OSDN Git Service

android-x86/external-llvm.git
5 years ago[globalisel] Restore comment explaining the nits of GISelChangeObserver::createdInstr()
Daniel Sanders [Mon, 11 Feb 2019 20:05:49 +0000 (20:05 +0000)]
[globalisel] Restore comment explaining the nits of GISelChangeObserver::createdInstr()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353741 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MemorySSA] Remove verifyClobberSanity.
Alina Sbirlea [Mon, 11 Feb 2019 19:51:21 +0000 (19:51 +0000)]
[MemorySSA] Remove verifyClobberSanity.

Summary:
This verification may fail after certain transformations due to
BasicAA's fragility. Added a small explanation and a testcase that
triggers the assert in checkClobberSanity (before its removal).
Addresses PR40509.

Reviewers: george.burgess.iv

Subscribers: sanjoy, jlebar, llvm-commits, Prazek

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D57973

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353739 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRefactor setAlreadyUnrolled() and setAlreadyVectorized().
Michael Kruse [Mon, 11 Feb 2019 19:45:44 +0000 (19:45 +0000)]
Refactor setAlreadyUnrolled() and setAlreadyVectorized().

Loop::setAlreadyUnrolled() and
LoopVectorizeHints::setLoopAlreadyUnrolled() both add loop metadata that
stops the same loop from being transformed multiple times. This patch
merges both implementations.

In doing so we fix 3 potential issues:

 * setLoopAlreadyUnrolled() kept the llvm.loop.vectorize/interleave.*
   metadata even though it will not be used anymore. This already caused
   problems such as http://llvm.org/PR40546. Change the behavior to the
   one of setAlreadyUnrolled which deletes this loop metadata.

 * setAlreadyUnrolled() used to create a new LoopID by calling
   MDNode::get with nullptr as the first operand, then replacing it by
   the returned references using replaceOperandWith. It is possible
   that MDNode::get would instead return an existing node (due to
   de-duplication) that then gets modified. To avoid, use a fresh
   TempMDNode that does not get uniqued with anything else before
   replacing it with replaceOperandWith.

 * LoopVectorizeHints::matchesHintMetadataName() only compares the
   suffix of the attribute to set the new value for. That is, when
   called with "enable", would erase attributes such as
   "llvm.loop.unroll.enable", "llvm.loop.vectorize.enable" and
   "llvm.loop.distribute.enable" instead of the one to replace.
   Fortunately, function was only called with "isvectorized".

Differential Revision: https://reviews.llvm.org/D57566

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353738 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Fix matchRotate bug when one operand is a ConstantExpr shift
Sanjay Patel [Mon, 11 Feb 2019 19:26:27 +0000 (19:26 +0000)]
[InstCombine] Fix matchRotate bug when one operand is a ConstantExpr shift

This bug seems to be harmless in release builds, but will cause an error in UBSAN
builds or an assertion failure in debug builds.

When it gets to this opcode comparison, it assumes both of the operands are BinaryOperators,
but the prior m_LogicalShift will also match a ConstantExpr. The cast<BinaryOperator> will
assert in a debug build, or reading an invalid value for BinaryOp from memory with
((BinaryOperator*)constantExpr)->getOpcode() will cause an error in a UBSAN build.

The test I added will fail without this change in debug/UBSAN builds, but not in release.

Patch by: @AndrewScheidecker (Andrew Scheidecker)

Differential Revision: https://reviews.llvm.org/D58049

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353736 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SelectionDAGBuilder] Add restrictions to EmitFuncArgumentDbgValue
Bjorn Pettersson [Mon, 11 Feb 2019 19:23:30 +0000 (19:23 +0000)]
[SelectionDAGBuilder] Add restrictions to EmitFuncArgumentDbgValue

Summary:
This patch fixes PR40587.

When a dbg.value instrinsic is emitted to the DAG
by using EmitFuncArgumentDbgValue the resulting
DBG_VALUE is hoisted to the beginning of the entry
block. I think the idea is to be able to locate
a formal argument already from the start of the
function.
However, EmitFuncArgumentDbgValue only checked that
the value that was used to describe a variable was
originating from a function parameter, not that the
variable itself actually was an argument to the
function. So when for example assigning a local
variable "local" the value from an argument "a",
the assocated DBG_VALUE instruction would be hoisted
to the beginning of the function, even if the scope
for "local" started somewhere else (or if "local"
was mapped to other values earlier in the function).

This patch adds some logic to EmitFuncArgumentDbgValue
to check that the variable being described actually
is an argument to the function. And that the dbg.value
being lowered already is in the entry block. Otherwise
we bail out, and the dbg.value will be handled as an
ordinary dbg.value (not as a "FuncArgumentDbgValue").

A tricky situation is when both the variable and
the value is related to function arguments, but not
neccessarily the same argument. We make sure that we
do not describe the same argument more than once as
a "FuncArgumentDbgValue". This solution works as long
as opt has injected a "first" dbg.value that corresponds
to the formal argument at the function entry.

Reviewers: jmorse, aprantl

Subscribers: jyknight, hiraditya, fedor.sergeev, dstenb, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D57702

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353735 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LICM&MSSA] Limit store hoisting.
Alina Sbirlea [Mon, 11 Feb 2019 19:07:15 +0000 (19:07 +0000)]
[LICM&MSSA] Limit store hoisting.

Summary:
If there is no clobbering access for a store inside the loop, that store
can only be hoisted if there are no interfearing loads.
A more general verification introduced here: there are no loads that are
not optimized to an access outside the loop.
Addresses PR40586.

Reviewers: george.burgess.iv

Subscribers: sanjoy, jlebar, Prazek, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D57967

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353734 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TargetLibraryInfo] Update run time support for Windows
Evandro Menezes [Mon, 11 Feb 2019 19:02:28 +0000 (19:02 +0000)]
[TargetLibraryInfo] Update run time support for Windows

It seems that the run time for Windows has changed and supports more math
functions than it used to, especially on AArch64, ARM, and AMD64.

