platform/upstream/llvm.git
7 years agoIntroduce KMP_DEVICE_THREAD_LIMIT
Jonathan Peyton [Wed, 26 Jul 2017 20:07:58 +0000 (20:07 +0000)]
Introduce KMP_DEVICE_THREAD_LIMIT

This change drops in KMP_DEVICE_THREAD_LIMIT to replace KMP_MAX_THREADS. It's
possible there will eventually be a OMP_DEVICE_THREAD_LIMIT, and we need
something to distinguish from OMP_THREAD_LIMIT, which is currently implemented
incorrectly (the fix for that will be added soon in a separate patch).
KMP_ALL_THREADS is deprecated here, but we can keep the "all" option on
KMP_DEVICE_THREAD_LIMIT to support that functionality. KMP_DEVICE_THREAD_LIMIT
now has priority over its deprecated rival KMP_ALL_THREADS. I also cleaned up
some comments that incorrectly referred to non-existent kmp_max_threads variable
instead of kmp_max_nth.

I've left the name of where this setting eventually ends up as
__kmp_max_nth, for now.

This change does not change much in the way of functionality. It does NOT change
OMP_THREAD_LIMIT. It's just cleaning up and setting up for that.

Patch by Terry Wilmarth

Differential Revision: https://reviews.llvm.org/D35860

llvm-svn: 309168

7 years agoAMDGPU/GlobalISel: Mark 32-bit G_OR as legal
Tom Stellard [Wed, 26 Jul 2017 20:00:53 +0000 (20:00 +0000)]
AMDGPU/GlobalISel: Mark 32-bit G_OR as legal

Reviewers: arsenm

Reviewed By: arsenm

Subscribers: kzhuravl, wdng, nhaehnle, yaxunl, rovka, kristof.beyls, igorb, dstuttard, tpr, t-tye, llvm-commits

Differential Revision: https://reviews.llvm.org/D35127

llvm-svn: 309165

7 years ago[Simplify] Fix invalid removal write for escaping values.
Michael Kruse [Wed, 26 Jul 2017 19:58:15 +0000 (19:58 +0000)]
[Simplify] Fix invalid removal write for escaping values.

A PHI node's incoming block is the user of its operand, not the PHI's parent.

Assuming the PHINode's parent being the user lead to the removal of a
MemoryAccesses because its use was assumed to be inside of the SCoP.

llvm-svn: 309164

7 years ago[GISel]: Avoid zero length array when building Instrs that don't have
Aditya Nandakumar [Wed, 26 Jul 2017 19:58:03 +0000 (19:58 +0000)]
[GISel]: Avoid zero length array when building Instrs that don't have
uses.

Also splitting the buildSources part allows more overloads such as
adding MachineOperands directly in the arguments for buildInstr.

llvm-svn: 309163

7 years agoTry to fix the windows build.
Rafael Espindola [Wed, 26 Jul 2017 19:57:40 +0000 (19:57 +0000)]
Try to fix the windows build.

llvm-svn: 309162

7 years agoTry to fix the windows build.
Rafael Espindola [Wed, 26 Jul 2017 19:39:40 +0000 (19:39 +0000)]
Try to fix the windows build.

llvm-svn: 309161

7 years agoTry to fix the windows build.
Rafael Espindola [Wed, 26 Jul 2017 19:37:58 +0000 (19:37 +0000)]
Try to fix the windows build.

llvm-svn: 309160

7 years agoChange CallLoweringInfo::CS to be an ImmutableCallSite instead of a pointer. NFCI.
Peter Collingbourne [Wed, 26 Jul 2017 19:15:29 +0000 (19:15 +0000)]
Change CallLoweringInfo::CS to be an ImmutableCallSite instead of a pointer. NFCI.

This was a use-after-free waiting to happen.

llvm-svn: 309159

7 years agoMigrate SimplifyLibCalls to new OptimizationRemarkEmitter
Adam Nemet [Wed, 26 Jul 2017 19:03:18 +0000 (19:03 +0000)]
Migrate SimplifyLibCalls to new OptimizationRemarkEmitter

Summary:
This changes SimplifyLibCalls to use the new OptimizationRemarkEmitter
API.

In fact, as SimplifyLibCalls is only ever called via InstCombine,
(as far as I can tell) the OptimizationRemarkEmitter is added there,
and then passed through to SimplifyLibCalls later.

I have avoided changing any remark text.

This closes PR33787

Patch by Sam Elliott!

Reviewers: anemet, davide

Reviewed By: anemet

Subscribers: davide, mehdi_amini, eraman, fhahn, llvm-commits

Differential Revision: https://reviews.llvm.org/D35608

llvm-svn: 309158

7 years agoThis patch returns proper value to indicate the case when instruction throughput...
Andrew V. Tischenko [Wed, 26 Jul 2017 18:55:14 +0000 (18:55 +0000)]
This patch returns proper value to indicate the case when instruction throughput can't be calculated.
Differential revision https://reviews.llvm.org/D35831

llvm-svn: 309156

7 years ago[OpenCL] Fix access qualifiers metadata for kernel arguments with typedef
Alexey Sotkin [Wed, 26 Jul 2017 18:49:54 +0000 (18:49 +0000)]
[OpenCL] Fix access qualifiers metadata for kernel arguments with typedef

Subscribers: cfe-commits, yaxunl, Anastasia

Differential Revision: https://reviews.llvm.org/D35420

llvm-svn: 309155

7 years agoDo a better job at emitting prefrabricated skeleton CUs.
Adrian Prantl [Wed, 26 Jul 2017 18:48:32 +0000 (18:48 +0000)]
Do a better job at emitting prefrabricated skeleton CUs.

