OSDN Git Service

android-x86/external-llvm.git
8 years ago[x86, SSE] update packed FP compare tests for direct translation from builtin to IR
Sanjay Patel [Wed, 15 Jun 2016 21:22:15 +0000 (21:22 +0000)]
[x86, SSE] update packed FP compare tests for direct translation from builtin to IR

The clang side of this was r272840:
http://reviews.llvm.org/rL272840

A follow-up step would be to auto-upgrade and remove these LLVM intrinsics completely.

Differential Revision: http://reviews.llvm.org/D21269

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272841 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAddress a few coding style issues. NFC.
Chad Rosier [Wed, 15 Jun 2016 21:14:02 +0000 (21:14 +0000)]
Address a few coding style issues. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272838 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix llvm-objdump when disassembling a stripped Mach-O binary with the -macho option.
Kevin Enderby [Wed, 15 Jun 2016 21:14:01 +0000 (21:14 +0000)]
Fix llvm-objdump when disassembling a stripped Mach-O binary with the -macho option.
It was printing out nothing in this case.

llvm-objdump tries to disassemble sections a symbol at a time.  In the case of a
fully stripped Mach-O executable the only symbol remaining in the (__TEXT,__text)
section is the special linker defined symbol __mh_execute_header . This
symbol is special in that while it is N_SECT symbol in the (__TEXT,__text)
its address is before the start of the (__TEXT,__text).  It’s address is the
start of the __TEXT segment which is where the mach header is statically
linked. So the code in DisassembleMachO() needs to deal with this case specially.

rdar://26778273

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272837 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Fix/simplify some conditional statements
Krzysztof Parzyszek [Wed, 15 Jun 2016 21:05:04 +0000 (21:05 +0000)]
[Hexagon] Fix/simplify some conditional statements

Fix for PR28138.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272836 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86]: Fix for uninitialized access introduced in r272797.
Kevin B. Smith [Wed, 15 Jun 2016 20:52:19 +0000 (20:52 +0000)]
[X86]: Fix for uninitialized access introduced in r272797.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272835 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[x86] delete unnecessary function declarations
Sanjay Patel [Wed, 15 Jun 2016 20:51:47 +0000 (20:51 +0000)]
[x86] delete unnecessary function declarations

Missed this in r272806, r272807.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272834 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CFLAA] Ignore non-pointers, move Attrs to graph nodes.
George Burgess IV [Wed, 15 Jun 2016 20:43:41 +0000 (20:43 +0000)]
[CFLAA] Ignore non-pointers, move Attrs to graph nodes.

This patch makes CFLAA ignore non-pointer values, since we can now
sanely do that with the escaping/unknown attributes. Additionally,
StratifiedAttrs make more sense to sit on nodes than edges (since
they're properties of values, and ultimately end up on the nodes of
StratifiedSets). So, this patch puts said attributes on nodes.

Patch by Jia Chen.

Differential Revision: http://reviews.llvm.org/D21387

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272833 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAArch64: stop trying to use 32-bit MOVZs when expanding patchpoints.
Tim Northover [Wed, 15 Jun 2016 20:33:36 +0000 (20:33 +0000)]
AArch64: stop trying to use 32-bit MOVZs when expanding patchpoints.

Of course the assembly was right but because the opcode was MOVZWi it was
encoded as "movz w16, #65535, lsl #32" which is an unallocated encoding and
would go horribly wrong on a CPU.

No idea how this bug survived this long. It seems nobody is using that aspect
of patchpoints.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272831 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAxe some trailing whitespace from my last commit
Reid Kleckner [Wed, 15 Jun 2016 20:32:42 +0000 (20:32 +0000)]
Axe some trailing whitespace from my last commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272830 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[codeview] Move deserialization methods out of line
Reid Kleckner [Wed, 15 Jun 2016 20:30:34 +0000 (20:30 +0000)]
[codeview] Move deserialization methods out of line

They aren't performance critical and don't need to be inline.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272829 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[x86] add folds for x86 vector compare nodes (PR27924)
Sanjay Patel [Wed, 15 Jun 2016 20:26:58 +0000 (20:26 +0000)]
[x86] add folds for x86 vector compare nodes (PR27924)

Ideally, we can get rid of most x86 LLVM intrinsics by transforming them to IR (and some of that happened
with http://reviews.llvm.org/rL272807), but it doesn't cost much to have some simple folds in the backend
too while we're working on that and as a backstop.

This fixes:
https://llvm.org/bugs/show_bug.cgi?id=27924

Differential Revision: http://reviews.llvm.org/D21356

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272828 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoStatistic: Add machine parseable json output
Matthias Braun [Wed, 15 Jun 2016 20:19:16 +0000 (20:19 +0000)]
Statistic: Add machine parseable json output

- We lacked a short unique identifier for a statistics, so I renamed the
  current "Name" field that just contained the DEBUG_TYPE name of the
  current file to DebugType and added a new "Name" field that contains
  the C++ identifier of the statistic variable.
- Add the -stats-json option which outputs statistics in json format.

Differential Revision: http://reviews.llvm.org/D20995

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272826 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LoopSimplify] Analyses do not need to be member variables.
Davide Italiano [Wed, 15 Jun 2016 18:51:25 +0000 (18:51 +0000)]
[LoopSimplify] Analyses do not need to be member variables.