Fixes PR40541.

Differential revision: https://reviews.llvm.org/D57625

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353733 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64][GlobalISel] Add isel support for a couple vector exts/truncs
Jessica Paquette [Mon, 11 Feb 2019 18:56:39 +0000 (18:56 +0000)]
[AArch64][GlobalISel] Add isel support for a couple vector exts/truncs

Add support for

- v4s16 <-> v4s32
- v2s64 <-> v2s32

And update tests that use them to show that we generate the correct
instructions.

Differential Revision: https://reviews.llvm.org/D57832

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353732 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel][AArch64] NFC: Remove unnecessary IR from select-fp-casts.mir
Jessica Paquette [Mon, 11 Feb 2019 18:41:22 +0000 (18:41 +0000)]
[GlobalISel][AArch64] NFC: Remove unnecessary IR from select-fp-casts.mir

The IR section in this test doesn't do anything, so there's no point in it
being there. Since it's redundant, just remove it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353731 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DebugInfo] Fix /usr/lib/debug llvm-symbolizer lookup with relative paths
Jordan Rupprecht [Mon, 11 Feb 2019 18:05:48 +0000 (18:05 +0000)]
[DebugInfo] Fix /usr/lib/debug llvm-symbolizer lookup with relative paths

Summary:
rL189250 added a realpath call, and rL352916 because realpath breaks assumptions with some build systems. However, the /usr/lib/debug case has been clarified, falling back to /usr/lib/debug is currently broken if the obj passed in is a relative path. Adding a call to use absolute paths when falling back to /usr/lib/debug fixes that while still not making any realpath assumptions.

This also adds a --fallback-debug-path command line flag for testing (since we probably can't write to /usr/lib/debug from buildbot environments), but was also verified manually:

```
$ rm -f path/to/dwarfdump-test.elf-x86-64
$ strace llvm-symbolizer --obj=relative/path/to/dwarfdump-test.elf-x86-64.debuglink 0x40113f |& grep dwarfdump
```

Lookups went to relative/path/to/dwarfdump-test.elf-x86-64, relative/path/to/.debug/dwarfdump-test.elf-x86-64, and then finally /usr/lib/debug/absolute/path/to/dwarfdump-test.elf-x86-64.

Reviewers: dblaikie, samsonov

Reviewed By: dblaikie

Subscribers: krytarowski, aprantl, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D57916

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353730 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA][Scheduler] Track resources that were found busy when issuing an instruction.
Andrea Di Biagio [Mon, 11 Feb 2019 17:55:47 +0000 (17:55 +0000)]
[MCA][Scheduler] Track resources that were found busy when issuing an instruction.

This is a follow up of r353706. When the scheduler fails to issue a ready
instruction to the underlying pipelines, it now updates a mask of 'busy resource
units'. That information will be used in future to obtain the set of
"problematic" resources in the case of bottlenecks caused by resource pressure.
No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353728 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PowerPC] Avoid scalarization of vector truncate
Roland Froese [Mon, 11 Feb 2019 17:29:14 +0000 (17:29 +0000)]
[PowerPC] Avoid scalarization of vector truncate

The PowerPC code generator currently scalarizes vector truncates that would fit in a vector register, resulting in vector extracts, scalar operations, and vector merges. This patch custom lowers a vector truncate that would fit in a register to a vector shuffle instead.

Differential Revision: https://reviews.llvm.org/D56507

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353724 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel][AArch64] Select G_FFLOOR
Jessica Paquette [Mon, 11 Feb 2019 17:22:58 +0000 (17:22 +0000)]
[GlobalISel][AArch64] Select G_FFLOOR

This teaches the legalizer about G_FFLOOR, and lets us select G_FFLOOR in
AArch64.

It updates the existing floating point tests, and adds a select-floor.mir test.

Differential Revision: https://reviews.llvm.org/D57486

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353722 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRecommit "[GlobalISel] Add IRTranslator support for G_FFLOOR"
Jessica Paquette [Mon, 11 Feb 2019 17:16:32 +0000 (17:16 +0000)]
Recommit "[GlobalISel] Add IRTranslator support for G_FFLOOR"

After the changes introduced in r353586, this instruction doesn't cause any
issues for any backend.

Original review: https://reviews.llvm.org/D57485

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353720 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoGlobalISel: Add G_FCANONICALIZE instruction
Matt Arsenault [Mon, 11 Feb 2019 17:05:20 +0000 (17:05 +0000)]
GlobalISel: Add G_FCANONICALIZE instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353719 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] fix atomic_optimizations_buffer.ll test after DPP combiner was enabled by...
Valery Pykhtin [Mon, 11 Feb 2019 16:28:42 +0000 (16:28 +0000)]
[AMDGPU] fix atomic_optimizations_buffer.ll test after DPP combiner was enabled by default.

Related commits: rL353691, rL353703.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353717 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Regenerate insertelement tests
Simon Pilgrim [Mon, 11 Feb 2019 16:16:09 +0000 (16:16 +0000)]
[X86] Regenerate insertelement tests

Add common X86/X64 prefixes (and use X86 instead of X32)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353716 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAdd recipes for migrating downstream branches of git mirrors
David Greene [Mon, 11 Feb 2019 15:40:02 +0000 (15:40 +0000)]
Add recipes for migrating downstream branches of git mirrors

Add some common recipes for downstream users developing on top of the
existing git mirrors. These instructions show how to migrate local
branches to the monorepo.

