platform/upstream/llvm.git
10 years ago[X86] Improve comments for r214888
Adam Nemet [Tue, 5 Aug 2014 17:58:49 +0000 (17:58 +0000)]
[X86] Improve comments for r214888

A rebase somehow ate my comments. This restores them.

llvm-svn: 214903

10 years agoR600/SI: Use register class instead of list of registers
Matt Arsenault [Tue, 5 Aug 2014 17:52:40 +0000 (17:52 +0000)]
R600/SI: Use register class instead of list of registers

I'm not sure if this has any consequence or not.

llvm-svn: 214902

10 years agoR600/SI: Add exec_lo and exec_hi subregisters.
Matt Arsenault [Tue, 5 Aug 2014 17:52:37 +0000 (17:52 +0000)]
R600/SI: Add exec_lo and exec_hi subregisters.

This allows accessing an SReg subregister with a normal subregister
index, instead of getting a machine verifier error.

Also be sure to include all of these subregisters in SReg_32.
This fixes inferring SGPR instead of SReg when finding a
super register class.

llvm-svn: 214901

10 years agoOn FreeBSD skip test that produces bogus output
Ed Maste [Tue, 5 Aug 2014 17:50:04 +0000 (17:50 +0000)]
On FreeBSD skip test that produces bogus output

The test produces lines that start with "<word>: " which confuses the
buildbot log parser.  Disable the test until either the test is fixed
or the buildbot can deal with the undesired output.

llvm.org/pr20545

llvm-svn: 214900

10 years agoBitcodeReader: Fix non-determinism in use-list order
Duncan P. N. Exon Smith [Tue, 5 Aug 2014 17:49:48 +0000 (17:49 +0000)]
BitcodeReader: Fix non-determinism in use-list order

`BasicBlockFwdRefs` (and `BlockAddrFwdRefs` before it) was being emptied
in a non-deterministic order.  When predicting use-list order I've
worked around this another way, but even when parsing lazily (and we
can't recreate use-list order) use-lists should be deterministic.

Make them so by using a side-queue of functions with forward-referenced
blocks that gets visited in order.

llvm-svn: 214899

10 years agoRemove dead zero store to calloc initialized memory
Philip Reames [Tue, 5 Aug 2014 17:48:20 +0000 (17:48 +0000)]
Remove dead zero store to calloc initialized memory

Optimize the following IR:

%1 = tail call noalias i8* @calloc(i64 1, i64 4)
%2 = bitcast i8* %1 to i32*
; This store is dead and should be removed
store i32 0, i32* %2, align 4

Memory returned by calloc is guaranteed to be zero initialized. If the value being stored is the constant zero (and the store is not otherwise observable across threads), we can delete the store.  If the store is to an out of bounds address, it is undefined and thus also removable.

Reviewed By: nicholas

Differential Revision: http://reviews.llvm.org/D3942

llvm-svn: 214897

10 years agoRevert r214881 because it broke lots of build-bots
Jonathan Roelofs [Tue, 5 Aug 2014 17:36:05 +0000 (17:36 +0000)]
Revert r214881 because it broke lots of build-bots

llvm-svn: 214893

10 years agoOptimize vector fabs of bitcasted constant integer values.
Sanjay Patel [Tue, 5 Aug 2014 17:35:22 +0000 (17:35 +0000)]
Optimize vector fabs of bitcasted constant integer values.

Allow vector fabs operations on bitcasted constant integer values to be optimized
in the same way that we already optimize scalar fabs.

So for code like this:
%bitcast = bitcast i64 18446744069414584320 to <2 x float> ; 0xFFFF_FFFF_0000_0000
%fabs = call <2 x float> @llvm.fabs.v2f32(<2 x float> %bitcast)
%ret = bitcast <2 x float> %fabs to i64

Instead of generating something like this:

movabsq (constant pool loadi of mask for sign bits)
vmovq   (move from integer register to vector/fp register)
vandps  (mask off sign bits)
vmovq   (move vector/fp register back to integer return register)

We should generate:

mov     (put constant value in return register)

I have also removed a redundant clause in the first 'if' statement:
N0.getOperand(0).getValueType().isInteger()

is the same thing as:
IntVT.isInteger()

Testcases for x86 and ARM added to existing files that deal with vector fabs.
One existing testcase for x86 removed because it is no longer ideal.

For more background, please see:
http://reviews.llvm.org/D4770

And:
http://llvm.org/bugs/show_bug.cgi?id=20354

Differential Revision: http://reviews.llvm.org/D4785

llvm-svn: 214892

10 years ago[AVX512] Add intrinsic for valignd/q
Adam Nemet [Tue, 5 Aug 2014 17:28:23 +0000 (17:28 +0000)]
[AVX512] Add intrinsic for valignd/q

Note that similar to palingr, we could further optimize these to emit
shufflevector when the shift count is <=64.  This however does not
change the overall design that unlike palignr we would still need the LLVM
intrinsic corresponding to this intruction to handle the >64 cases.  (palignr
uses the psrldq intrinsic in this case.)

llvm-svn: 214891

10 years ago[AVX512] Add masking variant and intrinsics for valignd/q
Adam Nemet [Tue, 5 Aug 2014 17:23:04 +0000 (17:23 +0000)]
[AVX512] Add masking variant and intrinsics for valignd/q

This is similar to what I did with the two-source permutation recently.  (It's
almost too similar so that we should consider generating the masking variants
with some tablegen help.)

