OSDN Git Service

android-x86/external-llvm.git
7 years ago[Solaris] Fix PR33228 - llvm::sys::fs::is_local_impl done right
Kamil Rytarowski [Thu, 1 Jun 2017 12:57:00 +0000 (12:57 +0000)]
[Solaris] Fix PR33228 - llvm::sys::fs::is_local_impl done right

Summary:
Solaris-specific implementation for llvm::sys::fs::is_local_impl.
FStype pattern matching might be a bit unreliable, but at least it fixes the build failure.

Reviewers: mgorny, nlopes, llvm-commits, krytarowski

Reviewed By: krytarowski

Subscribers: voskresensky.vladimir, krytarowski

Differential Revision: https://reviews.llvm.org/D33695

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304412 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoOnly generate addcarry node when it is legal.
Amaury Sechet [Thu, 1 Jun 2017 12:03:16 +0000 (12:03 +0000)]
Only generate addcarry node when it is legal.

Summary:
This is a problem uncovered by stage2 testing. ADDCARRY end up being generated on target that do not support it.

The patch that introduced the problem has other patches layed on top of it, so we want to fix the issue rather than revert it to avoid creating a lor of churn.

A regression test will be added shortly, but this is committed as this in order to get the build back to green promptly.

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33770

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304409 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM/ThinLTO] Port the ThinLTO pipeline (both components) to the new PM.
Chandler Carruth [Thu, 1 Jun 2017 11:39:39 +0000 (11:39 +0000)]
[PM/ThinLTO] Port the ThinLTO pipeline (both components) to the new PM.

Based on the original patch by Davide, but I've adjusted the API exposed
to just be different entry points rather than exposing more state
parameters. I've factored all the common logic out so that we don't have
any duplicate pipelines, we just stitch them together in different ways.
I think this makes the build easier to reason about and understand.

This adds a direct method for getting the module simplification pipeline
as well as a method to get the optimization pipeline. While not my
express goal, this seems nice and gives a good place comment about the
restrictions that are imposed on them.

I did make some minor changes to the way the pipelines are structured
here, but hopefully not ones that are significant or controversial:

1) I sunk the PGO indirect call promotion to only be run when we have
   PGO enabled (or as part of the special ThinLTO pipeline).

2) I made the extra GlobalOpt run in ThinLTO just happen all the time
   and at a slightly more powerful place (before we remove available
   externaly functions). This seems like general goodness and not a big
   compile time sink, so it didn't make sense to *only* use it in
   ThinLTO. Fewer differences in the pipeline makes everything simpler
   IMO.

3) I hoisted the ThinLTO stop point pre-link above the the RPO function
   attr inference. The RPO inference won't infer anything terribly
   meaningful pre-link (recursiveness?) so it didn't make a lot of
   sense. But if the placement of RPO inference starts to matter, we
   should move it to the canonicalization phase anyways which seems like
   a better place for it (and there is a FIXME to this effect!). But
   that seemed a bridge too far for this patch.

If we ever need to parameterize these pipelines more heavily, we can
always sink the logic to helper functions with parameters to keep those
parameters out of the public API. But the changes above seemed minor
that we could possible get away without the parameters entirely.

I added support for parsing 'thinlto' and 'thinlto-pre-link' names in
pass pipelines to make it easy to test these routines and play with them
in larger pipelines. I also added a really basic manifest of passes test
that will show exactly how the pipelines behave and work as well as
making updates to them clear.

Lastly, this factoring does introduce a nesting layer of module pass
managers in the default pipeline. I don't think this is a big deal and
the flexibility of decoupling the pipelines seems easily worth it.

Differential Revision: https://reviews.llvm.org/D33540

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304407 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Match bitcast of vxi1 to pmovmsk
Zvi Rackover [Thu, 1 Jun 2017 11:27:57 +0000 (11:27 +0000)]
[X86] Match bitcast of vxi1 to pmovmsk

Summary:
Add an early combine to match patterns such as:
  (i16 bitcast (v16i1 x))
  ->
  (i16 movmsk (v16i8 sext (v16i1 x)))

This combine needs to happen early enough before
type-legalization scalarizes the result of the setcc.

Reviewers: igorb, craig.topper, RKSimon

Subscribers: delena, llvm-commits

Differential Revision: https://reviews.llvm.org/D33311

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304406 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDo not legalize large setcc with setcce, introduce setcccarry and do it with usubo...
Amaury Sechet [Thu, 1 Jun 2017 11:14:17 +0000 (11:14 +0000)]
Do not legalize large setcc with setcce, introduce setcccarry and do it with usubo/setcccarry.

Summary:
This is a continuation of the work started in D29872 . Passing the carry down as a value rather than as a glue allows for further optimizations. Introducing setcccarry makes the use of addc/subc unecessary and we can start the removal process.

This patch only introduce the optimization strictly required to get the same level of optimization as was available before nothing more.

Reviewers: jyknight, nemanjai, mkuper, spatel, RKSimon, zvi, bkramer

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33374

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304404 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove ISD::SETCC match from combineX86ADD. It's done improperly and doesn't work.
Amaury Sechet [Thu, 1 Jun 2017 11:13:10 +0000 (11:13 +0000)]
Remove ISD::SETCC match from combineX86ADD. It's done improperly and doesn't work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304403 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAGCombine] Refactor common addcarry pattern.
Amaury Sechet [Thu, 1 Jun 2017 10:48:04 +0000 (10:48 +0000)]
[DAGCombine] Refactor common addcarry pattern.

