platform/upstream/llvm.git
5 years ago[clang] Respect TerseOutput when printing lambdas
Kadir Cetinkaya [Mon, 27 May 2019 16:20:45 +0000 (16:20 +0000)]
[clang] Respect TerseOutput when printing lambdas

Reviewers: ilya-biryukov, hokein, sammccall

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D62487

llvm-svn: 361771

5 years ago [AMDGPU] Fix for the address sanitizer failure caused by the ifollowing commit:
Alexander Timofeev [Mon, 27 May 2019 15:03:29 +0000 (15:03 +0000)]
[AMDGPU] Fix for the address sanitizer failure caused by the ifollowing commit:

    1a8b2ea611cf4ca7cb09562e0238cfefa27c05b5  Divergence driven ISel. Assign register class for cross block values according to the divergence.

llvm-svn: 361770

5 years agoWhen dumping the AST to JSON, dump the type information from a typeid expression...
Aaron Ballman [Mon, 27 May 2019 14:34:31 +0000 (14:34 +0000)]
When dumping the AST to JSON, dump the type information from a typeid expression with a type operand.

llvm-svn: 361769

5 years agoWhen dumping the AST to JSON, dump whether a function is variadic or not.
Aaron Ballman [Mon, 27 May 2019 14:29:10 +0000 (14:29 +0000)]
When dumping the AST to JSON, dump whether a function is variadic or not.

llvm-svn: 361768

5 years agoWhen dumping the AST to JSON, dump the declared name of a MemberExpr operand.
Aaron Ballman [Mon, 27 May 2019 14:25:04 +0000 (14:25 +0000)]
When dumping the AST to JSON, dump the declared name of a MemberExpr operand.

llvm-svn: 361767

5 years agoWhen dumping the AST to JSON, dump the argument name to a sizeof pack expression.
Aaron Ballman [Mon, 27 May 2019 14:17:32 +0000 (14:17 +0000)]
When dumping the AST to JSON, dump the argument name to a sizeof pack expression.

llvm-svn: 361766

5 years agoDWARFDebugArangeSet: Remove references to SymbolFileDWARF
Pavel Labath [Mon, 27 May 2019 14:16:15 +0000 (14:16 +0000)]
DWARFDebugArangeSet: Remove references to SymbolFileDWARF

This class does not depend on SymbolFileDWARF. Instead, include more
appropriate low-level headers.

llvm-svn: 361765

5 years agoAdd test cases for dumping AST expression nodes to JSON; NFC.
Aaron Ballman [Mon, 27 May 2019 14:12:48 +0000 (14:12 +0000)]
Add test cases for dumping AST expression nodes to JSON; NFC.

llvm-svn: 361764

5 years ago[AMDGPU][MC] Enabled constant expressions as operands of s_waitcnt
Dmitry Preobrazhensky [Mon, 27 May 2019 14:08:43 +0000 (14:08 +0000)]
[AMDGPU][MC] Enabled constant expressions as operands of s_waitcnt

See bug 40820: https://bugs.llvm.org/show_bug.cgi?id=40820

Reviewers: artem.tamazov, arsenm

Differential Revision: https://reviews.llvm.org/D61017

llvm-svn: 361763

5 years ago[MustExecute] Improve MustExecute to correctly handle loop nest
Xing Xue [Mon, 27 May 2019 13:57:28 +0000 (13:57 +0000)]
[MustExecute] Improve MustExecute to correctly handle loop nest

Summary:
for.outer:
  br for.inner
for.inner:
  LI <loop invariant load instruction>
for.inner.latch:
  br for.inner, for.outer.latch
for.outer.latch:
  br for.outer, for.outer.exit

LI is a loop invariant load instruction that post dominate for.outer, so LI should be able to move out of the loop nest. However, there is a bug in allLoopPathsLeadToBlock().

Current algorithm of allLoopPathsLeadToBlock()

  1. get all the transitive predecessors of the basic block LI belongs to (for.inner) ==> for.outer, for.inner.latch
  2. if any successors of any of the predecessors are not for.inner or for.inner's predecessors, then return false
  3. return true

Although for.inner.latch is for.inner's predecessor, but for.inner dominates for.inner.latch, which means if for.inner.latch is ever executed, for.inner should be as well. It should not return false for cases like this.

Author: Whitney (committed by xingxue)

Reviewers: kbarton, jdoerfert, Meinersbur, hfinkel, fhahn

Reviewed By: jdoerfert

Subscribers: hiraditya, jsji, llvm-commits, etiotto, bmahjour

Tags: #LLVM

Differential Revision: https://reviews.llvm.org/D62418

llvm-svn: 361762

5 years agoTest commit (NFC)
Nikola Prica [Mon, 27 May 2019 13:51:30 +0000 (13:51 +0000)]
Test commit (NFC)

Add blank line.

llvm-svn: 361761

5 years agoXFAIL prefer-debug-over-eh-frame.test on darwin
Pavel Labath [Mon, 27 May 2019 13:43:01 +0000 (13:43 +0000)]
XFAIL prefer-debug-over-eh-frame.test on darwin

debug_frame does not seem to work on darwin, so there is nothing to
prefer.

Adding `-g` to the compiler command line is enough to get the
__debug_frame section added to the dsym file. Though lldb then finds the
section, and correctly assigns the section type to it, this does not
seem to be enough to get lldb to actually use this section for
unwinding.

llvm-svn: 361760

5 years agoDWARF: Add a simple test exercising debug_loc parsing
Pavel Labath [Mon, 27 May 2019 13:23:23 +0000 (13:23 +0000)]
DWARF: Add a simple test exercising debug_loc parsing

llvm-svn: 361759

5 years agoFuncUnwinders: prefer debug_frame over eh_frame
Pavel Labath [Mon, 27 May 2019 11:53:24 +0000 (11:53 +0000)]
FuncUnwinders: prefer debug_frame over eh_frame

The two sections usually contain the same information, and we rarely
have both kinds of entries for a single function. However, in theory the
debug_frame plan can be more complete, whereas eh_frame is only required
to be correct at places where exceptions can be thrown.