This is a better fix than r308708 for the problem introduced in
r304020. It restores the skeleton CU testcases modified by that commit
to their original form and most importantly ensures that
frontend-generated skeleton CUs (such as used to point to Clang
modules) come after the regular CUs. This broke for DICompileUnit
nodes that don't have any immediate children because they are now
constructed lazily instead of the order in which they are listed in
!llvm.dbg.cu. After this commit we still don't guarantee that order,
but we do guarantee that empty skeletons come last.

Shipping versions of LLDB are very sensitive to the ordering of
CUs. I'll track a fix for LLDB to be more permissive separately.
This fixes a test failure in the LLDB testsuite.

rdar://problem/33357252

llvm-svn: 309154

7 years agoSimplify. NFC.
Rafael Espindola [Wed, 26 Jul 2017 18:47:49 +0000 (18:47 +0000)]
Simplify. NFC.

llvm-svn: 309153

7 years agoDetemplate SymbolTable.
Rafael Espindola [Wed, 26 Jul 2017 18:42:48 +0000 (18:42 +0000)]
Detemplate SymbolTable.

NFC, just makes it easier to access from non templated code.

llvm-svn: 309152

7 years agoUnlink nodes instead of copying, to avoid memory problems.
Eric Beckmann [Wed, 26 Jul 2017 18:33:21 +0000 (18:33 +0000)]
Unlink nodes instead of copying, to avoid memory problems.

llvm-svn: 309151

7 years ago[Dominators] Change Roots type to SmallVector
Jakub Kuderski [Wed, 26 Jul 2017 18:27:39 +0000 (18:27 +0000)]
[Dominators] Change Roots type to SmallVector

Summary: We can use the template parameter `IsPostDom` to pick an appropriate SmallVector size to store DomTree roots for dominators and postdominators. Before, the code would always allocate memory with `std::vector`.

Reviewers: dberlin, davide, sanjoy, grosser

Reviewed By: grosser

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35636

llvm-svn: 309148

7 years agoPartial fix for PR33858
Jonathan Roelofs [Wed, 26 Jul 2017 18:13:57 +0000 (18:13 +0000)]
Partial fix for PR33858

https://reviews.llvm.org/D35848

llvm-svn: 309147

7 years ago[Dominators] Move root-finding out of DomTreeBase and simplify it
Jakub Kuderski [Wed, 26 Jul 2017 18:07:40 +0000 (18:07 +0000)]
[Dominators] Move root-finding out of DomTreeBase and simplify it

Summary:
This patch moves root-finding logic from DominatorTreeBase to GenericDomTreeConstruction.h.
It makes the behavior simpler and more consistent by always adding a virtual root to PostDominatorTrees.

Reviewers: dberlin, davide, grosser, sanjoy

Reviewed By: dberlin

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35597

llvm-svn: 309146

7 years agoAdd bitrig removal to release notes
Erich Keane [Wed, 26 Jul 2017 18:04:45 +0000 (18:04 +0000)]
Add bitrig removal to release notes

llvm-svn: 309145

7 years agoUn-XFAIL some internal lit tests on Windows, they pass for me locally
Reid Kleckner [Wed, 26 Jul 2017 18:04:18 +0000 (18:04 +0000)]
Un-XFAIL some internal lit tests on Windows, they pass for me locally

llvm-svn: 309144

7 years agoDiffing against a file that is itself used in the test seems to be a bad
Eric Beckmann [Wed, 26 Jul 2017 17:47:44 +0000 (17:47 +0000)]
Diffing against a file that is itself used in the test seems to be a bad
idea, because it might get locked down and rendered unopenable.

llvm-svn: 309142

7 years agoSimplify. NFC.
Rafael Espindola [Wed, 26 Jul 2017 17:27:27 +0000 (17:27 +0000)]
Simplify. NFC.

llvm-svn: 309141

7 years agoFix LIT test breakage
George Karpenkov [Wed, 26 Jul 2017 17:19:36 +0000 (17:19 +0000)]
Fix LIT test breakage

Differential Revision: https://reviews.llvm.org/D35867

llvm-svn: 309140

7 years ago[X86][AVX512] Regenerated and cleaned up extension tests.
Simon Pilgrim [Wed, 26 Jul 2017 16:47:00 +0000 (16:47 +0000)]
[X86][AVX512] Regenerated and cleaned up extension tests.

llvm-svn: 309139

7 years ago[X86] Regenerate setcc tests
Simon Pilgrim [Wed, 26 Jul 2017 16:45:57 +0000 (16:45 +0000)]
[X86] Regenerate setcc tests

llvm-svn: 309138

7 years ago[X86][AVX512] Regenerate shuffle tests with broadcast comments.
Simon Pilgrim [Wed, 26 Jul 2017 16:41:18 +0000 (16:41 +0000)]
[X86][AVX512] Regenerate shuffle tests with broadcast comments.

llvm-svn: 309137

7 years ago[X86] Regenerate memset tests
Simon Pilgrim [Wed, 26 Jul 2017 16:39:07 +0000 (16:39 +0000)]
[X86] Regenerate memset tests

llvm-svn: 309136

7 years agoCorrectly enable the llvm-mt tests, now that build flags changed.
Eric Beckmann [Wed, 26 Jul 2017 16:35:44 +0000 (16:35 +0000)]
Correctly enable the llvm-mt tests, now that build flags changed.

llvm-svn: 309134

7 years agoQuote '?' in llvm-rc test
Reid Kleckner [Wed, 26 Jul 2017 16:25:48 +0000 (16:25 +0000)]
Quote '?' in llvm-rc test

Summary:
Bash interperets the '?' character as matching an arbitrary character.
On systems that have a file or directory with exactly one character in
their root directory, '/?' gets reinterpreted into that pathname, which
fails to match the expected Help text for llvm-rc.
This patch quotes the '/?' to avoid that edge case.