Both encoding and intrinsic tests are added as well.  For the latter, this is
what the IR that the intrinsic test on the clang side generates.

Part of <rdar://problem/17688758>

llvm-svn: 214890

10 years ago[X86] Increase X86_MAX_OPERANDS from 5 to 6
Adam Nemet [Tue, 5 Aug 2014 17:23:01 +0000 (17:23 +0000)]
[X86] Increase X86_MAX_OPERANDS from 5 to 6

This controls the number of operands in the disassembler's x86OperandSets
table.  The entries describe how the operand is encoded and its type.

Not to surprisingly 5 operands is insufficient for AVX512.  Consider
VALIGNDrrik in the next patch.  These are its operand specifiers:

  { /* 328 */
    { ENCODING_DUP, TYPE_DUP1 },
    { ENCODING_REG, TYPE_XMM512 },
    { ENCODING_WRITEMASK, TYPE_VK8 },
    { ENCODING_VVVV, TYPE_XMM512 },
    { ENCODING_RM_CD64, TYPE_XMM512 },
    { ENCODING_IB, TYPE_IMM8 },
  },

llvm-svn: 214889

10 years ago[X86] Add lowering to VALIGN
Adam Nemet [Tue, 5 Aug 2014 17:22:59 +0000 (17:22 +0000)]
[X86] Add lowering to VALIGN

This was currently part of lowering to PALIGNR with some special-casing to
make interlane shifting work.  Since AVX512F has interlane alignr (valignd/q)
and AVX512BW has vpalignr we need to support both of these *at the same time*,
e.g. for SKX.

This patch breaks out the common code and then add support to check both of
these lowering options from LowerVECTOR_SHUFFLE.

I also added some FIXMEs where I think the AVX512BW and AVX512VL additions
should probably go.

llvm-svn: 214888

10 years ago[X86] Separate DAG node for valign and palignr
Adam Nemet [Tue, 5 Aug 2014 17:22:55 +0000 (17:22 +0000)]
[X86] Separate DAG node for valign and palignr

They have different semantics (valign is interlane while palingr is intralane)
and palingr is still needed even in the AVX512 context.  According to the
latest spec AVX512BW provides these.

llvm-svn: 214887

10 years ago[AVX512] alignr: Use suffix rather than name argument to multiclass
Adam Nemet [Tue, 5 Aug 2014 17:22:52 +0000 (17:22 +0000)]
[AVX512] alignr: Use suffix rather than name argument to multiclass

Again no functional change.  This prepares for the suffix to be used with the
intrinsic matching.

llvm-svn: 214886

10 years ago[AVX512] Pull everything alignr-related into the multiclass
Adam Nemet [Tue, 5 Aug 2014 17:22:50 +0000 (17:22 +0000)]
[AVX512] Pull everything alignr-related into the multiclass

The packed integer pattern becomes the DAG pattern for rri and the packed
float, another Pat<> inside the multiclass.

No functional change.

llvm-svn: 214885

10 years agoWrap long lines
Adam Nemet [Tue, 5 Aug 2014 17:22:47 +0000 (17:22 +0000)]
Wrap long lines

llvm-svn: 214884

10 years agoReleaseNotes: try to fix links
Hans Wennborg [Tue, 5 Aug 2014 17:19:14 +0000 (17:19 +0000)]
ReleaseNotes: try to fix links

llvm-svn: 214883

10 years agoReleaseNotes: mention basic debug info and ASan support in the Windows blurb
Hans Wennborg [Tue, 5 Aug 2014 17:15:00 +0000 (17:15 +0000)]
ReleaseNotes: mention basic debug info and ASan support in the Windows blurb

llvm-svn: 214882

10 years agoFix return sequence on armv4 thumb
Jonathan Roelofs [Tue, 5 Aug 2014 17:13:17 +0000 (17:13 +0000)]
Fix return sequence on armv4 thumb

POP on armv4t cannot be used to change thumb state (unilke later non-m-class
architectures), therefore we need a different return sequence that uses 'bx'
instead:

  POP {r3}
  ADD sp, #offset
  BX r3

This patch also fixes an issue where the return value in r3 would get clobbered
for functions that return 128 bits of data. In that case, we generate this
sequence instead:

  MOV ip, r3
  POP {r3}
  ADD sp, #offset
  MOV lr, r3
  MOV r3, ip
  BX lr

http://reviews.llvm.org/D4748

llvm-svn: 214881

10 years agoPartially revert r214761 that asserted that all concrete debug info variables had...
David Blaikie [Tue, 5 Aug 2014 16:47:23 +0000 (16:47 +0000)]
Partially revert r214761 that asserted that all concrete debug info variables had DIEs, due to a failure on Darwin.

I'll work on a reduction and fix after this.

llvm-svn: 214880

10 years agoImprove test for merged global debug info by using llvm-dwarfdump.
David Blaikie [Tue, 5 Aug 2014 16:20:25 +0000 (16:20 +0000)]
Improve test for merged global debug info by using llvm-dwarfdump.

It's a bit of a tradeoff, since llvm-dwarfdump doesn't print the name of
the global symbol being used as an address in the addressing mode, but
this avoids the dependence on hardcoded set labels that keep changing
(5+ commits over the last few years that each update the set label as it
changes due to other, unrelated differences in output). This could've,
instead, been changed to match the set name then match the name in the
string pool but that would present other issues (needing to skip over
the sets that weren't of interest, etc) and checking that the addresses
(granted, without relocations applied - so it's not the whole story)
match in the two variable location descriptions seems sufficient and
fairly stable here.