Reviewers: jasonmolenda, clayborg

Subscribers: lldb-commits

Differential Revision: https://reviews.llvm.org/D62374

llvm-svn: 361758

5 years ago[OpenCL] Fix file-scope const sampler variable for 2.0
Yaxun Liu [Mon, 27 May 2019 11:19:07 +0000 (11:19 +0000)]
[OpenCL] Fix file-scope const sampler variable for 2.0

OpenCL spec v2.0 s6.13.14:

Samplers can also be declared as global constants in the program
source using the following syntax.

   const sampler_t <sampler name> = <value>
This works fine for OpenCL 1.2 but fails for 2.0, because clang duduces
address space of file-scope const sampler variable to be in global address
space whereas spec v2.0 s6.9.b forbids file-scope sampler variable to be
in global address space.

The fix is not to deduce address space for file-scope sampler variables.

Differential Revision: https://reviews.llvm.org/D62197

llvm-svn: 361757

5 years ago[ARM GlobalISel] Un-XFAIL some tests. NFC
Diana Picus [Mon, 27 May 2019 10:32:34 +0000 (10:32 +0000)]
[ARM GlobalISel] Un-XFAIL some tests. NFC

It turns out we support big endian now (probably since r332449, but I
haven't bisected to confirm).

llvm-svn: 361756

5 years ago[ARM GlobalISel] Cleanup CallLowering a bit
Diana Picus [Mon, 27 May 2019 10:30:33 +0000 (10:30 +0000)]
[ARM GlobalISel] Cleanup CallLowering a bit

We never actually use the Offsets produced by ComputeValueVTs, so remove
them until we need them.

llvm-svn: 361755

5 years agoDWARF: Remove cu_idx variables from parsing functions
Pavel Labath [Mon, 27 May 2019 10:10:59 +0000 (10:10 +0000)]
DWARF: Remove cu_idx variables from parsing functions

These variables were useful when looking up the compile unit index
required a binary search. Now that we can look up a compile unit index
in constant time, they are no longer needed.

llvm-svn: 361754

5 years ago[CodeComplete] Complete 'return true/false' in boolean functions
Ilya Biryukov [Mon, 27 May 2019 09:52:09 +0000 (09:52 +0000)]
[CodeComplete] Complete 'return true/false' in boolean functions

Reviewers: kadircet

Reviewed By: kadircet

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D62391

llvm-svn: 361753

5 years ago[ASTImporter] Added visibility context check for CXXRecordDecl.
Balazs Keri [Mon, 27 May 2019 09:36:00 +0000 (09:36 +0000)]
[ASTImporter] Added visibility context check for CXXRecordDecl.

Summary:
ASTImporter makes now difference between classes with same name in different
translation units if these are not visible outside. These classes are not linked
into one decl chain.

Reviewers: martong, a.sidorin, shafik

Reviewed By: shafik

Subscribers: rnkovacs, dkrupp, Szelethus, gamesh411, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D62312

llvm-svn: 361752

5 years agoCmake: allow using LLVM_EXTERNAL_PROJECTS with LLVM_ENABLE_PROJECTS
Hans Wennborg [Mon, 27 May 2019 09:03:00 +0000 (09:03 +0000)]
Cmake: allow using LLVM_EXTERNAL_PROJECTS with LLVM_ENABLE_PROJECTS

The current code iterates over the combination of LLVM_EXTERNAL_PROJECTS
and LLVM_ENABLE_PROJECTS, but then disables projects that are only in
the former. If a project is in LLVM_EXTERNAL_PROJECTS, it should be
enabled.

See also llvm-commits thread on r354060.

Differential revision: https://reviews.llvm.org/D62289

llvm-svn: 361751

5 years agoMake llvm-as --help great again
Serge Guelton [Mon, 27 May 2019 08:24:06 +0000 (08:24 +0000)]
Make llvm-as --help great again

This is a follow-up to https://reviews.llvm.org/D60411, but for llvm-as.

New output:

    OVERVIEW: llvm .ll -> .bc assembler

    USAGE: llvm-as [options] <input .llvm file>

    OPTIONS:

    Generic Options:

      -help                        - Display available options (-help-hidden for more)
      -help-list                   - Display list of available options (-help-list-hidden for more)
      -version                     - Display the version of this program

    llvm-as Options:

      -data-layout=<layout-string> - data layout string to use
      -disable-output              - Disable output
      -f                           - Enable binary output on terminals
      -module-hash                 - Emit module hash
      -o=<filename>                - Override output filename

Differential Revision: https://reviews.llvm.org/D60603

llvm-svn: 361750

5 years ago[clang-tidy] Fix unused-variable warning after r361647.
Haojian Wu [Mon, 27 May 2019 08:09:02 +0000 (08:09 +0000)]
[clang-tidy] Fix unused-variable warning after r361647.

Summary:
A range-for was added in r361647 where the range variable was only used in an
assertion.  As a result, it warned for Release builds. This revision
restructures the assertion to avoid the problem.

Patch by Yitzhak Mandelbaum.