Differential Revision: https://reviews.llvm.org/D56550

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353713 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoMove some classes into anonymous namespaces. NFC.
Benjamin Kramer [Mon, 11 Feb 2019 15:16:21 +0000 (15:16 +0000)]
Move some classes into anonymous namespaces. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353710 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA] Return a mask of busy resources from method ResourceManager::checkAvailability...
Andrea Di Biagio [Mon, 11 Feb 2019 14:53:04 +0000 (14:53 +0000)]
[MCA] Return a mask of busy resources from method ResourceManager::checkAvailability(). NFCI

In case of bottlenecks caused by pipeline pressure, we want to be able to
correctly report the set of problematic pipelines. This is a first step towards
adding support for bottleneck hints in llvm-mca (see PR37494). No functional
change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353706 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Remove unused variable
Benjamin Kramer [Mon, 11 Feb 2019 14:49:54 +0000 (14:49 +0000)]
[AMDGPU] Remove unused variable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353704 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Fix DPP sequence in atomic optimizer.
Neil Henning [Mon, 11 Feb 2019 14:44:14 +0000 (14:44 +0000)]
[AMDGPU] Fix DPP sequence in atomic optimizer.

This commit fixes the DPP sequence in the atomic optimizer (which was
previously missing the row_shr:3 step), and works around a read_register
exec bug by using a ballot instead.

Differential Revision: https://reviews.llvm.org/D57737

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353703 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert "[X86][SSE] Generalize X86ISD::BLENDI support to more value types"
Sam McCall [Mon, 11 Feb 2019 14:05:36 +0000 (14:05 +0000)]
Revert "[X86][SSE] Generalize X86ISD::BLENDI support to more value types"

This reverts commit r353610.
It causes a miscompile visible in macro expansion in a bootstrapped clang.

http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190211/626590.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353699 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TEST] Add missing opportunity test for PR39673
Max Kazantsev [Mon, 11 Feb 2019 12:58:18 +0000 (12:58 +0000)]
[TEST] Add missing opportunity test for PR39673

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353693 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Add v8m.base pattern for add negative imm
Sam Parker [Mon, 11 Feb 2019 11:35:42 +0000 (11:35 +0000)]
[ARM] Add v8m.base pattern for add negative imm

The v8m.base ISA contains movw, which can operate on an unsigned
16-bit value. Add the pattern that converts an add with a negative
value, that could fit into 16-bits when negated, into a sub with that
positive value.

Differential Revision: https://reviews.llvm.org/D57942

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353692 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Enable DPP combiner pass by default.
Valery Pykhtin [Mon, 11 Feb 2019 11:15:03 +0000 (11:15 +0000)]
[AMDGPU] Enable DPP combiner pass by default.

Related revisions: https://reviews.llvm.org/D55444, https://reviews.llvm.org/D55314

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353691 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][ARM] Simplify loop-indexing codegen test
Sam Parker [Mon, 11 Feb 2019 10:52:49 +0000 (10:52 +0000)]
[NFC][ARM] Simplify loop-indexing codegen test

Remove unnecessary offset checks, CHECK-BASE checks and add some
extra -NOT checks and TODO comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353689 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TEST] Add failing test from PR40454
Max Kazantsev [Mon, 11 Feb 2019 10:44:57 +0000 (10:44 +0000)]
[TEST] Add failing test from PR40454

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353688 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agotest-release.sh: Add option to use ninja
Diana Picus [Mon, 11 Feb 2019 10:30:22 +0000 (10:30 +0000)]
test-release.sh: Add option to use ninja

Allow the use of ninja instead of make. This is useful on some
platforms where we'd like to be able to limit the number of link jobs
without slowing down the other steps of the release.

This patch adds a -use-ninja command line option, which sets the
generator to Ninja both for LLVM and the test-suite. It also deals with
some differences between make and ninja:
* DESTDIR handling - ninja doesn't like this to be listed after the
  target, but both make and ninja can handle it before the command
* Verbose mode - ninja uses -v, make uses VERBOSE=1
* Keep going mode - make has a -k mode, which builds as much as possible
  even when failures are encountered; for ninja we need to set a hard
  limit (we use 100 since most people won't look at 100 failures anyway)

I haven't tested with gmake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353685 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAttempt to fix buildbot after r353679 #2
Eugene Leviant [Mon, 11 Feb 2019 10:17:17 +0000 (10:17 +0000)]
Attempt to fix buildbot after r353679 #2

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353683 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DWARF] LLVM ERROR: Broken function found, while removing Debug Intrinsics.
Carlos Alberto Enciso [Mon, 11 Feb 2019 10:16:38 +0000 (10:16 +0000)]
[DWARF] LLVM ERROR: Broken function found, while removing Debug Intrinsics.

Check that when SimplifyCFG is flattening a 'br', all their debug intrinsic instructions are removed, including any dbg.label referencing a label associated with the basic blocks being removed.

As the test case involves a CFG transformation, move it to the correct location.

Differential Revision: https://reviews.llvm.org/D57444

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353682 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAttempt to fix buildbot after r353679
Eugene Leviant [Mon, 11 Feb 2019 10:12:19 +0000 (10:12 +0000)]
Attempt to fix buildbot after r353679

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353681 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoSmall refactoring of FileError. NFC.
Eugene Leviant [Mon, 11 Feb 2019 09:49:37 +0000 (09:49 +0000)]
Small refactoring of FileError. NFC.

Differential revision: https://reviews.llvm.org/D57945

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353679 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] LoadStoreOptimizer: reoder limit
Sjoerd Meijer [Mon, 11 Feb 2019 09:37:42 +0000 (09:37 +0000)]
[ARM] LoadStoreOptimizer: reoder limit

The whole design of generating LDMs/STMs is fragile and unreliable: it depends on
rescheduling here in the LoadStoreOptimizer that isn't register pressure aware
and regalloc that isn't aware of generating LDMs/STMs.
This patch adds a (hidden) option to control the total number of instructions that
can be re-ordered. I appreciate this looks only a tiny bit better than a hard-coded
constant, but at least it allows more easy experimentation with different values
for now. Ideally we calculate this reorder limit based on some heuristics, and take
register pressure into account. I might be looking into that next.