There are a few similar other tests with similar label dependence that
I'll update soonish.

llvm-svn: 214878

10 years agoUpdate CREDITS.txt with the list of folks who worked on ARM EHABI support
Jonathan Roelofs [Tue, 5 Aug 2014 15:56:07 +0000 (15:56 +0000)]
Update CREDITS.txt with the list of folks who worked on ARM EHABI support

llvm-svn: 214877

10 years agoASTMatchersTests/matchesConditionallyWithCuda: Add -fno-ms-extensions, and get rid...
NAKAMURA Takumi [Tue, 5 Aug 2014 15:54:43 +0000 (15:54 +0000)]
ASTMatchersTests/matchesConditionallyWithCuda: Add -fno-ms-extensions, and get rid of initializer list.

I am not sure whether -xcuda might imply -fno-ms-extensions.

llvm-svn: 214876

10 years agoAdd accessors for the PPC 403 bank registers.
Joerg Sonnenberger [Tue, 5 Aug 2014 15:45:15 +0000 (15:45 +0000)]
Add accessors for the PPC 403 bank registers.

llvm-svn: 214875

10 years ago[clang-tidy] Simplify ast matcher.
Benjamin Kramer [Tue, 5 Aug 2014 15:33:46 +0000 (15:33 +0000)]
[clang-tidy] Simplify ast matcher.

Turns out there is a better way to do this. No functionality change.

llvm-svn: 214874

10 years agoFix a copy-pasta in r214573
Jonathan Roelofs [Tue, 5 Aug 2014 15:31:16 +0000 (15:31 +0000)]
Fix a copy-pasta in r214573

llvm-svn: 214873

10 years agoAdd tests for cp10/cp11 on ARMv5/6
Renato Golin [Tue, 5 Aug 2014 15:29:41 +0000 (15:29 +0000)]
Add tests for cp10/cp11 on ARMv5/6

Tests for ARMv7/8 are already on diagnostics.s

llvm-svn: 214872

10 years agoSpecify that the thumb setend and blx <immed> instructions are not valid on an m...
Keith Walker [Tue, 5 Aug 2014 15:11:59 +0000 (15:11 +0000)]
Specify that the thumb setend and blx <immed> instructions are not valid on an m-class target

llvm-svn: 214871

10 years agoASTMatchers.h: Fix an annotation. [-Wdocumentation-html]
NAKAMURA Takumi [Tue, 5 Aug 2014 15:01:31 +0000 (15:01 +0000)]
ASTMatchers.h: Fix an annotation. [-Wdocumentation-html]

Angle brackets, aka &lt; &gt;, should be escaped.

llvm-svn: 214870

10 years ago[CMake] clangCodeGen: Update libdeps to add clangLex for PPChainedCallbacks.
NAKAMURA Takumi [Tue, 5 Aug 2014 15:01:12 +0000 (15:01 +0000)]
[CMake] clangCodeGen: Update libdeps to add clangLex for PPChainedCallbacks.

llvm-svn: 214869

10 years agoDefine stc2/stc2l/ldc2/ldc2l as thumb2 instructions
Keith Walker [Tue, 5 Aug 2014 14:58:05 +0000 (14:58 +0000)]
Define stc2/stc2l/ldc2/ldc2l as thumb2 instructions

llvm-svn: 214868

10 years agoAccessors for SSR2 and SSR3 on PPC 403.
Joerg Sonnenberger [Tue, 5 Aug 2014 14:53:05 +0000 (14:53 +0000)]
Accessors for SSR2 and SSR3 on PPC 403.

llvm-svn: 214867

10 years agoR600/SI: Update MUBUF assembly string to match AMD proprietary compiler
Tom Stellard [Tue, 5 Aug 2014 14:48:12 +0000 (14:48 +0000)]
R600/SI: Update MUBUF assembly string to match AMD proprietary compiler

llvm-svn: 214866

10 years agoR600/SI: Avoid generating REGISTER_LOAD instructions.
Tom Stellard [Tue, 5 Aug 2014 14:40:52 +0000 (14:40 +0000)]
R600/SI: Avoid generating REGISTER_LOAD instructions.

SI doesn't use REGISTER_LOAD anymore, but it was still hitting this code
path for 8-bit and 16-bit private loads.

llvm-svn: 214865

10 years agoAdd dci/ici instructions for PPC 476 and friends.
Joerg Sonnenberger [Tue, 5 Aug 2014 14:40:32 +0000 (14:40 +0000)]
Add dci/ici instructions for PPC 476 and friends.

llvm-svn: 214864

10 years agoAdd mftblo and mftbhi for PPC 4xx.
Joerg Sonnenberger [Tue, 5 Aug 2014 14:18:16 +0000 (14:18 +0000)]
Add mftblo and mftbhi for PPC 4xx.

llvm-svn: 214863

10 years agoAdd lswi / stswi for assembler use with a warning to not add patterns
Joerg Sonnenberger [Tue, 5 Aug 2014 13:34:01 +0000 (13:34 +0000)]
Add lswi / stswi for assembler use with a warning to not add patterns
for them.

llvm-svn: 214862

10 years ago[ASan/Win] Handle SEH exceptions even with -GS
Timur Iskhodzhanov [Tue, 5 Aug 2014 13:26:26 +0000 (13:26 +0000)]
[ASan/Win] Handle SEH exceptions even with -GS

This is a follow-up to r213654, r213656, r213667 and r213668.

llvm-svn: 214861

10 years agoAArch64: Add support for instruction prefetch intrinsic
Yi Kong [Tue, 5 Aug 2014 12:46:47 +0000 (12:46 +0000)]
AArch64: Add support for instruction prefetch intrinsic

Instruction prefetch is not implemented for AArch64, it is incorrectly
translated into data prefetch instruction.

