platform/upstream/llvm.git
5 years ago[llvm-exegesis] Analysis: writeMeasurementValue(): don't alloc string for double...
Roman Lebedev [Mon, 19 Nov 2018 13:28:17 +0000 (13:28 +0000)]
[llvm-exegesis] Analysis: writeMeasurementValue(): don't alloc string for double each time.

Summary:
Test data: 500kLOC of benchmark.yaml, 23Mb. (that is a subset of the actual uops benchmark i was trying to analyze!)
Old time: (D54382)
```
 Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=100000 -benchmarks-file=/tmp/benchmarks.yaml -analysis-inconsistencies-output-file=/tmp/clusters.html' (16 runs):

       9024.354355      task-clock (msec)         #    1.000 CPUs utilized            ( +-  0.18% )
...
            9.0262 +- 0.0161 seconds time elapsed  ( +-  0.18% )
```
New time:
```
 Performance counter stats for './bin/llvm-exegesis -mode=analysis -analysis-epsilon=100000 -benchmarks-file=/tmp/benchmarks.yaml -analysis-inconsistencies-output-file=/tmp/clusters.html' (16 runs):

       8996.541057      task-clock (msec)         #    0.999 CPUs utilized            ( +-  0.19% )
...
            9.0045 +- 0.0172 seconds time elapsed  ( +-  0.19% )
```
-~0.3%, not that much. But this isn't the important part.

Old:
* calls to allocation functions: 2109712
* temporary allocations: 33112
* bytes allocated in total (ignoring deallocations): 4.43 GB

New:
* calls to allocation functions: 2095345 (-0.68%)
* temporary allocations: 18745 (-43.39% !!!)
* bytes allocated in total (ignoring deallocations): 4.31 GB (-2.71%)

Reviewers: courbet, MaskRay, RKSimon, gchatelet, john.brawn

Reviewed By: courbet

Subscribers: tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D54383

llvm-svn: 347199

5 years ago[llvm-exegesis] Analysis::writeSnippet(): be smarter about memory allocations.
Roman Lebedev [Mon, 19 Nov 2018 13:28:14 +0000 (13:28 +0000)]
[llvm-exegesis] Analysis::writeSnippet(): be smarter about memory allocations.

Summary:
Test data: 500kLOC of benchmark.yaml, 23Mb. (that is a subset of the actual uops benchmark i was trying to analyze!)
Old time: (D54381)
```
$ time ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=100000 -benchmarks-file=/tmp/benchmarks.yaml -analysis-inconsistencies-output-file=/tmp/clusters.html &> /dev/null

real    0m10.487s
user    0m9.745s
sys     0m0.740s
```
New time:
```
$ time ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=100000 -benchmarks-file=/tmp/benchmarks.yaml -analysis-inconsistencies-output-file=/tmp/clusters.html &> /dev/null

real    0m9.599s
user    0m8.824s
sys     0m0.772s

```
Not that much, around -9%. But that is not the good part yet, again.

Old:
* calls to allocation functions: 3347676
* temporary allocations: 277818
* bytes allocated in total (ignoring deallocations): 10.52 GB

New:
* calls to allocation functions: 2109712 (-36%)
* temporary allocations: 33112 (-88%)
* bytes allocated in total (ignoring deallocations): 4.43 GB (-58% *sic*)

Reviewers: courbet, MaskRay, RKSimon, gchatelet, john.brawn

Reviewed By: courbet, MaskRay

Subscribers: tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D54382

llvm-svn: 347198

5 years ago[llvm-exegesis] InstructionBenchmarkClustering::dbScan(): use llvm::SetVector<> inste...
Roman Lebedev [Mon, 19 Nov 2018 13:28:09 +0000 (13:28 +0000)]
[llvm-exegesis] InstructionBenchmarkClustering::dbScan(): use llvm::SetVector<> instead of ILLEGAL std::unordered_set<>

Summary:
Test data: 500kLOC of benchmark.yaml, 23Mb. (that is a subset of the actual uops benchmark i was trying to analyze!)
Old time:
```
$ time ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=100000 -benchmarks-file=/tmp/benchmarks.yaml -analysis-inconsistencies-output-file=/tmp/clusters.html &> /dev/null

real    0m24.884s
user    0m24.099s
sys     0m0.785s
```
New time:
```
$ time ./bin/llvm-exegesis -mode=analysis -analysis-epsilon=100000 -benchmarks-file=/tmp/benchmarks.yaml -analysis-inconsistencies-output-file=/tmp/clusters.html &> /dev/null

real    0m10.469s
user    0m9.797s
sys     0m0.672s
```
So -60%. And that isn't the good bit yet.

Old:
* calls to allocation functions: 106560180  (yes, 107 *million* allocations.)
* bytes allocated in total (ignoring deallocations): 12.17 GB

New:
* calls to allocation functions: 3347676  (-96.86%)  (just 3 mil)
* bytes allocated in total (ignoring deallocations): 10.52 GB (~2GB less)

---

Two points i want to raise:
* `std::unordered_set<>` should not have been used there in the first place.
  It is banned by the https://llvm.org/docs/ProgrammersManual.html#other-set-like-container-options
* There is no tests, so i'm not fully sure this is correct.
  Since it was unordered set, i guess there are zero restrictions on the order, and anything will be ok?
* I tried other containers suggested in https://llvm.org/docs/ProgrammersManual.html#set-like-containers-std-set-smallset-setvector-etc,
  this `llvm::SetVector<>` seems to be best here.

Reviewers: courbet, MaskRay, RKSimon, gchatelet, john.brawn

Reviewed By: courbet

Subscribers: kristina, bobsayshilol, tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D54381

llvm-svn: 347197

5 years agoFixed uninitialized variable issue.
Anastasia Stulova [Mon, 19 Nov 2018 12:43:39 +0000 (12:43 +0000)]
Fixed uninitialized variable issue.

