OSDN Git Service

android-x86/external-llvm.git
5 years ago[build] Rename clang-headers to clang-resource-headers
Shoaib Meenai [Mon, 4 Mar 2019 21:19:53 +0000 (21:19 +0000)]
[build] Rename clang-headers to clang-resource-headers

Summary:
The current install-clang-headers target installs clang's resource
directory headers. This is different from the install-llvm-headers
target, which installs LLVM's API headers. We want to introduce the
corresponding target to clang, and the natural name for that new target
would be install-clang-headers. Rename the existing target to
install-clang-resource-headers to free up the install-clang-headers name
for the new target, following the discussion on cfe-dev [1].

I didn't find any bots on zorg referencing install-clang-headers. I'll
send out another PSA to cfe-dev to accompany this rename.

[1] http://lists.llvm.org/pipermail/cfe-dev/2019-February/061365.html

Reviewers: beanz, phosek, tstellar, rnk, dim, serge-sans-paille

Subscribers: mgorny, javed.absar, jdoerfert, #sanitizers, openmp-commits, lldb-commits, cfe-commits, llvm-commits

Tags: #clang, #sanitizers, #lldb, #openmp, #llvm

Differential Revision: https://reviews.llvm.org/D58791

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355340 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel][AArch64] Legalize vector G_SELECT
Jessica Paquette [Mon, 4 Mar 2019 21:12:46 +0000 (21:12 +0000)]
[GlobalISel][AArch64] Legalize vector G_SELECT

Just scalarize it, and add a test showing it works.

Differential Revision: https://reviews.llvm.org/D58747

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355339 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix wrong enum value in switch.
Evgeniy Stepanov [Mon, 4 Mar 2019 21:00:28 +0000 (21:00 +0000)]
Fix wrong enum value in switch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355338 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ConstantHoisting] avoid hang/crash from unreachable blocks (PR40930)
Sanjay Patel [Mon, 4 Mar 2019 20:57:14 +0000 (20:57 +0000)]
[ConstantHoisting] avoid hang/crash from unreachable blocks (PR40930)

I'm not too familiar with this pass, so there might be a better
solution, but this appears to fix the degenerate:
PR40930
PR40931
PR40932
PR40934
...without affecting any real-world code.

As we've seen in several other passes, when we have unreachable blocks,
they can contain semi-bogus IR and/or cause unexpected conditions. We
would not typically expect these patterns to make it this far, but we
have to guard against them anyway.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355337 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PGO] Context sensitive PGO (part 3)
Rong Xu [Mon, 4 Mar 2019 20:21:27 +0000 (20:21 +0000)]
[PGO] Context sensitive PGO (part 3)

Part 3 of CSPGO changes (mostly related to PassMananger).

Differential Revision: https://reviews.llvm.org/D54175

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355330 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Add tests for add nsw + sadd.with.overflow; NFC
Nikita Popov [Mon, 4 Mar 2019 19:35:46 +0000 (19:35 +0000)]
[InstCombine] Add tests for add nsw + sadd.with.overflow; NFC

Baseline tests for D58881, which fixes part of PR38146.

Patch by Dan Robertson.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355328 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Subtarget] Follow up to r355167, add another set of curly braces to FeatureBitArray...
Craig Topper [Mon, 4 Mar 2019 19:23:37 +0000 (19:23 +0000)]
[Subtarget] Follow up to r355167, add another set of curly braces to FeatureBitArray initialization to satisfy older versions of clang.

Apparently older versions of clang like 3.6 require an extra set of curly braces around std::array initializations. I'm told the C++ language was changed regarding this by CWG 1270.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355327 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRe-commit r355104: "[AArch64][GlobalISel] Add support for 64 bit vector shuffle using...
Amara Emerson [Mon, 4 Mar 2019 19:16:00 +0000 (19:16 +0000)]
Re-commit r355104: "[AArch64][GlobalISel] Add support for 64 bit vector shuffle using TBL1."

The code to materialize a mask from a constant pool load tried to use a 128 bit
LDR to load a 64 bit constant pool entry, which was 8 byte aligned. This resulted
in a link failure in the NEON tests in the test suite since the LDR address was
unaligned. This change fixes that to instead emit a 64 bit LDR if the entry is
64 bit, before converting back to a 128 bit register for the TBL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355326 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MC] Teach ELFObjectWriter that parse-time variables do not appear in
Nirav Dave [Mon, 4 Mar 2019 19:12:56 +0000 (19:12 +0000)]
[MC] Teach ELFObjectWriter that parse-time variables do not appear in
symbol table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355325 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombiner][X86][SystemZ][AArch64] Combine some cases of (bitcast (build_vector...
Craig Topper [Mon, 4 Mar 2019 19:12:16 +0000 (19:12 +0000)]
[DAGCombiner][X86][SystemZ][AArch64] Combine some cases of (bitcast (build_vector constants)) between legalize types and legalize dag.

This patch enables combining integer bitcasts of integer build vectors when the new scalar type is legal. I've avoided floating point because the implementation bitcasts float to int along the way and we would need to check the intermediate types for legality

Differential Revision: https://reviews.llvm.org/D58884

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355324 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly] Add support for data sections in the assembler.
Wouter van Oortmerssen [Mon, 4 Mar 2019 17:18:04 +0000 (17:18 +0000)]
[WebAssembly] Add support for data sections in the assembler.

Summary:
This is quite minimal so far, introduce them with .section,
fill them with .int8 or .asciz, end with .size

Reviewers: dschuff, sbc100, aheejin

Subscribers: jgravelle-google, sunfish, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58660

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355321 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert duplicate check for DragonFly BSD
Kamil Rytarowski [Mon, 4 Mar 2019 15:51:02 +0000 (15:51 +0000)]
Revert duplicate check for DragonFly BSD

Summary: Revert duplicate check for DragonFly BSD

Submitted by tuxillo.

Reviewers: krytarowski

Reviewed By: krytarowski

Subscribers: mgorny, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58907

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355319 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Regenerate illegal type load test with non-undef load address.
Simon Pilgrim [Mon, 4 Mar 2019 14:49:02 +0000 (14:49 +0000)]
[X86] Regenerate illegal type load test with non-undef load address.