Reviewers: ilya-biryukov

Reviewed By: ilya-biryukov

Subscribers: xazax.hun, cfe-commits

Tags: #clang-tools-extra, #clang

Differential Revision: https://reviews.llvm.org/D62412

llvm-svn: 361749

5 years ago[test commit] Add my name to the CREDITS.TXT
Djordje Todorovic [Mon, 27 May 2019 07:48:28 +0000 (07:48 +0000)]
[test commit] Add my name to the CREDITS.TXT

This is my test commit. (NFC)

llvm-svn: 361748

5 years agoRemove elf::createSharedFile and move its code to SharedFile's ctor. NFC.
Rui Ueyama [Mon, 27 May 2019 07:26:13 +0000 (07:26 +0000)]
Remove elf::createSharedFile and move its code to SharedFile's ctor. NFC.

llvm-svn: 361747

5 years agoRevert r361356: "[MIR] Add simple PRE pass to MachineCSE"
David L. Jones [Mon, 27 May 2019 06:00:00 +0000 (06:00 +0000)]
Revert r361356: "[MIR] Add simple PRE pass to MachineCSE"

This is problematic on buildbots, as discussed here: https://reviews.llvm.org/rL361356

It seems like the plan already was to revert, but that hasn't happened yet.

llvm-svn: 361746

5 years ago[X86] Add test cases for D62444. NFC
Craig Topper [Mon, 27 May 2019 05:27:57 +0000 (05:27 +0000)]
[X86] Add test cases for D62444. NFC

llvm-svn: 361745

5 years agollvm-undname: Make demangling of MD5 names more robust
Nico Weber [Mon, 27 May 2019 00:48:59 +0000 (00:48 +0000)]
llvm-undname: Make demangling of MD5 names more robust

Demangler::parse() for MD5 names would:

1. Put all remaining text into the MD5 name sight unseen
2. Not modify MangledName

This meant that if the demangler recursively called parse() (e.g. in
demangleLocallyScopedNamePiece()), every recursive call that started on
an MD5 name would add all remaining bytes to the output buffer but
only advance the input by a byte.  For valid inputs, MD5 types are
never (well, see comments for 2 exceptions) nested, but for invalid
input this could cause memory use quadratic in the input size.

llvm-svn: 361744

5 years ago[LoopInterchange] Fix handling of LCSSA nodes defined in headers and latches.
Florian Hahn [Sun, 26 May 2019 23:38:25 +0000 (23:38 +0000)]
[LoopInterchange] Fix handling of LCSSA nodes defined in headers and latches.

The code to preserve LCSSA PHIs currently only properly supports
reduction PHIs and PHIs for values defined outside the latches.

This patch improves the LCSSA PHI handling to cover PHIs for values
defined in the latches.

Fixes PR41725.

Reviewers: efriedma, mcrosier, davide, jdoerfert

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D61576

llvm-svn: 361743

5 years ago[BPF] generate R_BPF_NONE relocation for BTF DataSec variables
Yonghong Song [Sun, 26 May 2019 21:26:06 +0000 (21:26 +0000)]
[BPF] generate R_BPF_NONE relocation for BTF DataSec variables

The variables in BTF DataSec type encode in-section offset.
R_BPF_NONE should be generated instead of R_BPF_64_32.

Signed-off-by: Yonghong Song <yhs@fb.com>
Differential Revision: https://reviews.llvm.org/D62460

llvm-svn: 361742

5 years ago [AMDGPU] Divergence driven ISel. Assign register class for cross block values...
Alexander Timofeev [Sun, 26 May 2019 20:33:26 +0000 (20:33 +0000)]
[AMDGPU] Divergence driven ISel. Assign register class for cross block values according to the divergence.

    Details: To make instruction selection really divergence driven it is necessary to assign
             the correct register classes to the cross block values beforehand. For the divergent targets
             same value type requires different register classes dependent on the value divergence.

    Reviewers: rampitec, nhaehnle

    Differential Revision: https://reviews.llvm.org/D59990

    This commit was reverted because of the build failure.
    The reason was mlformed patch.
    Build failure fixed.

llvm-svn: 361741

5 years ago[MCA][Scheduler] Improved critical memory dependency computation.
Andrea Di Biagio [Sun, 26 May 2019 19:50:31 +0000 (19:50 +0000)]
[MCA][Scheduler] Improved critical memory dependency computation.

This fixes a problem where back-pressure increases caused by register
dependencies were not correctly notified if execution was also delayed by memory
dependencies.

llvm-svn: 361740

5 years ago[SelectionDAG] GetDemandedBits - cleanup to more closely match SimplifyDemandedBits...
Simon Pilgrim [Sun, 26 May 2019 18:58:14 +0000 (18:58 +0000)]
[SelectionDAG] GetDemandedBits - cleanup to more closely match SimplifyDemandedBits. NFCI.

Prep work before adding demanded elts support.

llvm-svn: 361739

5 years ago[SelectionDAG] MaskedValueIsZero - add demanded elements implementation
Simon Pilgrim [Sun, 26 May 2019 18:43:44 +0000 (18:43 +0000)]
[SelectionDAG] MaskedValueIsZero - add demanded elements implementation

Will be used in an upcoming patch but I've updated the original implementation to call this to ensure test coverage.

llvm-svn: 361738

5 years ago[MCA] Refactor the logic that computes the critical memory dependency info. NFCI
Andrea Di Biagio [Sun, 26 May 2019 18:41:35 +0000 (18:41 +0000)]
[MCA] Refactor the logic that computes the critical memory dependency info. NFCI

CriticalRegDep has been renamed CriticalDependency, and it is now used by class
Instruction to store information about the critical register dependency and the
critical memory dependency. No functional change intendend.

llvm-svn: 361737

5 years ago[SimplifyCFG] back out all SwitchInst commits
Shawn Landden [Sun, 26 May 2019 18:15:51 +0000 (18:15 +0000)]
[SimplifyCFG] back out all SwitchInst commits

They caused the sanitizer builds to fail.

