oota-llvm.git
7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220599 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220594 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220592 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220584 91177308-0d34-0410-b5e6-96231b3b80d8

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220581 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220580 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220579 91177308-0d34-0410-b5e6-96231b3b80d8

7 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)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220578 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220577 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220575 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220570 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220568 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220567 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220565 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220564 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220561 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220556 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220553 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220548 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220545 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220544 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220542 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220541 91177308-0d34-0410-b5e6-96231b3b80d8

7 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>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220540 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220539 91177308-0d34-0410-b5e6-96231b3b80d8

7 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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220537 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220536 91177308-0d34-0410-b5e6-96231b3b80d8

7 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220533 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUpdate comment and fix typos in assert message. (NFC)
Ahmed Bougacha [Thu, 23 Oct 2014 22:40:34 +0000 (22:40 +0000)]
Update comment and fix typos in assert message. (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220531 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUpdate llvm.donothing documentation.
Juergen Ributzka [Thu, 23 Oct 2014 22:36:13 +0000 (22:36 +0000)]
Update llvm.donothing documentation.

llvm.donothing is no longer the only intrinsic that can be invoked.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220530 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoScheduleDAG: record PhysReg dependencies represented by CopyFromReg nodes
Tim Northover [Thu, 23 Oct 2014 22:31:48 +0000 (22:31 +0000)]
ScheduleDAG: record PhysReg dependencies represented by CopyFromReg nodes

x86's CMPXCHG -> EFLAGS consumer wasn't being recorded as a real EFLAGS
dependency because it was represented by a pair of CopyFromReg(EFLAGS) ->
CopyToReg(EFLAGS) nodes. ScheduleDAG was expecting the source to be an
implicit-def on the instruction, where the result numbers in the DAG and the
Uses list in TableGen matched up precisely.

The Copy notation seems much more robust, so this patch extends ScheduleDAG
rather than refactoring x86.

Should fix PR20376.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220529 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDebugInfo: Remove DwarfDebug::CurrentFnArguments since we have to handle argument...
David Blaikie [Thu, 23 Oct 2014 22:27:50 +0000 (22:27 +0000)]
DebugInfo: Remove DwarfDebug::CurrentFnArguments since we have to handle argument ordering of other arguments (abstract arguments) in the same way and already have code for that too.

While refactoring this code I was confused by both the name I had
introduced (addNonArgumentVariable... but it has all this logic to
handle argument numbering and keep things in order?) and by the
redundancy. Seems when I fixed the misordered inlined argument handling,
I didn't realize it was mostly redundant with the argument ordering code
(which I may've also written, I'm not sure). So let's just rely on the
more general case.

The only oddity in output this produces is that it means when we emit
all the variables for the current function, we don't track when we've
finished the argument variables and are about to start the local
variables and insert DW_AT_unspecified_parameters (for varargs
functions) there. Instead it ends up after the local variables, scopes,
etc. But this isn't invalid and doesn't cause DWARF consumers problems
that I know of... so we'll just go with that because it makes the code
nice & simple.

(though, let's see what the buildbots have to say about this - *crosses
fingers*)

There will be some cleanup commits to follow to remove the now trivial
wrappers, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220527 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoPR21189: Teach llvm-readobj to dump bits of COFF symbol subsections required to debug...
Timur Iskhodzhanov [Thu, 23 Oct 2014 22:25:31 +0000 (22:25 +0000)]
PR21189: Teach llvm-readobj to dump bits of COFF symbol subsections required to debug using VS2012+

Reviewed at http://reviews.llvm.org/D5755
Thanks to Andrey Guskov for his help investigating this!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220526 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDebugInfo: Sink DwarfDebug::addNonArgumentScopeVariable into DwarfFile.
David Blaikie [Thu, 23 Oct 2014 22:04:30 +0000 (22:04 +0000)]
DebugInfo: Sink DwarfDebug::addNonArgumentScopeVariable into DwarfFile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220520 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMachODump.cpp: fix MSVC build
Hans Wennborg [Thu, 23 Oct 2014 21:59:17 +0000 (21:59 +0000)]
MachODump.cpp: fix MSVC build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220518 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Improve mul w/ overflow codegen, to MUL8+SETO.
Ahmed Bougacha [Thu, 23 Oct 2014 21:55:31 +0000 (21:55 +0000)]
[X86] Improve mul w/ overflow codegen, to MUL8+SETO.