Reviewers: mnbvmar, ecbeckmann, rnk

Reviewed By: rnk

Subscribers: dyung, ruiu, llvm-commits

Differential Revision: https://reviews.llvm.org/D35852

llvm-svn: 309133

7 years ago[Hexagon] Mark raise_relocation_error as NORETURN.
Florian Hahn [Wed, 26 Jul 2017 16:07:51 +0000 (16:07 +0000)]
[Hexagon] Mark raise_relocation_error as NORETURN.

Summary:
This silences a couple of implicit fallthrough warnings with GCC 7.1 in
this file.

Reviewers: colinl, kparzysz

Reviewed By: kparzysz

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35889

llvm-svn: 309129

7 years agoUpdate the assertion to meet with the changes in r309121. (NFC)
Dehao Chen [Wed, 26 Jul 2017 15:47:00 +0000 (15:47 +0000)]
Update the assertion to meet with the changes in r309121. (NFC)

llvm-svn: 309125

7 years ago[X86] Add combineBT test failure because bits have multiple uses.
Simon Pilgrim [Wed, 26 Jul 2017 15:41:57 +0000 (15:41 +0000)]
[X86] Add combineBT test failure because bits have multiple uses.

llvm-svn: 309124

7 years ago[lit] Mark several of lit's tests XFAIL on Windows
Brian Gesiak [Wed, 26 Jul 2017 15:10:50 +0000 (15:10 +0000)]
[lit] Mark several of lit's tests XFAIL on Windows

Summary:
rL257221 attempted to run lit's own test suite continuously, but that
commit was reverted because lit's test suite does not pass on Windows.
Because lit's tests do not run continuously, they often regress.

In order to un-revert rL257221, mark lit tests that fail as XFAIL for
Windows platforms.

Test Plan:
On a Windows development environment, follow the instructions in
utils/lit/README.txt to run lit's test suite:

```
utils/lit/lit.py \
    --path /path/to/your/llvm/build/bin \
    utils/lit/tests
```

Verify that the test suite is run and a successful exit code is
returned.

Reviewers: mgorny, rnk, delcypher, beanz

Reviewed By: rnk

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35879

llvm-svn: 309123

7 years ago[lit] Fix type error for parallelism groups
Brian Gesiak [Wed, 26 Jul 2017 15:02:05 +0000 (15:02 +0000)]
[lit] Fix type error for parallelism groups

Summary:
Whereas rL299560 and rL309071 call `parallelism_groups.items()`, under the
assumption that `parallelism_groups` is a `dict` type, the default
parameter for that attribute is a `list`. Change the default to a
`dict` for type correctness.

This regression in the unit tests would have been caught if the
unit tests were being run continously. It also would have been caught
if the lit project used a Python type checker such as `mypy`.

Test Plan:
As per the instructions in `utils/lit/README.txt`, run the lit unit
test suite:

```
utils/lit/lit.py \
    --path /path/to/your/llvm/build/bin \
    utils/lit/tests
```

Verify that the test `lit :: unit/TestRunner.py` fails before applying this
patch, but passes once this patch is applied.

Reviewers: mgorny, rnk, rafael

Reviewed By: mgorny

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35878

llvm-svn: 309122

7 years agoMake new PM honor -fdebug-info-for-profiling
Dehao Chen [Wed, 26 Jul 2017 15:01:20 +0000 (15:01 +0000)]
Make new PM honor -fdebug-info-for-profiling

Summary: The new PM needs to invoke add-discriminator pass when building with -fdebug-info-for-profiling.

Reviewers: chandlerc, davidxl

Reviewed By: chandlerc

Subscribers: sanjoy, llvm-commits

Differential Revision: https://reviews.llvm.org/D35744

llvm-svn: 309121

7 years agoRevert "[lit] Remove dead code not referenced in the LLVM SVN repo."
Brian Gesiak [Wed, 26 Jul 2017 14:59:36 +0000 (14:59 +0000)]
Revert "[lit] Remove dead code not referenced in the LLVM SVN repo."

Summary:
This reverts rL306623, which removed `FileBasedTest`, an abstract base class,
but did not also remove the usages of that class in the lit unit tests.
The revert fixes four test failures in the lit unit test suite.

Test plan:
As per the instructions in `utils/lit/README.txt`, run the lit unit
test suite:

```
utils/lit/lit.py \
    --path /path/to/your/llvm/build/bin \
    utils/lit/tests
```

Verify that the following tests fail before applying this patch, and
pass once the patch is applied:

```
lit :: test-data.py
lit :: test-output.py
lit :: xunit-output.py
```

In addition, run `check-llvm` to make sure the existing LLVM test suite
executes normally.

Reviewers: george.karpenkov, mgorny, dlj

Reviewed By: mgorny

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35877

llvm-svn: 309120

7 years ago[ScheduleOptimizer] Translate to C++ bindings
Roman Gareev [Wed, 26 Jul 2017 14:59:15 +0000 (14:59 +0000)]
[ScheduleOptimizer] Translate to C++ bindings

Translate the ScheduleOptimizer to use the new isl C++ bindings.

Reviewed-by: Michael Kruse <llvm@meinersbur.de>
Differential Revision: https://reviews.llvm.org/D35845

llvm-svn: 309119

7 years agoRegression test for PR10856
Serge Pavlov [Wed, 26 Jul 2017 14:58:55 +0000 (14:58 +0000)]
Regression test for PR10856

llvm-svn: 309118

7 years ago[docs] change a few code-blocks to llvm from text
Nuno Lopes [Wed, 26 Jul 2017 14:11:23 +0000 (14:11 +0000)]
[docs] change a few code-blocks to llvm from text

llvm-svn: 309117

7 years agounguarded availability: add a fixit for the "annotate '...'
Alex Lorenz [Wed, 26 Jul 2017 13:58:02 +0000 (13:58 +0000)]
unguarded availability: add a fixit for the "annotate '...'
with an availability attribute to silence" note

rdar://33539233

Differential Revision: https://reviews.llvm.org/D35726

llvm-svn: 309116

7 years ago[CMake] Disable building libomptarget and add CMake switch
Jonas Hahnfeld [Wed, 26 Jul 2017 13:55:00 +0000 (13:55 +0000)]
[CMake] Disable building libomptarget and add CMake switch

Introduce OPENMP_ENABLE_LIBOMPTARGET which defaults to OFF at the moment.

libomptarget is not yet ready for prime time:
 - Offloading to NVIDIA GPUs is not completed yet (compiler, device RTL)
 - The generic ELF plugin for offloading to the host (meant for testing)
   uses a single instance of the OpenMP runtime (libomp). That is why
   omp_is_initial_device() returns 1 which makes the tests fail.
Because of these reasons, we want to disable building (and testing!)
for release 5.0.