Differential Revision: http://reviews.llvm.org/D4777

llvm-svn: 214860

10 years agoTeach the SLP Vectorizer that keeping some values live over a callsite can have a...
James Molloy [Tue, 5 Aug 2014 12:30:34 +0000 (12:30 +0000)]
Teach the SLP Vectorizer that keeping some values live over a callsite can have a cost.

Some types, such as 128-bit vector types on AArch64, don't have any callee-saved registers. So if a value needs to stay live over a callsite, it must be spilled and refilled. This cost is now taken into account.

llvm-svn: 214859

10 years agoclang-format: Add option to always break after a function's return type.
Daniel Jasper [Tue, 5 Aug 2014 12:16:31 +0000 (12:16 +0000)]
clang-format: Add option to always break after a function's return type.

This is required for GNU coding style, among others.

Also update the configuration documentation.

Modified from an original patch by Jarkko Hietaniemi, thank you!

llvm-svn: 214858

10 years agoclang-format: Break before 'else' in Stroustrup style.
Daniel Jasper [Tue, 5 Aug 2014 12:06:20 +0000 (12:06 +0000)]
clang-format: Break before 'else' in Stroustrup style.

Seems to be the desired thing to do according to:
  http://www.stroustrup.com/Programming/PPP-style-rev3.pdf

Patch by Jarkko Hietaniemi, thank you!

llvm-svn: 214857

10 years agoUpdating the documentation comment to list another parameter. No functional changes.
Aaron Ballman [Tue, 5 Aug 2014 11:48:21 +0000 (11:48 +0000)]
Updating the documentation comment to list another parameter. No functional changes.

llvm-svn: 214856

10 years agoThe test doesn't export ASAN_OPTIONS, so $ASAN_OPTIONS should not be used in RUN...
Alexander Kornienko [Tue, 5 Aug 2014 11:12:23 +0000 (11:12 +0000)]
The test doesn't export ASAN_OPTIONS, so $ASAN_OPTIONS should not be used in RUN lines.

llvm-svn: 214855

10 years agoMachO: set sizeof result type to what embedded headers expect.
Tim Northover [Tue, 5 Aug 2014 11:07:26 +0000 (11:07 +0000)]
MachO: set sizeof result type to what embedded headers expect.

Embedded systems seem to have inherited Darwin's choise of "unsigned long" for
size_t (via a bunch of headers), so we should respect that.

rdar://problem/17872787

llvm-svn: 214854

10 years ago[x86] Reformat some code I moved around in a prior commit but left
Chandler Carruth [Tue, 5 Aug 2014 10:35:30 +0000 (10:35 +0000)]
[x86] Reformat some code I moved around in a prior commit but left
poorly formatted. Sorry about that.

llvm-svn: 214853

10 years agoAdds AST matchers for matching CUDA declarations.
Manuel Klimek [Tue, 5 Aug 2014 09:45:53 +0000 (09:45 +0000)]
Adds AST matchers for matching CUDA declarations.

Patch by Jacques Pienaar.

llvm-svn: 214852

10 years agoAllow binary and for tblgen math.
Joerg Sonnenberger [Tue, 5 Aug 2014 09:43:25 +0000 (09:43 +0000)]
Allow binary and for tblgen math.

llvm-svn: 214851

10 years agoAfter three iterations of community review, we believe that this new
Jim Cownie [Tue, 5 Aug 2014 09:32:28 +0000 (09:32 +0000)]
After three iterations of community review, we believe that this new
CMAKE buld system should meet everyone's requirements.

Enhanced CMake Build System Commit

* Supports Linux, Mac, Windows, and Intel® Xeon Phi builds
* Supports building with gcc, icc, clang, and Visual Studio compilers
* Supports bulding "fat" libraries on OS/X with clang
* Details and documentation on how to use build system
  are in Build_With_CMake.txt
* To use the old CMake build system (corresponds to
  CMakeLists.txt.old), just rename CMakeLists.txt to
  CMakeLists.txt.other and rename CMakeLists.txt.old to
  CMakeLists.txt

llvm-svn: 214850

10 years ago[x86] Fix a crash and wrong-code bug in the new vector lowering all
Chandler Carruth [Tue, 5 Aug 2014 08:19:21 +0000 (08:19 +0000)]
[x86] Fix a crash and wrong-code bug in the new vector lowering all
found by a single test reduced out of a failure on llvm-stress.

The start of the problem (and the crash) came when we tried to use
a find of a non-used slot in the move-to half of the move-mask as the
target for two bad-half inputs. While if lucky this will be the first of
a pair of slots which we can place the bad-half inputs into, it isn't
actually guaranteed. This really isn't surprising, not sure what I was
thinking. The correct way to find the two unused slots is to look for
one of the *used* slots. We know it isn't that pair, and we can use some
modular arithmetic to find the other pair by masking off the odd bit and
adding 2 modulo 4. With this, we reliably found a viable pair of slots
for the bad-half inputs.