This would be affected by an upcoming patch without undoing some of the bugpoint reduction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355316 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoUnbreak shared library linkage on DragonFlyBSD.
Kamil Rytarowski [Mon, 4 Mar 2019 14:36:43 +0000 (14:36 +0000)]
Unbreak shared library linkage on DragonFlyBSD.

Patch submitted by rimvydas.

Reviewers: llvm-commits, krytarowski, mgorny

Reviewed By: krytarowski, mgorny

Subscribers: mgorny

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D35125

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355315 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA] Remove unused methods. NFC
Andrea Di Biagio [Mon, 4 Mar 2019 13:34:56 +0000 (13:34 +0000)]
[MCA] Remove unused methods. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355314 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU][MC] Enable lds_direct operand for v_readfirstlane_b32, v_readlane_b32 and...
Dmitry Preobrazhensky [Mon, 4 Mar 2019 12:48:32 +0000 (12:48 +0000)]
[AMDGPU][MC] Enable lds_direct operand for v_readfirstlane_b32, v_readlane_b32 and v_writelane_b32

See bug 40662: https://bugs.llvm.org/show_bug.cgi?id=40662

Reviewers: artem.tamazov, arsenm, rampitec

Differential Revision: https://reviews.llvm.org/D58713

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355312 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA] Correctly initialize struct SummaryView::BackPressureInfo.
Andrea Di Biagio [Mon, 4 Mar 2019 12:23:05 +0000 (12:23 +0000)]
[MCA] Correctly initialize struct SummaryView::BackPressureInfo.

This should appease the buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355309 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA] Highlight kernel bottlenecks in the summary view.
Andrea Di Biagio [Mon, 4 Mar 2019 11:52:34 +0000 (11:52 +0000)]
[MCA] Highlight kernel bottlenecks in the summary view.

This patch adds a new flag named -bottleneck-analysis to print out information
about throughput bottlenecks.

MCA knows how to identify and classify dynamic dispatch stalls. However, it
doesn't know how to analyze and highlight kernel bottlenecks.  The goal of this
patch is to teach MCA how to correlate increases in backend pressure to backend
stalls (and therefore, the loss of throughput).

From a Scheduler point of view, backend pressure is a function of the scheduler
buffer usage (i.e. how the number of uOps in the scheduler buffers changes over
time). Backend pressure increases (or decreases) when there is a mismatch
between the number of opcodes dispatched, and the number of opcodes issued in
the same cycle.  Since buffer resources are limited, continuous increases in
backend pressure would eventually leads to dispatch stalls. So, there is a
strong correlation between dispatch stalls, and how backpressure changed over
time.

This patch teaches how to identify situations where backend pressure increases
due to:
 - unavailable pipeline resources.
 - data dependencies.

Data dependencies may delay execution of instructions and therefore increase the
time that uOps have to spend in the scheduler buffers. That often translates to
an increase in backend pressure which may eventually lead to a bottleneck.
Contention on pipeline resources may also delay execution of instructions, and
lead to a temporary increase in backend pressure.

Internally, the Scheduler classifies instructions based on whether register /
memory operands are available or not.

An instruction is marked as "ready to execute" only if data dependencies are
fully resolved.
Every cycle, the Scheduler attempts to execute all instructions that are ready
to execute. If an instruction cannot execute because of unavailable pipeline
resources, then the Scheduler internally updates a BusyResourceUnits mask with
the ID of each unavailable resource.

ExecuteStage is responsible for tracking changes in backend pressure. If backend
pressure increases during a cycle because of contention on pipeline resources,
then ExecuteStage sends a "backend pressure" event to the listeners.
That event would contain information about instructions delayed by resource
pressure, as well as the BusyResourceUnits mask.

Note that ExecuteStage also knows how to identify situations where backpressure
increased because of delays introduced by data dependencies.

The SummaryView observes "backend pressure" events and prints out a "bottleneck
report".

Example of bottleneck report:

```
Cycles with backend pressure increase [ 99.89% ]
Throughput Bottlenecks:
  Resource Pressure       [ 0.00% ]
  Data Dependencies:      [ 99.89% ]
   - Register Dependencies [ 0.00% ]
   - Memory Dependencies   [ 99.89% ]
```

A bottleneck report is printed out only if increases in backend pressure
eventually caused backend stalls.

About the time complexity:

Time complexity is linear in the number of instructions in the
Scheduler::PendingSet.

The average slowdown tends to be in the range of ~5-6%.
For memory intensive kernels, the slowdown can be significant if flag
-noalias=false is specified. In the worst case scenario I have observed a
slowdown of ~30% when flag -noalias=false was specified.

We can definitely recover part of that slowdown if we optimize class LSUnit (by
doing extra bookkeeping to speedup queries). For now, this new analysis is
disabled by default, and it can be enabled via flag -bottleneck-analysis. Users
of MCA as a library can enable the generation of pressure events through the
constructor of ExecuteStage.

This patch partially addresses https://bugs.llvm.org/show_bug.cgi?id=37494

Differential Revision: https://reviews.llvm.org/D58728

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355308 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Avoid codegen changes when DBG_VALUE appears between lowered selects
Jeremy Morse [Mon, 4 Mar 2019 10:56:02 +0000 (10:56 +0000)]
[X86] Avoid codegen changes when DBG_VALUE appears between lowered selects

X86TargetLowering::EmitLoweredSelect presently detects sequences of CMOV pseudo
instructions without accounting for debug intrinsics. This leads to different
codegen with and without option -g, if a DBG_VALUE instruction lands in the
middle of several lowered selects.

Work around this by skipping over debug instructions when looking for CMOV
sequences, and sinking those debug insts into the EmitLoweredSelect sunk block.
This might slightly shift where variables appear in the instruction sequence,
but won't re-order assignments.

Differential Revision: https://reviews.llvm.org/D58672

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355307 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Fix selection of VLDR.16 instruction with imm offset
Oliver Stannard [Mon, 4 Mar 2019 09:17:38 +0000 (09:17 +0000)]
[ARM] Fix selection of VLDR.16 instruction with imm offset

The isScaledConstantInRange function takes upper and lower bounds which are
checked after dividing by the scale, so the bounds checks for half, single and
double precision should all be the same. Previously, we had wrong bounds checks
for half precision, so selected an immediate the instructions can't actually
represent.