Currently, @llvm.smul.with.overflow.i8 expands to 9 instructions, where
3 are really needed.

This adds X86ISD::UMUL8/SMUL8 SD nodes, and custom lowers them to
MUL8/IMUL8 + SETO.

i8 is a special case because there is no two/three operand variants of
(I)MUL8, so the first operand and return value need to go in AL/AX.

Also, we can't write patterns for these instructions: TableGen refuses
patterns where output operands don't match SDNode results. In this case,
instructions where the output operand is an implicitly defined register.

A related special case (and FIXME) exists for MUL8 (X86InstrArith.td):

  // FIXME: Used for 8-bit mul, ignore result upper 8 bits.
  // This probably ought to be moved to a def : Pat<> if the
  // syntax can be accepted.
  [(set AL, (mul AL, GR8:$src)), (implicit EFLAGS)]

Ideally, these go away with UMUL8, but we still need to improve TableGen
support of implicit operands in patterns.

Before this change:
  movsbl  %sil, %eax
  movsbl  %dil, %ecx
  imull   %eax, %ecx
  movb    %cl, %al
  sarb    $7, %al
  movzbl  %al, %eax
  movzbl  %ch, %esi
  cmpl    %eax, %esi
  setne   %al

After:
  movb    %dil, %al
  imulb   %sil
  seto    %al

Also, remove a made-redundant testcase for PR19858, and enable more FastISel
ALU-overflow tests for SelectionDAG too.

Differential Revision: http://reviews.llvm.org/D5809

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220516 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDebugInfo: Remove DwarfDebug::addCurrentFnArgument declaration now that it's moved...
David Blaikie [Thu, 23 Oct 2014 21:53:17 +0000 (21:53 +0000)]
DebugInfo: Remove DwarfDebug::addCurrentFnArgument declaration now that it's moved to DwarfFile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220515 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoHandle sqrt() shrinking in SimplifyLibCalls like any other call
Sanjay Patel [Thu, 23 Oct 2014 21:52:45 +0000 (21:52 +0000)]
Handle sqrt() shrinking in SimplifyLibCalls like any other call

This patch removes a chunk of special case logic for folding
(float)sqrt((double)x) -> sqrtf(x)
in InstCombineCasts and handles it in the mainstream path of SimplifyLibCalls.

No functional change intended, but I loosened the restriction on the existing
sqrt testcases to allow for this optimization even without unsafe-fp-math because
that's the existing behavior.

I also added a missing test case for not shrinking the llvm.sqrt.f64 intrinsic
in case the result is used as a double.

Differential Revision: http://reviews.llvm.org/D5919

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220514 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoChange the name of the field BindTable to bindtable to not over lap the type.
Kevin Enderby [Thu, 23 Oct 2014 19:53:12 +0000 (19:53 +0000)]
Change the name of the field BindTable to bindtable to not over lap the type.

Should fix the build bot issues from commit r220500.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220504 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake llvm-go test dependency optional.
Peter Collingbourne [Thu, 23 Oct 2014 19:51:40 +0000 (19:51 +0000)]
Make llvm-go test dependency optional.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220503 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake llvm-link behave a bit more like LTO.
Rafael Espindola [Thu, 23 Oct 2014 19:40:45 +0000 (19:40 +0000)]
Make llvm-link behave a bit more like LTO.

* Read modules lazily
* Don't treat the first file specially, instead merge all inputs into an empty
  module.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220501 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUpdate llvm-objdump’s Mach-O symbolizer code for Objective-C references.
Kevin Enderby [Thu, 23 Oct 2014 19:37:31 +0000 (19:37 +0000)]
Update llvm-objdump’s Mach-O symbolizer code for Objective-C references.

This prints disassembly comments for Objective-C references to CFStrings,
Selectors, Classes and method calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220500 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCleanup this test a bit.
Rafael Espindola [Thu, 23 Oct 2014 19:36:21 +0000 (19:36 +0000)]
Cleanup this test a bit.

