platform/upstream/llvm.git
3 years ago[RISCV] Unify the arch string parsing logic to to RISCVISAInfo.
Kito Cheng [Wed, 23 Jun 2021 08:18:54 +0000 (16:18 +0800)]
[RISCV] Unify the arch string parsing logic to to RISCVISAInfo.

How many place you need to modify when implementing a new extension for RISC-V?

At least 7 places as I know:

- Add new SubtargetFeature at RISCV.td
- -march parser in RISCV.cpp
- RISCVTargetInfo::initFeatureMap@RISCV.cpp for handling feature vector.
- RISCVTargetInfo::getTargetDefines@RISCV.cpp for pre-define marco.
- Arch string parser for ELF attribute in RISCVAsmParser.cpp
- ELF attribute emittion in RISCVAsmParser.cpp, and make sure it's in
  canonical order...
- ELF attribute emittion in RISCVTargetStreamer.cpp, and again, must in
  canonical order...

And now, this patch provide an unified infrastructure for handling (almost)
everything of RISC-V arch string.

After this patch, you only need to update 2 places for implement an extension
for RISC-V:
- Add new SubtargetFeature at RISCV.td, hmmm, it's hard to avoid.
- Add new entry to RISCVSupportedExtension@RISCVISAInfo.cpp or
  SupportedExperimentalExtensions@RISCVISAInfo.cpp .

Most codes are come from existing -march parser, but with few new feature/bug
fixes:
- Accept version for -march, e.g. -march=rv32i2p0.
- Reject version info with `p` but without minor version number like `rv32i2p`.

Differential Revision: https://reviews.llvm.org/D105168

3 years agoUse llvm::erase_value (NFC)
Kazu Hirata [Sun, 17 Oct 2021 06:31:20 +0000 (23:31 -0700)]
Use llvm::erase_value (NFC)

3 years agoFix a few warnings (signed/unsigned comparison in gtest, and missing field initializers)
David Blaikie [Sun, 17 Oct 2021 05:43:57 +0000 (22:43 -0700)]
Fix a few warnings (signed/unsigned comparison in gtest, and missing field initializers)

3 years ago[MLIR][LLVM] Add memset intrinsic
William S. Moses [Fri, 15 Oct 2021 19:34:23 +0000 (15:34 -0400)]
[MLIR][LLVM] Add memset intrinsic

Add memset intrinsic into LLVM dialect

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D111906

3 years agoRevert "[AArch64] Optimize add/sub with immediate"
Ben Shi [Sat, 16 Oct 2021 22:14:59 +0000 (22:14 +0000)]
Revert "[AArch64] Optimize add/sub with immediate"

This reverts commit 9bf6bef9951a1c230796ccad2c5c0195ce4c4dff.

3 years ago[Object] Simplify RELR decoding
Fangrui Song [Sat, 16 Oct 2021 22:03:14 +0000 (15:03 -0700)]
[Object] Simplify RELR decoding

3 years ago[NFC][sanitizer] Add StackDepotTestOnlyUnmap
Vitaly Buka [Sat, 16 Oct 2021 20:31:59 +0000 (13:31 -0700)]
[NFC][sanitizer] Add StackDepotTestOnlyUnmap

3 years ago[NFC][sanitizer] Rename stack depot tests
Vitaly Buka [Sat, 16 Oct 2021 20:31:59 +0000 (13:31 -0700)]
[NFC][sanitizer] Rename stack depot tests

3 years ago[X86] Add DAG combine for negation of CMOV absolute value pattern.
Craig Topper [Sat, 16 Oct 2021 20:31:41 +0000 (13:31 -0700)]
[X86] Add DAG combine for negation of CMOV absolute value pattern.

This patch detects the absolute value pattern on the RHS of a
subtract. If we find it we swap the CMOV true/false values and
replace the subtract with an ADD.

There may be a more generic way to do this, but I'm not sure.
Targets that don't have legal or custom ISD::ABS use a generic
expand in DAG combiner already when it sees (neg (abs(x))). I
haven't checked what happens if the neg is a more general subtract.

Fixes PR50991 for X86.

Reviewed By: RKSimon, spatel

Differential Revision: https://reviews.llvm.org/D111858

3 years ago[Builders.h] Silence a warning by adding a cast.
Chris Lattner [Sat, 16 Oct 2021 20:01:13 +0000 (13:01 -0700)]
[Builders.h] Silence a warning by adding a cast.

The no-result version of createOrFold calls 'tryFold' but
ignores the result since it doesn't matter what it produced.
Explicitly cast to void to silence this warning:

../llvm/mlir/include/mlir/IR/Builders.h:454:5: warning: ignoring return value of function declared with 'nodiscard' attribute [-Wunused-result]
    tryFold(op.getOperation(), unused);
    ^~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~

Differential Revision: https://reviews.llvm.org/D111951

3 years agoImprove fatal error message when an Attribute or Type wasn't initialized by a dialect...
Mehdi Amini [Sat, 16 Oct 2021 20:19:35 +0000 (20:19 +0000)]
Improve fatal error message when an Attribute or Type wasn't initialized by a dialect (NFC)

The existing message hints that the dialect may not be loaded, but there
is also the possibility that the dialect was loaded and the initialize()
method didn't include the Type/Attribute.