In preparation for porting this pass to the new PM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272818 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[codeview] Use ArrayRef instead of a non-const vector reference
Reid Kleckner [Wed, 15 Jun 2016 18:48:35 +0000 (18:48 +0000)]
[codeview] Use ArrayRef instead of a non-const vector reference

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272817 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[pdbdump] Verify LF_{CLASS,ENUM,INTERFACE,STRUCTURE,UNION} records.
Rui Ueyama [Wed, 15 Jun 2016 18:26:59 +0000 (18:26 +0000)]
[pdbdump] Verify LF_{CLASS,ENUM,INTERFACE,STRUCTURE,UNION} records.

Differential Revision: http://reviews.llvm.org/D21361

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272815 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86]: Updated r272801 to promote 16 bit compares with immediate operand
Kevin B. Smith [Wed, 15 Jun 2016 18:18:05 +0000 (18:18 +0000)]
[X86]: Updated r272801 to promote 16 bit compares with immediate operand
to 32 bits. This is in response to a comment by Eli Friedman.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272814 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CodeView] Add support for emitting S_UDT for typedefs
David Majnemer [Wed, 15 Jun 2016 18:00:01 +0000 (18:00 +0000)]
[CodeView] Add support for emitting S_UDT for typedefs

Emit a S_UDT record for typedefs.  We still need to do something for
class types.

Differential Revision: http://reviews.llvm.org/D21149

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272813 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd support for string attributes in the C API.
Amaury Sechet [Wed, 15 Jun 2016 17:50:39 +0000 (17:50 +0000)]
Add support for string attributes in the C API.

Summary: As per title. This completes the C API Attribute support.

Reviewers: Wallbraker, whitequark, echristo, rafael, jyknight

Subscribers: mehdi_amini

Differential Revision: http://reviews.llvm.org/D21365

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272811 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoTest commit after access grant. Modified comment by adding a period.
Pankaj Gode [Wed, 15 Jun 2016 17:24:52 +0000 (17:24 +0000)]
Test commit after access grant. Modified comment by adding a period.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272808 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[x86, SSE] remove the GCCBuiltins from the integer min/max intrinsics
Sanjay Patel [Wed, 15 Jun 2016 17:17:27 +0000 (17:17 +0000)]
[x86, SSE] remove the GCCBuiltins from the integer min/max intrinsics
This allows us to emit native IR in Clang (next commit).
Also, update the intrinsic tests to show that codegen already knows how to handle
the IR that Clang will soon produce.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272806 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[x86] fix function name; NFC
Sanjay Patel [Wed, 15 Jun 2016 17:12:29 +0000 (17:12 +0000)]
[x86] fix function name; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272805 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[TargetLibraryInfo] Teach isValidProtoForLibFunc about tan
David Majnemer [Wed, 15 Jun 2016 16:47:23 +0000 (16:47 +0000)]
[TargetLibraryInfo] Teach isValidProtoForLibFunc about tan

We would fail to validate the type of the tan function which would cause
downstream users of isValidProtoForLibFunc to assert.

This fixes PR28143.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272802 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86]: Quit promoting 8 and 16 bit compares to 32 bit.
Kevin B. Smith [Wed, 15 Jun 2016 16:37:46 +0000 (16:37 +0000)]
[X86]: Quit promoting 8 and 16 bit compares to 32 bit.
Differential Revision: http://reviews.llvm.org/D21144

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272801 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert "Preserve DebugInfo when replacing values in DAGCombiner"
Nirav Dave [Wed, 15 Jun 2016 16:08:50 +0000 (16:08 +0000)]
Revert "Preserve DebugInfo when replacing values in DAGCombiner"

Reverting due to assertion failure in
lib/CodeGen/SelectionDAG/InstrEmitter.cpp

This reverts commit r272792.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272799 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86]: Improve Liveness checking for X86FixupBWInsts.cpp
Kevin B. Smith [Wed, 15 Jun 2016 16:03:06 +0000 (16:03 +0000)]
[X86]: Improve Liveness checking for X86FixupBWInsts.cpp
Differential Revision: http://reviews.llvm.org/D21085

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272797 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LV] Add getter function for LoopVectorizationLegality::Strides. NFC
Adam Nemet [Wed, 15 Jun 2016 15:49:46 +0000 (15:49 +0000)]
[LV] Add getter function for LoopVectorizationLegality::Strides. NFC

This should help moving Strides to LAA later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272796 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips] Eliminate unused code for addrRegReg complex pattern. NFC.
Vasileios Kalintiris [Wed, 15 Jun 2016 15:30:07 +0000 (15:30 +0000)]
[mips] Eliminate unused code for addrRegReg complex pattern. NFC.

Reviewers: dsanders, sdardis

Subscribers: dsanders, sdardis, llvm-commits

Differential Revision: http://reviews.llvm.org/D21381

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272794 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd llvm_unreachable to silence a "not all control paths" false positive warning...
Aaron Ballman [Wed, 15 Jun 2016 15:27:53 +0000 (15:27 +0000)]
Add llvm_unreachable to silence a "not all control paths" false positive warning from MSVC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272793 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoPreserve DebugInfo when replacing values in DAGCombiner
Nirav Dave [Wed, 15 Jun 2016 14:50:08 +0000 (14:50 +0000)]
Preserve DebugInfo when replacing values in DAGCombiner

[DAG] Previously debug values would transfer debuginfo for the selected
start node for a replacement which allows for debug to be dropped.