This commit should fix failing bots.

llvm-svn: 347196

5 years ago[X86] Add codegen tests for slow-shld scalar funnel shifts
Simon Pilgrim [Mon, 19 Nov 2018 12:29:41 +0000 (12:29 +0000)]
[X86] Add codegen tests for slow-shld scalar funnel shifts

llvm-svn: 347195

5 years agoTest commit - delete trailing space.
Michael Platings [Mon, 19 Nov 2018 12:16:05 +0000 (12:16 +0000)]
Test commit - delete trailing space.

llvm-svn: 347194

5 years agoTest commit - delete a trailing space.
Michael Platings [Mon, 19 Nov 2018 12:10:07 +0000 (12:10 +0000)]
Test commit - delete a trailing space.

llvm-svn: 347193

5 years agoAMDGPU/InsertWaitcnts: Some more const-correctness
Nicolai Haehnle [Mon, 19 Nov 2018 12:03:11 +0000 (12:03 +0000)]
AMDGPU/InsertWaitcnts: Some more const-correctness

Reviewers: msearles, rampitec, scott.linder, kanarayan

Subscribers: arsenm, kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits, hakzsam

Differential Revision: https://reviews.llvm.org/D54225

llvm-svn: 347192

5 years ago[ARM] Remove trunc sinks in ARM CGP
Sam Parker [Mon, 19 Nov 2018 11:34:40 +0000 (11:34 +0000)]
[ARM] Remove trunc sinks in ARM CGP

Truncs are treated as sources if their produce a value of the same
type as the one we currently trying to promote. Truncs used to be
considered as a sink if their operand was the same value type.

We now allow smaller types in the search, so we should search through
truncs that produce a smaller value. These truncs can then be
converted to an AND mask.

This leaves sinks as being:
  - points where the value in the register is being observed, such as
    an icmp, switch or store.
  - points where value types have to match, such as calls and returns.
  - zext are included to ease the transformation and are generally
    removed later on.

During this change, it also became apart from truncating sinks was
broken: if a sink used a source, its type information had already
been lost by the time the truncation happens. So I've changed the
method of caching the type information.

Differential Revision: https://reviews.llvm.org/D54515

llvm-svn: 347191

5 years ago[LICM] Make LICM able to hoist phis
John Brawn [Mon, 19 Nov 2018 11:31:24 +0000 (11:31 +0000)]
[LICM] Make LICM able to hoist phis

The general approach taken is to make note of loop invariant branches, then when
we see something conditional on that branch, such as a phi, we create a copy of
the branch and (empty versions of) its successors and hoist using that.

This has no impact by itself that I've been able to see, as LICM typically
doesn't see such phis as they will have been converted into selects by the time
LICM is run, but once we start doing phi-to-select conversion later it will be
important.

Differential Revision: https://reviews.llvm.org/D52827

llvm-svn: 347190

5 years ago[OpenCL] Fix address space deduction in template args.
Anastasia Stulova [Mon, 19 Nov 2018 11:00:14 +0000 (11:00 +0000)]
[OpenCL] Fix address space deduction in template args.

Don't deduce address spaces for non-pointer-like types
in template args.

Fixes PR38603!

Differential Revision: https://reviews.llvm.org/D54634

llvm-svn: 347189

5 years agoRemove unused variable. NFC.
Benjamin Kramer [Mon, 19 Nov 2018 10:59:12 +0000 (10:59 +0000)]
Remove unused variable. NFC.

llvm-svn: 347188

5 years ago[MSP430] Optimize srl/sra in case of A >> (8 + N)
Anton Korobeynikov [Mon, 19 Nov 2018 10:43:02 +0000 (10:43 +0000)]
[MSP430] Optimize srl/sra in case of A >> (8 + N)

There is no variable-length shifts on MSP430. Therefore
"eat" 8 bits of shift via bswap & ext.

Path by Kristina Bessonova!

Differential Revision: https://reviews.llvm.org/D54623

llvm-svn: 347187

5 years agoFix disturbing warning - NFCI
Serge Guelton [Mon, 19 Nov 2018 10:05:28 +0000 (10:05 +0000)]
Fix disturbing warning - NFCI

llvm-svn: 347186

5 years ago[X86] Use a pcmpgt with 0 instead of psrad 31, to fill elements with the sign bit...
Craig Topper [Mon, 19 Nov 2018 07:22:26 +0000 (07:22 +0000)]
[X86] Use a pcmpgt with 0 instead of psrad 31, to fill elements with the sign bit in v4i32 MULH lowering.

The shift requires a copy to avoid clobbering a register. Comparing with 0 uses an xor to produce 0 that will be overwritten with the compare results. So still requires 2 instructions, but should be one byte shorter since it doesn't need to encode an immediate.

llvm-svn: 347185

5 years ago[LoopSimplifyCFG] Add requires: asserts after rL347183
Fangrui Song [Mon, 19 Nov 2018 06:28:15 +0000 (06:28 +0000)]
[LoopSimplifyCFG] Add requires: asserts after rL347183

llvm-svn: 347184

5 years ago[LoopSimplifyCFG] Teach LoopSimplifyCFG to constant-fold branches and switches
Max Kazantsev [Mon, 19 Nov 2018 05:54:38 +0000 (05:54 +0000)]
[LoopSimplifyCFG] Teach LoopSimplifyCFG to constant-fold branches and switches

This patch introduces infrastructure and the simplest case for constant-folding
of branch and switch instructions within loop into unconditional branches.
It is useful as a cleanup for such passes as loop unswitching that sometimes
produce such branches.

Only the simplest case supported in this patch: after the folding, no block
should become dead or stop being part of the loop. Support for more
sophisticated cases will go separately in follow-up patches.

Differential Revision: https://reviews.llvm.org/D54021
Reviewed By: anna

llvm-svn: 347183

5 years ago[ProfileSummary] Standardize methods and fix comment
Vedant Kumar [Mon, 19 Nov 2018 05:23:16 +0000 (05:23 +0000)]
[ProfileSummary] Standardize methods and fix comment

Every Analysis pass has a get method that returns a reference of the Result of
the Analysis, for example, BlockFrequencyInfo
&BlockFrequencyInfoWrapperPass::getBFI().  I believe that
ProfileSummaryInfo::getPSI() is the only exception to that, as it was returning
a pointer.

Another change is renaming isHotBB and isColdBB to isHotBlock and isColdBlock,
respectively.  Most methods use BB as the argument of variable names while
methods usually refer to Basic Blocks as Blocks, instead of BB.  For example,
Function::getEntryBlock, Loop:getExitBlock, etc.

I also fixed one of the comments.

Patch by Rodrigo Caetano Rocha!