See https://bugs.llvm.org/show_bug.cgi?id=33859

Differential Revision: https://reviews.llvm.org/D35719

llvm-svn: 309115

7 years ago[NFC] test commit.
Stefan Pintilie [Wed, 26 Jul 2017 13:44:59 +0000 (13:44 +0000)]
[NFC] test commit.

Added a comment to explain how to add a PPCISD node.

llvm-svn: 309114

7 years ago[Bash-autocompletion] Show HelpText with possible flags
Yuka Takahashi [Wed, 26 Jul 2017 13:36:58 +0000 (13:36 +0000)]
[Bash-autocompletion] Show HelpText with possible flags

Summary:
`clang --autocomplete=-std` will show
```
-std:   Language standard to compile for
-std=   Language standard to compile for
-stdlib=        C++ standard library to use
```
after this change.

However, showing HelpText with completion in bash seems super tricky, so
this feature will be used in other shells (fish, zsh...).

Reviewers: v.g.vassilev, teemperor, ruiu

Subscribers: cfe-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D35759

llvm-svn: 309113

7 years ago[Bash-completion] Fixed a bug that file doesn't autocompleted after =
Yuka Takahashi [Wed, 26 Jul 2017 13:30:36 +0000 (13:30 +0000)]
[Bash-completion] Fixed a bug that file doesn't autocompleted after =

Summary:
File path wasn't autocompleted after `-fmodule-cache-path=[tab]`, so
fixed this bug by checking if $flags contains only a newline or not.

Differential Revision: https://reviews.llvm.org/D35763

llvm-svn: 309112

7 years agoRevert r309094: [globalisel][tablegen] Fuse the generated tables together.
Daniel Sanders [Wed, 26 Jul 2017 13:28:40 +0000 (13:28 +0000)]
Revert r309094: [globalisel][tablegen] Fuse the generated tables together.

The ARM bots have started failing and while this patch should be an improvement
for these bots, it's also the only suspect in the blamelist. Reverting while
Diana and I investigate the problem.

llvm-svn: 309111

7 years ago[ScopInfo] Avoid use of getStmtFor(BB). NFC.
Michael Kruse [Wed, 26 Jul 2017 13:25:28 +0000 (13:25 +0000)]
[ScopInfo] Avoid use of getStmtFor(BB). NFC.

Since there will be no more a 1:1 correspondence between statements and
basic blocks, we would like to get rid of the method getStmtFor(BB)
and its uses. Here we remove one of its uses in ScopInfo by fetching
the statement in which the call instruction lies.

Contributed-by: Nandini Singhal <cs15mtech01004@iith.ac.in>
Differential Revision: https://reviews.llvm.org/D35691

llvm-svn: 309110

7 years ago[SCEVValidator] Loop exit values of loops before the SCoP are synthesizable.
Michael Kruse [Wed, 26 Jul 2017 13:05:45 +0000 (13:05 +0000)]
[SCEVValidator] Loop exit values of loops before the SCoP are synthesizable.

 In the following loop:

   int i;
   for (i = 0; i < func(); i+=1)
     ;
SCoP:
   for (int j = 0; j<n; j+=1)
     S(i, j)

The value i is synthesizable in the SCoP that includes only the j-loop.
This is because i is fixed within the SCoP, it is irrelevant whether
it originates from another loop.

This fixes a strange case where a PHI was synthesiable in a SCoP,
but not its incoming value, triggering an assertion.

This should fix MultiSource/Applications/sgefa/sgefa of the
perf-x86_64-penryn-O3-polly-before-vectorizer-unprofitable buildbot.

llvm-svn: 309109

7 years agoDAGCombiner: Extend reduceBuildVecToTrunc to handle non-zero offset
Zvi Rackover [Wed, 26 Jul 2017 12:57:03 +0000 (12:57 +0000)]
DAGCombiner: Extend reduceBuildVecToTrunc to handle non-zero offset

Summary:
Adding support for combining power2-strided build_vector's where the
first build_vectori's operand is extracted from a non-zero index.

Example:

 v4i32 build_vector((extract_elt V, 1),
                    (extract_elt V, 3),
                    (extract_elt V, 5),
                    (extract_elt V, 7))
 -->
 v4i32 truncate (bitcast (shuffle<1,u,3,u,5,u,7,u> V, u) to v4i64)

Reviewers: delena, RKSimon, guyblank

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35700

llvm-svn: 309108

7 years ago[X86] Regenerated BT tests
Simon Pilgrim [Wed, 26 Jul 2017 12:49:20 +0000 (12:49 +0000)]
[X86] Regenerated BT tests

Test on 32/64 bit targets where appropriate

llvm-svn: 309107

7 years agoRecommit r308327 2nd time: Add a warning for missing
Alex Lorenz [Wed, 26 Jul 2017 12:20:57 +0000 (12:20 +0000)]
Recommit r308327 2nd time: Add a warning for missing
'#pragma pack (pop)' and suspicious uses of '#pragma pack' in included files

The first recommit (r308441) caused a "non-default #pragma pack value might
change the alignment of struct or union members in the included file" warning
in LLVM itself. This recommit tweaks the added warning to avoid warnings for
#includes that don't have any records that are affected by the non-default
alignment. This tweak avoids the previously emitted warning in LLVM.