My suspicion is the change the countLeadingZeros().

llvm-svn: 361736

5 years agoDeleteNullPointerCheck now deletes until the end brace of the condition.
Mads Ravn [Sun, 26 May 2019 17:00:38 +0000 (17:00 +0000)]
DeleteNullPointerCheck now deletes until the end brace of the condition.

Patch by Jonathan Camilleri

Differential Revision https://reviews.llvm.org/D61861

llvm-svn: 361735

5 years ago[X86][SSE] Add shuffle combining support for ISD::ANY_EXTEND_VECTOR_INREG
Simon Pilgrim [Sun, 26 May 2019 16:00:35 +0000 (16:00 +0000)]
[X86][SSE] Add shuffle combining support for ISD::ANY_EXTEND_VECTOR_INREG

Reuses what we already have in place for ISD::ZERO_EXTEND_VECTOR_INREG just with a different sentinel

llvm-svn: 361734

5 years ago[SimplifyCFG] NFC, one more fixed test from previous push.
Shawn Landden [Sun, 26 May 2019 15:29:10 +0000 (15:29 +0000)]
[SimplifyCFG] NFC, one more fixed test from previous push.

The old test was checking for a stupid subtract one that is a transform that
makes the code woorse.

The constant-islands-jump-table.ll test wants the code a specific way,
that makes sense, so I will submit code to fix that one.

Sorry that I really didn't know how to run the test suite before this.

llvm-svn: 361733

5 years agoRevert rL361731 : [LLParser] Fix uninitialized variable warnings. NFCI.
Simon Pilgrim [Sun, 26 May 2019 15:08:45 +0000 (15:08 +0000)]
Revert rL361731 : [LLParser] Fix uninitialized variable warnings. NFCI.

These 3 variables cause quite a few warnings in the scan-build report on llvm.
........
Revert accidental commit.

llvm-svn: 361732

5 years ago[LLParser] Fix uninitialized variable warnings. NFCI.
Simon Pilgrim [Sun, 26 May 2019 15:05:12 +0000 (15:05 +0000)]
[LLParser] Fix uninitialized variable warnings. NFCI.

These 3 variables cause quite a few warnings in the scan-build report on llvm.

llvm-svn: 361731

5 years ago[SimplifyCFG] NFC, fix failing tests from last patches.
Shawn Landden [Sun, 26 May 2019 14:44:14 +0000 (14:44 +0000)]
[SimplifyCFG] NFC, fix failing tests from last patches.

No problems with the transforms.

llvm-svn: 361730

5 years ago[InstCombine] prevent crashing with invalid extractelement index
Sanjay Patel [Sun, 26 May 2019 14:03:50 +0000 (14:03 +0000)]
[InstCombine] prevent crashing with invalid extractelement index

This was found/reduced from a fuzzer report:
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=14956

llvm-svn: 361729

5 years ago[SimplifyCFG] ReduceSwitchRange: Improve on the case where the SubThreshold doesn...
Shawn Landden [Sun, 26 May 2019 13:55:52 +0000 (13:55 +0000)]
[SimplifyCFG] ReduceSwitchRange: Improve on the case where the SubThreshold doesn't trigger

llvm-svn: 361728

5 years ago[SimplifyCFG] Run ReduceSwitchRange unconditionally, generalize
Shawn Landden [Sun, 26 May 2019 13:55:14 +0000 (13:55 +0000)]
[SimplifyCFG] Run ReduceSwitchRange unconditionally, generalize

Rather than gating on "isSwitchDense" (resulting in necessesarily
sparse lookup tables even when they were generated), always run
this quite cheap transform.

This transform is useful not just for generating tables.
LowerSwitch also wants this: read LowerSwitch.cpp:257.

Be careful to not generate worse code, by introducing a
SubThreshold heuristic.

Instead of just sorting by signed, generalize the finding of the
best base.

And now that it is run unconditionally, do not replicate its
functionality in SwitchToLookupTable (which could use a Sub
when having a hole is smaller, hence the SubThreshold
heuristic located in a single place).
This simplifies SwitchToLookupTable, and fixes
some ugly corner cases due to the use of signed numbers,
such as a table containing i16 32768 and 32769, of which
32769 would be interpreted as -32768, and now the code thinks
the table is size 65536.

(We still use unconditional subtraction when building a single-register mask,
but I think this whole block should go when the more general sparse
map is added, which doesn't leave empty holes in the table.)

And the reason test4 and test5 did not trigger was documented wrong:
it was because they were not considered sufficiently "dense".

Also, fix generation of invalid LLVM-IR: shl by bit-width.

llvm-svn: 361727

5 years ago[SimpligyCFG] NFC, remove GCD that was only used for powers of two
Shawn Landden [Sun, 26 May 2019 13:54:04 +0000 (13:54 +0000)]
[SimpligyCFG] NFC, remove GCD that was only used for powers of two

and replace with an equilivent countTrailingZeros.

GCD is much more expensive than this, with repeated division.

This depends on D60823

llvm-svn: 361726

5 years ago[SimplifyCFG] NFC, update Switch tests to HEAD so I can see if my changes change...
Shawn Landden [Sun, 26 May 2019 13:52:41 +0000 (13:52 +0000)]
[SimplifyCFG] NFC, update Switch tests to HEAD so I can see if my changes change anything

Also add baseline tests to show effect of later patches.

llvm-svn: 361725

5 years ago[Support] make countLeadingZeros() and countTrailingZeros() return unsigned
Shawn Landden [Sun, 26 May 2019 13:49:58 +0000 (13:49 +0000)]
[Support] make countLeadingZeros() and countTrailingZeros() return unsigned

This matches countLeadingOnes() and countTrailingOnes(), and
APInt's countLeadingZeros() and countTrailingZeros().