Differential revision: https://reviews.llvm.org/D58822

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355305 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64/ARM] Fix two compiler warnings in InstructionSelector, NFCI
Jonas Hahnfeld [Mon, 4 Mar 2019 08:51:32 +0000 (08:51 +0000)]
[AArch64/ARM] Fix two compiler warnings in InstructionSelector, NFCI

1) GCC complains that KnownValid is set but not used.
2) In ARMInstructionSelector::selectGlobal() the code is mixing "enumeral
   and non-enumeral type in conditional expression". Solve this by casting
   to unsigned which is the final type anyway.

Differential Revision: https://reviews.llvm.org/D58834

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355304 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DebugInfo] Construct nested types on behalf of owner CU
Eugene Leviant [Mon, 4 Mar 2019 07:15:36 +0000 (07:15 +0000)]
[DebugInfo] Construct nested types on behalf of owner CU

Differential revision: https://reviews.llvm.org/D58786

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355303 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm] [Support] Revert "Reimplement getMainExecutable() using sysctl on NetBSD"
Michal Gorny [Mon, 4 Mar 2019 04:53:50 +0000 (04:53 +0000)]
[llvm] [Support] Revert "Reimplement getMainExecutable() using sysctl on NetBSD"

This apparently does not work reliably after all (non-reentrant?)
and causes test failures such as:

http://lab.llvm.org:8011/builders/netbsd-amd64/builds/19254/steps/run%20unit%20tests/logs/FAIL%3A%20libc%2B%2B%3A%3Asize.pass.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355302 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Mark debug values as unavailable after DCE.
Davide Italiano [Mon, 4 Mar 2019 04:38:58 +0000 (04:38 +0000)]
[InstCombine] Mark debug values as unavailable after DCE.

Fixes PR40838.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355301 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SubtargetFeatures] Add operator< for comparing SubtargetInfoKV objects. NFCI
Craig Topper [Mon, 4 Mar 2019 04:26:31 +0000 (04:26 +0000)]
[SubtargetFeatures] Add operator< for comparing SubtargetInfoKV objects. NFCI

Use instead of passing a lambda to std::is_sorted. This is more consistent with SubtargetFeatureKV.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355300 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SubtargetFeatures] Don't call ApplyFeatureFlag if the feature name is '+help'
Craig Topper [Mon, 4 Mar 2019 02:02:24 +0000 (02:02 +0000)]
[SubtargetFeatures] Don't call ApplyFeatureFlag if the feature name is '+help'

Just print the help and stop. Otherwise we'll print a message about it not being a real feature name after printing the help text.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355299 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SubtargetFeatuers] Simplify the code used to imply features from CPU name.
Craig Topper [Mon, 4 Mar 2019 02:02:22 +0000 (02:02 +0000)]
[SubtargetFeatuers] Simplify the code used to imply features from CPU name.

If we make SetImpliedBits OR features outside of its loop, we can reuse it for the first round of implying features for CPUs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355298 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly] Delete ThrowUnwindDest map from WasmEHFuncInfo
Heejin Ahn [Sun, 3 Mar 2019 22:35:56 +0000 (22:35 +0000)]
[WebAssembly] Delete ThrowUnwindDest map from WasmEHFuncInfo

Summary:
Before when we implemented the first EH proposal, 'catch <tag>'
instruction may not catch an exception so there were multiple EH pads an
exception can unwind to. That means a BB could have multiple EH pad
successors.

Now after we switched to the new proposal, every 'catch' instruction
catches an exception, and there is only one catchpad per catchswitch, so
we at most have one EH pad successor, making `ThrowUnwindDest` map in
`WasmEHInfo` unnecessary.

Keeping `ThrowUnwindDest` map in `WasmEHInfo` has its own problems,
because other optimization passes can split a BB that contains possibly
throwing calls (previously invokes), and we have to update the map every
time that happens, which is not easy for common CodeGen passes.

This also correctly updates successor info in LateEHPrepare when we add
a rethrow instruction.

Reviewers: dschuff

Subscribers: sbc100, jgravelle-google, sunfish, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58486

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355296 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Regenerate test to get the full FP operands printed. NFC
Craig Topper [Sun, 3 Mar 2019 20:28:52 +0000 (20:28 +0000)]
[X86] Regenerate test to get the full FP operands printed. NFC

Missed when I updated the printer to print implicit %st operand on binops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355295 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] remove stale FIXME comment from test; NFC
Sanjay Patel [Sun, 3 Mar 2019 19:08:54 +0000 (19:08 +0000)]
[InstCombine] remove stale FIXME comment from test; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355293 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ValueTracking] do not try to peek through bitcasts in computeKnownBitsFromAssume()
Sanjay Patel [Sun, 3 Mar 2019 18:59:33 +0000 (18:59 +0000)]
[ValueTracking] do not try to peek through bitcasts in computeKnownBitsFromAssume()

There are no tests for this case, and I'm not sure how it could ever work,
so I'm just removing this option from the matcher. This should fix PR40940:
https://bugs.llvm.org/show_bug.cgi?id=40940

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355292 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAdd extra ops in add to sub transform test in order to enforce proper operand orderin...
Amaury Sechet [Sun, 3 Mar 2019 15:11:13 +0000 (15:11 +0000)]
Add extra ops in add to sub transform test in order to enforce proper operand ordering. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355291 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DemandedBits] Remove some redundancy in the work list
Fangrui Song [Sun, 3 Mar 2019 14:50:01 +0000 (14:50 +0000)]
[DemandedBits] Remove some redundancy in the work list

InputIsKnownDead check is shared by all operands. Compute it once.