3 years agoRevert "[clang] Pass -clear-ast-before-backend in Clang::ConstructJob()"
Arthur Eubanks [Sat, 16 Oct 2021 19:05:41 +0000 (12:05 -0700)]
Revert "[clang] Pass -clear-ast-before-backend in Clang::ConstructJob()"

This reverts commit 47eb99aa44ab1d20327d67a49d6c47163de76387.

This causes crashes with -print-stats: PR52193.

3 years ago[APInt] Fix 1-bit edge case in smul_ov()
Nikita Popov [Sat, 16 Oct 2021 18:25:18 +0000 (20:25 +0200)]
[APInt] Fix 1-bit edge case in smul_ov()

The sdiv used to check for overflow can itself overflow if the
LHS is signed min and the RHS is -1. The code tried to account for
this by also checking the commuted version. However, for 1-bit
values, signed min and -1 are the same value, so both divisions
overflow. As such, the overflow for -1 * -1 was not detected
(which results in -1 rather than 1 for 1-bit values). Fix this by
explicitly checking for this case instead.

Noticed while adding exhaustive test coverage for smul_ov(),
which is also part of this commit.

3 years ago[OpenMP][deviceRTLs] Fix wrong return value of `__kmpc_is_spmd_exec_mode`
Shilei Tian [Sat, 16 Oct 2021 16:58:11 +0000 (12:58 -0400)]
[OpenMP][deviceRTLs] Fix wrong return value of `__kmpc_is_spmd_exec_mode`

D110279 introduced a bug to the device runtime. In `__kmpc_parallel_51`, we detect
whether we are already in parallel region by `__kmpc_parallel_level() > __kmpc_is_spmd_exec_mode()`.
It is based on the assumption that:
- In SPMD mode, parallel level is initialized to 1.
- In generic mode, parallel level is initialized to 0.
- `__kmpc_is_spmd_exec_mode` returns `1` for SPMD mode, 0 otherwise.

Because the return value type of `__kmpc_is_spmd_exec_mode` is `int8_t`, there
was an implicit cast from `bool` to `int8_t`. We can make sure it is either 0 or
1 since C++14. In D110279, the return value is the result of an `and` operation,
which is 2 in SPMD mode. This breaks the assumption in `__kmpc_parallel_51`.

Reviewed By: carlo.bertolli, dpalermo

Differential Revision: https://reviews.llvm.org/D111905

3 years ago[TTI][X86] Add v8i16 -> 2 x v4i16 stride 2 interleaved load costs
Simon Pilgrim [Sat, 16 Oct 2021 16:27:52 +0000 (17:27 +0100)]
[TTI][X86] Add v8i16 -> 2 x v4i16 stride 2 interleaved load costs

Split SSE2 and SSSE3 costs to correctly handle PSHUFB lowering - as was noted on D111938

3 years ago[libc++][doc] Adds more issue status labels.
Mark de Wever [Sat, 16 Oct 2021 10:22:25 +0000 (12:22 +0200)]
[libc++][doc] Adds more issue status labels.

A followup to D111458 adding more labels to LWG-issues. This should add
the labels for the not completed chrono, format, ranges, and spaceship
issues.

Some minor formatting cleanups along the way.

Reviewed By: #libc, Quuxplusone

Differential Revision: https://reviews.llvm.org/D111935

3 years ago[TTI][X86] Add SSE2 sub-128bit vXi16/32 and v2i64 stride 2 interleaved load costs
Simon Pilgrim [Sat, 16 Oct 2021 15:21:37 +0000 (16:21 +0100)]
[TTI][X86] Add SSE2 sub-128bit vXi16/32 and v2i64 stride 2 interleaved load costs

These cases use the same codegen as AVX2 (pshuflw/pshufd) for the sub-128bit vector deinterleaving, and unpcklqdq for v2i64.

It's going to take a while to add full interleaved cost coverage, but since these are the same for SSE2 -> AVX2 it should be an easy win.

Fixes PR47437

Differential Revision: https://reviews.llvm.org/D111938

3 years ago[NFC][X86][Codegen] Add missing interleaving tests after D111546
Roman Lebedev [Sat, 16 Oct 2021 14:48:09 +0000 (17:48 +0300)]
[NFC][X86][Codegen] Add missing interleaving tests after D111546

3 years agoUse llvm::is_contained (NFC)
Kazu Hirata [Sat, 16 Oct 2021 14:52:21 +0000 (07:52 -0700)]
Use llvm::is_contained (NFC)

3 years ago[Support] Add more Windows error codes to mapWindowsError
Martin Storsjö [Mon, 4 Oct 2021 21:03:57 +0000 (00:03 +0300)]
[Support] Add more Windows error codes to mapWindowsError

Also sort ERROR_BAD_NETPATH correctly.