Summary: This pattern is no very useful per se, but it exposes optimization for toehr patterns that wouldn't kick in otherwize. It's very common and worth optimizing for.

Reviewers: jyknight, nemanjai, mkuper, spatel, RKSimon, zvi, bkramer

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32756

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304402 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAGCombine] (add/uaddo X, Carry) -> (addcarry X, 0, Carry)
Amaury Sechet [Thu, 1 Jun 2017 10:42:39 +0000 (10:42 +0000)]
[DAGCombine] (add/uaddo X, Carry) -> (addcarry X, 0, Carry)

Summary:
This enables further transforms.

Depends on D32916

Reviewers: jyknight, nemanjai, mkuper, spatel, RKSimon, zvi, bkramer

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32925

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304401 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake mcpu=generic the default for armv7-a and armv8-a.
Kristof Beyls [Thu, 1 Jun 2017 07:31:43 +0000 (07:31 +0000)]
Make mcpu=generic the default for armv7-a and armv8-a.

As discussed in
http://lists.llvm.org/pipermail/llvm-dev/2017-May/113525.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304390 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Remove RecordVal constructor that takes a StringRef and Record::setName...
Craig Topper [Thu, 1 Jun 2017 06:56:16 +0000 (06:56 +0000)]
[TableGen] Remove RecordVal constructor that takes a StringRef and Record::setName(StringRef). Leave just the versions that take an Init.

They weren't used often enough to justify having two different interfaces. Push the responsiblity of creating a StringInit up to the caller.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304388 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Remove code for renaming anonymous register classes as it can never execute.
Craig Topper [Thu, 1 Jun 2017 06:56:13 +0000 (06:56 +0000)]
[TableGen] Remove code for renaming anonymous register classes as it can never execute.

It tried to detect 9 letters (the length of anonymous) followed by a period. But anonymous classes start with "anonymous_" rather than "anonymous." these days.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304387 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Use StringRef to capture getValueAsString in a couple more places. NFC
Craig Topper [Thu, 1 Jun 2017 06:56:11 +0000 (06:56 +0000)]
[TableGen] Use StringRef to capture getValueAsString in a couple more places. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304386 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ThinLTO] Migrate ThinLTOBitcodeWriter to the new PM.
Tim Shen [Thu, 1 Jun 2017 01:02:12 +0000 (01:02 +0000)]
[ThinLTO] Migrate ThinLTOBitcodeWriter to the new PM.

Summary: Also see D33429 for other ThinLTO + New PM related changes.

Reviewers: davide, chandlerc, tejohnson

Subscribers: mehdi_amini, Prazek, cfe-commits, inglorion, llvm-commits, eraman

Differential Revision: https://reviews.llvm.org/D33525

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304378 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PartialInlining] Reduce outlining overhead by removing unneeded live-out(s)
Xinliang David Li [Thu, 1 Jun 2017 00:12:41 +0000 (00:12 +0000)]
[PartialInlining] Reduce outlining overhead by removing unneeded live-out(s)

Differential Revision: http://reviews.llvm.org/D33694

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304375 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd LiveRangeShrink pass to shrink live range within BB.
Dehao Chen [Wed, 31 May 2017 23:25:25 +0000 (23:25 +0000)]
Add LiveRangeShrink pass to shrink live range within BB.

Summary: LiveRangeShrink pass moves instruction right after the definition with the same BB if the instruction and its operands all have more than one use. This pass is inexpensive and guarantees optimal live-range within BB.

Reviewers: davidxl, wmi, hfinkel, MatzeB, andreadb

Reviewed By: MatzeB, andreadb

Subscribers: hiraditya, jyknight, sanjoy, skatkov, gberry, jholewinski, qcolombet, javed.absar, krytarowski, atrick, spatel, RKSimon, andreadb, MatzeB, mehdi_amini, mgorny, efriedma, davide, dberlin, llvm-commits

Differential Revision: https://reviews.llvm.org/D32563

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304371 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[docs] Update name of vectorization interleave flag.
Eli Friedman [Wed, 31 May 2017 23:02:55 +0000 (23:02 +0000)]
[docs] Update name of vectorization interleave flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304370 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[EH] Recognize __(gxx|gcc)_personality_seh0 as the GNU EH personalities
Reid Kleckner [Wed, 31 May 2017 22:35:52 +0000 (22:35 +0000)]
[EH] Recognize __(gxx|gcc)_personality_seh0 as the GNU EH personalities

These are no-ops when there are no invokes. We don't need to emit LSDAs
for them.

Fixes PR33220.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304367 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoImplicitNullChecks: Clear kill/dead flags when moving instructions around
Matthias Braun [Wed, 31 May 2017 22:23:08 +0000 (22:23 +0000)]
ImplicitNullChecks: Clear kill/dead flags when moving instructions around

The values are marked as livein in the successor blocks so marking them
as killed or dead was wrong.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304366 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCheck hasPersonalityFn before calling getPersonalityFn
Reid Kleckner [Wed, 31 May 2017 22:21:20 +0000 (22:21 +0000)]
Check hasPersonalityFn before calling getPersonalityFn

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304365 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[EH] Fix the LSDA that we emit for unknown EH personalities
Reid Kleckner [Wed, 31 May 2017 22:18:49 +0000 (22:18 +0000)]
[EH] Fix the LSDA that we emit for unknown EH personalities

We should have a single call site entry with no landing pad. This
indicates that no EH action should be taken and the unwinder should
unwind to the next frame.