For non-integer instructions, use Visited.insert(I).second to replace a
find() and an insert().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355290 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRemove unused variable. NFCI.
Simon Pilgrim [Sun, 3 Mar 2019 14:23:07 +0000 (14:23 +0000)]
Remove unused variable. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355289 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] getShuffleScalarElt - peek through insert/extract subvector nodes.
Simon Pilgrim [Sun, 3 Mar 2019 14:11:05 +0000 (14:11 +0000)]
[X86] getShuffleScalarElt - peek through insert/extract subvector nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355288 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Pull out combineToConsecutiveLoads helper. NFCI.
Simon Pilgrim [Sun, 3 Mar 2019 13:53:27 +0000 (13:53 +0000)]
[X86] Pull out combineToConsecutiveLoads helper. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355287 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRemove large amount of empty lines mid-file. NFC
Kristina Brooks [Sun, 3 Mar 2019 13:21:38 +0000 (13:21 +0000)]
Remove large amount of empty lines mid-file. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355286 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DemandedBits] Optimize a find()+insert pattern with try_emplace and APInt::operator|=
Fangrui Song [Sun, 3 Mar 2019 11:12:57 +0000 (11:12 +0000)]
[DemandedBits] Optimize a find()+insert pattern with try_emplace and APInt::operator|=

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355284 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm] [Support] Reimplement getMainExecutable() using sysctl on NetBSD
Michal Gorny [Sun, 3 Mar 2019 10:06:40 +0000 (10:06 +0000)]
[llvm] [Support] Reimplement getMainExecutable() using sysctl on NetBSD

Use sysctl() to implement getMainExecutable() on NetBSD, rather than
trying to guess the correct path from argv[0].  This is one
of the fixes to recent clang-check-mac-libcxx-fixed-compilation-db.cpp
test failure on NetBSD.

This has been historically done on both FreeBSD and NetBSD in r303015,
and reverted in r303285 due to buggy implementation on FreeBSD.
However, FWIK the NetBSD implementation does not suffer from the same
bugs and is more reliable than playing with argv[0].

Differential Revision: https://reviews.llvm.org/D56975

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355283 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Prefer VPBLENDD for v2i64/v4i64 blends with AVX2.
Craig Topper [Sun, 3 Mar 2019 00:18:07 +0000 (00:18 +0000)]
[X86] Prefer VPBLENDD for v2i64/v4i64 blends with AVX2.

We were using VPBLENDW for v2i64 and VBLENDPD for v4i64. VPBLENDD has better throughput than VPBLENDW on some CPUs so it makes sense to use it when possible. VBLENDPD will probably become VBLENDD during execution domain fixing, but we might as well use integer in isel while we can.

This should work around some issues with the domain fixing pass prefering PBLENDW when we start with PBLENDW. There may still be some v8i16 cases that could use PBLENDD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355281 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAdd test case for add to sub transformation. NFC
Amaury Sechet [Sat, 2 Mar 2019 20:12:25 +0000 (20:12 +0000)]
Add test case for add to sub transformation. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355277 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Add a cfi/sources target.
Nico Weber [Sat, 2 Mar 2019 18:29:56 +0000 (18:29 +0000)]
gn build: Add a cfi/sources target.

This build target is currently unused, but after r355144 the sync script
started complaining about cfi.cpp not being listed, and this makes the
script happy again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355275 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] move add after smin/smax
Sanjay Patel [Sat, 2 Mar 2019 16:45:10 +0000 (16:45 +0000)]
[InstCombine] move add after smin/smax

Follow-up to rL355221.
This isn't specifically called for within PR14613,
but we'll get there eventually if it's not already
requested in some other bug report.

https://rise4fun.com/Alive/5b0

  Name: smax
  Pre: WillNotOverflowSignedSub(C1,C0)
  %a = add nsw i8 %x, C0
  %cond = icmp sgt i8 %a, C1
  %r = select i1 %cond, i8 %a, i8 C1
  =>
  %c2 = icmp sgt i8 %x, C1-C0
  %u2 = select i1 %c2, i8 %x, i8 C1-C0
  %r = add nsw i8 %u2, C0

  Name: smin
  Pre: WillNotOverflowSignedSub(C1,C0)
  %a = add nsw i32 %x, C0
  %cond = icmp slt i32 %a, C1
  %r = select i1 %cond, i32 %a, i32 C1
  =>
  %c2 = icmp slt i32 %x, C1-C0
  %u2 = select i1 %c2, i32 %x, i32 C1-C0
  %r = add nsw i32 %u2, C0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355272 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] add tests for add+smin/smax; NFC
Sanjay Patel [Sat, 2 Mar 2019 16:45:05 +0000 (16:45 +0000)]
[InstCombine] add tests for add+smin/smax; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355271 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAdd test case for add to sub transformation. NFC
Amaury Sechet [Sat, 2 Mar 2019 14:28:59 +0000 (14:28 +0000)]
Add test case for add to sub transformation. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355269 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix update_mir_test_checks.py to run on python3
Simon Pilgrim [Sat, 2 Mar 2019 11:14:01 +0000 (11:14 +0000)]
Fix update_mir_test_checks.py to run on python3

Split off from D58817

Differential Revision: https://reviews.llvm.org/D58820

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355268 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoUse SDValue::getConstantOperandAPInt helper where possible. NFCI.
Simon Pilgrim [Sat, 2 Mar 2019 11:11:22 +0000 (11:11 +0000)]
Use SDValue::getConstantOperandAPInt helper where possible. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355267 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Transforms] fix typo in test case. NFC.
Xing GUO [Sat, 2 Mar 2019 08:32:32 +0000 (08:32 +0000)]
[Transforms] fix typo in test case. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355265 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Codegen] fix typos in test case
Xing GUO [Sat, 2 Mar 2019 08:03:59 +0000 (08:03 +0000)]
[Codegen] fix typos in test case

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355264 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objdump] Should print unknown d_tag in hex format
Xing GUO [Sat, 2 Mar 2019 04:20:28 +0000 (04:20 +0000)]
[llvm-objdump] Should print unknown d_tag in hex format

Summary:
Currently, `llvm-objdump` prints "unknown" instead of d_tag value in hex format. Because getDynamicTagAsString returns "unknown" rather than empty
string.

Reviewers: grimar, jhenderson

Reviewed By: jhenderson

Subscribers: rupprecht, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58763

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355262 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly] Expand operations not supported by SIMD
Thomas Lively [Sat, 2 Mar 2019 03:32:25 +0000 (03:32 +0000)]
[WebAssembly] Expand operations not supported by SIMD

Summary:
This prevents crashes in instruction selection when these operations
are used. The tests check that the scalar version of the instruction
is used where applicable, although some expansions do not use the
scalar version.