Compared with the similar error code mapping in
libcxx/src/filesystem/operations.cpp, I'm leaving out
mappings for ERROR_NOT_SAME_DEVICE and ERROR_OPERATION_ABORTED.
They map nicely to std::errc::cross_device_link and
std::errc::operation_canceled, but those aren't available in
llvm::errc, as they aren't available across all platforms.

Also, the libcxx version maps ERROR_INVALID_NAME to
no_such_file_or_directory instead of invalid_argument.

Differential Revision: https://reviews.llvm.org/D111874

3 years ago[LV][X86] Add PR47437 test case
Simon Pilgrim [Sat, 16 Oct 2021 12:40:36 +0000 (13:40 +0100)]
[LV][X86] Add PR47437 test case

3 years ago[lldb] Split ParseSingleMember into Obj-C property and normal member/ivar parsing...
Raphael Isemann [Sat, 16 Oct 2021 11:46:21 +0000 (13:46 +0200)]
[lldb] Split ParseSingleMember into Obj-C property and normal member/ivar parsing code.

Right now DWARFASTParserClang::ParseSingleMember has two parts: One part parses
Objective-C properties and the other part parses C/C++ members/Objective-C
ivars. These parts are pretty much independent of each other (with one
historical exception, see below) and in practice they parse DIEs with different
tags/attributes: `DW_TAG_APPLE_property` and `DW_TAG_member`.

I don't see a good reason for keeping the different parsing code intertwined in
a single function, so instead split out the Objective-C property parser into its
own function.

Note that 90% of this commit is just unindenting nearly all of
`ParseSingleMember` which was inside a `if (tag == DW_TAG_member)` block. I.e.,
think of the old `ParseSingleMember` function as: The rest is just moving the
property parsing code into its own function and I added the ReportError
implementation in case we fail to resolve the property type (which before was
just a silent failure).

```
lang=c++
void DWARFASTParserClang::ParseSingleMember(...) {
  [...]
  if (tag == DW_TAG_member) {
    [...] // This huge block got unindented in this patch as the `if` above is gone.
  }
  if (property) {
    [...] // This is the property parsing code that is now its own function.
  }
}
```

There is one exception to the rule that the parsers are independent. Before 2012
Objective-C properties were encoded as `DW_TAG_member` with
`DW_AT_APPLE_property*` attributes describing the property. In 2012 this has
changed in a series of commits (see for example
c0449635b35b057c5a877343b0c5f14506c7cf02 which updates the docs) so that
`DW_TAG_APPLE_property` is now used for properties. With the old format we first
created an ivar and afterwards used the `DW_AT_APPLE_property*` attributes to
create the respective property, but there doesn't seem to be any way to create
such debug info with any clang from the last 9 years. So this is technically not
NFC in case some finds debug info from that time and tries to use properties.

Reviewed By: aprantl

Differential Revision: https://reviews.llvm.org/D111632

3 years ago[Symbolize] Demangle Rust symbols
Tomasz Miąsko [Fri, 15 Oct 2021 18:43:55 +0000 (20:43 +0200)]
[Symbolize] Demangle Rust symbols

Add support for demangling Rust v0 symbols to LLVM symbolizer by reusing
nonMicrosoftDemangle which supports both Itanium and Rust mangling.

Reviewed By: dblaikie, jhenderson

Part of https://reviews.llvm.org/D110664

3 years ago[llvm-cxxfilt] Use nonMicrosoftDemangle for demangling NFC
Tomasz Miąsko [Fri, 15 Oct 2021 18:43:27 +0000 (20:43 +0200)]
[llvm-cxxfilt] Use nonMicrosoftDemangle for demangling NFC

Reviewed By: dblaikie, jhenderson

Part of https://reviews.llvm.org/D110664

3 years ago[Demangle] Extract nonMicrosoftDemangle from llvm::demangle
Tomasz Miąsko [Fri, 15 Oct 2021 18:41:35 +0000 (20:41 +0200)]
[Demangle] Extract nonMicrosoftDemangle from llvm::demangle

Introduce a new demangling function that supports symbols using Itanium
mangling and Rust v0 mangling, and is expected in the near future to
include support for D mangling as well.

Unlike llvm::demangle, the function does not accept extra underscore
decoration. The callers generally know exactly when symbols should
include the extra decoration and so they should be responsible for
stripping it.

Functionally the only intended change is to allow demangling Rust
symbols with an extra underscore decoration through llvm::demangle,
which matches the existing behaviour for Itanium symbols.

Reviewed By: dblaikie, jhenderson

Part of https://reviews.llvm.org/D110664

3 years ago[docs] Mention DragonFlyBSD as a supported platform for LLVM.
Frederic Cambus [Sat, 16 Oct 2021 11:20:10 +0000 (13:20 +0200)]
[docs] Mention DragonFlyBSD as a supported platform for LLVM.

Differential Revision: https://reviews.llvm.org/D111758

3 years ago[Analysis] Replace assert(isa)/dyn_cast with cast. NFC.
Simon Pilgrim [Sat, 16 Oct 2021 10:40:19 +0000 (11:40 +0100)]
[Analysis] Replace assert(isa)/dyn_cast with cast. NFC.

cast<> will perform the assertion for us.

Removes a static analysis null dereference warning.