Differential Revision: https://reviews.llvm.org/D57954

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353678 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoMove CFLGraph and the AA summary code over to the new `CallBase`
Chandler Carruth [Mon, 11 Feb 2019 09:25:41 +0000 (09:25 +0000)]
Move CFLGraph and the AA summary code over to the new `CallBase`
instruction base class rather than the `CallSite` wrapper.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353676 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm] [cmake] Use current directory in GenerateVersionFromVCS
Michal Gorny [Mon, 11 Feb 2019 09:07:07 +0000 (09:07 +0000)]
[llvm] [cmake] Use current directory in GenerateVersionFromVCS

Find dependent scripts of GenerateVersionFromVCS in current directory
rather than ../../cmake/modules.  I do not see any reason why the former
would not work and The latter is incorrect when GenerateVersionFromVCS
is used from install directory (i.e. in stand-alone builds).

Differential Revision: https://reviews.llvm.org/D57996

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353674 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRemove `CallSite` from the CodeMetrics analysis, moving it to the new
Chandler Carruth [Mon, 11 Feb 2019 09:03:32 +0000 (09:03 +0000)]
Remove `CallSite` from the CodeMetrics analysis, moving it to the new
`CallBase` and simpler APIs therein.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353673 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRemove a declaration that is dead, and not even implemented any longer.
Chandler Carruth [Mon, 11 Feb 2019 09:03:26 +0000 (09:03 +0000)]
Remove a declaration that is dead, and not even implemented any longer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353672 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] LoadStoreOptimizer: just a clean-up. NFC.
Sjoerd Meijer [Mon, 11 Feb 2019 08:47:59 +0000 (08:47 +0000)]
[ARM] LoadStoreOptimizer: just a clean-up. NFC.

Differential Revision: https://reviews.llvm.org/D57955

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353670 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoUpdate more files added with the old header to the new one.
Chandler Carruth [Mon, 11 Feb 2019 08:25:56 +0000 (08:25 +0000)]
Update more files added with the old header to the new one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353667 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoUpdate new files added to llvm-objcopy to use the new file header.
Chandler Carruth [Mon, 11 Feb 2019 08:25:19 +0000 (08:25 +0000)]
Update new files added to llvm-objcopy to use the new file header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353666 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoUpdate files that were mistakenly added with the old file header to the
Chandler Carruth [Mon, 11 Feb 2019 08:07:38 +0000 (08:07 +0000)]
Update files that were mistakenly added with the old file header to the
new one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353665 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoUpdate files that were mistakenly added with the old file header.
Chandler Carruth [Mon, 11 Feb 2019 08:07:32 +0000 (08:07 +0000)]
Update files that were mistakenly added with the old file header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353664 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CallSite removal] Port InstSimplify over to use `CallBase` both in its
Chandler Carruth [Mon, 11 Feb 2019 07:54:10 +0000 (07:54 +0000)]
[CallSite removal] Port InstSimplify over to use `CallBase` both in its
interface and implementation.

Port code with: `cast<CallBase>(CS.getInstruction())`.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353662 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CallSite removal] Migrate ConstantFolding APIs and implementation to
Chandler Carruth [Mon, 11 Feb 2019 07:51:44 +0000 (07:51 +0000)]
[CallSite removal] Migrate ConstantFolding APIs and implementation to
`CallBase`.

Users have been updated. You can see how to update any out-of-tree
usages: pass `cast<CallBase>(CS.getInstruction())`.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353661 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CallSite removal] Migrate the statepoint GC infrastructure to use the
Chandler Carruth [Mon, 11 Feb 2019 07:42:30 +0000 (07:42 +0000)]
[CallSite removal] Migrate the statepoint GC infrastructure to use the
`CallBase` class rather than `CallSite` wrappers.

I pushed this change down through most of the statepoint infrastructure,
completely removing the use of CallSite where I could reasonably do so.
I ended up making a couple of cut-points: generic call handling
(instcombine, TLI, SDAG). As soon as it hit truly generic handling with
users outside the immediate code, I simply transitioned into or out of
a `CallSite` to make this a reasonable sized chunk.

Differential Revision: https://reviews.llvm.org/D56122

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353660 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Removed unused SDTypeProfile. NFC
Craig Topper [Mon, 11 Feb 2019 07:30:48 +0000 (07:30 +0000)]
[X86] Removed unused SDTypeProfile. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353659 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Fix clang-tidy dep on ClangSACheckers.
Nico Weber [Mon, 11 Feb 2019 03:09:57 +0000 (03:09 +0000)]
gn build: Fix clang-tidy dep on ClangSACheckers.

Patch by Mirko Bonadei <mbonadei@webrtc.org>!

Differential Revision: https://reviews.llvm.org/D57998

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353657 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] EltsFromConsecutiveLoads - replace SmallBitVector with APInt (NFC).
Simon Pilgrim [Sun, 10 Feb 2019 22:45:48 +0000 (22:45 +0000)]
[X86] EltsFromConsecutiveLoads - replace SmallBitVector with APInt (NFC).

Minor refactor to simplify some incoming patches to improve broadcast loads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353655 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel] Regex the opcodes in unit test to fix non-deterministic ordering
Mandeep Singh Grang [Sun, 10 Feb 2019 19:53:43 +0000 (19:53 +0000)]
[GlobalISel] Regex the opcodes in unit test to fix non-deterministic ordering

Differential Revision: https://reviews.llvm.org/D57988

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353652 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CodeGen][X86] Don't scalarize vector saturating add/sub
Nikita Popov [Sun, 10 Feb 2019 19:06:38 +0000 (19:06 +0000)]
[CodeGen][X86] Don't scalarize vector saturating add/sub

Now that we have vector support for [US](ADD|SUB)O we no longer
need to scalarize when expanding [US](ADD|SUB)SAT.

