [Wed, 2 Apr 2014 19:28:18 +0000 (19:28 +0000)]
Simplify resolveFrameIndex() signature.

Just pass a MachineInstr reference rather than an MBB iterator.
Creating a MachineInstr& is the first thing every implementation did
anyway.

llvm-svn: 205453