(as well as __builtin_clzll())

llvm-svn: 361724

5 years ago[ValueTracking] Base computeOverflowForUnsignedMul() on ConstantRange code; NFCI
Nikita Popov [Sun, 26 May 2019 13:22:01 +0000 (13:22 +0000)]
[ValueTracking] Base computeOverflowForUnsignedMul() on ConstantRange code; NFCI

The implementation in ValueTracking and ConstantRange are equally
powerful, reuse the one in ConstantRange, which will make this easier
to extend.

llvm-svn: 361723

5 years agogn build: Merge r361664
Nico Weber [Sun, 26 May 2019 13:06:48 +0000 (13:06 +0000)]
gn build: Merge r361664

llvm-svn: 361722

5 years ago[InstCombine] Refactor OptimizeOverflowCheck; NFCI
Nikita Popov [Sun, 26 May 2019 11:43:37 +0000 (11:43 +0000)]
[InstCombine] Refactor OptimizeOverflowCheck; NFCI

Extract method to compute overflow based on binop and signedness,
and then make the result handling code generic. This extends the
always-overflow handling to signed muls, but has currently no effect,
as we don't compute always overflow for them (thus NFC).

llvm-svn: 361721

5 years ago[InstCombine] Remove OverflowCheckFlavor; NFC
Nikita Popov [Sun, 26 May 2019 11:43:31 +0000 (11:43 +0000)]
[InstCombine] Remove OverflowCheckFlavor; NFC

Instead pass binary op and signedness. The extra enum only makes
things more complicated in this case.

llvm-svn: 361720

5 years ago[ARM] Select fp16 fma
David Green [Sun, 26 May 2019 11:34:30 +0000 (11:34 +0000)]
[ARM] Select fp16 fma

This adds a pattern for fma, similar to the float and double patterns.

Differential Revision: https://reviews.llvm.org/D62330

llvm-svn: 361719

5 years ago[ARM] Select a number of fp16 rounding functions
David Green [Sun, 26 May 2019 11:13:00 +0000 (11:13 +0000)]
[ARM] Select a number of fp16 rounding functions

This add patterns for fp16 round and ceil etc. Same as the float and double
patterns.

Differential Revision: https://reviews.llvm.org/D62326

llvm-svn: 361718

5 years ago[ARM] Promote various fp16 math intrinsics
David Green [Sun, 26 May 2019 10:59:21 +0000 (10:59 +0000)]
[ARM] Promote various fp16 math intrinsics

Promote a number of fp16 math intrinsics to float, so that the relevant float
math routines can be used. Copysign is expanded so as to be handled in-place.

Differential Revision: https://reviews.llvm.org/D62325

llvm-svn: 361717

5 years ago[X86][AVX] combineBitcastvxi1 - peek through bitops to determine size of original...
Simon Pilgrim [Sun, 26 May 2019 10:54:23 +0000 (10:54 +0000)]
[X86][AVX] combineBitcastvxi1 - peek through bitops to determine size of original vector

We were only testing for direct SETCC results - this allows us to peek through AND/OR/XOR combinations of the comparison results as well.

There's a missing SEXT(PACKSS) fold that I need to investigate for v8i1 cases before I can enable it there as well.

llvm-svn: 361716

5 years ago[ARM] Select fp16 fabs
David Green [Sun, 26 May 2019 10:51:58 +0000 (10:51 +0000)]
[ARM] Select fp16 fabs

This adds a pattern for the fabs intrinsic, the same as float and double.

Differential Revision: https://reviews.llvm.org/D62324

llvm-svn: 361715

5 years ago[ARM] Select fp16 fsqrt
David Green [Sun, 26 May 2019 10:42:24 +0000 (10:42 +0000)]
[ARM] Select fp16 fsqrt

This adds a pattern for the sqrt intrinsic, the same as float and double.

Differential Revision: https://reviews.llvm.org/D62322

llvm-svn: 361714

5 years ago[ARM] Promote fp16 frem
David Green [Sun, 26 May 2019 10:30:22 +0000 (10:30 +0000)]
[ARM] Promote fp16 frem

Promote fp16 frem operations on ARM to floats so they call fmodf.

Differential Revision: https://reviews.llvm.org/D62321

llvm-svn: 361713

5 years ago[ARM] Add some base fullfp16 tests. NFC
David Green [Sun, 26 May 2019 10:06:40 +0000 (10:06 +0000)]
[ARM] Add some base fullfp16 tests. NFC

llvm-svn: 361712

5 years ago[PowerPC] Add missing R_PPC_* relocation types
Fangrui Song [Sun, 26 May 2019 08:31:00 +0000 (08:31 +0000)]
[PowerPC] Add missing R_PPC_* relocation types

While people mostly care about 64-bit, some systems need basic lib32
support. The plan is to make lld (see PR40888) capable of linking some
applications (PR40888).

llvm-svn: 361711

5 years ago[Driver][RISCV] Simplify. NFC
Fangrui Song [Sun, 26 May 2019 07:43:45 +0000 (07:43 +0000)]
[Driver][RISCV] Simplify. NFC

llvm-svn: 361710

5 years ago[Driver] Update handling of c++ and runtime directories
Petr Hosek [Sun, 26 May 2019 03:39:07 +0000 (03:39 +0000)]
[Driver] Update handling of c++ and runtime directories

This is a follow up to r361432 and r361504 which addresses issues
introduced by those changes. Specifically, it avoids duplicating
file and runtime paths in case when the effective triple is the
same as the cannonical one. Furthermore, it fixes the broken multilib
setup in the Fuchsia driver and deduplicates some of the code.