This matches what the cost model already does.

Differential Revision: https://reviews.llvm.org/D57348

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353651 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64] Regenerate bswap tests
Simon Pilgrim [Sun, 10 Feb 2019 18:27:37 +0000 (18:27 +0000)]
[AArch64] Regenerate bswap tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353648 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add basic bitreverse/bswap combine tests
Simon Pilgrim [Sun, 10 Feb 2019 18:07:03 +0000 (18:07 +0000)]
[X86] Add basic bitreverse/bswap combine tests

Shows missing SimplifyDemandedBits support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353647 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAG] Add optional AllowUndefs to isNullOrNullSplat
Simon Pilgrim [Sun, 10 Feb 2019 17:42:15 +0000 (17:42 +0000)]
[DAG] Add optional AllowUndefs to isNullOrNullSplat

No change in default behaviour (AllowUndefs = false)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353646 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombine] Simplify funnel shifts with undef/zero args to bitshifts
Simon Pilgrim [Sun, 10 Feb 2019 17:04:00 +0000 (17:04 +0000)]
[DAGCombine] Simplify funnel shifts with undef/zero args to bitshifts

Now that we have SimplifyDemandedBits support for funnel shifts (rL353539), we need to simplify funnel shifts back to bitshifts in cases where either argument has been folded to undef/zero.

Differential Revision: https://reviews.llvm.org/D58009

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353645 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add masked variable tests for funnel undef/zero argument combines
Simon Pilgrim [Sun, 10 Feb 2019 15:46:32 +0000 (15:46 +0000)]
[X86] Add masked variable tests for funnel undef/zero argument combines

I've avoided 'modulo' masks as we'll SimplifyDemandedBits those in the future, and we just need to check that the shift variable is 'in range'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353644 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[x86] narrow 256-bit horizontal ops via demanded elements
Sanjay Patel [Sun, 10 Feb 2019 15:22:06 +0000 (15:22 +0000)]
[x86] narrow 256-bit horizontal ops via demanded elements

256-bit horizontal math ops are an x86 monstrosity (and thankfully have
not been extended to 512-bit AFAIK).

The two 128-bit halves operate on separate halves of the inputs. So if we
don't demand anything in the upper half of the result, we can extract the
low halves of the inputs, do the math, and then insert that result into a
256-bit output.

All of the extract/insert is free (ymm<-->xmm), so we're left with a
narrower (cheaper) version of the original op.

In the affected tests based on:
https://bugs.llvm.org/show_bug.cgi?id=33758
https://bugs.llvm.org/show_bug.cgi?id=38971
...we see that the h-op narrowing can result in further narrowing of other
math via existing generic transforms.

I originally drafted this patch as an exact pattern match starting from
extract_vector_elt, but I thought we might see diffs starting from
extract_subvector too, so I changed it to a more general demanded elements
solution. There are no extra existing regression test improvements from
that switch though, so we could go back.

Differential Revision: https://reviews.llvm.org/D57841

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353641 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add additional tests for funnel undef/zero argument combines
Simon Pilgrim [Sun, 10 Feb 2019 14:54:57 +0000 (14:54 +0000)]
[X86] Add additional tests for funnel undef/zero argument combines

As suggested on D58009

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353640 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TargetLowering] refactor setcc folds to fix another miscompile (PR40657)
Sanjay Patel [Sun, 10 Feb 2019 14:29:57 +0000 (14:29 +0000)]
[TargetLowering] refactor setcc folds to fix another miscompile (PR40657)

SimplifySetCC still has much room for improvement, but this should
fix the remaining problem examples from:
https://bugs.llvm.org/show_bug.cgi?id=40657

The initial fix for this problem was rL353615.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353639 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86][SSE] Add SimplifyDemandedBits test for BLENDVPD
Simon Pilgrim [Sun, 10 Feb 2019 12:55:44 +0000 (12:55 +0000)]
[X86][SSE] Add SimplifyDemandedBits test for BLENDVPD

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353638 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Local] Delete a redundant check. NFC
Fangrui Song [Sun, 10 Feb 2019 09:25:56 +0000 (09:25 +0000)]
[Local] Delete a redundant check. NFC

isInstructionTriviallyDead also performs the use_empty() check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353637 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2obj] - Fix .dynamic section entries writing for 32bit targets.
George Rimar [Sun, 10 Feb 2019 08:35:38 +0000 (08:35 +0000)]
[yaml2obj] - Fix .dynamic section entries writing for 32bit targets.

This was introduced by me in r353613.

I tried to fix Big-endian bot and replaced
uintX_t -> ELFT::Xword. But ELFT::Xword is a packed<uint64_t>,
so it is always 8 bytes and that was obviously incorrect.

My intention was to use something like packed<uint> actually, which
size is target dependent.

Patch fixes this bug and adds a test case, since no bots seems reported this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353636 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Move some vector InstAliases out from under unnecessary 'let Predicates'. NFCI
Craig Topper [Sun, 10 Feb 2019 02:34:31 +0000 (02:34 +0000)]
[X86] Move some vector InstAliases out from under unnecessary 'let Predicates'. NFCI

We don't have any assembler predicates for vector ISAs so this isn't necessary. It just adds extra lines and identation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353631 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Fix an unused variable warning.
Craig Topper [Sun, 10 Feb 2019 02:21:29 +0000 (02:21 +0000)]
[InstCombine] Fix an unused variable warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353630 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add tests for funnel undef argument combines
Simon Pilgrim [Sat, 9 Feb 2019 22:21:09 +0000 (22:21 +0000)]
[X86] Add tests for funnel undef argument combines

If one of the shifted arguments is undef we should be folding to a regular shift.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353628 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] CombineOr - fold to generic funnel shifts
Simon Pilgrim [Sat, 9 Feb 2019 20:34:59 +0000 (20:34 +0000)]
[X86] CombineOr - fold to generic funnel shifts

As discussed on D57389, this is a first step towards moving the SHLD/SHRD matching code to DAGCombiner using FSHL/FSHR instead.