Differential Revision: https://reviews.llvm.org/D54669

llvm-svn: 347182

5 years ago[X86] Use compare with 0 to fill an element with sign bits when sign extending to...
Craig Topper [Mon, 19 Nov 2018 04:33:20 +0000 (04:33 +0000)]
[X86] Use compare with 0 to fill an element with sign bits when sign extending to v2i64 pre-sse4.1

Previously we used an arithmetic shift right by 31, but that requires a copy to preserve the input. So we might as well materialize a zero and compare to it since the comparison will overwrite the register that contains the zeros. This should be one byte shorter.

llvm-svn: 347181

5 years ago[X86] Remove most of the SEXTLOAD Custom setOperationAction calls under -x86-experime...
Craig Topper [Mon, 19 Nov 2018 00:33:16 +0000 (00:33 +0000)]
[X86] Remove most of the SEXTLOAD Custom setOperationAction calls under -x86-experimental-vector-widening-legalization.

Leave just the v4i8->v4i64 and v8i8->v8i64, but only enable them on pre-sse4.1 targets when 64-bit mode is enabled. In those cases we end up creating sext loads that get scalarized to code that looks better than what we get from loading into a vector register and doing a multiple step sign extend using unpacks and shifts.

llvm-svn: 347180

5 years ago[PowerPC] Set the default PLT mode on OpenBSD/powerpc to Secure PLT.
Brad Smith [Mon, 19 Nov 2018 00:21:06 +0000 (00:21 +0000)]
[PowerPC] Set the default PLT mode on OpenBSD/powerpc to Secure PLT.

OpenBSD/powerpc only supports Secure PLT.

llvm-svn: 347179

5 years agoReplace the UTF-8 characters in the error message.
Brad Smith [Sun, 18 Nov 2018 22:30:58 +0000 (22:30 +0000)]
Replace the UTF-8 characters in the error message.

llvm-svn: 347178

5 years ago[X86][SSE] Add SimplifyDemandedVectorElts support for SSE packed i2fp conversions.
Simon Pilgrim [Sun, 18 Nov 2018 22:13:31 +0000 (22:13 +0000)]
[X86][SSE] Add SimplifyDemandedVectorElts support for SSE packed i2fp conversions.

llvm-svn: 347177

5 years ago[X86] Add custom type legalization for extending v4i8/v4i16->v4i64.
Craig Topper [Sun, 18 Nov 2018 21:28:50 +0000 (21:28 +0000)]
[X86] Add custom type legalization for extending v4i8/v4i16->v4i64.

Pre-SSE4.1 sext_invec for v2i64 is complicated because we don't have a v2i64 sra instruction. So instead we sign extend to i32 using unpack and sra, then copy the elements and do a v4i32 sra to fill with sign bits, then interleave the i32 sign extend and the sign bits. So really we're doing to two sign extends but only using half of the v4i32 intermediate result.

When the result is more than 128 bits, default type legalization would prefer to split the destination type all the way down to v2i64 with shuffles followed by v16i8/v8i16->v2i64 sext_inreg operations. This results in more instructions than necessary because we are only utilizing the lower 2 elements of the v4i32 intermediate result. Instead we can custom split a v4i8/v4i16->v4i64 sign_extend. Then we can sign extend v4i8/v4i16->v4i32 invec producing a full v4i32 result. Create the sign bit vector as a v4i32 then split and interleave with the sign bits using an punpackldq and punpackhdq.

llvm-svn: 347176

5 years ago[X86] Add a 32-bit command line with only sse2 to vector-sext.ll and vector-sext...
Craig Topper [Sun, 18 Nov 2018 21:28:47 +0000 (21:28 +0000)]
[X86] Add a 32-bit command line with only sse2 to vector-sext.ll and vector-sext.ll to show some of the scalarized load sequences without 64-bit scalar support.

Some of these sequeces look pretty bad since we have to copy the sign bit from a 32 bit register to a 64 bit register to finish a sign extend.

llvm-svn: 347175

5 years agoRevert "Implement basic DidAttach and DidLaunch for DynamicLoaderWindowsDYLD"
Zachary Turner [Sun, 18 Nov 2018 20:48:25 +0000 (20:48 +0000)]
Revert "Implement basic DidAttach and DidLaunch for DynamicLoaderWindowsDYLD"

This breaks many tests on Windows, which now all fail with an error such
as "Unable to read memory at address <xxxxxxxx>".

llvm-svn: 347174

5 years ago[X86][SSE] Add SimplifyDemandedVectorElts support for SSE splat-vector-shifts.
Simon Pilgrim [Sun, 18 Nov 2018 20:21:52 +0000 (20:21 +0000)]
[X86][SSE] Add SimplifyDemandedVectorElts support for SSE splat-vector-shifts.

SSE vector shifts only use the bottom 64-bits of the shift amount vector.

llvm-svn: 347173

5 years ago[X86] Disable combineToExtendVectorInReg under -x86-experimental-vector-widening...
Craig Topper [Sun, 18 Nov 2018 18:11:25 +0000 (18:11 +0000)]
[X86] Disable combineToExtendVectorInReg under -x86-experimental-vector-widening-legalization. Add custom type legalization for extends.

If we widen illegal types instead of promoting, we should be able to rely on the type legalizer to create the vector_inreg operations for us with some caveats.

This patch disables combineToExtendVectorInReg when we are using widening.

I've enabled custom legalization for v8i8->v8i64 extends under avx512f since the type legalizer would want to create a vector_inreg with a v64i8 input type which isn't legal without avx512bw. So we go to v16i8 with custom code using the relaxation of rules we get from D54346.

I've also enable custom legalization of v8i64 and v16i32 operations with with AVX. When the input type is 128 bits, the default splitting legalization would extend first 128->256, then do the a split to two 128 pieces. Extend each half to 256 and then concat the result. The custom legalization I've added instead uses a 128->256 bit vector_inreg extend that only reads the lower 64-bits for the low half of the split. Then shuffles the high 64-bits to the low 64-bits and does another vector_inreg extend.

llvm-svn: 347172

5 years ago[X86] Lower v16i16->v8i16 truncate using an 'and' with 255, an extract_subvector...
Craig Topper [Sun, 18 Nov 2018 17:59:28 +0000 (17:59 +0000)]
[X86] Lower v16i16->v8i16 truncate using an 'and' with 255, an extract_subvector, and a packuswb instruction.