We currently don't recognize __gxx_personality_seh0 as a known
personality, so we forcibly emit a table, and that table was wrong. This
was filed as PR33220. Now we emit a correct table for that personality.
The next step is to recognize that we can completely skip the table for
this personality.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304363 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[MachOObject] Fix bind opcode parser error on valid opcode sequence
Steven Wu [Wed, 31 May 2017 22:17:43 +0000 (22:17 +0000)]
[MachOObject] Fix bind opcode parser error on valid opcode sequence

BIND_OPCODE_SET_DYLIB_SPECIAL_IMM(0) is a valid way to setp library
ordinal. MachOObject should set LibraryOrdinalSet even when IMM is zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304362 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 22:16:24 +0000 (22:16 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304361 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 22:09:46 +0000 (22:09 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304358 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 22:02:05 +0000 (22:02 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304356 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 21:50:45 +0000 (21:50 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304355 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAvoid a UB pointer overflow in the ArrayRef unit test
Vedant Kumar [Wed, 31 May 2017 21:47:52 +0000 (21:47 +0000)]
Avoid a UB pointer overflow in the ArrayRef unit test

The intent of the test is to check that array lengths greater than
UINT_MAX work properly. Change the test to stress that scenario, without
triggering pointer overflow UB.

Caught by a WIP pointer overflow checker in clang.

Differential Revision: https://reviews.llvm.org/D33149

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304353 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert rL304050. It may break sanitizer bootstrap. Revert it for now while investigating.
Wei Mi [Wed, 31 May 2017 21:29:33 +0000 (21:29 +0000)]
Revert rL304050. It may break sanitizer bootstrap. Revert it for now while investigating.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304350 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTry to fix buildbots
Matthias Braun [Wed, 31 May 2017 21:25:03 +0000 (21:25 +0000)]
Try to fix buildbots

It seems not all of our bots have a std::vector::erase() taking a
const_iterator (even though that seems to be part of C++11) attempt to
workaround.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304349 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Adapt more places to getValueAsString now returning a StringRef instead...
Craig Topper [Wed, 31 May 2017 21:12:46 +0000 (21:12 +0000)]
[TableGen] Adapt more places to getValueAsString now returning a StringRef instead of a std::string.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304347 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoX86FloatingPoint: Fix livein lists
Matthias Braun [Wed, 31 May 2017 20:30:22 +0000 (20:30 +0000)]
X86FloatingPoint: Fix livein lists

After transforming FP to ST registers:
- Do not add the ST register to the livein lists, they are reserved so
  we do not need to track their liveness.
- Remove the FP registers from the livein lists, they don't have defs or
  uses anymore and so are not live.
- (The setKillFlags() call is moved to an earlier place as it relies on
   the FP registers still being present in the livein list.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304342 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoX86FloatingPoint: Add some static assert, cleanup; NFC
Matthias Braun [Wed, 31 May 2017 20:30:17 +0000 (20:30 +0000)]
X86FloatingPoint: Add some static assert, cleanup; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304341 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded missing break; added LLVM_FALLTHROUGH to address warning: this statement may...
Galina Kistanova [Wed, 31 May 2017 20:25:13 +0000 (20:25 +0000)]
Added missing break; added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304340 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[libFuzzer] fix a test to match the new sanitizer run-time
Kostya Serebryany [Wed, 31 May 2017 19:47:11 +0000 (19:47 +0000)]
[libFuzzer] fix a test to match the new sanitizer run-time

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304333 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 19:41:33 +0000 (19:41 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304332 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[IR] Add additional addParamAttr/removeParamAttr to AttributeList API
Reid Kleckner [Wed, 31 May 2017 19:23:09 +0000 (19:23 +0000)]
[IR] Add additional addParamAttr/removeParamAttr to AttributeList API

Summary:
Fairly straightforward patch to fill in some of the holes in the
attributes API with respect to accessing parameter/argument attributes.
The patch aims to step further towards encapsulating the
idx+FirstArgIndex pattern to access these attributes to within the
AttributeList.

Patch by Daniel Neilson!

Reviewers: rnk, chandlerc, pete, javed.absar, reames

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33355

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304329 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Make Record::getValueAsString and getValueAsListOfStrings return StringRef...
Craig Topper [Wed, 31 May 2017 19:01:11 +0000 (19:01 +0000)]
[TableGen] Make Record::getValueAsString and getValueAsListOfStrings return StringRefs instead of std::string

Internally both these methods just return the result of getValue on either a StringInit or a CodeInit object. In both cases this returns a StringRef pointing to a string allocated in the BumpPtrAllocator so its not going anywhere. So we can just pass that StringRef along.

This is a fairly naive patch that targets just the build failures caused by this change. There's additional work that can be done to avoid creating std::string at call sites that still think getValueAsString returns a std::string. I'll try to clean those up in future patches.

Differential Revision: https://reviews.llvm.org/D33710

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304325 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[BPF] Correct the file name of the -gen-asm-matcher output file to not start with...
Craig Topper [Wed, 31 May 2017 19:01:05 +0000 (19:01 +0000)]
[BPF] Correct the file name of the -gen-asm-matcher output file to not start with X86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304324 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ThinLTO] Reduce unnecessary map lookups during combined summary write
Teresa Johnson [Wed, 31 May 2017 18:58:11 +0000 (18:58 +0000)]
[ThinLTO] Reduce unnecessary map lookups during combined summary write

Summary:
Don't assign values to undefined references, simply don't emit those
reference edges as they are not useful (we were already not emitting
call edges to undefined refs).

Also, streamline the later lookup of value ids when writing the
summaries, by combining the check for value id existence with the access
of that value id.