There's a bit of work to do before I can do that, so this just folds to FSHL/FSHR in the existing code (handling the different SHRD/FSHR argument ordering), which fixes the issue we had with i16 shift amounts not being correctly masked.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353626 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[x86] add another test for setcc miscompile (PR40657); NFC
Sanjay Patel [Sat, 9 Feb 2019 20:06:11 +0000 (20:06 +0000)]
[x86] add another test for setcc miscompile (PR40657); NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353625 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Merge r353590
Nico Weber [Sat, 9 Feb 2019 17:58:16 +0000 (17:58 +0000)]
gn build: Merge r353590

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353621 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agollvm-lib: Implement /list flag
Nico Weber [Sat, 9 Feb 2019 17:33:04 +0000 (17:33 +0000)]
llvm-lib: Implement /list flag

Differential Revision: https://reviews.llvm.org/D57952

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353620 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TargetLowering] add tests to show effect of setcc sub->shift; NFC
Sanjay Patel [Sat, 9 Feb 2019 17:03:59 +0000 (17:03 +0000)]
[TargetLowering] add tests to show effect of setcc sub->shift; NFC

There's effectively no difference for the cases with variables.
We just trade a sub for an add on those. But the case with a
subtract from constant would require an extra move instruction
on x86, so this looks like a reasonable generic combine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353619 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[x86] add test for setcc sub->shift transform; NFC
Sanjay Patel [Sat, 9 Feb 2019 16:41:20 +0000 (16:41 +0000)]
[x86] add test for setcc sub->shift transform; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353618 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Regenerate test.
Simon Pilgrim [Sat, 9 Feb 2019 16:27:19 +0000 (16:27 +0000)]
[X86] Regenerate test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353616 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TargetLowering] avoid miscompile in setcc transform (PR40657)
Sanjay Patel [Sat, 9 Feb 2019 15:59:02 +0000 (15:59 +0000)]
[TargetLowering] avoid miscompile in setcc transform (PR40657)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353615 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2elf.cpp] - Fix compilation under linux.
George Rimar [Sat, 9 Feb 2019 15:18:52 +0000 (15:18 +0000)]
[yaml2elf.cpp] - Fix compilation under linux.

Fixes errors like:
/home/ssglocal/clang-cmake-x86_64-sde-avx512-linux/clang-cmake-x86_64-sde-avx512-linux/llvm/tools/yaml2obj/yaml2elf.cpp:597:5: error: need ‘typename’ before ‘ELFT:: Xword’ because ‘ELFT’ is a dependent scope
     ELFT::Xword Tag = (ELFT::Xword)DE.Tag;

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353614 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2elf] - An attemp to fix s390x BB after r353607.
George Rimar [Sat, 9 Feb 2019 15:03:19 +0000 (15:03 +0000)]
[yaml2elf] - An attemp to fix s390x BB after r353607.

s390x is big-endian and seems r353607 had an issue with endianess,
Bot was unhappy:
http://lab.llvm.org:8011/builders/clang-s390x-linux-lnt/builds/11168/steps/ninja%20check%201/logs/stdio

This should fix it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353613 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert "[SelectionDAG] Extract [US]MULO expansion into TL method; NFC"
Nikita Popov [Sat, 9 Feb 2019 13:54:02 +0000 (13:54 +0000)]
Revert "[SelectionDAG] Extract [US]MULO expansion into TL method; NFC"

This reverts commit r353611.

Triggers an assertion during the libcall expansion on ARM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353612 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SelectionDAG] Extract [US]MULO expansion into TL method; NFC
Nikita Popov [Sat, 9 Feb 2019 13:29:22 +0000 (13:29 +0000)]
[SelectionDAG] Extract [US]MULO expansion into TL method; NFC

In preparation for supporting vector expansion.

Also drop a variant of ExpandLibCall, of which the MULO expansions
were the only user.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353611 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86][SSE] Generalize X86ISD::BLENDI support to more value types
Simon Pilgrim [Sat, 9 Feb 2019 13:13:59 +0000 (13:13 +0000)]
[X86][SSE] Generalize X86ISD::BLENDI support to more value types

D42042 introduced the ability for the ExecutionDomainFixPass to more easily change between BLENDPD/BLENDPS/PBLENDW as the domains required.

With this ability, we can avoid most bitcasts/scaling in the DAG that was occurring with X86ISD::BLENDI lowering/combining, blend with the vXi32/vXi64 vectors directly and use isel patterns to lower to the float vector equivalent vectors.

This helps the shuffle combining and SimplifyDemandedVectorElts be more aggressive as we lose track of fewer UNDEF elements than when we go up/down through bitcasts.

I've introduced a basic blend(bitcast(x),bitcast(y)) -> bitcast(blend(x,y)) fold, there are more generalizations I can do there (e.g. widening/scaling and handling the tricky v16i16 repeated mask case).

The vector-reduce-smin/smax regressions will be fixed in a future improvement to SimplifyDemandedBits to peek through bitcasts and support X86ISD::BLENDV.

Differential Revision: https://reviews.llvm.org/D57888

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353610 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[lib/ObjectYAML] - Fix BB after r353607 [2]. NFC.
George Rimar [Sat, 9 Feb 2019 12:14:20 +0000 (12:14 +0000)]
[lib/ObjectYAML] - Fix BB after r353607 [2]. NFC.