Use simpler names and remove unnecessary fields.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220499 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCleanup this test a bit.
Rafael Espindola [Thu, 23 Oct 2014 19:23:42 +0000 (19:23 +0000)]
Cleanup this test a bit.

Use simpler names and remove unnecessary fields.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220498 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDebugInfo: Simplify/tidy/correct global variable decl/def emission handling.
David Blaikie [Thu, 23 Oct 2014 19:12:43 +0000 (19:12 +0000)]
DebugInfo: Simplify/tidy/correct global variable decl/def emission handling.

This fixes a bug (introduced by fixing the IR emitted from Clang where
the definition of a static member would be scoped within the class,
rather than within its lexical decl context) where the definition of a
static variable would be placed inside a class.

It also improves source fidelity by scoping static class member
definitions inside the lexical decl context in which tehy are written
(eg: namespace n { class foo { static int i; } int foo::i; } - the
definition of 'i' will be within the namespace 'n' in the DWARF output
now).

Lastly, and the original goal, this reduces debug info size slightly
(and makes debug info easier to read, etc) by placing the definitions of
non-member global variables within their namespace, rather than using a
separate namespace-scoped declaration along with a definition at global
scope.

Based on patches and discussion with Frédéric.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220497 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake this test a bit stricter.
Rafael Espindola [Thu, 23 Oct 2014 18:52:46 +0000 (18:52 +0000)]
Make this test a bit stricter.

This now:
* Forces the linker to include the internal definition.
* Checks the full output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220495 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake this test a bit stricter.
Rafael Espindola [Thu, 23 Oct 2014 18:44:07 +0000 (18:44 +0000)]
Make this test a bit stricter.

This now:
* Forces the linker to include the internal definition.
* Checks the full output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220494 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "Don't count inreg params when mangling fastcall functions"
Reid Kleckner [Thu, 23 Oct 2014 17:50:42 +0000 (17:50 +0000)]
Revert "Don't count inreg params when mangling fastcall functions"

This reverts commit r214981.

I'm not sure what I was thinking when I wrote this. Testing with MSVC
shows that this function is mangled to '@f@8':
  int __fastcall f(int a, int b);

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220492 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdding llvm-shlib to CMake build system with a few new bells and whistles
Chris Bieneman [Thu, 23 Oct 2014 17:22:14 +0000 (17:22 +0000)]
Adding llvm-shlib to CMake build system with a few new bells and whistles

Summary:
This patch adds a new CMake build setting LLVM_BUILD_LLVM_DYLIB, which defaults to OFF. When set to ON, this will generate a shared library containing most of LLVM. The contents of the shared library can be overriden by specifying LLVM_DYLIB_COMPONENTS. LLVM_DYLIB_COMPONENTS can be set to a semi-colon delimited list of any LLVM components that you llvm-config can resolve.

On Windows, unless you are using Cygwin, you must specify an explicit symbol export file using LLVM_EXPORTED_SYMBOL_FILE. On Cygwin and all unix-like platforms if you do not specify LLVM_EXPORTED_SYMBOL_FILE, an export file containing only the LLVM C API will be auto-generated from the list of LLVM components specified in LLVM_DYLIB_COMPONENTS.

Reviewers: rnk

Reviewed By: rnk

Subscribers: rnk, llvm-commits

Differential Revision: http://reviews.llvm.org/D5890

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220490 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove explicit (void) use of DwarfFile::DD that was accidentally left in r220452.
David Blaikie [Thu, 23 Oct 2014 16:12:58 +0000 (16:12 +0000)]
Remove explicit (void) use of DwarfFile::DD that was accidentally left in r220452.