Summary: This is an improvement over the two pshufbs and punpcklqdq we'd get otherwise.

Reviewers: RKSimon, spatel

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D54671

llvm-svn: 347171

5 years ago[DAG] add undef simplifications for select nodes
Sanjay Patel [Sun, 18 Nov 2018 17:36:23 +0000 (17:36 +0000)]
[DAG] add undef simplifications for select nodes

Sadly, this duplicates (twice) the logic from InstSimplify. There
might be some way to at least share the DAG versions of the code,
but copying the folds seems to be the standard method to ensure
that we don't miss these folds.

Unlike in IR, we don't run DAGCombiner to fixpoint, so there's no
way to ensure that we do these kinds of simplifications unless the
code is repeated at node creation time and during combines.

There were other tests that would become worthless with this
improvement that I changed as pre-commits:
rL347161
rL347164
rL347165
rL347166
rL347167

I'm not sure how to salvage the remaining tests (diffs in this patch).
So the x86 tests verify that the new code is working as intended.
The AMDGPU test is actually similar to my motivating case: we have
some undef value that has survived to machine IR in an x86 test, and
then it gets folded in some weird way, or we crash if we don't transfer
the undef flag. But we would have been better off never getting to that
point by doing these simplifications.

This will lead back to PR32023 someday...
https://bugs.llvm.org/show_bug.cgi?id=32023

llvm-svn: 347170

5 years agoRemove unused variable. NFCI.
Simon Pilgrim [Sun, 18 Nov 2018 17:24:59 +0000 (17:24 +0000)]
Remove unused variable. NFCI.

llvm-svn: 347169

5 years ago[X86][SSE] Split IsSplatValue into GetSplatValue and IsSplatVector
Simon Pilgrim [Sun, 18 Nov 2018 17:15:06 +0000 (17:15 +0000)]
[X86][SSE] Split IsSplatValue into GetSplatValue and IsSplatVector

Refactor towards making this recursive (necessary for PR38243 rotation splat detection).
IsSplatVector returns the original vector source of the splat and the splat index.
GetSplatValue returns the scalar splatted value as an extraction from IsSplatVector.

llvm-svn: 347168

5 years ago[x86] regenerate full checks; NFC
Sanjay Patel [Sun, 18 Nov 2018 16:56:17 +0000 (16:56 +0000)]
[x86] regenerate full checks; NFC

llvm-svn: 347167

5 years ago[SystemZ] make test immune to improvements in undef simplification
Sanjay Patel [Sun, 18 Nov 2018 16:50:44 +0000 (16:50 +0000)]
[SystemZ] make test immune to improvements in undef simplification

llvm-svn: 347166

5 years ago[Hexagon] make tests immune to improvements in undef simplification
Sanjay Patel [Sun, 18 Nov 2018 16:50:16 +0000 (16:50 +0000)]
[Hexagon] make tests immune to improvements in undef simplification

llvm-svn: 347165

5 years ago[ARM] make test immune to improvements in undef simplification
Sanjay Patel [Sun, 18 Nov 2018 16:49:42 +0000 (16:49 +0000)]
[ARM] make test immune to improvements in undef simplification

llvm-svn: 347164

5 years agoAdd the abseil-duration-factory-scale check.
Aaron Ballman [Sun, 18 Nov 2018 16:41:06 +0000 (16:41 +0000)]
Add the abseil-duration-factory-scale check.

This check removes unneeded scaling of arguments when calling Abseil Time factory functions.

Patch by Hyrum Wright.

llvm-svn: 347163

5 years ago[X86][SSE] Relax IsSplatValue - remove the 'variable shift' limit on subtracts.
Simon Pilgrim [Sun, 18 Nov 2018 15:52:08 +0000 (15:52 +0000)]
[X86][SSE] Relax IsSplatValue - remove the 'variable shift' limit on subtracts.

Means we don't use the per-lane-shifts as much when we can cheaply use the older splat-variable-shifts.

llvm-svn: 347162

5 years ago[x86] make tests immune to improvements in undef handling
Sanjay Patel [Sun, 18 Nov 2018 15:27:19 +0000 (15:27 +0000)]
[x86] make tests immune to improvements in undef handling

llvm-svn: 347161

5 years ago[SelectionDAG] simplify code; NFC
Sanjay Patel [Sun, 18 Nov 2018 14:39:03 +0000 (14:39 +0000)]
[SelectionDAG] simplify code; NFC

llvm-svn: 347160

5 years ago[X86][SSE] Add some generic masked gather codegen tests
Simon Pilgrim [Sun, 18 Nov 2018 14:35:57 +0000 (14:35 +0000)]
[X86][SSE] Add some generic masked gather codegen tests

llvm-svn: 347159

5 years ago[X86][SSE] Use raw shuffle mask decode in SimplifyDemandedVectorEltsForTargetNode...
Simon Pilgrim [Sun, 18 Nov 2018 13:34:53 +0000 (13:34 +0000)]
[X86][SSE] Use raw shuffle mask decode in SimplifyDemandedVectorEltsForTargetNode (PR39549)

We were using the 'normalized' shuffle mask from resolveTargetShuffleInputs, which replaces zero/undef inputs with sentinel values. For SimplifyDemandedVectorElts we need the raw mask so we can correctly demand those 'zero' inputs that got normalized away, this requires an extra bit of logic to locally normalize undef inputs.

llvm-svn: 347158

5 years ago[analyzer][NFC] Move CheckerOptInfo to CheckerRegistry.cpp, and make it local
Kristof Umann [Sun, 18 Nov 2018 12:47:03 +0000 (12:47 +0000)]
[analyzer][NFC] Move CheckerOptInfo to CheckerRegistry.cpp, and make it local

CheckerOptInfo feels very much out of place in CheckerRegistration.cpp, so I
moved it to CheckerRegistry.h.

Differential Revision: https://reviews.llvm.org/D54397

llvm-svn: 347157

5 years agoSwap order of discovering of -ltinfo and -lterminfo
Kamil Rytarowski [Sun, 18 Nov 2018 12:13:51 +0000 (12:13 +0000)]
Swap order of discovering of -ltinfo and -lterminfo

Summary:
NetBSD ships with native curses(3) and -ltinfo is a part of ncurses.
Set -lterminfo before -ltinfo, as it allows to prioritize native curses
libraries. Mixing curses and ncurses does not work well, especially
in software built on top of llvm.