The second and the last place it seems.

Error was:
[  4%] Building CXX object lib/Support/CMakeFiles/LLVMSupport.dir/Error.cpp.o
/Users/buildslave/as-bldslv9_new/lld-x86_64-darwin13/llvm.src/lib/ObjectYAML/ELFYAML.cpp:993:15: error: unused variable 'Object' [-Werror,-Wunused-variable]
  const auto *Object = static_cast<ELFYAML::Object *>(IO.getContext());

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353609 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[lib/ObjectYAML] - Fix BB after r353607. NFC.
George Rimar [Sat, 9 Feb 2019 12:04:39 +0000 (12:04 +0000)]
[lib/ObjectYAML] - Fix BB after r353607. NFC.

Error was:
[  4%] Building CXX object lib/Support/CMakeFiles/LLVMSupport.dir/DAGDeltaAlgorithm.cpp.o
/Users/buildslave/as-bldslv9_new/lld-x86_64-darwin13/llvm.src/lib/ObjectYAML/ELFYAML.cpp:666:15: error: unused variable 'Object' [-Werror,-Wunused-variable]
  const auto *Object = static_cast<ELFYAML::Object *>(IO.getContext());
(http://lab.llvm.org:8011/builders/lld-x86_64-darwin13/builds/29920)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353608 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2obj][obj2yaml] - Add support for dumping/parsing .dynamic sections.
George Rimar [Sat, 9 Feb 2019 11:34:28 +0000 (11:34 +0000)]
[yaml2obj][obj2yaml] - Add support for dumping/parsing .dynamic sections.

This teaches the tools to parse and dump
the .dynamic section and its dynamic tags.

Differential revision: https://reviews.llvm.org/D57691

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353606 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalOpt] Simplify __cxa_atexit elimination
Fangrui Song [Sat, 9 Feb 2019 09:18:37 +0000 (09:18 +0000)]
[GlobalOpt] Simplify __cxa_atexit elimination

cxxDtorIsEmpty checks callers recursively to determine if the
__cxa_atexit-registered function is empty, and eliminates the
__cxa_atexit call accordingly.

This recursive check is unnecessary as redundant instructions and
function calls can be removed by early-cse and inliner. In addition,
cxxDtorIsEmpty does not mark visited function and it may visit a
function exponential times (multiplication principle).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353603 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CMake] Don't set <PROJECT>_STANDALONE_BUILD
Petr Hosek [Sat, 9 Feb 2019 03:06:56 +0000 (03:06 +0000)]
[CMake] Don't set <PROJECT>_STANDALONE_BUILD

We shouldn't be treating runtimes builds as standalone builds since
we have enough of the context loaded into the runtimes environment.

Differential Revision: https://reviews.llvm.org/D57992

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353601 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MC] Clean up unused inline function and non-anchor defaulted destructors; NFCI
Hubert Tong [Sat, 9 Feb 2019 02:11:51 +0000 (02:11 +0000)]
[MC] Clean up unused inline function and non-anchor defaulted destructors; NFCI

Summary:
Take care of some missing clean-ups that belong with r249548 and some
other copy/paste that had happened. In particular, the destructors are
no longer vtable anchors after r249548; and `setSectionName` in
`MCSectionWasm` is private and unused since r313058 culled its only
caller. The destructors are now implicitly defined, and the unused
function is removed.

Reviewers: nemanjai, jasonliu, grosbach

Reviewed By: nemanjai

Subscribers: sbc100, aheejin, sunfish, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D57182

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353597 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoExtra processing for BitCast + PHI in InstCombine
Gabor Buella [Sat, 9 Feb 2019 01:44:28 +0000 (01:44 +0000)]
Extra processing for BitCast + PHI in InstCombine

For some specific cases with bitcast A->B->A with intervening PHI nodes InstCombiner::optimizeBitCastFromPhi transformation creates extra PHI nodes, which are actually a copy of already created PHI or in another words, they are redundant. These extra PHI nodes could lead to extra move instructions generated after DeSSA transformation. This happens when several conditions are met

 - SROA kicks in and creates new alloca;
 - there is a simple assignment L = R, which falls under 'canonicalize loads' done by combineLoadToOperationType (this transformation is by default). Exactly this transformation is the reason of bitcasts generated;
 - the alloca is then used in A->B->A + PHI chain;
 - there is a loop unrolling.

As a result optimizeBitCastFromPhi creates as many of PHI nodes for each new SROA alloca as loop unrolling factor is. These new extra PHI nodes are redundant actually except of one and should not be created. Moreover the idea of optimizeBitCastFromPhi is to get rid of the cast (when possible) but that doesn't happen in these conditions.

The proposed fix is to do the cast replacement for the whole calculated/accumulated PHI closure not for one cast only, which is an argument to the optimizeBitCastFromPhi. These will help to accomplish several things: 1) avoid extra PHI nodes generated as all casts which may trigger optimizeBitCastFromPhi transformation will be replaced, 3) bitcasts will be replaced, and 3) create more opportunities to remove dead code, which appears after the replacement.

A new test case shows that it's possible to get rid of all bitcasts completely and get quite good code reduction.

Author: Igor Tsimbalist <igor.v.tsimbalist@intel.com>

Reviewed By: Carrot

Differential Revision: https://reviews.llvm.org/D57053

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353595 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Split idot4/8 signed and unsigned tests. NFC.
Stanislav Mekhanoshin [Sat, 9 Feb 2019 01:02:28 +0000 (01:02 +0000)]
[AMDGPU] Split idot4/8 signed and unsigned tests. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353593 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoThis reverts commit 1440a848a635849b97f7a5cfa0ecc40d37451f5b.
Mikhail R. Gadelha [Sat, 9 Feb 2019 00:46:12 +0000 (00:46 +0000)]
This reverts commit 1440a848a635849b97f7a5cfa0ecc40d37451f5b.
and commit a1853e834c65751f92521f7481b15cf0365e796b.