Caught in post-commit review by Frédéric.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220487 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDo not emit intermediate register for zero FP immediate
Renato Golin [Thu, 23 Oct 2014 15:31:50 +0000 (15:31 +0000)]
Do not emit intermediate register for zero FP immediate

This updates check for double precision zero floating point constant to allow
use of instruction with immediate value rather than temporary register.
Currently "a == 0.0", where "a" is of "double" type generates:

vmov.i32        d16, #0x0
vcmpe.f64       d0, d16

With this change it becomes:

vcmpe.f64        d0, #0

Patch by Sergey Dmitrouk.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220486 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoclang-format two code snippets to make the next patch easy to read.
Rafael Espindola [Thu, 23 Oct 2014 15:20:05 +0000 (15:20 +0000)]
clang-format two code snippets to make the next patch easy to read.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220484 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd unittest for extreme alignments.
Rafael Espindola [Thu, 23 Oct 2014 14:45:19 +0000 (14:45 +0000)]
Add unittest for extreme alignments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220483 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoHexagon/Disassembler/LLVMBuild.txt: Update libdeps.
NAKAMURA Takumi [Thu, 23 Oct 2014 11:32:16 +0000 (11:32 +0000)]
Hexagon/Disassembler/LLVMBuild.txt: Update libdeps.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220482 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoHexagon/LLVMBuild.txt: Prune CRLF.
NAKAMURA Takumi [Thu, 23 Oct 2014 11:32:03 +0000 (11:32 +0000)]
Hexagon/LLVMBuild.txt: Prune CRLF.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220481 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Prune CRLF in CMakeLists.txt(s).
NAKAMURA Takumi [Thu, 23 Oct 2014 11:31:50 +0000 (11:31 +0000)]
[CMake] Prune CRLF in CMakeLists.txt(s).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220480 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Prune trailing whitespace.
NAKAMURA Takumi [Thu, 23 Oct 2014 11:31:33 +0000 (11:31 +0000)]
[CMake] Prune trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220479 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert r220427, "[Hexagon] Adding encoding bits for add opcode."
NAKAMURA Takumi [Thu, 23 Oct 2014 11:31:22 +0000 (11:31 +0000)]
Revert r220427, "[Hexagon] Adding encoding bits for add opcode."

It brought cyclic dependecy between HexagonAsmPrinter and HexagonDesc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220478 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[mips][microMIPS] Implement ADDIUR1SP instruction
Zoran Jovanovic [Thu, 23 Oct 2014 11:13:59 +0000 (11:13 +0000)]
[mips][microMIPS] Implement ADDIUR1SP instruction
Differential Revision: http://reviews.llvm.org/D5153

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220477 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agops][microMIPS] Implement ADDIUR2 instruction
Zoran Jovanovic [Thu, 23 Oct 2014 11:06:34 +0000 (11:06 +0000)]
ps][microMIPS] Implement ADDIUR2 instruction
Differential Revision: http://reviews.llvm.org/D5151

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220476 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agops][microMIPS] Implement LI16 instruction
Zoran Jovanovic [Thu, 23 Oct 2014 10:59:24 +0000 (10:59 +0000)]
ps][microMIPS] Implement LI16 instruction
Differential Revision: http://reviews.llvm.org/D5149

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220475 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[mips][microMIPS] Implement CodeGen support for SLL16 and SRL16 instructions
Zoran Jovanovic [Thu, 23 Oct 2014 10:42:01 +0000 (10:42 +0000)]
[mips][microMIPS] Implement CodeGen support for SLL16 and SRL16 instructions
Differential Revision: http://reviews.llvm.org/D5774

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220474 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Thumb2] Improve disassembly of memory hints
Oliver Stannard [Thu, 23 Oct 2014 08:52:58 +0000 (08:52 +0000)]
[Thumb2] Improve disassembly of memory hints

Currently, the ARM disassembler will disassemble the Thumb2 memory hint
instructions (PLD, PLDW and PLI), even for targets which do not have
these instructions. This patch adds the required checks to the
disassmebler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220472 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM, stack protector] If supported, use armv7 instructions.
Akira Hatanaka [Thu, 23 Oct 2014 04:17:05 +0000 (04:17 +0000)]
[ARM, stack protector] If supported, use armv7 instructions.