3 years ago[LazyValueInfo] getPredicateAt - remove unnecessary null pointer check. NFC.
Simon Pilgrim [Sat, 16 Oct 2021 10:20:07 +0000 (11:20 +0100)]
[LazyValueInfo] getPredicateAt - remove unnecessary null pointer check. NFC.

We already dereference the CxtI pointer several times before reaching the "if(CxtI)", we have no need to check it again.

Fixes a coverity warning.

3 years ago[ConstantFolding] ConstantFoldScalarCall2 - early-out if getLibFunc fails. NFC.
Simon Pilgrim [Sat, 16 Oct 2021 10:11:48 +0000 (11:11 +0100)]
[ConstantFolding] ConstantFoldScalarCall2 - early-out if getLibFunc fails. NFC.

3 years ago[ConstantFolding] Use getValueAPF const ref value where possible. NFC.
Simon Pilgrim [Sat, 16 Oct 2021 10:07:00 +0000 (11:07 +0100)]
[ConstantFolding] Use getValueAPF const ref value where possible. NFC.

Don't copy the value if we can avoid it.

3 years ago[ConstantFolding] ConstantFoldScalarCall1 - early-out if getLibFunc fails. NFC.
Simon Pilgrim [Sat, 16 Oct 2021 10:02:13 +0000 (11:02 +0100)]
[ConstantFolding] ConstantFoldScalarCall1 - early-out if getLibFunc fails. NFC.

3 years ago[X86][LV] X86 does *not* prefer vectorized addressing
Roman Lebedev [Sat, 16 Oct 2021 09:25:08 +0000 (12:25 +0300)]
[X86][LV] X86 does *not* prefer vectorized addressing

And another attempt to start untangling this ball of threads around gather.
There's `TTI::prefersVectorizedAddressing()`hoop, which confusingly defaults to `true`,
which tells LV to try to vectorize the addresses that lead to loads,
but X86 generally can not deal with vectors of addresses,
the only instructions that support that are GATHER/SCATTER,
but even those aren't available until AVX2, and aren't really usable until AVX512.

This specializes the hook for X86, to return true only if we have AVX512 or AVX2 w/ fast gather.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D111546

3 years ago[AArch64] Optimize add/sub with immediate
Ben Shi [Tue, 12 Oct 2021 09:03:16 +0000 (09:03 +0000)]
[AArch64] Optimize add/sub with immediate

Optimize ([add|sub] r, imm) -> ([ADD|SUB] ([ADD|SUB] r, #imm0, lsl #12), #imm1),
if imm == (imm0<<12)+imm1. and both imm0 and imm1 are non-zero 12-bit unsigned
integers.

Optimize ([add|sub] r, imm) -> ([SUB|ADD] ([SUB|ADD] r, #imm0, lsl #12), #imm1),
if imm == -(imm0<<12)-imm1, and both imm0 and imm1 are non-zero 12-bit unsigned
integers.

Reviewed By: jaykang10, dmgreen

Differential Revision: https://reviews.llvm.org/D111034

3 years ago[clang-tidy] Fix false positive in cppcoreguidelines-virtual-class-destructor
Carlos Galvez [Tue, 28 Sep 2021 08:37:32 +0000 (08:37 +0000)]
[clang-tidy] Fix false positive in cppcoreguidelines-virtual-class-destructor

Incorrectly triggers for template classes that inherit
from a base class that has virtual destructor.

Any class inheriting from a base that has a virtual destructor
will have their destructor also virtual, as per the Standard:

https://timsong-cpp.github.io/cppwp/n4140/class.dtor#9

> If a class has a base class with a virtual destructor,
> its destructor (whether user- or implicitly-declared) is virtual.

Added unit tests to prevent regression.

Fixes bug https://bugs.llvm.org/show_bug.cgi?id=51912

Differential Revision: https://reviews.llvm.org/D110614

3 years ago[mlir][linalg][bufferize] Relax rules for extract_slice/insert_slice matching
Matthias Springer [Sat, 16 Oct 2021 08:08:00 +0000 (17:08 +0900)]
[mlir][linalg][bufferize] Relax rules for extract_slice/insert_slice matching

The rules were too restrictive, causing out-of-place bufferization when the result of two ExtractSliceOp is fed into an InsertSliceOp.

Differential Revision: https://reviews.llvm.org/D111861

3 years ago[TableGen] Replace static_cast with llvm's cast. NFC
Craig Topper [Sat, 16 Oct 2021 07:26:59 +0000 (00:26 -0700)]
[TableGen] Replace static_cast with llvm's cast. NFC

These all appear next to an isa<> and cast<> is much more
common in these cases.

3 years agoFix missing failures in clang-ppc64be* and retry fixing clang-x64-windows-msvc
Juneyoung Lee [Sat, 16 Oct 2021 07:20:14 +0000 (16:20 +0900)]
Fix missing failures in clang-ppc64be* and retry fixing clang-x64-windows-msvc

3 years ago[MLIR] Generalize Affine dependence analysis using Affine Relations
Groverkss [Sat, 16 Oct 2021 06:45:31 +0000 (12:15 +0530)]
[MLIR] Generalize Affine dependence analysis using Affine Relations

This patch removes code very specific to affine dependence analysis and
refactors it as a FlatAfffineRelation.