Reviewers: aheejin

Subscribers: dschuff, sbc100, jgravelle-google, hiraditya, sunfish, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58859

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355261 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Improve use of SHLD/SHRD
Amaury Sechet [Sat, 2 Mar 2019 02:44:16 +0000 (02:44 +0000)]
[X86] Improve use of SHLD/SHRD

Summary:
This extends the variety of pattern that can generate a SHLD instead of using two shifts.

This fixes a regression that would be introduced by D57367 or D33587

Reviewers: RKSimon, craig.topper

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D57389

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355260 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SCEV] Handle case where MaxBECount is less precise than ExactBECount for OR.
Florian Hahn [Sat, 2 Mar 2019 02:31:44 +0000 (02:31 +0000)]
[SCEV] Handle case where MaxBECount is less precise than ExactBECount for OR.

In some cases, MaxBECount can be less precise than ExactBECount for AND
and OR (the AND case was PR26207). In the OR test case, both ExactBECounts are
undef, but MaxBECount are different, so we hit the assertion below. This
patch uses the same solution the AND case already uses.

Assertion failed:
   ((isa<SCEVCouldNotCompute>(ExactNotTaken) || !isa<SCEVCouldNotCompute>(MaxNotTaken))
     && "Exact is not allowed to be less precise than Max"), function ExitLimit

This patch also consolidates test cases for both AND and OR in a single
test case.

Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=13245

Reviewers: sanjoy, efriedma, mkazantsev

Reviewed By: sanjoy

Differential Revision: https://reviews.llvm.org/D58853

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355259 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAdd test case for truncate funnel shifts. NFC
Amaury Sechet [Sat, 2 Mar 2019 02:24:36 +0000 (02:24 +0000)]
Add test case for truncate funnel shifts. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355258 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SCEV] Remove undef check for SCEVConstant (NFC)
Florian Hahn [Sat, 2 Mar 2019 01:57:28 +0000 (01:57 +0000)]
[SCEV] Remove undef check for SCEVConstant (NFC)

The value stored in SCEVConstant is of type ConstantInt*, which can
never be UndefValue. So we should never hit that code.

Reviewers: mkazantsev, sanjoy

Reviewed By: sanjoy

Differential Revision: https://reviews.llvm.org/D58851

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355257 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert "[DWARFFormValue] Cleanup DWARFFormValue interface. (2/2) (NFC)"
Vlad Tsyrklevich [Sat, 2 Mar 2019 01:10:00 +0000 (01:10 +0000)]
Revert "[DWARFFormValue] Cleanup DWARFFormValue interface. (2/2) (NFC)"

This reverts commit r355233, it was causing UBSan failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355255 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert "[WebAssembly][WIP] Expand operations not supported by SIMD"
Thomas Lively [Sat, 2 Mar 2019 00:55:16 +0000 (00:55 +0000)]
Revert "[WebAssembly][WIP] Expand operations not supported by SIMD"

This was accidentally committed without tests or review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355254 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ProfileData] Sort FuncData before iteration to remove non-determinism
Mandeep Singh Grang [Sat, 2 Mar 2019 00:47:43 +0000 (00:47 +0000)]
[ProfileData] Sort FuncData before iteration to remove non-determinism

Reviewers: rsmith, bogner, dblaikie

Reviewed By: dblaikie

Subscribers: Hahnfeld, jdoerfert, vsk, dblaikie, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D57986

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355252 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly][WIP] Expand operations not supported by SIMD
Thomas Lively [Sat, 2 Mar 2019 00:18:07 +0000 (00:18 +0000)]
[WebAssembly][WIP] Expand operations not supported by SIMD

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355247 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm] Fix typo: 's/analsyis/analysis/' [NFC]
Mandeep Singh Grang [Sat, 2 Mar 2019 00:14:10 +0000 (00:14 +0000)]
[llvm] Fix typo: 's/analsyis/analysis/' [NFC]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355246 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[tblgen] Track CodeInit origins when possible
Daniel Sanders [Sat, 2 Mar 2019 00:12:57 +0000 (00:12 +0000)]
[tblgen] Track CodeInit origins when possible

Summary:
Add an SMLoc to CodeInit that records the source line it originated from.
This allows tablegen to point precisely at portions of code when reporting
errors within the CodeInit. For example, in the upcoming GlobalISel
combiner, it can report undefined expansions and point at the instance of
the expansion. This is achieved using something like:
  SMLoc::getFromPointer(SMLoc::getPointer() +
                        (StringRef - CodeInit::getValue()))

The location is lost when producing a CodeInit by string concatenation so
a fallback SMLoc is required (e.g. the Record::getLoc()) but that's pretty
rare for CodeInits.

There's a reasonable case for extending tracking of a couple other Init
objects, for example StringInit's are often parsed and it would be good to
point inside the string when reporting errors about that. However, location
tracking also harms de-duplication. This is fine for CodeInit where there's
only a few hundred of them (~160 for X86) and it may be worth it for
StringInit (~86k up to ~1.9M for roughly 15MB increase for X86).
However the origin tracking would be a _terrible_ idea for IntInit, BitInit,
and UnsetInit. I haven't measured either of those three but BitInit would
most likely be on the order of increasing the current 2 BitInit values up
to billions.

Reviewers: volkan, aditya_nandakumar, bogner, paquette, aemerson

Reviewed By: paquette

Subscribers: javed.absar, kristof.beyls, dexonsmith, llvm-commits, kristina

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58141

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355245 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agollvm-dwarfdump: Add new variable, parameter and inlining statistics; also function...
Caroline Tice [Fri, 1 Mar 2019 23:51:54 +0000 (23:51 +0000)]
llvm-dwarfdump: Add new variable, parameter and inlining statistics; also function source location statistics.

Add statistics for abstract origins, function, variable and parameter
locations; break the 'variable' counts down into variables and
parameters. Also update call site counting to check for
DW_AT_call_{file,line} in addition to DW_TAG_call_site.

Differential revision: https://reviews.llvm.org/D58849

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355243 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoTry to fix Windows bots after r355226.
Paul Robinson [Fri, 1 Mar 2019 22:28:13 +0000 (22:28 +0000)]
Try to fix Windows bots after r355226.