Differential Revision: https://reviews.llvm.org/D62442

llvm-svn: 361709

5 years agoAdd missing newline at end of file
Duncan P. N. Exon Smith [Sat, 25 May 2019 22:38:02 +0000 (22:38 +0000)]
Add missing newline at end of file

llvm-svn: 361708

5 years ago[SimplifyCFG] Added condition assumption for unreachable blocks
David Bolvansky [Sat, 25 May 2019 22:34:27 +0000 (22:34 +0000)]
[SimplifyCFG] Added condition assumption for unreachable blocks

Summary: PR41688

Reviewers: spatel, efriedma, craig.topper, hfinkel, reames

Reviewed By: hfinkel

Subscribers: javed.absar, dmgreen, fhahn, hfinkel, reames, nikic, lebedev.ri, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D61409

llvm-svn: 361707

5 years ago[X86] lowerBuildVectorToBitOp - support build_vector(shift()) -> shift(build_vector...
Simon Pilgrim [Sat, 25 May 2019 18:02:17 +0000 (18:02 +0000)]
[X86] lowerBuildVectorToBitOp - support build_vector(shift()) -> shift(build_vector(),C)

Commonly occurs in sign-extension cases

llvm-svn: 361706

5 years ago[LLVM-C] Add Accessor for Mach-O Universal Binary Slices
Robert Widmann [Sat, 25 May 2019 16:47:27 +0000 (16:47 +0000)]
[LLVM-C] Add Accessor for Mach-O Universal Binary Slices

Summary: Allow for retrieving an object file corresponding to an architecture-specific slice in a Mach-O universal binary file.

Reviewers: whitequark, deadalnix

Reviewed By: whitequark

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D60378

llvm-svn: 361705

5 years ago[X86] Combine fminnum/fmaxnum with non-nan operand to fmin/fmax
Nikita Popov [Sat, 25 May 2019 16:44:29 +0000 (16:44 +0000)]
[X86] Combine fminnum/fmaxnum with non-nan operand to fmin/fmax

If we have a known non-nan operand, place it in the second operand
of fmin/fmax that is returned if either operand is nan.

Differential Revision: https://reviews.llvm.org/D62448

llvm-svn: 361704

5 years ago[LVI][CVP] Add support for saturating add/sub
Nikita Popov [Sat, 25 May 2019 16:44:14 +0000 (16:44 +0000)]
[LVI][CVP] Add support for saturating add/sub

Adds support for the uadd.sat family of intrinsics in LVI, based on
ConstantRange methods from D60946.

Differential Revision: https://reviews.llvm.org/D62447

llvm-svn: 361703

5 years ago[X86][SSE] vector-sext - cleanup prefix lists
Simon Pilgrim [Sat, 25 May 2019 16:33:17 +0000 (16:33 +0000)]
[X86][SSE] vector-sext - cleanup prefix lists

Add X32-SSE common prefix to merge some checks

llvm-svn: 361702

5 years ago[SelectionDAG] define binops as a superset of commutative binops
Sanjay Patel [Sat, 25 May 2019 15:28:55 +0000 (15:28 +0000)]
[SelectionDAG] define binops as a superset of commutative binops

The test diffs show improved vector narrowing for integer min/max opcodes because
those were all absent from the list. I'm not sure if we can expose functional diffs
for all of the moved/added opcodes though.

It seems like we are missing an AVX512 opportunity to use 256-bit ops in place of
512-bit ops on some tests/targets, but I think that can be a follow-up.

Preliminary steps to make sure the callers are not misusing these queries:
rL361268
rL361547

Differential Revision: https://reviews.llvm.org/D62191

llvm-svn: 361701

5 years ago[X86] Add tests for min/maxnum with const operand; NFC
Nikita Popov [Sat, 25 May 2019 15:06:54 +0000 (15:06 +0000)]
[X86] Add tests for min/maxnum with const operand; NFC

llvm-svn: 361700

5 years ago[LoopVectorize] Fix test by regenerating checks
Nikita Popov [Sat, 25 May 2019 14:33:30 +0000 (14:33 +0000)]
[LoopVectorize] Fix test by regenerating checks

llvm-svn: 361699

5 years ago[CVP] Remove unnecessary checks for empty GNWR; NFC
Nikita Popov [Sat, 25 May 2019 14:11:55 +0000 (14:11 +0000)]
[CVP] Remove unnecessary checks for empty GNWR; NFC

The guaranteed no-wrap region is never empty, it always contains at
least zero, so these optimizations don't ever apply.

To make this more obviously true, replace the conversative return
in makeGNWR with an assertion.

llvm-svn: 361698

5 years ago[NFC] Make tests more robust for new optimizations
David Bolvansky [Sat, 25 May 2019 14:10:20 +0000 (14:10 +0000)]
[NFC] Make tests more robust for new optimizations

llvm-svn: 361697

5 years ago[SelectionDAG] soften assertion when legalizing narrow vector FP ops
Sanjay Patel [Sat, 25 May 2019 13:48:07 +0000 (13:48 +0000)]
[SelectionDAG] soften assertion when legalizing narrow vector FP ops

The test based on PR42010:
https://bugs.llvm.org/show_bug.cgi?id=42010
...may show an inaccuracy for PPC's target defs, but we should not
be so aggressive with an assert here. There's no telling what out-of-tree
targets look like.

llvm-svn: 361696

5 years ago[NFC] Update test checks
David Bolvansky [Sat, 25 May 2019 13:11:22 +0000 (13:11 +0000)]
[NFC] Update test checks

llvm-svn: 361695

5 years ago[CVP] Add tests for saturating add/sub ranges; NFC
Nikita Popov [Sat, 25 May 2019 09:53:51 +0000 (09:53 +0000)]
[CVP] Add tests for saturating add/sub ranges; NFC

llvm-svn: 361694

5 years ago[LVI][CVP] Calculate with.overflow result range
Nikita Popov [Sat, 25 May 2019 09:53:45 +0000 (09:53 +0000)]
[LVI][CVP] Calculate with.overflow result range

In LVI, calculate the range of extractvalue(op.with.overflow(%x, %y), 0)
as the range of op(%x, %y). This is mainly useful in conjunction with
D60650: If the result of the operation is extracted in a branch guarded
against overflow, then the value of %x will be appropriately constrained
and the result range of the operation will be calculated taking that
into account.