A FlatAffineRelation represents a set of ordered pairs (domain -> range) where
"domain" and "range" are tuples of identifiers. These relations are used to
represent an "access relation" for memory access on a memref.  An access
relation maps elements of an iteration domain to the element(s) of an array
domain accessed by that iteration of the associated statement through some
array reference.  The dependence relation representing the dependence
constraints between two memory accesses can be built by composing the access
relation of the destination access by the inverse of the access relation of
source access.

This patch does not change the functionality of the existing dependence
analysis in checkMemrefAccessDependence, but refactors it to use
FlatAffineRelations to deduplicate code and enable code reuse for future
development of features like scheduling, value-based dependence analysis, etc.

Reviewed By: bondhugula

Differential Revision: https://reviews.llvm.org/D110563

3 years agoFix lit test failures in clang-ppc* and clang-x64-windows-msvc
Juneyoung Lee [Sat, 16 Oct 2021 05:33:59 +0000 (14:33 +0900)]
Fix lit test failures in clang-ppc* and clang-x64-windows-msvc

3 years agoResolve lit failures in clang after 8ca4b3e's land
Juneyoung Lee [Sat, 16 Oct 2021 04:51:50 +0000 (13:51 +0900)]
Resolve lit failures in clang after 8ca4b3e's land

3 years ago[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn...
Juneyoung Lee [Fri, 15 Oct 2021 10:45:30 +0000 (19:45 +0900)]
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default (2)

This patch updates test files after D105169.
Autogenerated test codes are changed by `utils/update_cc_test_checks.py,` and non-autogenerated test codes are changed as follows:

(1) I wrote a python script that (partially) updates the tests using regex: {F18594904} The script is not perfect, but I believe it gives hints about which patterns are updated to have `noundef` attached.

(2) The remaining tests are updated manually.

Reviewed By: eugenis

Differential Revision: https://reviews.llvm.org/D108453

3 years ago[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn...
Juneyoung Lee [Fri, 15 Oct 2021 10:26:07 +0000 (19:26 +0900)]
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default

Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions.
I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default.

Test updates are made as a separate patch: D108453

Reviewed By: eugenis

Differential Revision: https://reviews.llvm.org/D105169

3 years ago[Polly][docs] Fix Sphinx warning.
Michael Kruse [Sat, 16 Oct 2021 02:11:23 +0000 (21:11 -0500)]
[Polly][docs] Fix Sphinx warning.

ReStructured Text is not Markdown.

3 years ago[X86] Add more tests for D111858. NFC
Craig Topper [Sat, 16 Oct 2021 00:39:00 +0000 (17:39 -0700)]
[X86] Add more tests for D111858. NFC

Add tests with sub instead of neg.

3 years ago[WebAssembly] Add prototype relaxed laneselect instructions
Zhi An Ng [Sat, 16 Oct 2021 00:45:08 +0000 (17:45 -0700)]
[WebAssembly] Add prototype relaxed laneselect instructions

Add i8x16, i16x8, i32x4, i64x2 laneselect instructions. These are only
exposed as builtins, and require user opt-in.

3 years ago[mlir] Add folder for shape.add
Jacques Pienaar [Sat, 16 Oct 2021 00:30:17 +0000 (17:30 -0700)]
[mlir] Add folder for shape.add

3 years ago[mlir][sparse] run less combinations of SpMM in test (to reduce runtime)
Aart Bik [Fri, 15 Oct 2021 17:07:45 +0000 (10:07 -0700)]
[mlir][sparse] run less combinations of SpMM in test (to reduce runtime)

This revision also adds a few passes to the sparse compiler part to unify the transformation sequence with all other paths we currently use.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111900

3 years ago[MLIR][TOSA] Drop "OnTensors" suffix
Geoffrey Martin-Noble [Fri, 15 Oct 2021 22:40:42 +0000 (15:40 -0700)]
[MLIR][TOSA] Drop "OnTensors" suffix

This is the only lowering to Linalg Tosa has, so it's needlessly
verbose. Likely this was a carry over from IREE's usage where we
originally lowered to linalg on buffers (the only linalg that existed at
the time), so the everything on tensors needed the suffix. We're dropping
it in IREE also, having transitioned entirely to using Linalg on
tensors.

Reviewed By: sjarus

Differential Revision: https://reviews.llvm.org/D111911

3 years ago[ELF] Require two-dash form for --pack-dyn-relocs
Fangrui Song [Fri, 15 Oct 2021 22:36:30 +0000 (15:36 -0700)]
[ELF] Require two-dash form for --pack-dyn-relocs

LLD specific options can be more rigid.
Also add a test.

3 years ago[clang] fix typo correction not looking for candidates in base classes.
Matheus Izvekov [Thu, 14 Oct 2021 19:49:22 +0000 (21:49 +0200)]
[clang] fix typo correction not looking for candidates in base classes.

RecordMemberExprValidator was not looking through ElaboratedType
nodes when looking for candidates which occur in base classes.

Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Reviewed By: rsmith

Differential Revision: https://reviews.llvm.org/D111830

3 years agoRevert "[HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols"
Anshil Gandhi [Fri, 15 Oct 2021 22:15:45 +0000 (16:15 -0600)]
Revert "[HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols"

This reverts commit 03375a3fb33b11e1249d9c88070b7f33cb97802a.

3 years agoFix Xcode project for debugserver
Lawrence D'\''Anna [Fri, 15 Oct 2021 22:05:46 +0000 (15:05 -0700)]
Fix Xcode project for debugserver

It seems StringConvert.cpp was moved, and the Xcode project file
wasn't updated.

Reviewed By: JDevlieghere

Differential Revision: https://reviews.llvm.org/D111910

3 years ago[ConstantRange] Compute precise shl range for single elements
Nikita Popov [Fri, 15 Oct 2021 19:25:39 +0000 (21:25 +0200)]
[ConstantRange] Compute precise shl range for single elements

For the common case where the shift amount is constant (a single
element range) we can easily compute a precise range (up to
unsigned envelope), so do that.

3 years ago[HIP] Relax conditions for address space cast in builtin args
Anshil Gandhi [Fri, 15 Oct 2021 21:16:32 +0000 (15:16 -0600)]
[HIP] Relax conditions for address space cast in builtin args

Allow (implicit) address space casting between LLVM-equivalent
target address spaces.

Reviewed By: yaxunl, tra

Differential Revision: https://reviews.llvm.org/D111734

3 years ago[NFC] Make Assume2KnowledgeMap's typedef more precise
Arthur Eubanks [Fri, 15 Oct 2021 21:33:13 +0000 (14:33 -0700)]
[NFC] Make Assume2KnowledgeMap's typedef more precise

3 years ago[InstCombine] generalize fold for mask-with-signbit-splat, part 2
Sanjay Patel [Fri, 15 Oct 2021 21:09:02 +0000 (17:09 -0400)]
[InstCombine] generalize fold for mask-with-signbit-splat, part 2

This removes an over-specified fold. The more general transform
was added with:
727e642e970d

There's a difference on an existing test that shows a potentially
unnecessary use limit on an icmp fold.

That fold is in InstCombinerImpl::foldICmpSubConstant(), and IIRC
there was some back-and-forth on it and similar folds because they
could cause analysis/passes (SCEV, LSR?) to miss optimizations.

Differential Revision: https://reviews.llvm.org/D111410

3 years ago[AMDGPU] Precommit fused-bitlogic.ll test. NFC.
Stanislav Mekhanoshin [Fri, 15 Oct 2021 20:50:23 +0000 (13:50 -0700)]
[AMDGPU] Precommit fused-bitlogic.ll test. NFC.

3 years ago[ConstantRange] Support checking optimality for subset of inputs (NFC)
Nikita Popov [Fri, 15 Oct 2021 20:44:52 +0000 (22:44 +0200)]
[ConstantRange] Support checking optimality for subset of inputs (NFC)

We always want to check correctness, but for some operations we
can only guarantee optimality for a subset of inputs. Accept an
additional predicate that determines whether optimality for a
given pair of ranges should be checked.

3 years agoRevert "[HIP] Relax conditions for address space cast in builtin args"
Anshil Gandhi [Fri, 15 Oct 2021 20:41:41 +0000 (14:41 -0600)]
Revert "[HIP] Relax conditions for address space cast in builtin args"

This reverts commit 3b48e1170dc623a95ff13a1e34c839cc094bf321.

3 years ago[InstCombine] generalize fold for mask-with-signbit-splat
Sanjay Patel [Fri, 15 Oct 2021 20:22:59 +0000 (16:22 -0400)]
[InstCombine] generalize fold for mask-with-signbit-splat

(iN X s>> (N-1)) & Y --> (X < 0) ? Y : 0

https://alive2.llvm.org/ce/z/qeYhdz

I was looking at a missing abs() transform and found my way to this
generalization of an existing fold that was added with D67799.
As discussed in that review, we want to make sure codegen handles
this difference well, and for all of the targets/types that I
spot-checked, it looks good.

I am leaving the existing fold in place in this commit because
it covers a potentially missing icmp fold, but I plan to remove
that as a follow-up commit as suggested during review.

Differential Revision: https://reviews.llvm.org/D111410

3 years ago[HIP] Relax conditions for address space cast in builtin args
Anshil Gandhi [Fri, 15 Oct 2021 19:44:38 +0000 (13:44 -0600)]
[HIP] Relax conditions for address space cast in builtin args

Allow (implicit) address space casting between LLVM-equivalent
target address spaces.

Reviewed By: yaxunl

Differential Revision: https://reviews.llvm.org/D111734

3 years ago[BasicAA] Rename ExtendedValue to CastedValue (NFC)
Nikita Popov [Fri, 15 Oct 2021 19:56:54 +0000 (21:56 +0200)]
[BasicAA] Rename ExtendedValue to CastedValue (NFC)

As suggested on D110977, rename ExtendedValue to CastedValue,
because it will contain more than just extensions in the future.