Windows has two path separator characters.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355235 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DWARFFormValue] Cleanup DWARFFormValue interface. (2/2) (NFC)
Jonas Devlieghere [Fri, 1 Mar 2019 22:14:24 +0000 (22:14 +0000)]
[DWARFFormValue] Cleanup DWARFFormValue interface. (2/2) (NFC)

Continues the work started in r354941. Changes (all but one) uses of the
extractValue to static createFromData.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355233 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove IntrArgMemOnly from target specific gather/scatter intrinsics
Craig Topper [Fri, 1 Mar 2019 21:02:40 +0000 (21:02 +0000)]
[X86] Remove IntrArgMemOnly from target specific gather/scatter intrinsics

IntrArgMemOnly implies that only memory pointed to by pointer typed arguments will be accessed. But these intrinsics allow you to pass null to the pointer argument and put the full address into the index argument. Other passes won't be able to understand this.

A colleague found that ISPC was creating gathers like this and then dead store elimination removed some stores because it didn't understand what the gather was doing since the pointer argument was null.

Differential Revision: https://reviews.llvm.org/D58805

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355228 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add test case for D58805. NFC
Craig Topper [Fri, 1 Mar 2019 21:02:34 +0000 (21:02 +0000)]
[X86] Add test case for D58805. NFC

This demonstrates dead store elimination removing a store that may alias a gather that uses null as its base.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355227 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DWARF] Make -g with empty assembler source work better.
Paul Robinson [Fri, 1 Mar 2019 20:58:04 +0000 (20:58 +0000)]
[DWARF] Make -g with empty assembler source work better.

This was sometimes causing clang or llvm-mc to crash, and in other
cases could emit a bogus DWARF line-table header. I did an interim
patch in r352541; this patch should be a cleaner and more complete
fix, and retains the test.

Addresses PR40538.

Differential Revision: https://reviews.llvm.org/D58750

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355226 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TableGen][SelectionDAG][X86] Add specific isel matchers for immAllZerosV/immAllOnesV...
Craig Topper [Fri, 1 Mar 2019 20:18:38 +0000 (20:18 +0000)]
[TableGen][SelectionDAG][X86] Add specific isel matchers for immAllZerosV/immAllOnesV. Remove bitcasts from X86 patterns that are no longer necessary.

Previously we had build_vector PatFrags that called ISD::isBuildVectorAllZeros/Ones. Internally the ISD::isBuildVectorAllZeros/Ones look through bitcasts, but we aren't able to take advantage of that in isel. Instead of we have to canonicalize the types of the all zeros/ones build_vectors and insert bitcasts. Then we have to pattern match those exact bitcasts.

By emitting specific matchers for these 2 nodes, we can make isel look through any bitcasts without needing to explicitly match them. We should also be able to remove the canonicalization to vXi32 from lowering, but I've left that for a follow up.

This removes something like 40,000 bytes from the X86 isel table.

Differential Revision: https://reviews.llvm.org/D58595

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355224 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ValueTracking] Known bits support for unsigned saturating add/sub
Nikita Popov [Fri, 1 Mar 2019 20:07:04 +0000 (20:07 +0000)]
[ValueTracking] Known bits support for unsigned saturating add/sub

We have two sources of known bits:

1. For adds leading ones of either operand are preserved. For sub
leading zeros of LHS and leading ones of RHS become leading zeros in
the result.

2. The saturating math is a select between add/sub and an all-ones/
zero value. As such we can carry out the add/sub known bits
calculation, and only preseve the known one/zero bits respectively.

Differential Revision: https://reviews.llvm.org/D58329

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355223 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Extend saturating idempotent atomicrmw transform to FP
Philip Reames [Fri, 1 Mar 2019 19:50:36 +0000 (19:50 +0000)]
[InstCombine] Extend saturating idempotent atomicrmw transform to FP

I'm assuming that the nan propogation logic for InstructonSimplify's handling of fadd and fsub is correct, and applying the same to atomicrmw.

Differential Revision: https://reviews.llvm.org/D58836

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355222 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] move add after umin/umax
Sanjay Patel [Fri, 1 Mar 2019 19:42:40 +0000 (19:42 +0000)]
[InstCombine] move add after umin/umax

In the motivating cases from PR14613:
https://bugs.llvm.org/show_bug.cgi?id=14613
...moving the add enables us to narrow the
min/max which eliminates zext/trunc which
enables signficantly better vectorization.
But that bug is still not completely fixed.

https://rise4fun.com/Alive/5KQ

  Name: umax
  Pre: C1 u>= C0
  %a = add nuw i8 %x, C0
  %cond = icmp ugt i8 %a, C1
  %r = select i1 %cond, i8 %a, i8 C1
  =>
  %c2 = icmp ugt i8 %x, C1-C0
  %u2 = select i1 %c2, i8 %x, i8 C1-C0
  %r = add nuw i8 %u2, C0

  Name: umin
  Pre: C1 u>= C0
  %a = add nuw i32 %x, C0
  %cond = icmp ult i32 %a, C1
  %r = select i1 %cond, i32 %a, i32 C1
  =>
  %c2 = icmp ult i32 %x, C1-C0
  %u2 = select i1 %c2, i32 %x, i32 C1-C0
  %r = add nuw i32 %u2, C0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355221 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] add tests for umin/umax narrowing (PR14613); NFC
Sanjay Patel [Fri, 1 Mar 2019 19:42:34 +0000 (19:42 +0000)]
[InstCombine] add tests for umin/umax narrowing (PR14613); NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355220 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert "[MIPS GlobalISel] Fix mul operands"
Vlad Tsyrklevich [Fri, 1 Mar 2019 18:58:22 +0000 (18:58 +0000)]
Revert "[MIPS GlobalISel] Fix mul operands"

This reverts commit r355178, it is causing ASan failures on the
sanitizer bots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355219 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoELF: Change FileSize back to a uint64_t.
Peter Collingbourne [Fri, 1 Mar 2019 18:53:41 +0000 (18:53 +0000)]
ELF: Change FileSize back to a uint64_t.

This lets us detect file size overflows when creating a 64-bit binary on
a 32-bit machine.

Differential Revision: https://reviews.llvm.org/D58840

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355218 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LICM] Infer proper alignment from loads during scalar promotion
Philip Reames [Fri, 1 Mar 2019 18:45:05 +0000 (18:45 +0000)]
[LICM] Infer proper alignment from loads during scalar promotion

This patch fixes an issue where we would compute an unnecessarily small alignment during scalar promotion when no store is not to be guaranteed to execute, but we've proven load speculation safety. Since speculating a load requires proving the existing alignment is valid at the new location (see Loads.cpp), we can use the alignment fact from the load.