This commit enables using movt/movw to load the stack guard address:

movw r0, :lower16:(L_g3$non_lazy_ptr-(LPC0_0+8))
movt r0, :upper16:(L_g3$non_lazy_ptr-(LPC0_0+8))
ldr r0, [pc, r0]

Previously a pc-relative load was emitted:

ldr r0, LCPI0_0
ldr r0, [pc, r0]

rdar://problem/18740489

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220470 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAssert that ValueHandleBase::ValueIsRAUWd doesn't change the tracked Value type.
Frederic Riss [Thu, 23 Oct 2014 04:08:42 +0000 (04:08 +0000)]
Assert that ValueHandleBase::ValueIsRAUWd doesn't change the tracked Value type.

This invariant is enforced in Value::replaceAllUsesWith, thus it seems
logical to apply it also to ValueHandles. This commit fixes InstCombine
to not trigger the assertion during the removal of constant bitcasts in
call instructions.

Differential Revision: http://reviews.llvm.org/D5828

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220468 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoModernize doxygen comments in Support/Dwarf.h
Frederic Riss [Thu, 23 Oct 2014 04:08:38 +0000 (04:08 +0000)]
Modernize doxygen comments in Support/Dwarf.h

In post-commit review of r219442, Rafael pointed out that the comment style
of the newly introduced helper didn't follow LLVM's coding standard.
Modernize the whole file to the new standards.

Differential Revision: http://reviews.llvm.org/D5918

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220467 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[dwarfdump] Dump DW_AT_ranges values inline in the debug_info dump.
Frederic Riss [Thu, 23 Oct 2014 04:08:34 +0000 (04:08 +0000)]
[dwarfdump] Dump DW_AT_ranges values inline in the debug_info dump.

The output looks like that:
                      DW_AT_ranges [FORM_data4]    (0x00000000
                         [0x00000001000024a0 - 0x00000001000024c2)
                         [0x0000000100002505 - 0x000000010000268b))

Differential Revision: http://reviews.llvm.org/D5712

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220466 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd llvm-go tool.
Peter Collingbourne [Thu, 23 Oct 2014 02:33:23 +0000 (02:33 +0000)]
Add llvm-go tool.

This tool lets us build LLVM components within the tree by setting up a
$GOPATH that resembles a tree fetched in the normal way with "go get".

It is intended that components such as the Go frontend will be built in-tree
using this tool.

Differential Revision: http://reviews.llvm.org/D5902

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220462 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[msan] Emit checks for constant shadow values under an experimental flag.
Evgeniy Stepanov [Thu, 23 Oct 2014 01:05:46 +0000 (01:05 +0000)]
[msan] Emit checks for constant shadow values under an experimental flag.

Does not change the default behavior.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220457 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DebugInfo] Sink DwarfDebug::addCurrentFnArgument down into DwarfFile.
David Blaikie [Thu, 23 Oct 2014 00:16:05 +0000 (00:16 +0000)]
[DebugInfo] Sink DwarfDebug::addCurrentFnArgument down into DwarfFile.

Variable handling will be sunk into DwarfFile so that abstract variables
and the like can be shared across multiple CUs (to handle cross-CU
inlining, for example).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220453 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DebugInfo] Add DwarfDebug& to DwarfFile.
David Blaikie [Thu, 23 Oct 2014 00:16:03 +0000 (00:16 +0000)]
[DebugInfo] Add DwarfDebug& to DwarfFile.

Use the DwarfDebug in one function that previously took it as a
parameter, and lay the foundation for use this for other operations
coming soon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220452 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DebugInfo] Remove LexicalScopes::isCurrentFunctionScope and CSE a use of LexicalScop...
David Blaikie [Thu, 23 Oct 2014 00:06:27 +0000 (00:06 +0000)]
[DebugInfo] Remove LexicalScopes::isCurrentFunctionScope and CSE a use of LexicalScopes::getCurrentFunctionScope

Now that we're sure the only root (non-abstract) scope is the current
function scope, there's no need for isCurrentFunctionScope, the property
can be tested directly instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220451 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix Mips nacl-mask test for new bundle-aligned label behavior
Derek Schuff [Wed, 22 Oct 2014 23:32:00 +0000 (23:32 +0000)]
Fix Mips nacl-mask test for new bundle-aligned label behavior

After r220439 the behavior of labels in bundle-align mode changed,
and I neglected to update this test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220447 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[MCJIT] Make repeat calls to MCJIT::getPointerToFunction for declarations safe.
Lang Hames [Wed, 22 Oct 2014 23:18:42 +0000 (23:18 +0000)]
[MCJIT] Make repeat calls to MCJIT::getPointerToFunction for declarations safe.