3 years ago[ConstantRange] Better diagnostic for correctness test failure (NFC)
Nikita Popov [Fri, 15 Oct 2021 19:35:57 +0000 (21:35 +0200)]
[ConstantRange] Better diagnostic for correctness test failure (NFC)

Print a friendly error message including the inputs, result and
not-contained element if an exhaustive correctness test fails,
same as we do if the optimality test fails.

3 years ago[modules] Make a module map referenced by a system map a system one too.
Volodymyr Sapsai [Sat, 9 Oct 2021 00:10:18 +0000 (17:10 -0700)]
[modules] Make a module map referenced by a system map a system one too.

Mimic the behavior of including headers where a system includer makes an
includee a system header too.

rdar://84049469

Differential Revision: https://reviews.llvm.org/D111476

3 years ago[VectorCombine] Add option to only run scalarization transforms.
Florian Hahn [Fri, 15 Oct 2021 18:27:23 +0000 (19:27 +0100)]
[VectorCombine] Add option to only run scalarization transforms.

This patch adds a pass option to only run transforms that scalarize
vector operations and do not create new vector instructions.

When running VectorCombine early in the pipeline introducing new vector
operations can have negative effects, like blocking loop or SLP
vectorization. To avoid regressions, restrict the early VectorCombine
run (when using -enable-matrix) to only perform scalarization and not
introduce new vector operations.

This is done as option to the pass directly, which is then set when
adding the pass to the pipeline. This is done for the new pass manager
only.

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D111800

3 years ago[compiler-rt/profile] Hide __llvm_profile_raw_version
Pirama Arumuga Nainar [Fri, 15 Oct 2021 18:56:16 +0000 (11:56 -0700)]
[compiler-rt/profile] Hide __llvm_profile_raw_version

Hide __llvm_profile_raw_version so as not to resolve reference from a
dependent shared object.  Since libclang_rt.profile is added later in
the command line, a definition of __llvm_profile_raw_version is not
included if it is provided from an earlier object, e.g.  from a shared
dependency.

This causes an extra dependence edge where if libA.so depends on libB.so
and both are coverage-instrumented, libA.so uses libB.so's definition of
__llvm_profile_raw_version.  This leads to a runtime link failure if the
libB.so available at runtime does not provide this symbol (but provides
the other dependent symbols).  Such a scenario can occur in Android's
mainline modules.
E.g.:
  ld -o libB.so libclang_rt.profile-x86_64.a
  ld -o libA.so -l B libclang_rt.profile-x86_64.a

libB.so has a global definition of __llvm_profile_raw_version.  libA.so
uses libB.so's definition of __llvm_profile_raw_version.  At runtime,
libB.so may not be coverage-instrumented (i.e. not export
__llvm_profile_raw_version) so runtime linking of libA.so will fail.

Marking this symbol as hidden forces each binary to use the definition
of __llvm_profile_raw_version from libclang_rt.profile.

Differential Revision: https://reviews.llvm.org/D111759

3 years ago[WebAssembly] Add import info to `dylink` section of shared libraries
Sam Clegg [Thu, 7 Oct 2021 19:17:15 +0000 (12:17 -0700)]
[WebAssembly] Add import info to `dylink` section of shared libraries

See https://github.com/WebAssembly/tool-conventions/pull/175

Differential Revision: https://reviews.llvm.org/D111345

3 years ago[SelectionDAG] Fix typo in option help
Mingming Liu [Fri, 15 Oct 2021 18:27:40 +0000 (11:27 -0700)]
[SelectionDAG] Fix typo in option help

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D111867

3 years ago[HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols
Anshil Gandhi [Fri, 15 Oct 2021 17:13:12 +0000 (11:13 -0600)]
[HIP] [AlwaysInliner] Disable AlwaysInliner to eliminate undefined symbols

By default clang emits complete contructors as alias of base constructors if they are the same.
The backend is supposed to emit symbols for the alias, otherwise it causes undefined symbols.
@yaxunl observed that this issue is related to the llvm options `-amdgpu-early-inline-all=true`
and `-amdgpu-function-calls=false`. This issue is resolved by only inlining global values
with internal linkage. The `getCalleeFunction()` in AMDGPUResourceUsageAnalysis also had
to be extended to support aliases to functions. inline-calls.ll was corrected appropriately.

Reviewed By: yaxunl, #amdgpu

Differential Revision: https://reviews.llvm.org/D109707

3 years ago[lld/mac] Mark private externs with GOT relocs as LOCAL in indirect symbtab
Nico Weber [Thu, 14 Oct 2021 22:32:10 +0000 (18:32 -0400)]
[lld/mac] Mark private externs with GOT relocs as LOCAL in indirect symbtab

prepareSymbolRelocation() in Writer.cpp adds both symbols that need binding and
symbols relocated with a pointer relocation to the got.

Pointer relocations are emitted for non-movq GOTPCREL(%rip) loads.  (movqs
become GOT_LOADs so that the linker knows they can be relaxed to leaqs, while
others, such as addq, become just GOT -- a pointer relocation -- since they
can't be relaxed in that way).