Sadly, that wasn't enough. We also had a wrong code bug that surfaced
when I reduced the test case for this where we would use the same slot
twice for the two bad inputs. This is because both of the bad inputs
could be in odd slots originally and thus the mod-2 mapping would
actually be the same. The whole point of the weird indexing into the
pair of empty slots was to try to leverage when the end result needed
the two bad-half inputs to be paired in a dword and pre-pair them in the
correct orrientation. This is less important with the powerful combining
we're now doing, and also easier and more reliable to achieve be noting
that we add the bad-half inputs in order. Thus, if they are in a dword
pair, the low part of that will be the first input in the sequence.
Always putting that in the low element will just do the right thing in
addition to computing the correct result.

Test case added. =]

llvm-svn: 214849

10 years ago[FastIsel][AArch64] Fix previous commit r214844 (Don't perform sign-/zero-extension...
Juergen Ributzka [Tue, 5 Aug 2014 07:31:30 +0000 (07:31 +0000)]
[FastIsel][AArch64] Fix previous commit r214844 (Don't perform sign-/zero-extension for function arguments that have already been sign-/zero-extended.)

The original code would fail for unsupported value types like i1, i8, and i16.
This fix changes the code to only create a sub-register copy for i64 value types
and all other types (i1/i8/i16/i32) just use the source register without any
modifications.

getRegClassFor() is now guarded by the i64 value type check, that guarantees
that we always request a register for a valid value type.

llvm-svn: 214848

10 years agoMS ABI: Make the alias template mangling more correct
David Majnemer [Tue, 5 Aug 2014 06:42:40 +0000 (06:42 +0000)]
MS ABI: Make the alias template mangling more correct

llvm-svn: 214847

10 years ago[FastISel][AArch64] Implement the FastLowerArguments hook.
Juergen Ributzka [Tue, 5 Aug 2014 05:43:48 +0000 (05:43 +0000)]
[FastISel][AArch64] Implement the FastLowerArguments hook.

This implements basic argument lowering for AArch64 in FastISel. It only
handles a small subset of the C calling convention. It supports simple
arguments that can be passed in GPR and FPR registers.

This should cover most of the trivial cases without falling back to
SelectionDAG.

This fixes <rdar://problem/17890986>.

llvm-svn: 214846

10 years agoRevert "r214832 - MachineCombiner Pass for selecting faster instruction"
Kevin Qin [Tue, 5 Aug 2014 05:43:47 +0000 (05:43 +0000)]
Revert "r214832 - MachineCombiner Pass for selecting faster instruction"

It broke compiling of most Benchmark and internal test, as clang got
clashed by segmentation fault or assertion.

llvm-svn: 214845

10 years ago[FastISel][AArch64] Don't perform sign-/zero-extension for function arguments that...
Juergen Ributzka [Tue, 5 Aug 2014 05:43:44 +0000 (05:43 +0000)]
[FastISel][AArch64] Don't perform sign-/zero-extension for function arguments that have already been sign-/zero-extended.

llvm-svn: 214844

10 years agoProvide convenient access to the zext/sext attributes of function arguments. NFC.
Juergen Ributzka [Tue, 5 Aug 2014 05:43:41 +0000 (05:43 +0000)]
Provide convenient access to the zext/sext attributes of function arguments. NFC.

llvm-svn: 214843

10 years agoHave MachineFunction cache a pointer to the subtarget to make lookups
Eric Christopher [Tue, 5 Aug 2014 02:39:49 +0000 (02:39 +0000)]
Have MachineFunction cache a pointer to the subtarget to make lookups
shorter/easier and have the DAG use that to do the same lookup. This
can be used in the future for TargetMachine based caching lookups from
the MachineFunction easily.

Update the MIPS subtarget switching machinery to update this pointer
at the same time it runs.

llvm-svn: 214838

10 years agoIf you found a step through plan stop looking up the stack for a step out plan.
Jim Ingham [Tue, 5 Aug 2014 01:59:20 +0000 (01:59 +0000)]
If you found a step through plan stop looking up the stack for a step out plan.

llvm-svn: 214837

10 years agoAdd some useful logging to the step log.
Jim Ingham [Tue, 5 Aug 2014 01:58:14 +0000 (01:58 +0000)]
Add some useful logging to the step log.

llvm-svn: 214836

10 years ago[PECOFF] addDeadStripRoot is not thread-safe.
Rui Ueyama [Tue, 5 Aug 2014 01:44:43 +0000 (01:44 +0000)]
[PECOFF] addDeadStripRoot is not thread-safe.

llvm-svn: 214835

10 years agodit pointed out on IRC that '__i = _VSTD::next(__i)' was a very long-winded way of...
Marshall Clow [Tue, 5 Aug 2014 01:34:12 +0000 (01:34 +0000)]
dit pointed out on IRC that '__i = _VSTD::next(__i)' was a very long-winded way of writing '++__i'. Since I hate being thought of as long-winded (this checkin comment notwithstanding), I fixed it. No functionality change.

llvm-svn: 214834

10 years ago[UBSan] Allow to suppress reports from vptr checker for specified types.
Alexey Samsonov [Tue, 5 Aug 2014 01:24:22 +0000 (01:24 +0000)]
[UBSan] Allow to suppress reports from vptr checker for specified types.