Push debug value transfer to occur with node/value replacement in
SelectionDAG, remove now extraneous transfers of debug values.

This refixes PR9817 which was being incompletely checked in the
testsuite.

Reviewers: jyknight

Subscribers: dblaikie, llvm-commits

Differential Revision: http://reviews.llvm.org/D21037

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272792 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoReverting r272778 because there's an assertion
Ranjeet Singh [Wed, 15 Jun 2016 14:23:29 +0000 (14:23 +0000)]
Reverting r272778 because there's an assertion
failure when running the test CodeGen/ARM/intrinsics-coprocessor.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272791 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AVX512] Remove the GCCBuiltins from the mask pcmpeq/pcmpgt intrinsics so we can...
Craig Topper [Wed, 15 Jun 2016 14:06:28 +0000 (14:06 +0000)]
[AVX512] Remove the GCCBuiltins from the mask pcmpeq/pcmpgt intrinsics so we can emit native IR from clang.

The intrinsics themselves can be removed in a future commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272786 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AMDGPU] Fix few coding style issues. NFC.
Valery Pykhtin [Wed, 15 Jun 2016 13:55:09 +0000 (13:55 +0000)]
[AMDGPU] Fix few coding style issues. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272785 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips] Missing test case
Simon Dardis [Wed, 15 Jun 2016 13:49:58 +0000 (13:49 +0000)]
[mips] Missing test case

Add missing testcase from r272666.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272784 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LV] Remove more unused functions. NFC
Adam Nemet [Wed, 15 Jun 2016 12:26:15 +0000 (12:26 +0000)]
[LV] Remove more unused functions.  NFC

LoopVectorizationLegality::strides_begin/end are also unused.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272781 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LV] Remove unused function. NFC
Adam Nemet [Wed, 15 Jun 2016 12:26:11 +0000 (12:26 +0000)]
[LV] Remove unused function. NFC

LoopVectorizationLegality::mustCheckStrides is unused.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272780 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[ARM] Add support for mrrc/mrrc2 intrinsics.
Ranjeet Singh [Wed, 15 Jun 2016 11:32:24 +0000 (11:32 +0000)]
[ARM] Add support for mrrc/mrrc2 intrinsics.

Differential Revision: http://reviews.llvm.org/D21178

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272778 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoWork around MSVC "friend" semantics.
Sean Silva [Wed, 15 Jun 2016 10:51:40 +0000 (10:51 +0000)]
Work around MSVC "friend" semantics.

The error on clang-x86-win2008-selfhost is:

C:\buildbot\slave-config\clang-x86-win2008-selfhost\llvm\lib\Transforms\Vectorize\SLPVectorizer.cpp(955) : error C2248: 'llvm::slpvectorizer::BoUpSLP::ScheduleData' : cannot access private struct declared in class 'llvm::slpvectorizer::BoUpSLP'
        C:\buildbot\slave-config\clang-x86-win2008-selfhost\llvm\lib\Transforms\Vectorize\SLPVectorizer.cpp(608) : see declaration of 'llvm::slpvectorizer::BoUpSLP::ScheduleData'
        C:\buildbot\slave-config\clang-x86-win2008-selfhost\llvm\lib\Transforms\Vectorize\SLPVectorizer.cpp(337) : see declaration of 'llvm::slpvectorizer::BoUpSLP'

I reproduced this locally with both MSVC 2013 and MSVC 2015.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272772 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips] Replace AdditionalRequires<[IsGP64bit]> with GPR_64. NFC.
Daniel Sanders [Wed, 15 Jun 2016 10:36:16 +0000 (10:36 +0000)]
[mips] Replace AdditionalRequires<[IsGP64bit]> with GPR_64. NFC.

Summary: Also fixed one case where HasMips64 was being used instead of IsGP64bit.

Reviewers: sdardis

Subscribers: dsanders, llvm-commits, sdardis

Differential Revision: http://reviews.llvm.org/D21028

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272771 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUse FPasses in opt exactly when it is initialized.
Patrik Hagglund [Wed, 15 Jun 2016 10:32:00 +0000 (10:32 +0000)]
Use FPasses in opt exactly when it is initialized.

Previously, there was a discrepancy between the population of function
passes in FPasses, and their invocation. Function passes specified on
the command line, after an optimizaton level was simply discared. This
fix PR27509.

Patch by Jesper Antonsson.

Differential Review: http://reviews.llvm.org/D20725

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272770 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips] Removed invalid test from o32_cc.ll
Daniel Sanders [Wed, 15 Jun 2016 09:47:27 +0000 (09:47 +0000)]
[mips] Removed invalid test from o32_cc.ll

MIPS32R1 cannot implement a 64-bit FPU because this was introduced in MIPS32R2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272769 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips] clang-format Mips16ISelDAGToDAG.{cpp,h}
Daniel Sanders [Wed, 15 Jun 2016 09:44:22 +0000 (09:44 +0000)]
[mips] clang-format Mips16ISelDAGToDAG.{cpp,h}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272768 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSpeculative buildbot fix.
Sean Silva [Wed, 15 Jun 2016 09:00:33 +0000 (09:00 +0000)]
Speculative buildbot fix.