For non-atomics, this is a performance problem. For atomics, this is a correctness issue, though an *incredibly* rare one to see in practice. For atomics, we might not be able to lower an improperly aligned load or store (i.e. i32 align 1). If such an instruction makes it all the way to codegen, we *may* fail to codegen the operation, or we may simply generate a slow call to a library function. The part that makes this super hard to see in practice is that the memory location actually *is* well aligned, and instcombine knows that. So, to see a failure, you have to have a) hit the bug in LICM, b) somehow hit a depth limit in InstCombine/ValueTracking to avoid fixing the alignment, and c) then have generated an instruction which fails codegen rather than simply emitting a slow libcall. All around, pretty hard to hit.

Differential Revision: https://reviews.llvm.org/D58809

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355217 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PGO] Remove the default parameter in ProfileSummary API. NFC
Rong Xu [Fri, 1 Mar 2019 18:36:24 +0000 (18:36 +0000)]
[PGO] Remove the default parameter in ProfileSummary API. NFC

Remove the default parameter in setProfileSummary() and getSummary().
This is a follow-up of r355131.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355216 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Tests] More missing atomicrmw combines
Philip Reames [Fri, 1 Mar 2019 18:24:05 +0000 (18:24 +0000)]
[Tests] More missing atomicrmw combines

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355215 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Tests] Add tests for missed optimizations of saturating and idempotent FP atomicrmws
Philip Reames [Fri, 1 Mar 2019 18:10:37 +0000 (18:10 +0000)]
[Tests] Add tests for missed optimizations of saturating and idempotent FP atomicrmws

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355212 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Extend "idempotent" atomicrmw optimizations to floating point
Philip Reames [Fri, 1 Mar 2019 18:00:07 +0000 (18:00 +0000)]
[InstCombine] Extend "idempotent" atomicrmw optimizations to floating point

An idempotent atomicrmw is one that does not change memory in the process of execution.  We have already added handling for the various integer operations; this patch extends the same handling to floating point operations which were recently added to IR.

Note: At the moment, we canonicalize idempotent fsub to fadd when ordering requirements prevent us from using a load.  As discussed in the review, I will be replacing this with canonicalizing both floating point ops to integer ops in the near future.

Differential Revision: https://reviews.llvm.org/D58251

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355210 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert "[WebAssembly] Lower SIMD shifts since they are fixed in V8"
Thomas Lively [Fri, 1 Mar 2019 17:43:55 +0000 (17:43 +0000)]
Revert "[WebAssembly] Lower SIMD shifts since they are fixed in V8"

They weren't fixed in V8. Oops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355208 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj] Display section names for STT_SECTION symbols.
Matt Davis [Fri, 1 Mar 2019 17:31:32 +0000 (17:31 +0000)]
[llvm-readobj] Display section names for STT_SECTION symbols.

Summary:
This patch will obtain the section name for symbols that refer to a section.  Prior to this patch the Name field for STT_SECTIONs was blank, now it is populated.

Before:
```
Symbol table '.symtab' contains 6 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 0000000000000000     0 SECTION LOCAL  DEFAULT    1
     2: 0000000000000000     0 SECTION LOCAL  DEFAULT    3
     3: 0000000000000000     0 SECTION LOCAL  DEFAULT    4
     4: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT  UND _GLOBAL_OFFSET_TABLE_
     5: 0000000000000000     0 TLS     GLOBAL DEFAULT  UND sym
```

With this patch:
```
Symbol table '.symtab' contains 6 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 0000000000000000     0 SECTION LOCAL  DEFAULT    1 .text
     2: 0000000000000000     0 SECTION LOCAL  DEFAULT    3 .data
     3: 0000000000000000     0 SECTION LOCAL  DEFAULT    4 .bss
     4: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT  UND _GLOBAL_OFFSET_TABLE_
     5: 0000000000000000     0 TLS     GLOBAL DEFAULT  UND sym
```

This fixes PR40788

Reviewers: jhenderson, rupprecht, espindola

Reviewed By: rupprecht

Subscribers: emaste, javed.absar, arichardson, MaskRay, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58796

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355207 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] add tests for add+umin/umax canonicalization; NFC
Sanjay Patel [Fri, 1 Mar 2019 17:29:10 +0000 (17:29 +0000)]
[InstCombine] add tests for add+umin/umax canonicalization; NFC

Fixing this should solve the biggest part of the vector problems seen in:
https://bugs.llvm.org/show_bug.cgi?id=14613

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355206 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoHide two unused debugging methods, NFCI.
Jonas Hahnfeld [Fri, 1 Mar 2019 17:15:21 +0000 (17:15 +0000)]
Hide two unused debugging methods, NFCI.

GCC correctly moans that PlainCFGBuilder::isExternalDef(llvm::Value*) and
StackSafetyDataFlowAnalysis::verifyFixedPoint() are defined but not used
in Release builds. Hide them behind 'ifndef NDEBUG'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355205 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoTry to fix NetBSD buildbot breakage introduced in D57463.
Manman Ren [Fri, 1 Mar 2019 15:25:24 +0000 (15:25 +0000)]
Try to fix NetBSD buildbot breakage introduced in D57463.

By including the header file in the source.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355202 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Fix FP16 stack loads/stores for Thumb2 with frame pointer
Oliver Stannard [Fri, 1 Mar 2019 14:20:28 +0000 (14:20 +0000)]
[ARM] Fix FP16 stack loads/stores for Thumb2 with frame pointer

The new addressing mode added for the v8.2A FP16 instructions uses bit 8 of the
immediate to encode the sign of the offset, like the other FP loads/stores, so
need to be treated the same way.

Differential revision: https://reviews.llvm.org/D58816

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355201 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Consider undefined-on-NaN conditions in checkVSELConstraints
Oliver Stannard [Fri, 1 Mar 2019 13:58:25 +0000 (13:58 +0000)]
[ARM] Consider undefined-on-NaN conditions in checkVSELConstraints

This function was not checking for the condition code variants which are
undefined if either input is NaN, so we were missing selection of the VSEL
instruction in some cases when using -fno-honor-nans or -ffast-math.