Based on http://reviews.llvm.org/D4702 by Byoungyoung Lee!

llvm-svn: 214833

10 years agoMachineCombiner Pass for selecting faster instruction
Gerolf Hoflehner [Tue, 5 Aug 2014 01:16:13 +0000 (01:16 +0000)]
MachineCombiner Pass for selecting faster instruction
 sequence on AArch64

Re-commit of r214669 without changes to test cases
LLVM::CodeGen/AArch64/arm64-neon-mul-div.ll and
LLVM:: CodeGen/AArch64/dp-3source.ll
This resolves the reported compfails of the original commit.

llvm-svn: 214832

10 years ago[Sanitizer] Turn SuppressionContext::Init() into InitIfNecessary().
Alexey Samsonov [Tue, 5 Aug 2014 00:43:23 +0000 (00:43 +0000)]
[Sanitizer] Turn SuppressionContext::Init() into InitIfNecessary().

Suppression context might be used in multiple sanitizers working
simultaneously (e.g. LSan and UBSan) and not knowing about each other.

llvm-svn: 214831

10 years agoReleaseNotes: add blurb about Windows support
Hans Wennborg [Tue, 5 Aug 2014 00:21:23 +0000 (00:21 +0000)]
ReleaseNotes: add blurb about Windows support

llvm-svn: 214830

10 years agoCodeGenObjCXX: Add a test for r214699
David Majnemer [Tue, 5 Aug 2014 00:01:15 +0000 (00:01 +0000)]
CodeGenObjCXX: Add a test for r214699

I forgot to add this with the changes for r214699.

llvm-svn: 214829

10 years agoMS ABI: Aligned tentative definitions don't have CommonLinkage
David Majnemer [Tue, 5 Aug 2014 00:01:13 +0000 (00:01 +0000)]
MS ABI: Aligned tentative definitions don't have CommonLinkage

int __declspec(align(16)) foo; is a tentative definition but the storage
for that variable should not have CommonLinkage.

llvm-svn: 214828

10 years agoAdd TCR register access
Joerg Sonnenberger [Mon, 4 Aug 2014 23:53:42 +0000 (23:53 +0000)]
Add TCR register access

llvm-svn: 214826

10 years agoAdd PPC 603's tlbld and tlbli instructions.
Joerg Sonnenberger [Mon, 4 Aug 2014 23:49:45 +0000 (23:49 +0000)]
Add PPC 603's tlbld and tlbli instructions.

llvm-svn: 214825

10 years ago[PECOFF] Fix /include option in .drectve section.
Rui Ueyama [Mon, 4 Aug 2014 23:48:57 +0000 (23:48 +0000)]
[PECOFF] Fix /include option in .drectve section.

/INCLUDE arguments passed as command line options are handled in the
same way as Unix -u. All option values are converted to an undefined
symbol and added to a dummy input file, so that the specified symbols
are resolved.

One tricky thing on Windows is that the option is also allowed to
appear in the object file's directive section. At the time when
it's being read, all (regular) command line options have already
been processed. We cannot add undefined atoms to the dummy file
anymore.

Previously, we added such /INCLUDE to a set that has already been
processed. As a result the options were ignored.

This patch fixes the issue. Now, /INCLUDE symbols in the directive
section are handled as real undefined symbol in the COFF file.
We create an undefined symbol for each /INCLUDE argument and add
it to the file being parsed.

llvm-svn: 214824

10 years agoTeach ProcessWindows plugin to support stdio i/o redirection.
Zachary Turner [Mon, 4 Aug 2014 23:31:21 +0000 (23:31 +0000)]
Teach ProcessWindows plugin to support stdio i/o redirection.

llvm-svn: 214816

10 years agoAllow CP10/CP11 operations on ARMv5/v6
Renato Golin [Mon, 4 Aug 2014 23:21:56 +0000 (23:21 +0000)]
Allow CP10/CP11 operations on ARMv5/v6

Those registers are VFP/NEON and vector instructions should be used instead,
but old cores rely on those co-processors to enable VFP unwinding. This change
was prompted by the libc++abi's unwinding routine and is also present in many
legacy low-level bare-metal code that we ought to compile/assemble.

Fixing bug PR20025 and allowing PR20529 to proceed with a fix in libc++abi.

llvm-svn: 214802

10 years ago[PPC64LE] Fix wrong IR for vec_sld and vec_vsldoi
Bill Schmidt [Mon, 4 Aug 2014 23:21:26 +0000 (23:21 +0000)]
[PPC64LE] Fix wrong IR for vec_sld and vec_vsldoi

My original LE implementation of the vsldoi instruction, with its
altivec.h interfaces vec_sld and vec_vsldoi, produces incorrect
shufflevector operations in the LLVM IR.  Correct code is generated
because the back end handles the incorrect shufflevector in a
consistent manner.

This patch and a companion patch for LLVM correct this problem by
removing the fixup from altivec.h and the corresponding fixup from the
PowerPC back end.  Several test cases are also modified to reflect the
now-correct LLVM IR.