This wasn't failing for me with clang as the compiler. I think GCC may
disagree with clang about whether a friend declaration introduces a
declaration in the enclosing namespace (or something).

Example error:

/home/uweigand/sandbox/buildbot/clang-s390x-linux/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp:950:77: error: â€˜llvm::raw_ostream& llvm::slpvectorizer::operator<<(llvm::raw_ostream&, const llvm::slpvectorizer::BoUpSLP::ScheduleData&)’ should have been declared inside â€˜llvm::slpvectorizer’
                                              const BoUpSLP::ScheduleData &SD) {
                                                                             ^

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272767 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PM] Port SLPVectorizer to the new PM
Sean Silva [Wed, 15 Jun 2016 08:43:40 +0000 (08:43 +0000)]
[PM] Port SLPVectorizer to the new PM

This uses the "runImpl" approach to share code with the old PM.

Porting to the new PM meant abandoning the anonymous namespace enclosing
most of SLPVectorizer.cpp which is a bit of a bummer (but not a big deal
compared to having to pull the pass class into a header which the new PM
requires since it calls the constructor directly).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272766 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][msa] Fix register/register-class mismatches in emitINSERT_DF_VIDX().
Daniel Sanders [Wed, 15 Jun 2016 08:43:23 +0000 (08:43 +0000)]
[mips][msa] Fix register/register-class mismatches in emitINSERT_DF_VIDX().

Reviewers: sdardis

Subscribers: dsanders, sdardis, llvm-commits

Differential Revision: http://reviews.llvm.org/D21068

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272765 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][microMIPS] Add CodeGen support for AND*, OR16, OR*, XOR*, NOT16 and NOR instru...
Zlatko Buljan [Wed, 15 Jun 2016 07:46:24 +0000 (07:46 +0000)]
[mips][microMIPS] Add CodeGen support for AND*, OR16, OR*, XOR*, NOT16 and NOR instructions
Differential Revision: http://reviews.llvm.org/D16719

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272764 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AVX512] Fix BLENDM lowering patterns. Operands should be swapped to match SELECT...
Igor Breger [Wed, 15 Jun 2016 07:30:38 +0000 (07:30 +0000)]
[AVX512] Fix BLENDM lowering patterns. Operands should be swapped to match SELECT behavior.
Use BLENDM instead of masked move instruction.

Differential Revision: http://reviews.llvm.org/D21001

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272763 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoPush a dependent computation into the assert that uses it; NFC
Sanjoy Das [Wed, 15 Jun 2016 07:27:04 +0000 (07:27 +0000)]
Push a dependent computation into the assert that uses it; NFC

... instead of explicitly conditioning on NDEBUG.  Also use an easier to
read conditional expression.

(Addresses post-commit review from David Blaikie.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272762 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Fix MUBUF offset bugs affecting llvm.amdgcn.buffer.* intrinsics
Nicolai Haehnle [Wed, 15 Jun 2016 07:13:05 +0000 (07:13 +0000)]
AMDGPU: Fix MUBUF offset bugs affecting llvm.amdgcn.buffer.* intrinsics

Summary:
This fixes two related bugs. First, the generic optimization passes
unfortunately generate negative constant offsets but the hardware treats
SOffset as an unsigned value.

Second, there is a hardware bug on SI and CI, where address clamping in MUBUF
instructions does not work correctly when SOffset is larger than the buffer
size. This patch works around this bug by never using SOffset.

An alternative workaround would be to do the clamping manually when SOffset
is too large, but generating the required code sequence during instruction
selection would be rather involved, and in any case the resulting code would
probably be worse.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96360

Reviewers: arsenm, tstellarAMD

Subscribers: arsenm, llvm-commits, kzhuravl

Differential Revision: http://reviews.llvm.org/D21326

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272761 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix unused variable warning; NFC
Sanjoy Das [Wed, 15 Jun 2016 06:53:59 +0000 (06:53 +0000)]
Fix unused variable warning; NFC

TailCallReturnAddrDelta is used only in an assert, so put it under
defined(NDEBUG).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272760 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SCEV] Use dyn_cast<T> instead of dyn_cast<const T>; NFC
Sanjoy Das [Wed, 15 Jun 2016 06:53:55 +0000 (06:53 +0000)]
[SCEV] Use dyn_cast<T> instead of dyn_cast<const T>; NFC

The const is unnecessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272759 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SCEV] Use cast<> instead of dyn_cast; NFC
Sanjoy Das [Wed, 15 Jun 2016 06:53:51 +0000 (06:53 +0000)]
[SCEV] Use cast<> instead of dyn_cast; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272758 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PM] Port AlignmentFromAssumptions to the new PM.
Sean Silva [Wed, 15 Jun 2016 06:18:01 +0000 (06:18 +0000)]
[PM] Port AlignmentFromAssumptions to the new PM.

This uses the "runImpl" pattern to share code between the old and new PM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272757 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDon't force SP-relative addressing for statepoints
Sanjoy Das [Wed, 15 Jun 2016 05:35:14 +0000 (05:35 +0000)]
Don't force SP-relative addressing for statepoints

Summary:
...  when the offset is not statically known.

Prioritize addresses relative to the stack pointer in the stackmap, but
fallback gracefully to other modes of addressing if the offset to the
stack pointer is not a known constant.

Patch by Oscar Blumberg!