Differential revision: https://reviews.llvm.org/D58812

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355199 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Regenerate legalize test files
Simon Pilgrim [Fri, 1 Mar 2019 13:13:40 +0000 (13:13 +0000)]
[X86] Regenerate legalize test files

Noticed while getting update_mir_test_checks.py to work on python3

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355198 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Thumb] Add some integer abs testcases for different typesizes.
Simon Pilgrim [Fri, 1 Mar 2019 12:08:50 +0000 (12:08 +0000)]
[Thumb] Add some integer abs testcases for different typesizes.

Committed on behalf of @ikulagin (Ivan Kulagin)

Differential Revision: https://reviews.llvm.org/D52138

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355197 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2obj] - Allow setting custom sh_info for RawContentSection sections.
George Rimar [Fri, 1 Mar 2019 10:18:16 +0000 (10:18 +0000)]
[yaml2obj] - Allow setting custom sh_info for RawContentSection sections.

This is for tweaking SHT_SYMTAB sections.
Their sh_info contains the (number of symbols + 1) usually.
But for creating invalid inputs for test cases it would be convenient
to allow explicitly override this field from YAML.

Differential revision: https://reviews.llvm.org/D58779

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355193 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago Attempt to fix buildbot after r354972 [#2]. NFCI.
Alexey Lapshin [Fri, 1 Mar 2019 10:15:18 +0000 (10:15 +0000)]
Attempt to fix buildbot after r354972 [#2]. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355192 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM GlobalISel] Support G_CTLZ for Thumb2
Diana Picus [Fri, 1 Mar 2019 10:12:28 +0000 (10:12 +0000)]
[ARM GlobalISel] Support G_CTLZ for Thumb2

Same as ARM mode but with different opcode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355191 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM GlobalISel] Check target flags in test. NFCI
Diana Picus [Fri, 1 Mar 2019 10:01:22 +0000 (10:01 +0000)]
[ARM GlobalISel] Check target flags in test. NFCI

There was a time when we couldn't dump target-specific flags such as
arm-sbrel etc, so the tests didn't check for them. We can now be more
specific in our tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355189 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Tablegen] Add support for the !mul operator.
Nicola Zaghen [Fri, 1 Mar 2019 09:46:29 +0000 (09:46 +0000)]
[Tablegen] Add support for the !mul operator.

This is a small addition to arithmetic operations that improves
expressiveness of the language.

Differential Revision: https://reviews.llvm.org/D58775

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355187 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Add armv8a triple to test check updaters
Oliver Stannard [Fri, 1 Mar 2019 09:26:21 +0000 (09:26 +0000)]
[ARM] Add armv8a triple to test check updaters

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355186 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CommandLine] Allow grouping options which can have values.
Igor Kudrin [Fri, 1 Mar 2019 09:22:42 +0000 (09:22 +0000)]
[CommandLine] Allow grouping options which can have values.

This patch allows all forms of values for options to be used at the end
of a group. With the fix, it is possible to follow the way GNU binutils
tools handle grouping options better. For example, the -j option can be
used with objdump in any of the following ways:

$ objdump -d -j .text a.o
$ objdump -d -j.text a.o
$ objdump -dj .text a.o
$ objdump -dj.text a.o

Differential Revision: https://reviews.llvm.org/D58711

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355185 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CommandLine] Do not crash if an option has both ValueRequired and Grouping.
Igor Kudrin [Fri, 1 Mar 2019 09:20:56 +0000 (09:20 +0000)]
[CommandLine] Do not crash if an option has both ValueRequired and Grouping.

If an option, which requires a value, has a `cl::Grouping` formatting
modifier, it works well as far as it is used at the end of a group,
or as a separate argument. However, if the option appears accidentally
in the middle of a group, the program just crashes. This patch prints
an error message instead.

Differential Revision: https://reviews.llvm.org/D58499

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355184 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj]Add test showing behaviour of thin archive member path printing
James Henderson [Fri, 1 Mar 2019 09:12:38 +0000 (09:12 +0000)]
[llvm-readobj]Add test showing behaviour of thin archive member path printing

This was a test requested in https://reviews.llvm.org/D58677.

Reviewed by: rupprecht, grimar, Higuoxing

Differential Revision: https://reviews.llvm.org/D5877

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355183 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Mark ds instructions as meybeAtomic
Stanislav Mekhanoshin [Fri, 1 Mar 2019 07:59:17 +0000 (07:59 +0000)]
[AMDGPU] Mark ds instructions as meybeAtomic

These were not recognized as potential atomics by memory legalizer.
The test was working not because legalizer did a right thing, but
because it has skipped all these instructions. When I have fixed
DS desciption test started to fail because region address has
changed from 4 to 2 a while ago.

Differential Revision: https://reviews.llvm.org/D58802

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355179 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIPS GlobalISel] Fix mul operands
Petar Avramovic [Fri, 1 Mar 2019 07:35:57 +0000 (07:35 +0000)]
[MIPS GlobalISel] Fix mul operands

Unsigned mul high for MIPS32 is selected into two PseudoInstructions:
PseudoMULTu and PseudoMFHI that use accumulator register class ACC64 for
some of its operands. Registers in this class have appropriate hi and lo
register as subregisters: $lo0 and $hi0 are subregisters of $ac0 etc.
mul instruction implicit-defs $lo0 and $hi0 according to MipsInstrInfo.td.
In functions where mul and PseudoMULTu are present fastRegisterAllocator
will "run out of registers during register allocation" because
'calcSpillCost' for $ac0 will return spillImpossible because subregisters
$lo0 and $hi0 of $ac0 are reserved by mul instruction above. A solution is
to mark implicit-defs of $lo0 and $hi0 as dead in mul instruction.

Differential Revision: https://reviews.llvm.org/D58715

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355178 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIPS GlobalISel] Select G_UMULH
Petar Avramovic [Fri, 1 Mar 2019 07:25:44 +0000 (07:25 +0000)]
[MIPS GlobalISel] Select G_UMULH

Legalize G_UMULO and select G_UMULH for MIPS32.

Differential Revision: https://reviews.llvm.org/D58714

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355177 91177308-0d34-0410-b5e6-96231b3b80d8