The vec_sums and vec_vsumsws interfaces in altivec.h are also fixed,
because they used vec_perm calls intended to be recognized as vsldoi
instructions.  These vec_perm calls are now replaced with code that
more clearly shows the intent of the transformation.

llvm-svn: 214801

10 years ago[PPC64LE] Fix wrong IR for vec_sld and vec_vsldoi
Bill Schmidt [Mon, 4 Aug 2014 23:21:01 +0000 (23:21 +0000)]
[PPC64LE] Fix wrong IR for vec_sld and vec_vsldoi

My original LE implementation of the vsldoi instruction, with its
altivec.h interfaces vec_sld and vec_vsldoi, produces incorrect
shufflevector operations in the LLVM IR.  Correct code is generated
because the back end handles the incorrect shufflevector in a
consistent manner.

This patch and a companion patch for Clang correct this problem by
removing the fixup from altivec.h and the corresponding fixup from the
PowerPC back end.  Several test cases are also modified to reflect the
now-correct LLVM IR.

llvm-svn: 214800

10 years agoEnable Darwin vararg parameters support in assembler macros.
Kevin Enderby [Mon, 4 Aug 2014 23:14:37 +0000 (23:14 +0000)]
Enable Darwin vararg parameters support in assembler macros.

Duplicate the vararg tests for linux and add a tests which mixed
vararg arguments with darwin positional parameters.

Patch by: Janne Grunau <j@jannau.net>

llvm-svn: 214799

10 years agoChanged the liveness tracking in the RegisterScavenger
Pedro Artigas [Mon, 4 Aug 2014 23:07:49 +0000 (23:07 +0000)]
Changed the liveness tracking in the RegisterScavenger
to use register units instead of registers.

reviewed by Jakob Stoklund Olesen.

llvm-svn: 214798

10 years agoAdd simplified aliases for access to DCCR, ICCR, DEAR and ESR
Joerg Sonnenberger [Mon, 4 Aug 2014 22:56:42 +0000 (22:56 +0000)]
Add simplified aliases for access to DCCR, ICCR, DEAR and ESR

llvm-svn: 214797

10 years agoThreadSafety.cpp: Prune bogus \param(s). [-Wdocumentation]
NAKAMURA Takumi [Mon, 4 Aug 2014 22:48:46 +0000 (22:48 +0000)]
ThreadSafety.cpp: Prune bogus \param(s). [-Wdocumentation]

llvm-svn: 214796

10 years agoThreadSafetyAnalysis: Fix annotations in class members. [-Wdocumentation]
NAKAMURA Takumi [Mon, 4 Aug 2014 22:48:36 +0000 (22:48 +0000)]
ThreadSafetyAnalysis: Fix annotations in class members. [-Wdocumentation]

llvm-svn: 214795

10 years agoclang/test/Sema/struct-packed-align.c: Add the case in MS mode that alignment doesn...
NAKAMURA Takumi [Mon, 4 Aug 2014 22:48:19 +0000 (22:48 +0000)]
clang/test/Sema/struct-packed-align.c: Add the case in MS mode that alignment doesn't affect packing.

Also, revert a couple of suppressions.
  r214298, "Suppress clang/test/Sema/struct-packed-align.c for targeting LLP64."
  r214301, "Suppress clang/test/Sema/struct-packed-align.c also on msvc for investigating."

llvm-svn: 214794

10 years agoTrailing linefeed.
NAKAMURA Takumi [Mon, 4 Aug 2014 22:48:09 +0000 (22:48 +0000)]
Trailing linefeed.

llvm-svn: 214793

10 years agoRemove separator parameter from static local naming code
Reid Kleckner [Mon, 4 Aug 2014 22:35:30 +0000 (22:35 +0000)]
Remove separator parameter from static local naming code

It was always set to ".", which was duplicated in a few places.

llvm-svn: 214792

10 years ago[PECOFF] Remove unused paraemter.
Rui Ueyama [Mon, 4 Aug 2014 22:19:53 +0000 (22:19 +0000)]
[PECOFF] Remove unused paraemter.

llvm-svn: 214791

10 years agoFix SmallDenseMap assignment operator.
Andrew Trick [Mon, 4 Aug 2014 22:18:25 +0000 (22:18 +0000)]
Fix SmallDenseMap assignment operator.

Self assignment would lead to buckets of garbage, causing quadratic probing to hang.

llvm-svn: 214790

10 years agoThread Safety Analysis: add a -Wthread-safety-negative flag that warns whenever
DeLesley Hutchins [Mon, 4 Aug 2014 22:13:06 +0000 (22:13 +0000)]
Thread Safety Analysis: add a -Wthread-safety-negative flag that warns whenever
a mutex is acquired, but corresponding mutex is not provably not-held.  This
is based on the earlier negative requirements patch.

llvm-svn: 214789

10 years ago[FastISel][AArch64] Fix shift lowering for i8 and i16 value types.
Juergen Ributzka [Mon, 4 Aug 2014 21:49:51 +0000 (21:49 +0000)]
[FastISel][AArch64] Fix shift lowering for i8 and i16 value types.

This fix changes the parameters #r and #s that are passed to the UBFM/SBFM
instruction to get the zero/sign-extension for free.

The original problem was that the shift left would use the 32-bit shift even for
i8/i16 value types, which could leave the upper bits set with "garbage" values.

The arithmetic shift right on the other side would use the wrong MSB as sign-bit
to determine what bits to shift into the value.