Reviewers: sanjoy

Subscribers: llvm-commits, majnemer, rnk, sanjoy, thanm

Differential Revision: http://reviews.llvm.org/D21259

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272756 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd support for callsite in the new C API for attributes
Amaury Sechet [Wed, 15 Jun 2016 05:14:29 +0000 (05:14 +0000)]
Add support for callsite in the new C API for attributes

Summary: The second consumer of attributes.

Reviewers: Wallbraker, whitequark, echristo, rafael, jyknight

Subscribers: mehdi_amini

Differential Revision: http://reviews.llvm.org/D21266

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272754 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SCEV] clang-format some sections
Sanjoy Das [Wed, 15 Jun 2016 04:37:50 +0000 (04:37 +0000)]
[SCEV] clang-format some sections

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272753 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SCEV] Change the interface for SolveQuadraticEquation; NFC
Sanjoy Das [Wed, 15 Jun 2016 04:37:47 +0000 (04:37 +0000)]
[SCEV] Change the interface for SolveQuadraticEquation; NFC

Use Optional<T> to denote the absence of a solution, not
SCEVCouldNotCompute.  This makes the usage of SolveQuadraticEquation
somewhat simpler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272752 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/SI: Correctly encode constant expressions
Tom Stellard [Wed, 15 Jun 2016 03:09:39 +0000 (03:09 +0000)]
AMDGPU/SI: Correctly encode constant expressions

Summary:
We we have an MCConstantExpr, we can encode it directly into the instruction
instead of emitting fixups.

Reviewers: artem.tamazov, vpykhtin, SamWot, nhaustov, arsenm

Subscribers: arsenm, llvm-commits, kzhuravl

Differential Revision: http://reviews.llvm.org/D21236

Change-Id: I88b3edf288d48e65c5d705fc4850d281f8e36948

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272750 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[MBP] add comments and bug fix
Xinliang David Li [Wed, 15 Jun 2016 03:03:30 +0000 (03:03 +0000)]
[MBP] add comments and bug fix

Document the new parameter and threshod computation
model.  Also fix a bug when the threshold parameter
is set to be different from the default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272749 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/AsmParser: Add support for parsing symbol operands
Tom Stellard [Wed, 15 Jun 2016 02:54:14 +0000 (02:54 +0000)]
AMDGPU/AsmParser: Add support for parsing symbol operands

Summary:
We can now reference symbols directly in operands, like this:
s_mov_b32 s0, global

Reviewers: artem.tamazov, vpykhtin, SamWot, nhaustov

Subscribers: arsenm, llvm-commits, kzhuravl

Differential Revision: http://reviews.llvm.org/D21038

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272748 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LibFuzzer] Fix ``FuzzerMutate.ShuffleBytes2`` unit test on OSX.
Dan Liew [Wed, 15 Jun 2016 01:40:02 +0000 (01:40 +0000)]
[LibFuzzer] Fix ``FuzzerMutate.ShuffleBytes2`` unit test on OSX.

The ``FuzzerMutate.ShuffleBytes2`` unit test was failing on
OSX due to the implementation of ``std::random_shuffle()``
being different between libcxx and libstdc++.

@kcc has decided (see http://reviews.llvm.org/D21218) it is acceptable
for there to be different mutation behavior on different platforms so
this commit just adjusts the test to perform the minimum number of
iterations (that is a power of 2) to see all the mutations the unit test
is looking for.

Differential Revision: http://reviews.llvm.org/D21359

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272743 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRecommit [LV] Enable vectorization of loops where the IV has an external use
Michael Kuperstein [Wed, 15 Jun 2016 00:35:26 +0000 (00:35 +0000)]
Recommit [LV] Enable vectorization of loops where the IV has an external use

r272715 broke libcxx because it did not correctly handle cases where the
last iteration of one IV is the second-to-last iteration of another.

Original commit message:
Vectorizing loops with "escaping" IVs has been disabled since r190790, due to
PR17179. This re-enables it, with support for external use of both
"post-increment" (last iteration) and "pre-increment" (second-to-last iteration)
IVs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272742 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMake DetectDeadLanes preserve CFG
Matt Arsenault [Wed, 15 Jun 2016 00:25:09 +0000 (00:25 +0000)]
Make DetectDeadLanes preserve CFG

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272740 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LoopUnroll] Don't crash trying to unroll loop with EH pad exit
David Majnemer [Wed, 15 Jun 2016 00:19:56 +0000 (00:19 +0000)]
[LoopUnroll] Don't crash trying to unroll loop with EH pad exit

We do not support splitting cleanuppad or catchswitches.  This is
problematic for passes which assume that a loop is in loop simplify
form (the loop would have a dedicated exit block instead of sharing it).

While it isn't great that we don't support this for cleanups, we still
cannot make loop-simplify form an assertable precondition because
indirectbr will also disable these sorts of CFG cleanups.

This fixes PR28132.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272739 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CodeView] Don't emit debuginfo for imported symbols
David Majnemer [Wed, 15 Jun 2016 00:19:52 +0000 (00:19 +0000)]
[CodeView] Don't emit debuginfo for imported symbols

Emitting symbol information requires us to have a definition for the
symbol.  A symbol reference is insufficient.