Original message:

This commit adds a new -Wpragma-pack warning. It warns in the following cases:

- When a translation unit is missing terminating #pragma pack (pop) directives.
- When entering an included file if the current alignment value as determined
  by '#pragma pack' directives is different from the default alignment value.
- When leaving an included file that changed the state of the current alignment
  value.

rdar://10184173

Differential Revision: https://reviews.llvm.org/D35484

llvm-svn: 309106

7 years ago[COFF, ARM64] Fix symbol offsets in ADRP/ADD/LDR/STR relocations
Martin Storsjo [Wed, 26 Jul 2017 11:19:17 +0000 (11:19 +0000)]
[COFF, ARM64] Fix symbol offsets in ADRP/ADD/LDR/STR relocations

In COFF, a symbol offset can't be stored in the relocation (as is
done in ELF or MachO), but is stored as the immediate in the
instruction itself. The immediate in the ADRP thus is the symbol
offset in bytes, not in pages. For the PAGEOFFSET_12A/L relocations,
ignore any offset outside of the lowest 12 bits; they won't have any
effect on the ADD/LDR/STR instruction itself but only on the associated
ADRP.

This is similar to how the same issue is handled for MOVW/MOVT
instructions in ELF (see e.g. SVN r307713, and r307728 in lld).

This fixes "fixup out of range" errors while building larger object
files, where temporary symbols end up as a plain section symbol and
an offset, and fixes any cases where the symbol offset mean that
the actual target ended up on a different page than the symbol
itself.

Differential Revision: https://reviews.llvm.org/D35791

llvm-svn: 309105

7 years ago[X86] Add urem vector test for non-uniform pow2 constants
Simon Pilgrim [Wed, 26 Jul 2017 11:07:45 +0000 (11:07 +0000)]
[X86] Add urem vector test for non-uniform pow2 constants

llvm-svn: 309104

7 years ago[X86] Regenerated urem pow2 tests on 32/64 bit targets
Simon Pilgrim [Wed, 26 Jul 2017 11:05:16 +0000 (11:05 +0000)]
[X86] Regenerated urem pow2 tests on 32/64 bit targets

llvm-svn: 309103

7 years ago[X86] Regenerated umul overflow tests on 32/64 bit targets
Simon Pilgrim [Wed, 26 Jul 2017 11:04:18 +0000 (11:04 +0000)]
[X86] Regenerated umul overflow tests on 32/64 bit targets

llvm-svn: 309102

7 years ago[ARM] GlobalISel: Map G_GLOBAL_VALUE to GPR
Diana Picus [Wed, 26 Jul 2017 11:01:13 +0000 (11:01 +0000)]
[ARM] GlobalISel: Map G_GLOBAL_VALUE to GPR

A G_GLOBAL_VALUE is basically a pointer, so it should live in the GPR.

llvm-svn: 309101

7 years ago[X86][AVX] Regenerated and cleaned up AVX1 intrinsic tests.
Simon Pilgrim [Wed, 26 Jul 2017 10:54:51 +0000 (10:54 +0000)]
[X86][AVX] Regenerated and cleaned up AVX1 intrinsic tests.

Cleaned up triple settings, added 32-bit/64-bit targets where useful, added broadcast comments

llvm-svn: 309100

7 years ago[X86][AVX2] Regenerated and cleaned up broadcast tests.
Simon Pilgrim [Wed, 26 Jul 2017 10:47:51 +0000 (10:47 +0000)]
[X86][AVX2] Regenerated and cleaned up broadcast tests.

llvm-svn: 309099

7 years ago[X86][AVX512] Regenerated and added 32-bit targets to select tests
Simon Pilgrim [Wed, 26 Jul 2017 10:39:55 +0000 (10:39 +0000)]
[X86][AVX512] Regenerated and added 32-bit targets to select tests

llvm-svn: 309098

7 years ago[X86][AVX] Regenerated and cleaned up masked gather/scatter tests.
Simon Pilgrim [Wed, 26 Jul 2017 10:37:12 +0000 (10:37 +0000)]
[X86][AVX] Regenerated and cleaned up masked gather/scatter tests.

Remove unused KNL checks and triple settings, added broadcast comments

llvm-svn: 309097

7 years ago[ELF] - Ignore --warn-once option.
George Rimar [Wed, 26 Jul 2017 10:25:49 +0000 (10:25 +0000)]
[ELF] - Ignore --warn-once option.

Was discussed in D35797 thread.

llvm-svn: 309096

7 years ago[X86][AVX] Regenerate lzcnt test.
Simon Pilgrim [Wed, 26 Jul 2017 10:22:56 +0000 (10:22 +0000)]
[X86][AVX] Regenerate lzcnt test.

Tidied up triples and checks.

llvm-svn: 309095

7 years ago[globalisel][tablegen] Fuse the generated tables together.
Daniel Sanders [Wed, 26 Jul 2017 10:20:56 +0000 (10:20 +0000)]
[globalisel][tablegen] Fuse the generated tables together.

Summary:
Now that we have control flow in place, fuse the per-rule tables into a
single table. This is a compile-time saving at this point. However, this will
also enable the optimization of a table so that similar instructions can be
tested together, reducing the time spent on the matching the code.

This is NFC in terms of externally visible behaviour but some internals have
changed slightly. State.MIs is no longer reset between each rule that is
attempted because it's not necessary to do so. As a consequence of this the
restriction on the order that instructions are added to State.MIs has been
relaxed to only affect recorded instructions that require new elements to be
added to the vector. GIM_RecordInsn can now write to any element from 1 to
State.MIs.size() instead of just State.MIs.size().