Differential Revision: https://reviews.llvm.org/D60656

llvm-svn: 361693

5 years ago[LVI] Extract helper for binary range calculations; NFC
Nikita Popov [Sat, 25 May 2019 09:53:37 +0000 (09:53 +0000)]
[LVI] Extract helper for binary range calculations; NFC

llvm-svn: 361692

5 years ago[X86FixupLEAs] Turn optIncDec into a generic two address LEA optimizer. Support LEA64...
Craig Topper [Sat, 25 May 2019 06:17:47 +0000 (06:17 +0000)]
[X86FixupLEAs] Turn optIncDec into a generic two address LEA optimizer. Support LEA64_32r properly.

INC/DEC is really a special case of a more generic issue. We should also turn leas into add reg/reg or add reg/imm regardless of the slow lea flags.

This also supports LEA64_32 which has 64 bit input registers and 32 bit output registers. So we need to convert the 64 bit inputs to their 32 bit equivalents to check if they are equal to base reg.

One thing to note, the original code preserved the kill flags by adding operands to the new instruction instead of using addReg. But I think tied operands aren't supposed to have the kill flag set. I dropped the kill flags, but I could probably try to preserve it in the add reg/reg case if we think its important. Not sure which operand its supposed to go on for the LEA64_32r instruction due to the super reg implicit uses. Though I'm also not sure those are needed since they were probably just created by an INSERT_SUBREG from a 32-bit input.

Differential Revision: https://reviews.llvm.org/D61472

llvm-svn: 361691

5 years ago[X86] Add zero idioms to the haswell, broadwell, and skylake schedule models. Add...
Craig Topper [Sat, 25 May 2019 04:47:49 +0000 (04:47 +0000)]
[X86] Add zero idioms to the haswell, broadwell, and skylake schedule models. Add 256-bit fp xor to sandybridge zero idioms

This copies the Sandy Bridge zero idiom support to later CPUs. Adding the AVX2 and AVX512F/VL instructions as appropriate.

Differential Revision: https://reviews.llvm.org/D62360

llvm-svn: 361690

5 years ago[X86][llvm-mca] Add zero idiom tests for Intel CPUs. NFC
Craig Topper [Sat, 25 May 2019 04:47:42 +0000 (04:47 +0000)]
[X86][llvm-mca] Add zero idiom tests for Intel CPUs. NFC

This pre-commits tests for D62360

llvm-svn: 361689

5 years agoRevert r361644, "[AMDGPU] Divergence driven ISel. Assign register class for cross...
Peter Collingbourne [Sat, 25 May 2019 01:52:38 +0000 (01:52 +0000)]
Revert r361644, "[AMDGPU] Divergence driven ISel. Assign register class for cross block values according to the divergence."

Broke sanitizer bots:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux/builds/21694/steps/bootstrap%20clang/logs/stdio
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/32478/steps/check-llvm%20asan/logs/stdio

llvm-svn: 361688

5 years ago[clangd] tweaks: Add clangBasic dependency to LINK_LIBS
Heejin Ahn [Sat, 25 May 2019 01:35:14 +0000 (01:35 +0000)]
[clangd] tweaks: Add clangBasic dependency to LINK_LIBS

This is necessary to make builds with `-DBUILD_SHARED_LIBS=ON` work.

llvm-svn: 361687

5 years agoPermit static local structured bindings to be named from arbitrary scopes inside...
Richard Smith [Sat, 25 May 2019 01:04:17 +0000 (01:04 +0000)]
Permit static local structured bindings to be named from arbitrary scopes inside their declaring scope.

llvm-svn: 361686

5 years agoRevert "[Analysis] Link library dependencies to Analysis plugins"
Akira Hatanaka [Sat, 25 May 2019 00:50:03 +0000 (00:50 +0000)]
Revert "[Analysis] Link library dependencies to Analysis plugins"

This reverts commit r361340. The following builder has been broken for
the past few days because of this commit:

http://green.lab.llvm.org/green/job/clang-stage2-cmake-RgSan/

Also revert r361399, which was committed to fix r361340.

llvm-svn: 361685

5 years agoRename clangToolingRefactor to clangToolingRefactoring for consistency with its directory
Nico Weber [Sat, 25 May 2019 00:27:19 +0000 (00:27 +0000)]
Rename clangToolingRefactor to clangToolingRefactoring for consistency with its directory

See "[cfe-dev] The name of clang/lib/Tooling/Refactoring".

Differential Revision: https://reviews.llvm.org/D62420

llvm-svn: 361684

5 years agollvm-dwarfdump: Don't error on mixed units using/not using str_offsets
David Blaikie [Sat, 25 May 2019 00:07:22 +0000 (00:07 +0000)]
llvm-dwarfdump: Don't error on mixed units using/not using str_offsets

This lead to errors when dumping binaries with v4 and v5 units linked
together (but could've also errored on v5 units that did/didn't use
str_offsets).

Also improves error handling and messages around invalid str_offsets
contributions.

llvm-svn: 361683

5 years ago[analyzer] Add a prunable note for skipping vbase inits in subclasses.
Artem Dergachev [Fri, 24 May 2019 23:37:11 +0000 (23:37 +0000)]
[analyzer] Add a prunable note for skipping vbase inits in subclasses.

When initialization of virtual base classes is skipped, we now tell the user
about it, because this aspect of C++ isn't very well-known.

The implementation is based on the new "note tags" feature (r358781).
In order to make use of it, allow note tags to produce prunable notes,
and move the note tag factory to CoreEngine.