Original patch by Ryo Onodera (NetBSD) in pkgsrc.

Reviewers: labath, dim, mgorny

Reviewed By: dim, mgorny

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D54650

llvm-svn: 347156

5 years ago[WebAssembly] Add null streamer support
Heejin Ahn [Sun, 18 Nov 2018 11:58:47 +0000 (11:58 +0000)]
[WebAssembly] Add null streamer support

Summary: Now `llc -filetype=null` works.

Reviewers: eush

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

Differential Revision: https://reviews.llvm.org/D54660

llvm-svn: 347155

5 years ago[WebAssembly] Add equality comparison operators for WasmEventType
Heejin Ahn [Sun, 18 Nov 2018 11:53:35 +0000 (11:53 +0000)]
[WebAssembly] Add equality comparison operators for WasmEventType

Summary:
This was missing in D54096. Independent tests for this is not available
here, because these are used in lld.

Reviewers: sbc100

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

Differential Revision: https://reviews.llvm.org/D54662

llvm-svn: 347154

5 years ago[analyzer][UninitializedObjectChecker] Uninit regions are only reported once
Kristof Umann [Sun, 18 Nov 2018 11:34:10 +0000 (11:34 +0000)]
[analyzer][UninitializedObjectChecker] Uninit regions are only reported once

Especially with pointees, a lot of meaningless reports came from uninitialized
regions that were already reported. This is fixed by storing all reported fields
to the GDM.

Differential Revision: https://reviews.llvm.org/D51531

llvm-svn: 347153

5 years agocmake: z3: Remove EXACT from 4.7.1 after being compatible with 4.8.1
Jan Kratochvil [Sun, 18 Nov 2018 09:36:36 +0000 (09:36 +0000)]
cmake: z3: Remove EXACT from 4.7.1 after being compatible with 4.8.1

After check-in of D54391 a comment there by @mikhail.ramalho says:
Since we're supporting version 4.8.1 now, the cmake file should be changed to
"minimum" instead of "exact".

Differential Revision: https://reviews.llvm.org/D54535

llvm-svn: 347152

5 years ago[X86] Add -x86-experimental-vector-widening-legalization check to combineSelect and...
Craig Topper [Sun, 18 Nov 2018 08:30:09 +0000 (08:30 +0000)]
[X86] Add -x86-experimental-vector-widening-legalization check to combineSelect and combineSetCC to cover vXi16/vXi8 promotion without BWI.

I don't yet have any test cases for this, but its the right thing to do based on log file inspection.

llvm-svn: 347151

5 years ago[X86] Rename WidenMaskArithmetic->PromoteMaskArithmetic since we usually use widen...
Craig Topper [Sun, 18 Nov 2018 07:35:08 +0000 (07:35 +0000)]
[X86] Rename WidenMaskArithmetic->PromoteMaskArithmetic since we usually use widen to refer to adding elements not making elements larger. NFC

llvm-svn: 347150

5 years ago[X86] Don't use a pmaddwd for vXi32 multiply if the inputs are zero extends from...
Craig Topper [Sun, 18 Nov 2018 05:53:21 +0000 (05:53 +0000)]
[X86] Don't use a pmaddwd for vXi32 multiply if the inputs are zero extends from i8 or smaller without SSE4.1. Prefer to shrink the mul instead.

The zero extend will require two stages of unpacks to implement. So its better to shrink the multiply using pmullw and then extend that result back to v4i32 using a single unpack.

llvm-svn: 347149

5 years agotighten up a couple of assertions. hitting the BitPosition == BitWidth case that...
John Regehr [Sun, 18 Nov 2018 01:51:43 +0000 (01:51 +0000)]
tighten up a couple of assertions. hitting the BitPosition == BitWidth case that was previously not caught resulted in nasty corruption of APInts that (on my system at least) could not be detected using UBSan, ASan, or Valgrind. this patch does not cause any extra failures in a check-all nor does it interfere with bootstrapping. David Blaikie informally approved this change.

llvm-svn: 347148

5 years ago[CorrelatedValuePropagation] Preserve debug locations (PR38178)
Vedant Kumar [Sun, 18 Nov 2018 00:29:58 +0000 (00:29 +0000)]
[CorrelatedValuePropagation] Preserve debug locations (PR38178)

Fix all of the missing debug location errors in CVP found by debugify.

This includes the missing-location-after-udiv-truncation case described
in llvm.org/PR38178.

llvm-svn: 347147

5 years agoFix bot failure from r347145
Teresa Johnson [Sat, 17 Nov 2018 20:41:45 +0000 (20:41 +0000)]
Fix bot failure from r347145

The #if check around the statistics computation gave an error about
the statistic being an unused variable. Instead, guard with
AreStatisticsEnabled().

llvm-svn: 347146

5 years ago[ThinLTO] Add some stats for read only variable internalization
Teresa Johnson [Sat, 17 Nov 2018 20:03:22 +0000 (20:03 +0000)]
[ThinLTO] Add some stats for read only variable internalization

Summary:
Follow up to D49362 ([ThinLTO] Internalize read only globals). Add a
statistic on the number of read only variables (only counting live
variables since dead variables will be dropped anyway).

Reviewers: evgeny777

Subscribers: mehdi_amini, inglorion, eraman, steven_wu, dexonsmith, arphaman, llvm-commits

Differential Revision: https://reviews.llvm.org/D54642

llvm-svn: 347145

5 years ago[Clang] Add options -fprofile-filter-files and -fprofile-exclude-files to filter...
Calixte Denizet [Sat, 17 Nov 2018 19:41:39 +0000 (19:41 +0000)]
[Clang] Add options -fprofile-filter-files and -fprofile-exclude-files to filter the files to instrument with gcov (after revert https://reviews.llvm.org/rL346659)

Summary:
the previous patch (https://reviews.llvm.org/rC346642) has been reverted because of test failure under windows.
So this patch fix the test cfe/trunk/test/CodeGen/code-coverage-filter.c.