Reviewers: ab, t.p.northover, qcolombet, rovka, aditya_nandakumar

Reviewed By: rovka

Subscribers: kristof.beyls, igorb, llvm-commits

Differential Revision: https://reviews.llvm.org/D35681

llvm-svn: 309094

7 years ago[X86][FMA] Regenerate test with broadcast comments.
Simon Pilgrim [Wed, 26 Jul 2017 10:20:49 +0000 (10:20 +0000)]
[X86][FMA] Regenerate test with broadcast comments.

llvm-svn: 309093

7 years agoFollow up for r307085: a better fix for a dangling StringRef.
Alexander Kornienko [Wed, 26 Jul 2017 10:14:55 +0000 (10:14 +0000)]
Follow up for r307085: a better fix for a dangling StringRef.

llvm-svn: 309092

7 years ago[ELF] - Change way how we handle --noinhibit-exec
George Rimar [Wed, 26 Jul 2017 09:46:59 +0000 (09:46 +0000)]
[ELF] - Change way how we handle --noinhibit-exec

Previously we handled this option implicitly, only
for infering unresolved symbols handling policy.

ld man says: "--noinhibit-exec Retain the executable
output file whenever it is still usable",
and we may want to handle other cases too.

Differential revision: https://reviews.llvm.org/D35793

llvm-svn: 309091

7 years ago[ARM] GlobalISel: Mark G_GLOBAL_VALUE as legal
Diana Picus [Wed, 26 Jul 2017 09:25:15 +0000 (09:25 +0000)]
[ARM] GlobalISel: Mark G_GLOBAL_VALUE as legal

llvm-svn: 309090

7 years ago[ELF] - Print options aliases in --help
George Rimar [Wed, 26 Jul 2017 09:21:10 +0000 (09:21 +0000)]
[ELF] - Print options aliases in --help

This is PR30422,
previously LLD did not render all option aliases in --help.
With this patch it will.

Differential revision: https://reviews.llvm.org/D35477

llvm-svn: 309089

7 years agoUpdate after LLVM change r309087
George Rimar [Wed, 26 Jul 2017 09:10:17 +0000 (09:10 +0000)]
Update after LLVM change r309087

llvm-svn: 309088

7 years ago[libOption] - Add flag allowing to print options aliases in help text.
George Rimar [Wed, 26 Jul 2017 09:09:56 +0000 (09:09 +0000)]
[libOption] - Add flag allowing to print options aliases in help text.

By default, we display only options that are not
hidden and have help texts. This patch adds flag
allowing to display aliases that have no help text.
In this case help text of aliased option used instead.

Differential revision: https://reviews.llvm.org/D35476

llvm-svn: 309087

7 years ago[X86][LLVM]Expanding Supports lowerInterleavedStore() in X86InterleavedAccess.
Michael Zuckerman [Wed, 26 Jul 2017 08:10:14 +0000 (08:10 +0000)]
[X86][LLVM]Expanding Supports lowerInterleavedStore() in X86InterleavedAccess.

This patch expands the support of lowerInterleavedStore to 32x8i stride 4.

LLVM creates suboptimal shuffle code-gen for AVX2. In overall, this patch is a specific fix for the pattern (Strid=4 VF=32) and we plan to include more patterns in the future. To reach our goal of "more patterns". We include two mask creators. The first function creates shuffle's mask equivalent to unpacklo/unpackhi instructions. The other creator creates mask equivalent to a concat of two half vectors(high/low).

The patch goal is to optimize the following sequence:
At the end of the computation, we have ymm2, ymm0, ymm12 and ymm3 holding
each 32 chars:

c0, c1, , c31
m0, m1, , m31
y0, y1, , y31
k0, k1, ., k31

And these need to be transposed/interleaved and stored like so:

c0 m0 y0 k0 c1 m1 y1 k1 c2 m2 y2 k2 c3 m3 y3 k3 ....

Reviewers:
dorit
Farhana
RKSimon
guyblank
DavidKreitzer

Differential Revision: https://reviews.llvm.org/D34601

llvm-svn: 309086

7 years agoTargetLowering: Change isShuffleMaskLegal's mask argument type to ArrayRef<int>....
Zvi Rackover [Wed, 26 Jul 2017 08:06:58 +0000 (08:06 +0000)]
TargetLowering: Change isShuffleMaskLegal's mask argument type to ArrayRef<int>. NFCI.

Changing mask argument type from const SmallVectorImpl<int>& to
ArrayRef<int>.

This came up in D35700 where a mask is received as an ArrayRef<int> and
we want to pass it to TargetLowering::isShuffleMaskLegal().
Also saves a few lines of code.

llvm-svn: 309085

7 years ago[X86][LLVM]Expanding Supports lowerInterleavedStore() in X86InterleavedAccess part1.
Michael Zuckerman [Wed, 26 Jul 2017 07:45:02 +0000 (07:45 +0000)]
[X86][LLVM]Expanding Supports lowerInterleavedStore() in X86InterleavedAccess part1.

splitting patch D34601 into two part. This part changes the location of two functions.
The second part will be based on that patch. This was requested by @RKSimon.

Reviewers:
1. dorit
2. Farhana
3. RKSimon
4. guyblank
5. DavidKreitzer

llvm-svn: 309084

7 years agoRevert "[sanitizer] Support compiler-rt builtins"
Petr Hosek [Wed, 26 Jul 2017 06:46:11 +0000 (06:46 +0000)]
Revert "[sanitizer] Support compiler-rt builtins"

This reverts commit fd63314d6770e0da62572a3fea2c41c4cc0fc58a.

llvm-svn: 309083

7 years agoRevert "[sanitizer] Support libc++abi in addition to libstdc++"
Petr Hosek [Wed, 26 Jul 2017 06:46:10 +0000 (06:46 +0000)]
Revert "[sanitizer] Support libc++abi in addition to libstdc++"

This reverts commit d1997bff31cf6b484eb59c2ee1fc3155442e338c.

llvm-svn: 309082

7 years ago[clang] Add ARM64 support to armintr.h for MSVC compatibility
Mandeep Singh Grang [Wed, 26 Jul 2017 05:29:40 +0000 (05:29 +0000)]
[clang] Add ARM64 support to armintr.h for MSVC compatibility

Summary: This fixes compiling with headers from the Windows SDK for ARM64.