This fixes PR28123.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272738 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRemove the ScalarReplAggregates pass
David Majnemer [Wed, 15 Jun 2016 00:19:09 +0000 (00:19 +0000)]
Remove the ScalarReplAggregates pass

Nearly all the changes to this pass have been done while maintaining and
updating other parts of LLVM.  LLVM has had another pass, SROA, which
has superseded ScalarReplAggregates for quite some time.

Differential Revision: http://reviews.llvm.org/D21316

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272737 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Run pointer optimization passes
Matt Arsenault [Wed, 15 Jun 2016 00:11:01 +0000 (00:11 +0000)]
AMDGPU: Run pointer optimization passes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272736 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoVerifier: check that functions have at most a single !prof attachment.
Peter Collingbourne [Tue, 14 Jun 2016 23:13:15 +0000 (23:13 +0000)]
Verifier: check that functions have at most a single !prof attachment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272734 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix a test case to match its intention
Xinliang David Li [Tue, 14 Jun 2016 23:05:46 +0000 (23:05 +0000)]
Fix a test case to match its intention

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272733 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRemove unused import. NFC
Amaury Sechet [Tue, 14 Jun 2016 22:30:50 +0000 (22:30 +0000)]
Remove unused import. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272731 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoReverting r272715 since it broke libcxx.
Michael Kuperstein [Tue, 14 Jun 2016 22:30:41 +0000 (22:30 +0000)]
Reverting r272715 since it broke libcxx.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272730 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSet machine block placement hot prob threshold for both static and runtime profile.
Dehao Chen [Tue, 14 Jun 2016 22:27:17 +0000 (22:27 +0000)]
Set machine block placement hot prob threshold for both static and runtime profile.

Summary: With runtime profile, we have more confidence in branch probability, thus during basic block layout, we set a lower hot prob threshold so that blocks can be layouted optimally.

Reviewers: djasper, davidxl

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D20991

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272729 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[pdbdump] Verify TPI hash for LF_ENUM type records.
Rui Ueyama [Tue, 14 Jun 2016 22:25:07 +0000 (22:25 +0000)]
[pdbdump] Verify TPI hash for LF_ENUM type records.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272728 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix list of deprecated C API attribute functions
Amaury Sechet [Tue, 14 Jun 2016 22:23:29 +0000 (22:23 +0000)]
Fix list of deprecated C API attribute functions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272727 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert "Enable libFuzzer's afl_driver to append stderr to a file."
Vitaly Buka [Tue, 14 Jun 2016 22:09:00 +0000 (22:09 +0000)]
Revert "Enable libFuzzer's afl_driver to append stderr to a file."

Crashes with AddressSanitizer: SEGV on unknown address
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fuzzer/builds/9924/steps/annotate/logs/stdio

This reverts commit r272706.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272726 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRename AttributeSetImpl::NumAttrs and AttributeSetImpl::getNumAttributes to reflect...
Amaury Sechet [Tue, 14 Jun 2016 22:04:16 +0000 (22:04 +0000)]
Rename AttributeSetImpl::NumAttrs and AttributeSetImpl::getNumAttributes to reflect that they work on slots rather than attributes. NFC

Summary: The current naming not only doesn't convey the meaning of what this does, but worse, it convey the wrong meaning. This was a major source of confusion understanding the code, so I'm applying the boy scout rule here and making it better after I leave.

Reviewers: void, bkramer, whitequark

Differential Revision: http://reviews.llvm.org/D21264

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272725 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PM] Port WholeProgramDevirt to the new pass manager.
Davide Italiano [Tue, 14 Jun 2016 21:44:19 +0000 (21:44 +0000)]
[PM] Port WholeProgramDevirt to the new pass manager.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272721 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LV] Enable vectorization of loops where the IV has an external use
Michael Kuperstein [Tue, 14 Jun 2016 21:27:27 +0000 (21:27 +0000)]
[LV] Enable vectorization of loops where the IV has an external use

Vectorizing loops with "escaping" IVs has been disabled since r190790, due to
PR17179. This re-enables it, with support for external use of both
"post-increment" (last iteration) and "pre-increment" (second-to-last iteration)
IVs.

Differential Revision: http://reviews.llvm.org/D21048

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272715 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[x86] add current codegen tests for PR27924
Sanjay Patel [Tue, 14 Jun 2016 21:25:46 +0000 (21:25 +0000)]
[x86] add current codegen tests for PR27924

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272714 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd a missing test.
Evgeniy Stepanov [Tue, 14 Jun 2016 21:24:13 +0000 (21:24 +0000)]
Add a missing test.

This is a test for r272421: Disable MSan-hostile loop unswitching.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272713 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[MemorySSA] Set CFGOnly correctly for MemorySSAWrapperPass
Geoff Berry [Tue, 14 Jun 2016 21:19:40 +0000 (21:19 +0000)]
[MemorySSA] Set CFGOnly correctly for MemorySSAWrapperPass

Subscribers: mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D21344

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272712 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoIR: Introduce local_unnamed_addr attribute.
Peter Collingbourne [Tue, 14 Jun 2016 21:01:22 +0000 (21:01 +0000)]
IR: Introduce local_unnamed_addr attribute.

If a local_unnamed_addr attribute is attached to a global, the address
is known to be insignificant within the module. It is distinct from the
existing unnamed_addr attribute in that it only describes a local property
of the module rather than a global property of the symbol.