Reviewers: marco-c

Reviewed By: marco-c

Subscribers: cfe-commits, sylvestre.ledru

Differential Revision: https://reviews.llvm.org/D54600

llvm-svn: 347144

5 years ago[X86] Add support for matching PACKUSWB from a v64i8 shuffle.
Craig Topper [Sat, 17 Nov 2018 18:54:43 +0000 (18:54 +0000)]
[X86] Add support for matching PACKUSWB from a v64i8 shuffle.

llvm-svn: 347143

5 years ago[X86] Add test case to show missed opportunity to use PACKUSWB in v64i8 shuffle lowering.
Craig Topper [Sat, 17 Nov 2018 18:54:41 +0000 (18:54 +0000)]
[X86] Add test case to show missed opportunity to use PACKUSWB in v64i8 shuffle lowering.

llvm-svn: 347142

5 years agoSink BuryPointer from Clang into LLVM for reuse there
David Blaikie [Sat, 17 Nov 2018 18:04:13 +0000 (18:04 +0000)]
Sink BuryPointer from Clang into LLVM for reuse there

llvm-svn: 347141

5 years agoMove BuryPointer from Clang to LLVM for use in other LLVM tools
David Blaikie [Sat, 17 Nov 2018 18:03:47 +0000 (18:03 +0000)]
Move BuryPointer from Clang to LLVM for use in other LLVM tools

Specifically planning to use this in llvm-symbolizer to remove the cost
of cleanup there.

llvm-svn: 347140

5 years ago[X86][SSE] Add shuffle demanded elts test case for PR39549
Simon Pilgrim [Sat, 17 Nov 2018 14:06:03 +0000 (14:06 +0000)]
[X86][SSE] Add shuffle demanded elts test case for PR39549

llvm-svn: 347139

5 years ago[AST][NFC] Pack CXXDefaultInitExpr
Bruno Ricci [Sat, 17 Nov 2018 13:02:47 +0000 (13:02 +0000)]
[AST][NFC] Pack CXXDefaultInitExpr

Use the newly available space in the bit-fields of Stmt.
This saves one pointer per CXXDefaultInitExpr.

llvm-svn: 347138

5 years ago[AST][NFC] Pack CXXDefaultArgExpr
Bruno Ricci [Sat, 17 Nov 2018 12:56:30 +0000 (12:56 +0000)]
[AST][NFC] Pack CXXDefaultArgExpr

Use the newly available space in the bit-fields of Stmt.
This saves one pointer per CXXDefaultArgExpr.

llvm-svn: 347137

5 years ago[AST][NFC] Pack CXXThrowExpr
Bruno Ricci [Sat, 17 Nov 2018 12:53:56 +0000 (12:53 +0000)]
[AST][NFC] Pack CXXThrowExpr

Use the newly available space in the bit-fields of Stmt.
This saves 8 bytes per CXXThrowExpr.

llvm-svn: 347136

5 years ago[llvm-objdump] Print a blank row at the end of sections
Xing GUO [Sat, 17 Nov 2018 08:12:48 +0000 (08:12 +0000)]
[llvm-objdump] Print a blank row at the end of sections

Summary:
When using option `-x` (--all-headers), it will print `Sections`, `Symbol Table`, `Program Header` ...
`Sections` and `Symbol Table` will be connected together.

Before:
```
Sections:
Idx Name          Size      Address          Type
  0               00000000 0000000000000000
  ...
  29 .shstrtab     0000011a 0000000000000000
SYMBOL TABLE:
  ...
```

After:
```
Sections:
Idx Name          Size      Address          Type
  0               00000000 0000000000000000
  ...
  29 .shstrtab     0000011a 0000000000000000

SYMBOL TABLE:
  ...
```

Reviewers: Higuoxing

Reviewed By: Higuoxing

Subscribers: llvm-commits, jhenderson

Differential Revision: https://reviews.llvm.org/D54665

llvm-svn: 347135

5 years agollvm-symbolizer: Avoid calling getFromOffset when the index entry is already available
David Blaikie [Sat, 17 Nov 2018 05:57:58 +0000 (05:57 +0000)]
llvm-symbolizer: Avoid calling getFromOffset when the index entry is already available

Especially for symbolizer it can be efficient to have to search through
the entire index when it isn't needed - llvm-symbolizer looks up only a
few CUs & already has an index available in getUnitForEntry, once it's
passed down to DWARFUnitHeader::extract then there's no need for it to
call getFromOffset.

llvm-svn: 347134

5 years agoFix unused variable warning.
David L. Jones [Sat, 17 Nov 2018 04:48:54 +0000 (04:48 +0000)]
Fix unused variable warning.

llvm-svn: 347133

5 years ago[clang-tidy/checks] Implement a clang-tidy check to verify Google Objective-C functio...
Stephane Moore [Sat, 17 Nov 2018 02:37:21 +0000 (02:37 +0000)]
[clang-tidy/checks] Implement a clang-tidy check to verify Google Objective-C function naming conventions ðŸ“œ

Summary:
§1 Description

This check finds function names in function declarations in Objective-C files that do not follow the naming pattern described in the Google Objective-C Style Guide. Function names should be in UpperCamelCase and functions that are not of static storage class should have an appropriate prefix as described in the Google Objective-C Style Guide. The function `main` is a notable exception. Function declarations in expansions in system headers are ignored.

Example conforming function definitions:
```
static bool IsPositive(int i) { return i > 0; }
static bool ABIsPositive(int i) { return i > 0; }
bool ABIsNegative(int i) { return i < 0; }
```

A fixit hint is generated for functions of static storage class but otherwise the check does not generate a fixit hint because an appropriate prefix for the function cannot be determined.

§2 Test Notes
* Verified clang-tidy tests pass successfully.
* Used check_clang_tidy.py to verify expected output of processing google-objc-function-naming.m

Reviewers: benhamilton, hokein, Wizard, aaron.ballman

Reviewed By: benhamilton

Subscribers: Eugene.Zelenko, mgorny, xazax.hun, cfe-commits

Tags: #clang-tools-extra

Differential Revision: https://reviews.llvm.org/D51575

llvm-svn: 347132

5 years ago[X86] Don't extend v32i8 multiplies to v32i16 with avx512bw and prefer-vector-width...
Craig Topper [Sat, 17 Nov 2018 02:36:07 +0000 (02:36 +0000)]
[X86] Don't extend v32i8 multiplies to v32i16 with avx512bw and prefer-vector-width=256.

llvm-svn: 347131

5 years ago[X86] Add test cases to show incorrect use of a 512 bit vector in v32i8 multiply...
Craig Topper [Sat, 17 Nov 2018 02:36:02 +0000 (02:36 +0000)]
[X86] Add test cases to show incorrect use of a 512 bit vector in v32i8 multiply lowering with prefer-vector-width=256.