Reviewers: compnerd, ruiu, mstorsjo

Reviewed By: compnerd, mstorsjo

Subscribers: mgorny, aemerson, javed.absar, kristof.beyls, llvm-commits, cfe-commits

Differential Revision: https://reviews.llvm.org/D35862

llvm-svn: 309081

7 years ago[SCEV] Cache results of computeExitLimit
Max Kazantsev [Wed, 26 Jul 2017 04:55:54 +0000 (04:55 +0000)]
[SCEV] Cache results of computeExitLimit

This patch adds a cache for computeExitLimit to save compilation time. A lot of examples of
tests that take extensive time to compile are attached to the bug 33494.

Differential Revision: https://reviews.llvm.org/D35827

llvm-svn: 309080

7 years ago[X86] Prevent selecting masked aligned load instructions if the load should be non...
Craig Topper [Wed, 26 Jul 2017 04:31:04 +0000 (04:31 +0000)]
[X86] Prevent selecting masked aligned load instructions if the load should be non-temporal

Summary: The aligned load predicates don't  suppress themselves if the load is non-temporal the way the unaligned predicates do. For the most part this isn't a problem because the aligned predicates are mostly used for instructions that only load the the non-temporal loads have priority over those. The exception are masked loads.

Reviewers: RKSimon, zvi

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35712

llvm-svn: 309079

7 years agoAdd test coverage for new PM PGOOpt handling.
Dehao Chen [Wed, 26 Jul 2017 02:00:43 +0000 (02:00 +0000)]
Add test coverage for new PM PGOOpt handling.

Summary: This patch adds flags and tests to cover the PGOOpt handling logic in new PM.

Reviewers: chandlerc, davide

Reviewed By: chandlerc

Subscribers: sanjoy, llvm-commits

Differential Revision: https://reviews.llvm.org/D35807

llvm-svn: 309076

7 years ago[gold] Enable data-sections by default for the gold-plugin.
Davide Italiano [Wed, 26 Jul 2017 01:47:17 +0000 (01:47 +0000)]
[gold] Enable data-sections by default for the gold-plugin.

Follow up to r309056.

llvm-svn: 309075

7 years ago[sanitizer] Support libc++abi in addition to libstdc++
Petr Hosek [Wed, 26 Jul 2017 01:43:02 +0000 (01:43 +0000)]
[sanitizer] Support libc++abi in addition to libstdc++

This change adds sanitizer support for LLVM's libunwind and libc++abi
as an alternative to libstdc++. This allows using the in tree version
of libunwind and libc++abi which is useful when building a toolchain
for different target.

Differential Revision: https://reviews.llvm.org/D34501

llvm-svn: 309074

7 years agoAdd "REQUIRES: asserts" for test unswitch-equality-undef.ll.
Wei Mi [Wed, 26 Jul 2017 01:34:46 +0000 (01:34 +0000)]
Add "REQUIRES: asserts" for test unswitch-equality-undef.ll.

llvm-svn: 309073

7 years ago[SCEV] Remove unnecessary call to forgetMemoizedResults
Sanjoy Das [Wed, 26 Jul 2017 01:32:19 +0000 (01:32 +0000)]
[SCEV] Remove unnecessary call to forgetMemoizedResults

`SCEVUnknown::allUsesReplacedWith` does not need to call `forgetMemoizedResults`
since RAUW does a value-equivalent replacement by assumption.  If this
assumption was false then the later setValPtr(New) call would be incorrect too.

This is a non-trivial performance optimization for functions with a large number
of loops since `forgetMemoizedResults` walks all loop backedge taken counts to
see if any of them use the SCEVUnknown being RAUWed.  However, this improvement
is difficult to demonstrate without checking in an excessively large IR file.

llvm-svn: 309072

7 years ago[lit] Attempt to fix Python unittest adaptor logic
Reid Kleckner [Wed, 26 Jul 2017 01:27:18 +0000 (01:27 +0000)]
[lit] Attempt to fix Python unittest adaptor logic

llvm-svn: 309071

7 years agoMove manifest utils into separate lib, to reduce libxml2 deps.
Eric Beckmann [Wed, 26 Jul 2017 01:21:55 +0000 (01:21 +0000)]
Move manifest utils into separate lib, to reduce libxml2 deps.

Summary:
Previously were in support.  Since many many things depend on support,
were all forced to also depend on libxml2, which we only want in a few cases.
This puts all the libxml2 deps in a separate lib to be used only in a few
places.

Reviewers: ruiu, thakis, rnk

Subscribers: mgorny, hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D35819

llvm-svn: 309070

7 years ago[PDB] Remove stale GSI.h header that I intended to remove in the previous commit
Reid Kleckner [Wed, 26 Jul 2017 00:58:49 +0000 (00:58 +0000)]
[PDB] Remove stale GSI.h header that I intended to remove in the previous commit

llvm-svn: 309069

7 years ago[DWARF] Generalized verification of .apple_names accelerator table to be applicable...
Spyridoula Gravani [Wed, 26 Jul 2017 00:52:31 +0000 (00:52 +0000)]
[DWARF] Generalized verification of .apple_names accelerator table to be applicable to any acceleration table. Added verification for .apple_types, .apple_namespaces and .apple_objc sections.