This fixes <rdar://problem/17907720>.

llvm-svn: 214788

10 years agoIR: Fix up doxygen comment for LLVMContext::diagnose
Justin Bogner [Mon, 4 Aug 2014 21:49:15 +0000 (21:49 +0000)]
IR: Fix up doxygen comment for LLVMContext::diagnose

This comment was referring to the DiagnosticSeverity with RS_
prefixes, but they're actually DS_. I've also modernized the comment
style since I was changing it anyway.

llvm-svn: 214787

10 years agoUpdate for LLVM API change.
Eric Christopher [Mon, 4 Aug 2014 21:33:42 +0000 (21:33 +0000)]
Update for LLVM API change.

llvm-svn: 214786

10 years ago[SDAG] Fix a really, really terrible bug in the DAG combiner.
Chandler Carruth [Mon, 4 Aug 2014 21:29:59 +0000 (21:29 +0000)]
[SDAG] Fix a really, really terrible bug in the DAG combiner.

This code is completely wrong. It is also dead, as if it were to *ever*
run, it would crash. Fortunately, after my work to the combiner, it is
at least *possible* to reach the code, and llvm-stress has found a test
case. Thanks to Patrick for reporting.

It would be really good if anyone who remembers how this code works and
what it was intended to do could add some more obvious test coverage
instead of my completely contrived and reduced test case. My test case
was so brittle I left a bread crumb comment in it to help the next
person to stumble on it and not know what it was actually testing for.

llvm-svn: 214785

10 years agotlbre / tlbwe / tlbsx / tlbsx. variants for the PPC 4xx CPUs.
Joerg Sonnenberger [Mon, 4 Aug 2014 21:28:22 +0000 (21:28 +0000)]
tlbre / tlbwe / tlbsx / tlbsx. variants for the PPC 4xx CPUs.

llvm-svn: 214784

10 years agoAdd code to AssemblyParse_x86::get_non_call_site_unwind_plan
Jason Molenda [Mon, 4 Aug 2014 21:26:55 +0000 (21:26 +0000)]
Add code to AssemblyParse_x86::get_non_call_site_unwind_plan
to recognize an epilogue that ends with a jmp to
objc_retainAutoreleaseReturnValue instead of a ret instruction.
<rdar://problem/17889928>

llvm-svn: 214783

10 years agoReorder to keep data and routines separate and to keep a couple of
Eric Christopher [Mon, 4 Aug 2014 21:25:44 +0000 (21:25 +0000)]
Reorder to keep data and routines separate and to keep a couple of
similar routines close to each other.

llvm-svn: 214782

10 years agoRemove the TargetMachine forwards for TargetSubtargetInfo based
Eric Christopher [Mon, 4 Aug 2014 21:25:23 +0000 (21:25 +0000)]
Remove the TargetMachine forwards for TargetSubtargetInfo based
information and update all callers. No functional change.

llvm-svn: 214781

10 years agoReimplement the temporary non-const getSubtargetImpl routine so
Eric Christopher [Mon, 4 Aug 2014 21:24:07 +0000 (21:24 +0000)]
Reimplement the temporary non-const getSubtargetImpl routine so
that we can avoid implementing it on every target. Thanks to Richard
Smith for the suggestions!

llvm-svn: 214780

10 years ago[AArch64] Extend the number of scalar instructions supported in the AdvSIMD
Chad Rosier [Mon, 4 Aug 2014 21:20:25 +0000 (21:20 +0000)]
[AArch64] Extend the number of scalar instructions supported in the AdvSIMD
scalar integer instruction pass.

This is a patch I had lying around from a few months ago.  The pass is
currently disabled by default, so nothing to interesting.

llvm-svn: 214779

10 years agoUsersManual: update clang-cl options
Hans Wennborg [Mon, 4 Aug 2014 21:07:58 +0000 (21:07 +0000)]
UsersManual: update clang-cl options

llvm-svn: 214777

10 years agoMC uses .lcomm now, so adjust.
Joerg Sonnenberger [Mon, 4 Aug 2014 21:06:00 +0000 (21:06 +0000)]
MC uses .lcomm now, so adjust.

llvm-svn: 214776

10 years agoFix failure to invoke exception handler on Win64
Reid Kleckner [Mon, 4 Aug 2014 21:05:27 +0000 (21:05 +0000)]
Fix failure to invoke exception handler on Win64

When the last instruction prior to a function epilogue is a call, we
need to emit a nop so that the return address is not in the epilogue IP
range.  This is consistent with MSVC's behavior, and may be a workaround
for a bug in the Win64 unwinder.

Differential Revision: http://reviews.llvm.org/D4751

Patch by Vadim Chugunov!

llvm-svn: 214775

10 years agoDon't drop dllimport from qualified friend redeclarations (PR20512)
Hans Wennborg [Mon, 4 Aug 2014 20:54:39 +0000 (20:54 +0000)]
Don't drop dllimport from qualified friend redeclarations (PR20512)

This matches MSVC's logic, which seems to be that when the friend
declaration is qualified, it cannot be a declaration of a new symbol
and so the dll linkage doesn't change.

Differential Revision: http://reviews.llvm.org/D4764

llvm-svn: 214774

10 years agoNit picky const-correctness with auto. No functional changes.
Aaron Ballman [Mon, 4 Aug 2014 20:49:47 +0000 (20:49 +0000)]
Nit picky const-correctness with auto. No functional changes.

llvm-svn: 214773