MCJIT::getPointerForFunction adds the resulting address to the global mapping.
This should be done via updateGlobalMapping rather than addGlobalMapping, since
the latter asserts if a mapping already exists.

MCJIT::getPointerToFunction is actually deprecated - hopefully we can remove it
(or more likely re-task it) entirely soon. In the mean time it should at least
work as advertised.

<rdar://problem/18727946>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220444 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAttempt to fix the build after r220439
David Majnemer [Wed, 22 Oct 2014 22:46:05 +0000 (22:46 +0000)]
Attempt to fix the build after r220439

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220440 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[MC] Attach labels to existing fragments instead of using a separate fragment
Derek Schuff [Wed, 22 Oct 2014 22:38:06 +0000 (22:38 +0000)]
[MC] Attach labels to existing fragments instead of using a separate fragment

Summary:
Currently when emitting a label, a new data fragment is created for it if the
current fragment isn't a data fragment.
This change instead enqueues the label and attaches it to the next fragment
(e.g. created for the next instruction) if possible.

When bundle alignment is not enabled, this has no functionality change (it
just results in fewer extra fragments being created). For bundle alignment,
previously labels would point to the beginning of the bundle padding instead
of the beginning of the emitted instruction. This was not only less efficient
(e.g. jumping to the nops instead of past them) but also led to miscalculation
of the address of the GOT (since MC uses a label difference rather than
emitting a "." symbol).

Fixes https://code.google.com/p/nativeclient/issues/detail?id=3982

Test Plan: regression test attached

Reviewers: jvoung, eliben

Subscribers: jfb, llvm-commits

Differential Revision: http://reviews.llvm.org/D5915

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220439 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Adding encoding bits for add opcode.
Colin LeMahieu [Wed, 22 Oct 2014 20:58:35 +0000 (20:58 +0000)]
[Hexagon] Adding encoding bits for add opcode.
Adding llvm-mc tests.
Removing unit tests.

http://reviews.llvm.org/D5624

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220427 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64] Add support for the .inst directive.
Chad Rosier [Wed, 22 Oct 2014 20:35:57 +0000 (20:35 +0000)]
[AArch64] Add support for the .inst directive.

This has been implement using the MCTargetStreamer interface as is done in the
ARM, Mips and PPC backends.

Phabricator: http://reviews.llvm.org/D5891
PR20964

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220422 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoGo: add binding for LLVMSetUnnamedAddr.
Peter Collingbourne [Wed, 22 Oct 2014 20:20:27 +0000 (20:20 +0000)]
Go: add binding for LLVMSetUnnamedAddr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220416 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoStrength reduce constant-sized vectors into arrays. No functionality change.
Benjamin Kramer [Wed, 22 Oct 2014 19:55:26 +0000 (19:55 +0000)]
Strength reduce constant-sized vectors into arrays. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220412 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDo not add -gsplit-dwarf to LLVM_DEFINITIONS.
Peter Collingbourne [Wed, 22 Oct 2014 19:49:19 +0000 (19:49 +0000)]
Do not add -gsplit-dwarf to LLVM_DEFINITIONS.

This would cause the flag to appear in the output of "llvm-config --cppflags",
which should contain only preprocessor flags. The -gsplit-dwarf flag in
particular can cause problems with certain downstream users such as cgo.

Differential Revision: http://reviews.llvm.org/D5895

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220410 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoLoopVectorize: Simplify code. No functionality change.
Benjamin Kramer [Wed, 22 Oct 2014 19:13:54 +0000 (19:13 +0000)]
LoopVectorize: Simplify code. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220405 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoShorten auto iterators for function basic blocks.
Diego Novillo [Wed, 22 Oct 2014 18:39:50 +0000 (18:39 +0000)]
Shorten auto iterators for function basic blocks.

Use consistent naming for basic block instances.