This attribute is intended to be used by the code generator and LTO to allow
the linker to decide whether the global needs to be in the symbol table. It is
possible to exclude a global from the symbol table if three things are true:
- This attribute is present on every instance of the global (which means that
  the normal rule that the global must have a unique address can be broken without
  being observable by the program by performing comparisons against the global's
  address)
- The global has linkonce_odr linkage (which means that each linkage unit must have
  its own copy of the global if it requires one, and the copy in each linkage unit
  must be the same)
- It is a constant or a function (which means that the program cannot observe that
  the unique-address rule has been broken by writing to the global)

Although this attribute could in principle be computed from the module
contents, LTO clients (i.e. linkers) will normally need to be able to compute
this property as part of symbol resolution, and it would be inefficient to
materialize every module just to compute it.

See:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160509/356401.html
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160516/356738.html
for earlier discussion.

Part of the fix for PR27553.

Differential Revision: http://reviews.llvm.org/D20348

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272709 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoResubmit "[pdb] Actually write a PDB to disk from YAML.""
Zachary Turner [Tue, 14 Jun 2016 20:48:36 +0000 (20:48 +0000)]
Resubmit "[pdb] Actually write a PDB to disk from YAML.""

Reviewed By: ruiu
Differential Revision: http://reviews.llvm.org/D21220

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272708 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoEnable libFuzzer's afl_driver to append stderr to a file.
Vitaly Buka [Tue, 14 Jun 2016 20:42:05 +0000 (20:42 +0000)]
Enable libFuzzer's afl_driver to append stderr to a file.

Summary:
[libFuzzer] Enable afl_driver to append stderr to a user specified file.

Append stderr of afl_driver to the file specified by the environmental variable
AFL_DRIVER_STDERR_DUPLICATE_FILENAME if it is set. This lets users see outputs
on crashes without rerunning crashing test cases (which won't work for crashes
that are difficult to reproduce). Before this patch, stderr would only be sent to afl-fuzz
and users would have no way of seeing it.

Reviewers: llvm-commits, aizatsky, kcc, vitalybuka

Subscribers: vitalybuka

Differential Revision: http://reviews.llvm.org/D21194

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272706 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/SI: Refactor fixup handling for constant addrspace variables
Tom Stellard [Tue, 14 Jun 2016 20:29:59 +0000 (20:29 +0000)]
AMDGPU/SI: Refactor fixup handling for constant addrspace variables

Summary:
We now use a standard fixup type applying the pc-relative address of
constant address space variables, and we have the GlobalAddress lowering
code add the required 4 byte offset to the global address rather than
doing it as part of the fixup.

This refactoring will make it easier to use the same code for global
address space variables and also simplifies the code.

Re-commit this after fixing a bug where we were trying to use a
reference to a Triple object that had already been destroyed.

Reviewers: arsenm, kzhuravl

Subscribers: arsenm, kzhuravl, llvm-commits

Differential Revision: http://reviews.llvm.org/D21154

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272705 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMake sure attribute kind and attributes are named respectively Kind and Attr consiste...
Amaury Sechet [Tue, 14 Jun 2016 20:27:35 +0000 (20:27 +0000)]
Make sure attribute kind and attributes are named respectively Kind and Attr consistently. Historically they used to be the same the terminology is very confused in the codebase. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272704 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[ValueTracking] Calls to @llvm.assume always return
Sanjoy Das [Tue, 14 Jun 2016 20:23:16 +0000 (20:23 +0000)]
[ValueTracking] Calls to @llvm.assume always return

This change teaches llvm::isGuaranteedToTransferExecutionToSuccessor
that calls to @llvm.assume always terminate.  Most other relevant
intrinsics should be covered by the "CS.onlyReadsMemory() ||
CS.onlyAccessesArgMemory()" bit but we were missing @llvm.assumes
because we state that it clobbers memory.

Added an LICM test case, but this change is not specific to LICM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272703 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[TailDuplication] Split up NumInstrDups statistic.
Geoff Berry [Tue, 14 Jun 2016 19:40:10 +0000 (19:40 +0000)]
[TailDuplication] Split up NumInstrDups statistic.

Summary:
Split NumInstrDups statistic into separate added/removed counts to avoid
negative stat being printed as unsigned.

Subscribers: mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D21335

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272700 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LibFuzzer] Disable the ``fuzzer-trace-pc.test`` test on non-linux platforms.
Dan Liew [Tue, 14 Jun 2016 19:15:13 +0000 (19:15 +0000)]
[LibFuzzer] Disable the ``fuzzer-trace-pc.test`` test on non-linux platforms.

On OSX this test sometimes fails due to the
``LLVMFuzzer-FullCoverageSetTest-TracePC`` program going over the
default 2GiB memory limit. This shouldn't be happening and needs
investigating. For now just disable the test so we can set up an
OSX buildbot.

Differential Revision: http://reviews.llvm.org/D21319

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272696 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Reduce the width of multiplification when its operands are extended from i8...
Wei Mi [Tue, 14 Jun 2016 18:53:20 +0000 (18:53 +0000)]
[X86] Reduce the width of multiplification when its operands are extended from i8 or i16

For <N x i32> type mul, pmuludq will be used for targets without SSE41, which
often introduces many extra pack and unpack instructions in vectorized loop
body because pmuludq generates <N/2 x i64> type value. However when the operands
of <N x i32> mul are extended from smaller size values like i8 and i16, the type
of mul may be shrunk to use pmullw + pmulhw/pmulhuw instead of pmuludq, which
generates better code. For targets with SSE41, pmulld is supported so no
shrinking is needed.