Reviewers: pcc

Subscribers: Prazek, llvm-commits, inglorion

Differential Revision: https://reviews.llvm.org/D33634

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304323 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ScheduleDAG] Deal with already scheduled loads in ScheduleDAG.
Nirav Dave [Wed, 31 May 2017 18:43:17 +0000 (18:43 +0000)]
[ScheduleDAG] Deal with already scheduled loads in ScheduleDAG.

Summary:
If we attempt to unfold an SUnit in ScheduleDAG that results in
finding an already scheduled load, we must should abort the
unfold as it will not improve scheduling.

This fixes PR32610.

Reviewers: jmolloy, sunfish, bogner, spatel

Subscribers: llvm-commits, MatzeB

Differential Revision: https://reviews.llvm.org/D32911

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304321 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTargetMachine: Indicate whether machine verifier passes.
Matthias Braun [Wed, 31 May 2017 18:41:23 +0000 (18:41 +0000)]
TargetMachine: Indicate whether machine verifier passes.

This adds a callback to the LLVMTargetMachine that lets target indicate
that they do not pass the machine verifier checks in all cases yet.

This is intended to be a temporary measure while the targets are fixed
allowing us to enable the machine verifier by default with
EXPENSIVE_CHECKS enabled!

Differential Revision: https://reviews.llvm.org/D33696

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304320 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[sanitizer-coverage] remove stale code (old coverage); llvm part
Kostya Serebryany [Wed, 31 May 2017 18:27:33 +0000 (18:27 +0000)]
[sanitizer-coverage] remove stale code (old coverage); llvm  part

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304319 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Correctly specify the cache line size for Power 7, 8 and 9.
Sean Fertile [Wed, 31 May 2017 18:20:17 +0000 (18:20 +0000)]
[PowerPC] Correctly specify the cache line size for Power 7, 8 and 9.

Fixes PPCTTIImpl::getCacheLineSize() returning the wrong cache line size for
newer ppc processors.

Commiting on behalf of Stefan Pintilie.
Differential Revision: https://reviews.llvm.org/D33656

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304317 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "[Atomics][LoopIdiom] Recognize unordered atomic memcpy"
Anna Thomas [Wed, 31 May 2017 17:20:51 +0000 (17:20 +0000)]
Revert "[Atomics][LoopIdiom] Recognize unordered atomic memcpy"

This reverts commit r304310.

It caused build failures in polly and mingw
due to undefined reference to
llvm::RTLIB::getMEMCPY_ELEMENT_ATOMIC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304315 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PPC] Inline expansion of memcmp
Zaara Syeda [Wed, 31 May 2017 17:12:38 +0000 (17:12 +0000)]
[PPC] Inline expansion of memcmp

This patch does an inline expansion of memcmp.
It changes the memcmp library call into an inline expansion when the size is
known at compile time and is under a target specified threshold.
This expansion is implemented in CodeGenPrepare and expands into straight line
code. The target specifies a maximum load size and the expansion works by using
this size to load the two sources, compare, and exit early if a difference is
found. It also has a special case when the memcmp result is used in a compare
to zero equality.

Differential Revision: https://reviews.llvm.org/D28637

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304313 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 17:10:03 +0000 (17:10 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304312 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AMDGPU] Fix bugs in new waitcnt pass. Add test.
Mark Searles [Wed, 31 May 2017 16:44:23 +0000 (16:44 +0000)]
[AMDGPU] Fix bugs in new waitcnt pass. Add test.
- new waitcnt pass remains off by default; -enable-si-insert-waitcnts=1 to enable it
- fix handling of PERMUTE ops
- fix insertion of waitcnt instrs at function begin/end ( port of analogous code that was added to old waitcnt pass )
- add new test

  Differential Revision: https://reviews.llvm.org/D33114

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304311 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Atomics][LoopIdiom] Recognize unordered atomic memcpy
Anna Thomas [Wed, 31 May 2017 16:39:52 +0000 (16:39 +0000)]
[Atomics][LoopIdiom] Recognize unordered atomic memcpy

Summary:
Expanding the loop idiom test for memcpy to also recognize unordered atomic memcpy.
The only difference for recognizing
an unordered atomic memcpy and instead of a normal memcpy is
that the loads and/or stores involved are unordered atomic operations.
Background:  http://lists.llvm.org/pipermail/llvm-dev/2017-May/112779.html

Patch by Daniel Neilson!

Reviewers: reames, anna, skatkov

Reviewed By: reames

Subscribers: llvm-commits, mzolotukhin

Differential Revision: https://reviews.llvm.org/D33243

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304310 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AMDGPU][MC] New syntax for ds_swizzle_b32 offset
Dmitry Preobrazhensky [Wed, 31 May 2017 16:26:47 +0000 (16:26 +0000)]
[AMDGPU][MC] New syntax for ds_swizzle_b32 offset

See Bug 28601: https://bugs.llvm.org//show_bug.cgi?id=28601

Reviewers: artem.tamazov, vpykhtin

Differential Revision: https://reviews.llvm.org/D33542

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304309 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64] Enable FeatureFuseAES on Cortex-A53.
Florian Hahn [Wed, 31 May 2017 15:50:03 +0000 (15:50 +0000)]
[AArch64] Enable FeatureFuseAES on Cortex-A53.

It improves performance on Cortex-A53.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304307 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64] Enable FeatureFuseAES on Cortex-A73.
Florian Hahn [Wed, 31 May 2017 15:25:25 +0000 (15:25 +0000)]
[AArch64] Enable FeatureFuseAES on Cortex-A73.