Differential Revision: https://reviews.llvm.org/D61817

llvm-svn: 361682

5 years ago[CFG] Add branch to skip vbase inits when they're handled by superclass.
Artem Dergachev [Fri, 24 May 2019 23:37:08 +0000 (23:37 +0000)]
[CFG] Add branch to skip vbase inits when they're handled by superclass.

This patch adds the run-time CFG branch that would skip initialization of
virtual base classes depending on whether the constructor is called from a
superclass constructor or not. Previously the Static Analyzer was already
skipping virtual base-class initializers in such constructors, but it wasn't
skipping their arguments and their potential side effects, which was causing
pr41300 (and was generally incorrect). The previous skipping behavior is
now replaced with a hard assertion that we're not even getting there due
to how our CFG works.

The new CFG element is under a CFG build option so that not to break other
consumers of the CFG by this change. Static Analyzer support for this change
is implemented.

Differential Revision: https://reviews.llvm.org/D61816

llvm-svn: 361681

5 years agoFix crash deserializing a CUDAKernelCallExpr with a +Asserts binary.
Richard Smith [Fri, 24 May 2019 23:26:07 +0000 (23:26 +0000)]
Fix crash deserializing a CUDAKernelCallExpr with a +Asserts binary.

The assertion in setConfig read from the (uninitialized) CONFIG
expression.

llvm-svn: 361680

5 years ago[GlobalISel][AArch64] Make FP constraint checks consider possible use/def banks
Jessica Paquette [Fri, 24 May 2019 23:08:45 +0000 (23:08 +0000)]
[GlobalISel][AArch64] Make FP constraint checks consider possible use/def banks

In a few places in getInstrMapping, we check if use/def instructions for the
instruction we're mapping have floating point constraints.

We can improve this check and reduce the number of copies in GISel-compiled code
if we make a couple observations:

- For a def instruction, it only matters if the def instruction must always
  output a value stored on a FPR

- For a use instruction, it only matters if the use instruction must always
  only take in values stored in FPRs

This adds two new functions:

- onlyUsesFP
- onlyDefinesFP

Then we can use those when we're checking the uses/defs instead.

Without this patch, the load, unmerge, store, and select in the added test
would have unnecessary copies.

Differential Revision: https://reviews.llvm.org/D62426

llvm-svn: 361679

5 years ago[WebAssembly] Relax signature checking for undefined functions that are not called...
Sam Clegg [Fri, 24 May 2019 22:45:08 +0000 (22:45 +0000)]
[WebAssembly] Relax signature checking for undefined functions that are not called directly

When function signatures don't match and the undefined function is not
called directly (i.e. only has its address taken) we don't issue a
warning or create a runtime thunk for the undefined function.

Instead in this case we simply use the defined version of the function.
This is possible since checking signatures of dynamic calls happens
at runtime so any invalid usage will still result in a runtime error.

This is needed to allow C++ programs to link without generating
warnings.  Its not uncommon in C++ for vtables to be populated by
function address whee the signature of the function is not known in the
compilation unit.  In this case clang declares the method as void(void)
and relies on the vtable caller casting the data back to the correct
signature.

Fixes: https://bugs.llvm.org/show_bug.cgi?id=40412

Differential Revision: https://reviews.llvm.org/D62153

llvm-svn: 361678

5 years ago[GlobalISel][AArch64] NFC: Factor out HasFPConstraints into a proper function
Jessica Paquette [Fri, 24 May 2019 22:12:21 +0000 (22:12 +0000)]
[GlobalISel][AArch64] NFC: Factor out HasFPConstraints into a proper function

Factor it out into a function, and replace places where we had the same check
with the new function.

Differential Revision: https://reviews.llvm.org/D62421

llvm-svn: 361677

5 years agoRevert Xcode scheme changes from 361675
Greg Clayton [Fri, 24 May 2019 22:12:01 +0000 (22:12 +0000)]
Revert Xcode scheme changes from 361675

llvm-svn: 361676

5 years agoCleanup fixed form sizes.
Greg Clayton [Fri, 24 May 2019 22:08:50 +0000 (22:08 +0000)]
Cleanup fixed form sizes.

The fix form sizes use to have two arrays: one for 4 byte addresses and in for 8 byte addresses. The table had an issue where DW_FORM_flag_present wasn't being represented as a fixed size form because its actual size _is_ zero and zero was used to indicate the form isn't fixed in size. Any code that needed to quickly access the DWARF had to get a FixedFormSizes instance using the address byte size.

This fix cleans things up by adding a DWARFFormValue::GetFixedSize() both as a static method and as a member function on DWARFFormValue. It correctly can indicate if a form size is zero. This cleanup is a precursor to a follow up patch where I hope to speed up DWARF parsing.

I verified performance doesn't regress by loading hundreds of DWARF files and setting a breakpoint by file and line and by name in files that do not have DWARF indexes. Performance remained consistent between the two approaches.

Differential Revision: https://reviews.llvm.org/D62416

llvm-svn: 361675

5 years agoMark tests as x86.
Alina Sbirlea [Fri, 24 May 2019 21:49:27 +0000 (21:49 +0000)]
Mark tests as x86.

llvm-svn: 361674

5 years ago[Target] Make Processes' GetLanguageRuntime non-virtual
Alex Langford [Fri, 24 May 2019 21:27:37 +0000 (21:27 +0000)]
[Target] Make Processes' GetLanguageRuntime non-virtual

llvm-svn: 361673

5 years ago[DWARFExpression] Remove commented-out code (NFC)
Jonas Devlieghere [Fri, 24 May 2019 21:26:30 +0000 (21:26 +0000)]
[DWARFExpression] Remove commented-out code (NFC)

llvm-svn: 361672