Differential Revision: http://reviews.llvm.org/D20931

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272694 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert "[pdb] Actually write a PDB to disk from YAML."
Zachary Turner [Tue, 14 Jun 2016 18:51:35 +0000 (18:51 +0000)]
Revert "[pdb] Actually write a PDB to disk from YAML."

This reverts commit 879139e1c6577b09df52de56a6bab856a19ed185.

This was committed accidentally when I blindly typed git svn
dcommit instead of the command to generate a patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272693 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[pdb] Actually write a PDB to disk from YAML.
Zachary Turner [Tue, 14 Jun 2016 18:49:36 +0000 (18:49 +0000)]
[pdb] Actually write a PDB to disk from YAML.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272692 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CFLAA] Tag arguments as escaped instead of unknown.
George Burgess IV [Tue, 14 Jun 2016 18:12:28 +0000 (18:12 +0000)]
[CFLAA] Tag arguments as escaped instead of unknown.

This patch also includes some refactoring.

Prior to this patch, we tagged all CFLAA attributes as unknown. This is
suboptimal, since it meant that any Value used as an argument would be
considered to alias any other Value that existed.

Now that we have the machinery to tag sets below the set for an
arbitrary value with attributes, it's okay to be less conservative with
arguments. (Specifically, we still tag the set under an argument with
unknown).

Patch by Jia Chen.

Differential Revision: http://reviews.llvm.org/D21262

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272690 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CFLAA] Refactor graph-building code. NFC.
George Burgess IV [Tue, 14 Jun 2016 18:02:27 +0000 (18:02 +0000)]
[CFLAA] Refactor graph-building code. NFC.

This patch refactors CFLAA's graph building code. This makes keeping
track of common state (TargetLibraryInfo, ...) easier.

Patch by Jia Chen.

Differential Revision: http://reviews.llvm.org/D21261

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272688 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[docs] Update AMDGPU relocation information
Konstantin Zhuravlyov [Tue, 14 Jun 2016 17:20:29 +0000 (17:20 +0000)]
[docs] Update AMDGPU relocation information
  - Added new notation for specifying relocation calculation
  - Renamed:
    - R_AMDGPU_32_LOW -> R_AMDGPU_ABS32_LO
    - R_AMDGPU_32_HIGH -> R_AMDGPU_ABS32_HI
    - R_AMDGPU_64 -> R_AMDGPU_ABS64
  - Added:
    - R_AMDGPU_REL32
    - R_AMDGPU_REL64
    - R_AMDGPU_ABS32
  - Updated calculations for relative relocations

Differential Revision: http://reviews.llvm.org/D21215

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272684 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[FileCheck] Document --check-prefixes.
Daniel Sanders [Tue, 14 Jun 2016 16:42:05 +0000 (16:42 +0000)]
[FileCheck] Document --check-prefixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272683 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix some typos in the Kaleidoscope tutorial (PR28120)
Hans Wennborg [Tue, 14 Jun 2016 16:05:12 +0000 (16:05 +0000)]
Fix some typos in the Kaleidoscope tutorial (PR28120)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272681 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert "AMDGPU/SI: Refactor fixup handling for constant addrspace variables"
Tom Stellard [Tue, 14 Jun 2016 15:16:35 +0000 (15:16 +0000)]
Revert "AMDGPU/SI: Refactor fixup handling for constant addrspace variables"

This reverts commit r272675.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272677 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/SI: Refactor fixup handling for constant addrspace variables
Tom Stellard [Tue, 14 Jun 2016 15:11:01 +0000 (15:11 +0000)]
AMDGPU/SI: Refactor fixup handling for constant addrspace variables

Summary:
We now use a standard fixup type applying the pc-relative address of
constant address space variables, and we have the GlobalAddress lowering
code add the required 4 byte offset to the global address rather than
doing it as part of the fixup.

This refactoring will make it easier to use the same code for global
address space variables and also simplifies the code.

Reviewers: arsenm, kzhuravl

Subscribers: arsenm, kzhuravl, llvm-commits

Differential Revision: http://reviews.llvm.org/D21154

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272675 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix BSS global handling in AsmPrinter
Nirav Dave [Tue, 14 Jun 2016 15:09:30 +0000 (15:09 +0000)]
Fix BSS global handling in AsmPrinter

Change EmitGlobalVariable to check final assembler section is in BSS
before using .lcomm/.comm directive. This prevents globals from being
put into .bss erroneously when -data-sections is used.

This fixes PR26570.

Reviewers: echristo, rafael

Subscribers: llvm-commits, mehdi_amini

Differential Revision: http://reviews.llvm.org/D21146

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272674 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AMDGPU][llvm-mc] Predefined symbols to access -mcpu from the assembly source (.optio...
Artem Tamazov [Tue, 14 Jun 2016 15:03:59 +0000 (15:03 +0000)]
[AMDGPU][llvm-mc] Predefined symbols to access -mcpu from the assembly source (.option.machine_version...)

The feature allows for conditional assembly etc.
TODO: make those symbols read-only.
Test added.

Differential Revision: http://reviews.llvm.org/D21238

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272673 91177308-0d34-0410-b5e6-96231b3b80d8