It improves performance on Cortex-A73.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304304 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix assertion when merging multiple empty AttributeLists
Reid Kleckner [Wed, 31 May 2017 14:24:06 +0000 (14:24 +0000)]
Fix assertion when merging multiple empty AttributeLists

Patch by Nicholas Wilson!

Differential Revision: https://reviews.llvm.org/D33627

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304300 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAG] Avoid use of stale store.
Nirav Dave [Wed, 31 May 2017 13:36:17 +0000 (13:36 +0000)]
[DAG] Avoid use of stale store.

Correct references to alignment of store which may be deleted in a
previous iteration of merge. Instead use first store that would be
merged.

Corrects pr33172's use-after-poison caught by ASan.

Reviewers: spatel, hfinkel, RKSimon

Reviewed By: RKSimon

Subscribers: thegameg, javed.absar, llvm-commits

Differential Revision: https://reviews.llvm.org/D33686

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304299 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Fix a performance bug for PPC::XXPERMDI.
Tony Jiang [Wed, 31 May 2017 13:09:57 +0000 (13:09 +0000)]
[PowerPC] Fix a performance bug for PPC::XXPERMDI.

There are some VectorShuffle Nodes in SDAG which can be selected to XXPERMDI
Instruction, this patch recognizes them and does the selection to improve
the PPC performance.

Differential Revision: https://reviews.llvm.org/D33404

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304298 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRegenerate xchg-nofold.ll expected results. NFC.
Amaury Sechet [Wed, 31 May 2017 09:44:08 +0000 (09:44 +0000)]
Regenerate xchg-nofold.ll expected results. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304291 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Eliminate integer compare instructions - vol. 3
Nemanja Ivanovic [Wed, 31 May 2017 08:04:07 +0000 (08:04 +0000)]
[PowerPC] Eliminate integer compare instructions - vol. 3

This patch builds upon https://reviews.llvm.org/rL302810 to add
handling for the 64-bit SETEQ patterns.

Differential Revision: https://reviews.llvm.org/D33369

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304286 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVR] Fix a big in shift operator lowering; Authored by Dr. Gergo Erdi
Dylan McKay [Wed, 31 May 2017 06:27:46 +0000 (06:27 +0000)]
[AVR] Fix a big in shift operator lowering; Authored by Dr. Gergo Erdi

When generating code for a shift loop, check the shift
 amount against the literal value 0, not R0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304284 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVR] CPIRdK can only work with r16..r31; Authored by Dr. Gergo Erdi
Dylan McKay [Wed, 31 May 2017 06:10:59 +0000 (06:10 +0000)]
[AVR] CPIRdK can only work with r16..r31; Authored by Dr. Gergo Erdi
(https://github.com/avr-rust/rust/issues/50)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304283 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Eliminate integer compare instructions - vol. 2
Nemanja Ivanovic [Wed, 31 May 2017 05:40:25 +0000 (05:40 +0000)]
[PowerPC] Eliminate integer compare instructions - vol. 2

This patch builds upon https://reviews.llvm.org/rL302810 to add
handling for bitwise logical operations in general purpose registers.
The idea is to keep the values in GPRs as long as possible - only
extracting them to a condition register bit when no further operations
are to be done.

Differential Revision: https://reviews.llvm.org/D31851

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304282 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Implement non-const versions of Record::getValue by delegating to the...
Craig Topper [Wed, 31 May 2017 05:12:36 +0000 (05:12 +0000)]
[TableGen] Implement non-const versions of Record::getValue by delegating to the const versions to avoid duplicate code. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304281 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Make one of RecordVal's constructors delegate to the other to reduce dupli...
Craig Topper [Wed, 31 May 2017 05:12:33 +0000 (05:12 +0000)]
[TableGen] Make one of RecordVal's constructors delegate to the other to reduce duplicate code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304280 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ObjectYAML] Split CodeViewYAML into 3 pieces.
Zachary Turner [Wed, 31 May 2017 04:17:13 +0000 (04:17 +0000)]
[ObjectYAML] Split CodeViewYAML into 3 pieces.

The code was a mess and disorganized due to the sheer amount
of it being in one file.  So I'm splitting this into three files.
One for CodeView types, one for CodeView symbols, and one for
CodeView debug subsections.  NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304278 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[coroutines] Call initializePass in coroutine pass constructors
Gor Nishanov [Wed, 31 May 2017 03:12:42 +0000 (03:12 +0000)]
[coroutines] Call initializePass in coroutine pass constructors

Summary:

Fixes: https://bugs.llvm.org/show_bug.cgi?id=33226

Reviewers: chandlerc, davide, majnemer, dblaikie

Reviewed By: chandlerc

Subscribers: EricWF, llvm-commits

Differential Revision: https://reviews.llvm.org/D33701

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304277 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CFLAA] Add missing break; note things are broken.
George Burgess IV [Wed, 31 May 2017 02:35:26 +0000 (02:35 +0000)]
[CFLAA] Add missing break; note things are broken.

Thanks to Galina Kistanova for finding the missing break!