No functional changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220404 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix number of operands in documentation for minnum / maxnum
Matt Arsenault [Wed, 22 Oct 2014 18:25:02 +0000 (18:25 +0000)]
Fix number of operands in documentation for minnum / maxnum

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220402 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agotest: Make this test runnable in directories with @ in their names
Justin Bogner [Wed, 22 Oct 2014 18:18:54 +0000 (18:18 +0000)]
test: Make this test runnable in directories with @ in their names

Jenkins likes to use directories with names involving the '@'
character, which breaks the sed expression in this test. Switch to use
'|' on the assumption that it's less likely to show up in a path.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220401 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix VS2012 build; C++11 type aliases are not supported.
Hans Wennborg [Wed, 22 Oct 2014 17:47:49 +0000 (17:47 +0000)]
Fix VS2012 build; C++11 type aliases are not supported.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220399 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAmmending 220393 - Removing unused decoding tables.
Colin LeMahieu [Wed, 22 Oct 2014 17:23:01 +0000 (17:23 +0000)]
Ammending 220393 - Removing unused decoding tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220397 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAmmending 220393 - Removing unused functions.
Colin LeMahieu [Wed, 22 Oct 2014 17:03:19 +0000 (17:03 +0000)]
Ammending 220393 - Removing unused functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220396 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PATCH] Support select-cc for VSFRC when VSX is enabled
Bill Schmidt [Wed, 22 Oct 2014 16:58:20 +0000 (16:58 +0000)]
[PATCH] Support select-cc for VSFRC when VSX is enabled

A previous patch enabled SELECT_VSRC and SELECT_CC_VSRC for VSX to
handle <2 x double> cases.  This patch adds SELECT_VSFRC and
SELECT_CC_VSFRC to allow use of all 64 vector-scalar registers for the
f64 type when VSX is enabled.  The changes are analogous to those in
the previous patch.  I've added a new variant to vsx.ll to test the
code generation.

(I also cleaned up a little formatting in PPCInstrVSX.td from the
previous patch.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220395 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse auto iteration in lib/Transforms/Scalar/SampleProfile.cpp. No functional changes.
Diego Novillo [Wed, 22 Oct 2014 16:51:50 +0000 (16:51 +0000)]
Use auto iteration in lib/Transforms/Scalar/SampleProfile.cpp. No functional changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220394 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Adding basic disassembler.
Colin LeMahieu [Wed, 22 Oct 2014 16:49:14 +0000 (16:49 +0000)]
[Hexagon] Adding basic disassembler.
Marking all instructions as CodeGenOnly since encoding bits are not set yet.
http://reviews.llvm.org/D5829?vs=on&id=15023&whitespace=ignore-all#toc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220393 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoPreserving 'nonnull' metadata in SimplifyCFG
Philip Reames [Wed, 22 Oct 2014 16:37:13 +0000 (16:37 +0000)]
Preserving 'nonnull' metadata in SimplifyCFG

When we hoist two loads above an if, we can preserve the nonnull metadata.  We could also do the same for sinking them, but we appear to not handle metadata at all in that case.

Thanks to Hal for the review.

Differential Revision: http://reviews.llvm.org/D5910

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220392 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoShrinkify libcalls: use float versions of double libm functions with fast-math (bug...
Sanjay Patel [Wed, 22 Oct 2014 15:29:23 +0000 (15:29 +0000)]
Shrinkify libcalls: use float versions of double libm functions with fast-math (bug 17850)

When a call to a double-precision libm function has fast-math semantics
(via function attribute for now because there is no IR-level FMF on calls),
we can avoid fpext/fptrunc operations and use the float version of the call
if the input and output are both float.

We already do this optimization using a command-line option; this patch just
adds the ability for fast-math to use the existing functionality.

I moved the cl::opt from InstructionCombining into SimplifyLibCalls because
it's only ever used internally to that class.

Modified the existing test cases to use the unsafe-fp-math attribute rather
than repeating all tests.