They broke arm and aarch64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353590 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRecommit "[GlobalISel] Introduce a generic floating point floor opcode, G_FFLOOR""
Jessica Paquette [Sat, 9 Feb 2019 00:37:31 +0000 (00:37 +0000)]
Recommit "[GlobalISel] Introduce a generic floating point floor opcode, G_FFLOOR""

After r353586, we won't fail on the AMDGPU floor pattern that was killing the
importer before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353589 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Split dot-insts feature
Stanislav Mekhanoshin [Sat, 9 Feb 2019 00:34:21 +0000 (00:34 +0000)]
[AMDGPU] Split dot-insts feature

Differential Revision: https://reviews.llvm.org/D57971

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353587 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel] Skip patterns that define complex suboperands twice instead of dying
Jessica Paquette [Sat, 9 Feb 2019 00:29:13 +0000 (00:29 +0000)]
[GlobalISel] Skip patterns that define complex suboperands twice instead of dying

If we run into a pattern that looks like this:

add
  (complex $x, $y)
  (complex $x, $z)

We should skip the pattern instead of asserting/doing something unpredictable.

This makes us return an Error in that case, and adds a testcase for skipped
patterns.

Differential Revision: https://reviews.llvm.org/D57980

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353586 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Merge r353566
Nico Weber [Sat, 9 Feb 2019 00:21:06 +0000 (00:21 +0000)]
gn build: Merge r353566

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353585 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC] Avoid passing blocks vector to the OutlineRegionInfo constructor by value.
Sergey Dmitriev [Fri, 8 Feb 2019 23:52:15 +0000 (23:52 +0000)]
[NFC] Avoid passing blocks vector to the OutlineRegionInfo constructor by value.

Reviewers: vsk, fhahn, davidxl

Reviewed By: vsk

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D57957

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353582 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[x86] add test for miscompiling setcc transform (PR40657); NFC
Sanjay Patel [Fri, 8 Feb 2019 23:34:57 +0000 (23:34 +0000)]
[x86] add test for miscompiling setcc transform (PR40657); NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353580 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRe-apply r353553 "[GISel][NFC]: Add missing call to record CSE hits in the CSEMIRBuilder"
Francis Visoiu Mistrih [Fri, 8 Feb 2019 23:34:11 +0000 (23:34 +0000)]
Re-apply r353553 "[GISel][NFC]: Add missing call to record CSE hits in the CSEMIRBuilder"

With a fix after r353563 that adds some more opcodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353579 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert r353553 "[GISel][NFC]: Add missing call to record CSE hits in the CSEMIRBuilder"
Francis Visoiu Mistrih [Fri, 8 Feb 2019 22:49:43 +0000 (22:49 +0000)]
Revert r353553 "[GISel][NFC]: Add missing call to record CSE hits in the CSEMIRBuilder"

This reverts commit r353553.

This breaks CodeGen/AArch64/GlobalISel/legalize-ext-csedebug-output.mir:

http://green.lab.llvm.org/green/job/clang-stage1-cmake-RA-incremental/57963/console

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353575 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Docs] Use code-block:: text for part of the callbr documentation to attempt to make...
Craig Topper [Fri, 8 Feb 2019 21:09:33 +0000 (21:09 +0000)]
[Docs] Use code-block:: text for part of the callbr documentation to attempt to make the bot happy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353567 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add FPCW as an implicit use on floating point load instructions.
Craig Topper [Fri, 8 Feb 2019 20:50:09 +0000 (20:50 +0000)]
[X86] Add FPCW as an implicit use on floating point load instructions.

These instructions can generate a stack overflow exception so technically they read the stack overflow exception mask bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353564 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoImplementation of asm-goto support in LLVM
Craig Topper [Fri, 8 Feb 2019 20:48:56 +0000 (20:48 +0000)]
Implementation of asm-goto support in LLVM

This patch accompanies the RFC posted here:
http://lists.llvm.org/pipermail/llvm-dev/2018-October/127239.html

This patch adds a new CallBr IR instruction to support asm-goto
inline assembly like gcc as used by the linux kernel. This
instruction is both a call instruction and a terminator
instruction with multiple successors. Only inline assembly
usage is supported today.

This also adds a new INLINEASM_BR opcode to SelectionDAG and
MachineIR to represent an INLINEASM block that is also
considered a terminator instruction.

There will likely be more bug fixes and optimizations to follow
this, but we felt it had reached a point where we would like to
switch to an incremental development model.

Patch by Craig Topper, Alexander Ivchenko, Mikhail Dvoretckii

Differential Revision: https://reviews.llvm.org/D53765

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353563 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CodeExtractor] Restore outputs after creating exit stubs
Vedant Kumar [Fri, 8 Feb 2019 20:48:04 +0000 (20:48 +0000)]
[CodeExtractor] Restore outputs after creating exit stubs

When CodeExtractor saves the result of InvokeInst at the first insertion
point of the 'normal destination' basic block, this block can be omitted
in the outlined region, so store is placed outside of the function. The
suggested solution is to process saving outputs after creating exit
stubs for new function, and stores will be placed in that blocks before
return in this case.

Patch by Sergei Kachkov!

Fixes llvm.org/PR40455.

Differential Revision: https://reviews.llvm.org/D57919

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353562 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAMDGPU/GlobalISel: Fix broken tests
Matt Arsenault [Fri, 8 Feb 2019 19:59:39 +0000 (19:59 +0000)]
AMDGPU/GlobalISel: Fix broken tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353559 91177308-0d34-0410-b5e6-96231b3b80d8