On the min-legal-vector-width test this actually causes some of the v32i16 operations we emitted to be scalarized.

llvm-svn: 347130

5 years agoReverted r347092 due to the following build fails:
Vyacheslav Zakharin [Sat, 17 Nov 2018 02:26:34 +0000 (02:26 +0000)]
Reverted r347092 due to the following build fails:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap/builds/8662
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/26263

llvm-svn: 347129

5 years agoAdd initial scaffolding for the GN build.
Nico Weber [Sat, 17 Nov 2018 02:21:53 +0000 (02:21 +0000)]
Add initial scaffolding for the GN build.

See "GN build roundtable summary; adding GN build files to the repo" on
llvm-dev and cfe-dev for discussion.

In particular, this build is completely unsupported. People adding new files to
LLVM are not expected to update the GN build files, and reviewers are not
supposed to request the gn build files to be updated.

This adds just enough to be able to build llvm/lib/Demangle. It requires using
a monorepo.

This adds a few build config options you can set in args.gn
(`gn args out/foo --list` for all):
- is_debug = true to enable debug builds (defaults to release)
- llvm_enable_assertions to toggle assertions (defaults to true)
- clang_base_path, if set an absolute path to a locally-built clang to be used
  as host compiler

Differential Revision: https://reviews.llvm.org/D54345

llvm-svn: 347128

5 years ago[X86] Use getUnpackl/getUnpackh instead of hardcoding a shuffle mask.
Craig Topper [Sat, 17 Nov 2018 02:18:12 +0000 (02:18 +0000)]
[X86] Use getUnpackl/getUnpackh instead of hardcoding a shuffle mask.

llvm-svn: 347127

5 years agoUse llvm::copy. NFC
Fangrui Song [Sat, 17 Nov 2018 01:44:25 +0000 (01:44 +0000)]
Use llvm::copy. NFC

llvm-svn: 347126

5 years agoJust don't even attempt to invoke sed on Windows.
Adrian Prantl [Sat, 17 Nov 2018 01:27:47 +0000 (01:27 +0000)]
Just don't even attempt to invoke sed on Windows.

llvm-svn: 347125

5 years agoRevert "Makefile.rules: Fix the windows-sed-quoting issue harder."
Adrian Prantl [Sat, 17 Nov 2018 01:27:46 +0000 (01:27 +0000)]
Revert "Makefile.rules: Fix the windows-sed-quoting issue harder."

llvm-svn: 347124

5 years ago[llvm-objcopy] Use llvm::all_of and rename the variables "Segment" to avoid confusion...
Fangrui Song [Sat, 17 Nov 2018 01:15:55 +0000 (01:15 +0000)]
[llvm-objcopy] Use llvm::all_of and rename the variables "Segment" to avoid confusion with the type of the same name

llvm-svn: 347123

5 years agoadd PdbSymUid.cpp
Jason Molenda [Sat, 17 Nov 2018 01:08:01 +0000 (01:08 +0000)]
add PdbSymUid.cpp

llvm-svn: 347122

5 years ago[hwasan] don't check tail magic when in right_align mode (should fix the bot)
Kostya Serebryany [Sat, 17 Nov 2018 01:02:09 +0000 (01:02 +0000)]
[hwasan] don't check tail magic when in right_align mode (should fix the bot)

llvm-svn: 347121

5 years agoMakefile.rules: Fix the windows-sed-quoting issue harder.
Adrian Prantl [Sat, 17 Nov 2018 00:45:58 +0000 (00:45 +0000)]
Makefile.rules: Fix the windows-sed-quoting issue harder.

llvm-svn: 347120

5 years ago[clangd] Fix crash hovering on non-decltype trailing return
Marc-Andre Laperle [Sat, 17 Nov 2018 00:41:14 +0000 (00:41 +0000)]
[clangd] Fix crash hovering on non-decltype trailing return

Summary:
More specifically, hovering on "auto" in
auto main() -> int {
    return 0;
}

Signed-off-by: Marc-Andre Laperle <malaperle@gmail.com>
Reviewers: ilya-biryukov

Reviewed By: ilya-biryukov

Subscribers: ilya-biryukov, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits

Differential Revision: https://reviews.llvm.org/D54553

llvm-svn: 347119

5 years ago[hwasan] make the heap-buffer-overflow.c test more robust and re-enable it. With...
Kostya Serebryany [Sat, 17 Nov 2018 00:40:34 +0000 (00:40 +0000)]
[hwasan] make the heap-buffer-overflow.c test more robust and re-enable it. With malloc_align_right the relative offsets of heap chunks are less predictable to simply don't test for them.

llvm-svn: 347118

5 years agoMake TestAppleSimulatorOSType.py more flexible
Frederic Riss [Sat, 17 Nov 2018 00:40:31 +0000 (00:40 +0000)]
Make TestAppleSimulatorOSType.py more flexible

Different versions of Xcode have different outputs for the simctl command

llvm-svn: 347117

5 years ago[hwasan] implement free_checks_tail_magic=1
Kostya Serebryany [Sat, 17 Nov 2018 00:25:17 +0000 (00:25 +0000)]
[hwasan] implement free_checks_tail_magic=1

Summary:
With free_checks_tail_magic=1 (default) HWASAN
writes magic bytes to the tail of every heap allocation
(last bytes of the last granule, if the last granule is not fully used)
and checks these bytes on free().

This feature will detect buffer overwires within the last granule
at the time of free().

This is an alternative to malloc_align_right=[1289] that should have
fewer compatibility issues. It is also weaker since it doesn't
detect read overflows and reports bugs at free() instead of at access.