Differential Revision: https://reviews.llvm.org/D35853

llvm-svn: 309068

7 years ago[clang-tidy] Do not issue fixit for explicit template specializations
Felix Berger [Wed, 26 Jul 2017 00:45:41 +0000 (00:45 +0000)]
[clang-tidy] Do not issue fixit for explicit template specializations

Summary:

Do not issue fixit in UnnecessaryValueParamCheck if the function is an explicit template specialization as this could cause build breakages.

Reviewers: alexfh

Subscribers: JDevlieghere, xazax.hun, cfe-commits

Differential Revision: https://reviews.llvm.org/D35718

llvm-svn: 309067

7 years ago[PDB] Improve GSI hash table dumping for publics and globals
Reid Kleckner [Wed, 26 Jul 2017 00:40:36 +0000 (00:40 +0000)]
[PDB] Improve GSI hash table dumping for publics and globals

The PDB "symbol stream" actually contains symbol records for the publics
and the globals stream. The globals and publics streams are essentially
hash tables that point into a single stream of records. In order to
match cvdump's behavior, we need to only dump symbol records referenced
from the hash table. This patch implements that, and then implements
global stream dumping, since it's just a subset of public stream
dumping.

Now we shouldn't see S_PROCREF or S_GDATA32 records when dumping
publics, and instead we should see those record in the globals stream.

llvm-svn: 309066

7 years agoReapply "llvm-mt: implement simple merging of manifests, not factoring namespaces.
Eric Beckmann [Wed, 26 Jul 2017 00:25:12 +0000 (00:25 +0000)]
Reapply "llvm-mt: implement simple merging of manifests, not factoring namespaces.

This time with correct #if.

This reverts commit 9cf4eca0e0383040c1ff1416815c7f649650c2a0.

llvm-svn: 309064

7 years ago[AArch64] Fix some Clang-tidy modernize-use-using and Include What You Use warnings...
Eugene Zelenko [Tue, 25 Jul 2017 23:51:02 +0000 (23:51 +0000)]
[AArch64] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC).

llvm-svn: 309062

7 years ago[StaticAnalyzer] LoopUnrolling - Attempt #2 to fix a crash in r309006.
Peter Szecsi [Tue, 25 Jul 2017 23:49:16 +0000 (23:49 +0000)]
[StaticAnalyzer] LoopUnrolling - Attempt #2 to fix a crash in r309006.

llvm-svn: 309061

7 years ago[sanitizer] Support compiler-rt builtins
Petr Hosek [Tue, 25 Jul 2017 23:38:25 +0000 (23:38 +0000)]
[sanitizer] Support compiler-rt builtins

This change adds support for compiler-rt builtins as an alternative
compiler runtime to libgcc.

Differential Revision: https://reviews.llvm.org/D35165

llvm-svn: 309060

7 years agoDisable loop unswitching for some patterns containing equality comparison with undef.
Wei Mi [Tue, 25 Jul 2017 23:37:17 +0000 (23:37 +0000)]
Disable loop unswitching for some patterns containing equality comparison with undef.

This is a workaround for the bug described in PR31652 and
http://lists.llvm.org/pipermail/llvm-dev/2017-July/115497.html. The temporary
solution is to add a function EqualityPropUnSafe. In EqualityPropUnSafe, for
some simple patterns we can know the equality comparison may contains undef,
so we regard such comparison as unsafe and will not do loop-unswitching for
them. We also need to disable the select simplification when one of select
operand is undef and its result feeds into equality comparison.

The patch cannot clear the safety issue caused by the bug, but it can suppress
the issue from happening to some extent.

Differential Revision: https://reviews.llvm.org/D35811

llvm-svn: 309059

7 years ago[CodeGen] Correctly model std::byte's aliasing properties
David Majnemer [Tue, 25 Jul 2017 23:33:58 +0000 (23:33 +0000)]
[CodeGen] Correctly model std::byte's aliasing properties

std::byte, when defined as an enum, needs to be given special treatment
with regards to its aliasing properties. An array of std::byte is
allowed to be used as storage for other types.

This fixes PR33916.

Differential Revision: https://reviews.llvm.org/D35824

llvm-svn: 309058

7 years agoDebug Info: Support fragmented variables in the MMI side table
Adrian Prantl [Tue, 25 Jul 2017 23:32:59 +0000 (23:32 +0000)]
Debug Info: Support fragmented variables in the MMI side table

This reapplies commit r309034 with a bugfix+test for inlined variables.

llvm-svn: 309057

7 years ago[gold] Enable function-sections by default.
Davide Italiano [Tue, 25 Jul 2017 23:32:50 +0000 (23:32 +0000)]
[gold] Enable function-sections by default.

This is needed, among others, to respect --section-ordering-file
with LTO. I'll follow up with a similar change for data sections.

I hope every version of gold available on the bots has support for
--section-ordering file.

llvm-svn: 309056

7 years agoSimplify ignored options.
Rui Ueyama [Tue, 25 Jul 2017 23:32:05 +0000 (23:32 +0000)]
Simplify ignored options.

Since the flag is ignored anyway, it doesn't matter whether
it is an alias or not.

llvm-svn: 309055

7 years agoReorder tests to match latest SD-6 draft.
Richard Smith [Tue, 25 Jul 2017 23:31:42 +0000 (23:31 +0000)]
Reorder tests to match latest SD-6 draft.

llvm-svn: 309054

7 years agoSimplify. NFC.
Rafael Espindola [Tue, 25 Jul 2017 23:23:40 +0000 (23:23 +0000)]
Simplify. NFC.

llvm-svn: 309053

7 years agoclang/StaticAnalyzer/Core/PathSensitive/LoopUnrolling.h: Add a forward decl AnalysisM...
NAKAMURA Takumi [Tue, 25 Jul 2017 23:23:17 +0000 (23:23 +0000)]
clang/StaticAnalyzer/Core/PathSensitive/LoopUnrolling.h: Add a forward decl AnalysisManager, to unbreak modules build.

llvm-svn: 309052