When trying to make a test for this, I realized our logic for handling
extractvalue/insertvalue/... is somewhat broken. This makes constructing
a test-case for this missing break nontrivial.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304275 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoX86FrameLowering: No need to mark FP as live-in everywhere
Matthias Braun [Wed, 31 May 2017 02:11:10 +0000 (02:11 +0000)]
X86FrameLowering: No need to mark FP as live-in everywhere

The frame pointer (when used as frame pointer) is a reserved register.
We do not track liveness of reserved registers and hence do not need to
add them to the basic block livein lists.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304274 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 01:54:18 +0000 (01:54 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304273 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoNewGVN: Fix PR 33185 by checking whether we need to recursively
Daniel Berlin [Wed, 31 May 2017 01:47:32 +0000 (01:47 +0000)]
NewGVN: Fix PR 33185 by checking whether we need to recursively
generate a phi of ops, which we don't currently support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304272 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix test that wasn't update_test_check'd
Daniel Berlin [Wed, 31 May 2017 01:47:29 +0000 (01:47 +0000)]
Fix test that wasn't update_test_check'd

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304271 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoInstructionSimplify: Remove now-redundant reachability tests, as dominates() already...
Daniel Berlin [Wed, 31 May 2017 01:47:24 +0000 (01:47 +0000)]
InstructionSimplify: Remove now-redundant reachability tests, as dominates() already does them

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304270 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMark a test as requiring a default triple
Vedant Kumar [Wed, 31 May 2017 01:42:55 +0000 (01:42 +0000)]
Mark a test as requiring a default triple

This test assumes that llc can infer a default triple. I'm not sure why
exactly, but the Verify MachineInstrs bot requires tests to be explicit
about this dependency.

This commit follows the lead from r248452 and adds in 'REQUIRES:
default_triple' to omit-empty.ll.

Bot URL: http://lab.llvm.org:8080/green/job/Verify-Machineinstrs_AArch64/7500

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304269 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.
Galina Kistanova [Wed, 31 May 2017 01:33:39 +0000 (01:33 +0000)]
Added LLVM_FALLTHROUGH to address warning: this statement may fall through. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304268 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoARM: Fix cmpxchg O0 expansion
Matthias Braun [Wed, 31 May 2017 01:21:35 +0000 (01:21 +0000)]
ARM: Fix cmpxchg O0 expansion

This is the equivalent of r304048 for ARM:

- Rewrite livein calculation to use the computeLiveIns() helper
  function. This is slightly less efficient but easier to reason about
  and doesn't unnecessarily add pristine and reserved registers[1]
- Zero the status register at the beginning of the loop to make sure it
  has a defined value.
- Remove kill flags of values that need to stay alive throughout the loop.

[1] An upcoming commit of mine will tighten the MachineVerifier to catch
    these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304267 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoARM: Do not add reserved registers to block livein lists; NFC
Matthias Braun [Wed, 31 May 2017 01:21:30 +0000 (01:21 +0000)]
ARM: Do not add reserved registers to block livein lists; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304266 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CodeGen] Fix some Clang-tidy modernize-use-using and Include What You Use warnings...
Eugene Zelenko [Wed, 31 May 2017 01:10:10 +0000 (01:10 +0000)]
[CodeGen] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304265 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix CodeView-related modules build failures post-r304248
Vedant Kumar [Wed, 31 May 2017 01:08:43 +0000 (01:08 +0000)]
Fix CodeView-related modules build failures post-r304248

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304264 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ObjectYAML] Clean up the CodeView headers a bit.
Zachary Turner [Wed, 31 May 2017 01:08:36 +0000 (01:08 +0000)]
[ObjectYAML] Clean up the CodeView headers a bit.

CodeViewYAML.h attempts to hide the details of many of the
CodeView yaml structures and types, but at the same time it
exposes the mapping traits for them to external users of the
header.

This patch just hides these in the implementation files so that
the interface is kept as simple as possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304263 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix misspelling
Alina Sbirlea [Wed, 31 May 2017 01:00:51 +0000 (01:00 +0000)]
Fix misspelling

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304262 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd latency info for Exynos interleaved Load/Store instructions.
Abderrazek Zaafrani [Wed, 31 May 2017 00:20:55 +0000 (00:20 +0000)]
Add latency info for Exynos interleaved Load/Store instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304259 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTry to fix build again.
Zachary Turner [Tue, 30 May 2017 23:57:46 +0000 (23:57 +0000)]
Try to fix build again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304257 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CodeView] Move CodeView symbol yaml logic to ObjectYAML.
Zachary Turner [Tue, 30 May 2017 23:50:44 +0000 (23:50 +0000)]
[CodeView] Move CodeView symbol yaml logic to ObjectYAML.

This continues the effort to get the CodeView YAML parsing logic
into ObjectYAML.  After this patch, the only missing piece will
be the CodeView debug symbol subsections.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304256 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix bug on Big-Endian system, due to reference to vector out of scope.
Eric Beckmann [Tue, 30 May 2017 23:10:57 +0000 (23:10 +0000)]
Fix bug on Big-Endian system, due to reference to vector out of scope.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304255 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMachineInstr: Do not skip dead def operands when printing.
Matthias Braun [Tue, 30 May 2017 23:09:21 +0000 (23:09 +0000)]
MachineInstr: Do not skip dead def operands when printing.

This was introduced a long time ago in r86583 when regmask operands
didn't exist. Nowadays the behavior hurts more than it helps. This
removes it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304254 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoThis patch should fix various clang warnings and a use of to_string
Eric Beckmann [Tue, 30 May 2017 22:29:06 +0000 (22:29 +0000)]
This patch should fix various clang warnings and a use of to_string
which isn't support before c++11.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304252 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AntiDepBreaker] Revert r299124 and add a test.
Tim Shen [Tue, 30 May 2017 22:26:52 +0000 (22:26 +0000)]
[AntiDepBreaker] Revert r299124 and add a test.