Reviewers: eugenis

Subscribers: kubamracek, delcypher, #sanitizers, llvm-commits

Differential Revision: https://reviews.llvm.org/D54656

llvm-svn: 347116

5 years agoMoved dag-combine-select-undef.ll into amdgpu. NFC.
Stanislav Mekhanoshin [Sat, 17 Nov 2018 00:17:15 +0000 (00:17 +0000)]
Moved dag-combine-select-undef.ll into amdgpu. NFC.

Tests really needs target arch to be specified.

llvm-svn: 347115

5 years agoAdd missing triple from llvm-mc command line.
Zachary Turner [Sat, 17 Nov 2018 00:05:06 +0000 (00:05 +0000)]
Add missing triple from llvm-mc command line.

llvm-svn: 347114

5 years agoMake git-llvm python3 compatible again. Hopefully. :)
James Y Knight [Fri, 16 Nov 2018 23:59:23 +0000 (23:59 +0000)]
Make git-llvm python3 compatible again. Hopefully. :)

llvm-svn: 347113

5 years agoFixed test after r347110
Stanislav Mekhanoshin [Fri, 16 Nov 2018 23:40:04 +0000 (23:40 +0000)]
Fixed test after r347110

Comments in llc outputs are printed differently on different
platforms, some with '#', some with '##'. Removed non-essential
part of the checks.

llvm-svn: 347112

5 years agoAdd missing test for r347072 -gcodeview-ghash
Reid Kleckner [Fri, 16 Nov 2018 23:17:11 +0000 (23:17 +0000)]
Add missing test for r347072 -gcodeview-ghash

llvm-svn: 347111

5 years agoDAG combiner: fold (select, C, X, undef) -> X
Stanislav Mekhanoshin [Fri, 16 Nov 2018 23:13:38 +0000 (23:13 +0000)]
DAG combiner: fold (select, C, X, undef) -> X

Differential Revision: https://reviews.llvm.org/D54646

llvm-svn: 347110

5 years agoRewrite stop-hook tests as a couple of FileCheck tests
Frederic Riss [Fri, 16 Nov 2018 23:07:28 +0000 (23:07 +0000)]
Rewrite stop-hook tests as a couple of FileCheck tests

Those tests were using pexpect and being flaky on some of ours bots.
This patch reimplmeents the tests usinf FileCheck, and it also
extends the test coverage to a few more stop-hook options.

llvm-svn: 347109

5 years ago[CMake] Use lld and llvm-objcopy for first stage compiler in Fuchsia
Petr Hosek [Fri, 16 Nov 2018 23:07:03 +0000 (23:07 +0000)]
[CMake] Use lld and llvm-objcopy for first stage compiler in Fuchsia

When cross-compiling the second stage to a different target, we need to
make sure that the first-stage compiler can produce binaries for that
target. Using lld and llvm-objcopy as the default linker and objcopy
tool eliminates some of the dependencies on the host toolchain.

Differential Revision: https://reviews.llvm.org/D54655

llvm-svn: 347108

5 years ago[hwasan] use reads instead of writes in a test
Kostya Serebryany [Fri, 16 Nov 2018 23:01:42 +0000 (23:01 +0000)]
[hwasan] use reads instead of writes in a test

llvm-svn: 347107

5 years agoRevert "Cast the 2nd argument of _Unwind_SetIP() to _Unwind_Ptr"
Kamil Rytarowski [Fri, 16 Nov 2018 23:00:28 +0000 (23:00 +0000)]
Revert "Cast the 2nd argument of _Unwind_SetIP() to _Unwind_Ptr"

_Unwind_Ptr is unknown on some targets.

Detected on green-dragon-21 (MacPro Late 2013 | OS X 10.14(18A391) | Xcode 10.1(10B61)).

llvm-svn: 347106

5 years ago[X86] Add custom promotion of narrow fp_to_uint/fp_to_sint operations under -x86...
Craig Topper [Fri, 16 Nov 2018 22:53:00 +0000 (22:53 +0000)]
[X86] Add custom promotion of narrow fp_to_uint/fp_to_sint operations under -x86-experimental-vector-widening-legalization.

This tries to force the result type to vXi32 followed by a truncate. This can help avoid scalarization that would otherwise occur.

There's some annoying examples of an avx512 truncate instruction followed by a packus where we should really be able to just use one truncate. But overall this is still a net improvement.

llvm-svn: 347105

5 years agoMakefile.rules: Degrade gracefully on Windows machines without sed.
Adrian Prantl [Fri, 16 Nov 2018 22:44:57 +0000 (22:44 +0000)]
Makefile.rules: Degrade gracefully on Windows machines without sed.

llvm-svn: 347104

5 years agoSpeed up git-llvm script by only svn up'ing affected directories.
James Y Knight [Fri, 16 Nov 2018 22:36:17 +0000 (22:36 +0000)]
Speed up git-llvm script by only svn up'ing affected directories.

Also, support modifications to toplevel files in git (which need to be
committed to "monorepo-root" in svn).

Differential Revision: https://reviews.llvm.org/D54341

llvm-svn: 347103

5 years agoCast the 2nd argument of _Unwind_SetIP() to _Unwind_Ptr
Kamil Rytarowski [Fri, 16 Nov 2018 22:17:36 +0000 (22:17 +0000)]
Cast the 2nd argument of _Unwind_SetIP() to _Unwind_Ptr

This modification is require for NetBSD with GCC, as there is a
custom unwind.h header implementation with different types.

No functional change intended for others.

Cherry-picked chunk from D33878.

llvm-svn: 347102

5 years agoCast _Unwind_GetIP() and _Unwind_GetRegionStart() to uintptr_t
Kamil Rytarowski [Fri, 16 Nov 2018 22:15:09 +0000 (22:15 +0000)]
Cast _Unwind_GetIP() and _Unwind_GetRegionStart() to uintptr_t

This modification is require for NetBSD with GCC, as there is a
custom unwind.h header implementation with different types.

No functional change intended for others.

Cherry-picked chunk from D33878.

llvm-svn: 347101

5 years ago[X86] Qualify part of the masked gather handling in ReplaceNodeResults with a getType...
Craig Topper [Fri, 16 Nov 2018 22:04:29 +0000 (22:04 +0000)]
[X86] Qualify part of the masked gather handling in ReplaceNodeResults with a getTypeAction call to know if we can use default legalization.

If we managed to switch to -x86-experimental-vector-widening-legalization this block can be removed.

llvm-svn: 347100