This patch should solve: http://llvm.org/bugs/show_bug.cgi?id=17850

Differential Revision: http://reviews.llvm.org/D5893

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220390 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake two helper functions static.
Rafael Espindola [Wed, 22 Oct 2014 15:05:51 +0000 (15:05 +0000)]
Make two helper functions static.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220389 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoChange error to warning when a profile cannot be found.
Diego Novillo [Wed, 22 Oct 2014 13:36:35 +0000 (13:36 +0000)]
Change error to warning when a profile cannot be found.

When the profile for a function cannot be applied, we use to emit an
error. This seems extreme. The compiler can continue, it's just that the
optimization opportunities won't include profile information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220386 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Support select-cc for VSX
Bill Schmidt [Wed, 22 Oct 2014 13:13:40 +0000 (13:13 +0000)]
[PowerPC] Support select-cc for VSX

The tests test/CodeGen/Generic/select-cc.ll and
test/CodeGen/PowerPC/select-cc.ll both fail with VSX enabled.  The
problem is that the lowering logic for the SELECT and SELECT_CC
operations doesn't currently support the VSX registers.  This patch
fixes that.

In lib/Target/PowerPC/PPCInstrInfo.td, we have pseudos to handle this
for other register classes.  Similar pseudos are added in
PPCInstrVSX.td (they must be there, because the "vsrc" register class
definition appears there) for the VSRC register class.  The
SELECT_VSRC pseudo is then used in pattern matching for SELECT_CC.

The rest of the patch just adds logic for SELECT_VSRC wherever similar
logic appears for SELECT_VRRC.

There are no new test cases because the existing tests above test
this, along with a variant in test/CodeGen/PowerPC/vsx.ll.

After discussion with Hal, a future patch will add similar _VSFRC
variants to override f64 type handling (currently using F8RC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220385 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFixing a -Wsign-compare warning; NFC.
Aaron Ballman [Wed, 22 Oct 2014 13:09:43 +0000 (13:09 +0000)]
Fixing a -Wsign-compare warning; NFC.

I think it might make sense to make COFF::MaxNumberOfSections16 be a uint32_t, however, that may have wider-reaching implications in other projects, which is why I did not change that declaration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220384 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoSupport using sample profiles with partial debug info.
Diego Novillo [Wed, 22 Oct 2014 12:59:00 +0000 (12:59 +0000)]
Support using sample profiles with partial debug info.

Summary:
When using a profile, we used to require the use -gmlt so that we could
get access to the line locations. This is used to match line numbers in
the input profile to the line numbers in the function's IR.

But this is actually not necessary. The driver can provide source
location tracking without the emission of debug information. In these
cases, the annotation 'llvm.dbg.cu' is missing from the IR, but the
actual line location annotations are still present.

This patch adds a new way of looking for the start of the current
function. Instead of looking through the compile units in llvm.dbg.cu,
we can walk up the scope for the first instruction in the function with
a debug loc. If that describes the function, we use it. Otherwise, we
keep looking until we find one.

If no such instruction is found, we then give up and produce an error.

Reviewers: echristo, dblaikie

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5887

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220382 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64] Cleanup A57PBQPConstraints
Arnaud A. de Grandmaison [Wed, 22 Oct 2014 12:40:20 +0000 (12:40 +0000)]
[AArch64] Cleanup A57PBQPConstraints

And add a long awaited testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220381 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstSimplify] Support constant folding to vector of pointers
Bruno Cardoso Lopes [Wed, 22 Oct 2014 12:18:48 +0000 (12:18 +0000)]
[InstSimplify] Support constant folding to vector of pointers

ConstantFolding crashes when trying to InstSimplify the following load:

@a = private unnamed_addr constant %mst {
     i8* inttoptr (i64 -1 to i8*),
     i8* inttoptr (i64 -1 to i8*)
}, align 8

%x = load <2 x i8*>* bitcast (%mst* @a to <2 x i8*>*), align 8

This patch fix this by adding support to this type of folding:

%x = load <2 x i8*>* bitcast (%mst* @a to <2 x i8*>*), align 8
==> gets folded to:
  %x = <2 x i8*> <i8* inttoptr (i64 -1 to i8*), i8* inttoptr (i64 -1 to i8*)>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220380 91177308-0d34-0410-b5e6-96231b3b80d8