Summary:
AntiDepBreaker intends to add all live-outs, including the implicit
CSRs, in StartBlock. r299124 was done without understanding that
intention.

Now with the live-ins propagated correctly (D32464), we can revert this change.

Reviewers: MatzeB, qcolombet

Subscribers: nemanjai, llvm-commits

Differential Revision: https://reviews.llvm.org/D33697

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304251 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMIR: update test for noVRegs removal.
Tim Northover [Tue, 30 May 2017 22:02:19 +0000 (22:02 +0000)]
MIR: update test for noVRegs removal.

I think I hadn't git pulled recently enough to bring it in.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304250 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTry to fix build.
Zachary Turner [Tue, 30 May 2017 22:00:37 +0000 (22:00 +0000)]
Try to fix build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304249 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CodeView] Move CodeView YAML code to ObjectYAML.
Zachary Turner [Tue, 30 May 2017 21:53:05 +0000 (21:53 +0000)]
[CodeView] Move CodeView YAML code to ObjectYAML.

This is the beginning of an effort to move the codeview yaml
reader / writer into ObjectYAML so that it can be shared.
Currently the only consumer / producer of CodeView YAML is
llvm-pdbdump, but CodeView can exist outside of PDB files, and
indeed is put into object files and passed to the linker to
produce PDB files.  Furthermore, there are subtle differences
in the types of records that show up in object file CodeView
vs PDB file CodeView, but they are otherwise 99% the same.

By having this code in ObjectYAML, we can have llvm-pdbdump
reuse this code, while teaching obj2yaml and yaml2obj to use
this syntax for dealing with object files that can contain
CodeView.

This patch only adds support for CodeView type information
to ObjectYAML.  Subsequent patches will add support for
CodeView symbol information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304248 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTargetPassConfig: Keep a reference to an LLVMTargetMachine; NFC
Matthias Braun [Tue, 30 May 2017 21:36:41 +0000 (21:36 +0000)]
TargetPassConfig: Keep a reference to an LLVMTargetMachine; NFC

TargetPassConfig is not useful for targets that do not use the CodeGen
library, so we may just as well store a pointer to an
LLVMTargetMachine instead of just to a TargetMachine.

While at it, also change the constructor to take a reference instead of a
pointer as the TM must not be nullptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304247 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMIR: remove explicit "noVRegs" property.
Tim Northover [Tue, 30 May 2017 21:28:57 +0000 (21:28 +0000)]
MIR: remove explicit "noVRegs" property.

We can infer this from the incoming MIR, so there's no reason to
represent it with a special flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304246 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PartialInlining] Shrinkwrap allocas with live range contained in outline region.
Xinliang David Li [Tue, 30 May 2017 21:22:18 +0000 (21:22 +0000)]
[PartialInlining] Shrinkwrap allocas with live range contained in outline region.

Differential Revision: http://reviews.llvm.org/D33618

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304245 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Localizer] Don't trick to be smart for the insertion point
Quentin Colombet [Tue, 30 May 2017 20:53:06 +0000 (20:53 +0000)]
[Localizer] Don't trick to be smart for the insertion point

There is no guarantee that the first use of a constant that is traversed
is actually the first in the related basic block. Thus, if we use that
as the insertion point we may end up with definitions that don't
dominate there use.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304244 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[llvm-config] Fix cflags test looking for "warning"
Ben Langmuir [Tue, 30 May 2017 20:21:47 +0000 (20:21 +0000)]
[llvm-config] Fix cflags test looking for "warning"

This will fail if you configure with e.g. -Wno-unknown-warning-option.
Change it to check for 'warning:' just like we did for 'error:' in
r289484.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304239 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LV] Reapply r303763 with fix for PR33193
Matthew Simpson [Tue, 30 May 2017 19:55:57 +0000 (19:55 +0000)]
[LV] Reapply r303763 with fix for PR33193

r303763 caused build failures in some out-of-tree tests due to an assertion in
TTI. The original patch updated cost estimates for induction variable update
instructions marked for scalarization. However, it didn't consider that the
incoming value of an induction variable phi node could be a cast instruction.
This caused queries for cast instruction costs with a mix of vector and scalar
types. This patch includes a fix for cast instructions and the test case from
PR33193.

The fix was suggested by Jonas Paulsson <paulsson@linux.vnet.ibm.com>.

Reference: https://bugs.llvm.org/show_bug.cgi?id=33193
Original Differential Revision: https://reviews.llvm.org/D33457

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304235 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Object] Remove unused field + constructor.
Benjamin Kramer [Tue, 30 May 2017 19:37:02 +0000 (19:37 +0000)]
[Object] Remove unused field + constructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304233 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Object] Fix pessimizing move.
Benjamin Kramer [Tue, 30 May 2017 19:36:58 +0000 (19:36 +0000)]
[Object] Fix pessimizing move.

Returning the Error by value triggers copy elision, the move is more
expensive. Clang rightfully warns about it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304232 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "This patch closes PR28513: an optimization of multiplication by different...
Vedant Kumar [Tue, 30 May 2017 19:25:22 +0000 (19:25 +0000)]
Revert "This patch closes PR28513: an optimization of multiplication by different constants. It's implemented on DAG combiner level."

This reverts commit r304209.

I think this change is responsible for a tablgen failure in stage2 builds:

  http://green.lab.llvm.org/green/job/clang-stage2-configure-Rthinlto_build/2171/

I reproduced the failure locally (without ThinLTO), reverted the commit, rebuilt the stage1 clang, rebuilt the stage2 llvm-tblgen tool, and found that the crash disappears when the commit is reverted. Here is the stack trace:

FAILED: lib/Target/ARM/ARMGenRegisterBank.inc.tmp
cd /Volumes/Builds/pz-master-stage2-RA/lib/Target/ARM && /Volumes/Builds/pz-master-stage2-RA/bin/llvm-tblgen -gen-register-bank -I /Users/vk/llvm/lib/Target/ARM -I /Users/vk/llvm/include -I /Users/vk/llvm/lib/Target /Users/vk/llvm/lib/Target/ARM/ARM.td -o /Volumes
/Builds/pz-master-stage2-RA/lib/Target/ARM/ARMGenRegisterBank.inc.tmp
0  llvm-tblgen              0x0000000106fc9568 llvm::sys::PrintStackTrace(llvm::raw_ostream&) + 40
1  llvm-tblgen              0x0000000106fc9be6 SignalHandler(int) + 422
2  libsystem_platform.dylib 0x00000001076a7fba _sigtramp + 26
3  libsystem_platform.dylib 0x00007fff58deb468 _sigtramp + 1366570184
4  llvm-tblgen              0x0000000106e89cc7 llvm::CodeGenRegBank::getCompositeSubRegIndex(llvm::CodeGenSubRegIndex*, llvm::CodeGenSubRegIndex*) + 615
5  llvm-tblgen              0x0000000106e88be6 llvm::CodeGenRegister::computeSubRegs(llvm::CodeGenRegBank&) + 2182
6  llvm-tblgen              0x0000000106e8e9f0 llvm::CodeGenRegBank::CodeGenRegBank(llvm::RecordKeeper&) + 2192
7  llvm-tblgen              0x0000000106f384a1 llvm::EmitRegisterBank(llvm::RecordKeeper&, llvm::raw_ostream&) + 65
8  llvm-tblgen              0x0000000106f72c64 (anonymous namespace)::LLVMTableGenMain(llvm::raw_ostream&, llvm::RecordKeeper&) + 1172
9  llvm-tblgen              0x0000000106fcb15f llvm::TableGenMain(char*, bool (*)(llvm::raw_ostream&, llvm::RecordKeeper&)) + 3599
10 llvm-tblgen              0x0000000106f727a6 main + 134
11 libdyld.dylib            0x000000010733c6a5 start + 1
Stack dump:
0.      Program arguments: /Volumes/Builds/pz-master-stage2-RA/bin/llvm-tblgen -gen-register-bank -I /Users/vk/llvm/lib/Target/ARM -I /Users/vk/llvm/include -I /Users/vk/llvm/lib/Target /Users/vk/llvm/lib/Target/ARM/ARM.td -o /Volumes/Builds/pz-master-stage2-RA/lib/Target/ARM/ARMGenRegisterBank.inc.tmp
/bin/sh: line 1: 41986 Segmentation fault: 11  /Volumes/Builds/pz-master-stage2-RA/bin/llvm-tblgen -gen-register-bank -I /Users/vk/llvm/lib/Target/ARM -I /Users/vk/llvm/include -I /Users/vk/llvm/lib/Target /Users/vk/llvm/lib/Target/ARM/ARM.td -o /Volumes/Builds/pz
-master-stage2-RA/lib/Target/ARM/ARMGenRegisterBank.inc.tmp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304231 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdded missing break.
Galina Kistanova [Tue, 30 May 2017 19:02:49 +0000 (19:02 +0000)]
Added missing break.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304230 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "[Cloning] Take another pass at properly cloning debug info"
Keno Fischer [Tue, 30 May 2017 18:56:26 +0000 (18:56 +0000)]
Revert "[Cloning] Take another pass at properly cloning debug info"

At least one build bot is complaining. Will investigate after lunch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304228 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoARM: Add missing flags to TBB_[JH]T pseudo instructions
Matthias Braun [Tue, 30 May 2017 18:52:33 +0000 (18:52 +0000)]
ARM: Add missing flags to TBB_[JH]T pseudo instructions

NFC except for calming down the machine verifier in some cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304227 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Cloning] Take another pass at properly cloning debug info
Keno Fischer [Tue, 30 May 2017 18:28:30 +0000 (18:28 +0000)]
[Cloning] Take another pass at properly cloning debug info

Summary:
In rL302576, DISubprograms gained the constraint that a !dbg attachments to functions must
have a 1:1 mapping to DISubprograms. As part of that change, the function cloning support
was adjusted to attempt to enforce this invariant during cloning. However, there
were several problems with the implementation. Part of these were fixed in rL304079.
However, there was a more fundamental problem with these changes, namely that it
bypasses the matadata value map, causing the cloned metadata to be a mix of metadata
pointing to the new suprogram (where manual code was added to fix those up) and the
old suprogram (where this was not the case). This mismatch could cause a number of
different assertion failures in the DWARF emitter. Some of these are given at
https://github.com/JuliaLang/julia/issues/22069, but some others have been observed
as well. Attempt to rectify this by partially reverting the manual DI metadata fixup,
and instead using the standard value map approach. To retain the desired semantics
of not duplicating the compilation unit and inlined subprograms, explicitly freeze
these in the value map.

Reviewers: dblaikie, aprantl, GorNishanov, echristo

Reviewed By: aprantl

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33655

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304226 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdding parsing ability for .res file.
Eric Beckmann [Tue, 30 May 2017 18:19:06 +0000 (18:19 +0000)]
Adding parsing ability for .res file.

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D33566

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304225 91177308-0d34-0410-b5e6-96231b3b80d8