For example, this C file produces a private_extern GOT relocation when
compiled with -O2 with clang:

    extern const char kString[];
    const char* g(int a) { return kString + a; }

Linkers need to put pointer-relocated symbols into the GOT, but ld64 marks them
as LOCAL in the indirect symbol table. This matters, since `strip -x` looks at
the indirect symbol table when deciding what to strip.

The indirect symtab emitting code was assuming that only symbols that need
binding are in the GOT, but pointer relocations where there too. Hence, the
code needs to explicitly check if a symbol is a private extern.

Fixes https://crbug.com/1242638, which has some more information in comments 14
and 15. With this patch, the output of `nm -U` on Chromium Framework after
stripping now contains just two symbols when using lld, just like with ld64.

Differential Revision: https://reviews.llvm.org/D111852

3 years ago[amdgpu] Fix a crash case when preserving MDT in SILowerControlFlow
Michael Liao [Thu, 14 Oct 2021 20:10:15 +0000 (16:10 -0400)]
[amdgpu] Fix a crash case when preserving MDT in SILowerControlFlow

- When a redundant MBB is being erased from MDT, check whether its
  single successor is dominiated by it. If yes, update that successor's
  idom before erasing MBB; otherwise, it implies MBB is a leaf node and
  could be erased directly.

Reviewed By: foad

Differential Revision: https://reviews.llvm.org/D111831

3 years ago[ubsan] Remove REQUIRED from some TestCases
Vitaly Buka [Fri, 15 Oct 2021 02:04:23 +0000 (19:04 -0700)]
[ubsan] Remove REQUIRED from some TestCases

It's not obvious why they are needed, and tests pass.

Reviewed By: lebedev.ri

Differential Revision: https://reviews.llvm.org/D111859

3 years ago[clang] Pass -clear-ast-before-backend in Clang::ConstructJob()
Arthur Eubanks [Wed, 6 Oct 2021 20:57:29 +0000 (13:57 -0700)]
[clang] Pass -clear-ast-before-backend in Clang::ConstructJob()

This clears the memory used for the Clang AST before we run LLVM passes.

https://llvm-compile-time-tracker.com/compare.php?from=d0a5f61c4f6fccec87fd5207e3fcd9502dd59854&to=b7437fee79e04464dd968e1a29185495f3590481&stat=max-rss
shows significant memory savings with no slowdown (in fact -O0 slightly speeds up).

For more background, see
https://lists.llvm.org/pipermail/cfe-dev/2021-September/068930.html.

Turn this off for the interpreter since it does codegen multiple times.

Differential Revision: https://reviews.llvm.org/D111270

3 years ago[SystemZ] Handle huge immediates in SystemZInstrInfo::loadImmediate().
Jonas Paulsson [Fri, 15 Oct 2021 15:43:39 +0000 (17:43 +0200)]
[SystemZ] Handle huge immediates in SystemZInstrInfo::loadImmediate().

This is needed during isel pseudo expansion in order not to crash on huge
immediates.

Review: Ulrich Weigand

3 years ago[clang] Use llvm::is_contained (NFC)
Kazu Hirata [Fri, 15 Oct 2021 17:07:08 +0000 (10:07 -0700)]
[clang] Use llvm::is_contained (NFC)

3 years agoNFC: Remove wayward MIR tests from lib/Target
Jessica Paquette [Fri, 15 Oct 2021 16:59:00 +0000 (09:59 -0700)]
NFC: Remove wayward MIR tests from lib/Target

These were put in lib/Target instead of tests.

Thankfully dupes of them already existed in the tests directory.

So, just delete them.

3 years ago[lldb] Harden TestCompletion against new settings in 'target.process'
Raphael Isemann [Fri, 15 Oct 2021 16:48:17 +0000 (18:48 +0200)]
[lldb] Harden TestCompletion against new settings in 'target.process'

This test starts failing when people add a setting starting with
`target.process.t` which of course can easily happen. Make it a bit more
resistant by only requiring that `target.process.thr` has a unique completion.

3 years ago[SLP]Add a test for shrink shuffle after reorder, NFC.
Alexey Bataev [Fri, 15 Oct 2021 16:42:14 +0000 (09:42 -0700)]
[SLP]Add a test for shrink shuffle after reorder, NFC.

3 years ago[DebugInfo] retainedTypes should not have subprograms
Ellis Hoag [Fri, 15 Oct 2021 16:37:12 +0000 (12:37 -0400)]
[DebugInfo] retainedTypes should not have subprograms

After D80369, the retainedTypes in CU's should not have any subprograms
so we should not handle that case when emitting debug info.

Differential Revision: https://reviews.llvm.org/D111593

3 years ago[DebugInfo] Limit the size of DIExpressions that we will salvage up to
Stephen Tozer [Tue, 21 Sep 2021 14:11:22 +0000 (15:11 +0100)]
[DebugInfo] Limit the size of DIExpressions that we will salvage up to

Fixes: https://bugs.llvm.org/show_bug.cgi?id=51841

This patch places an arbitrary limit on the size of DIExpressions that
we will produce via salvaging, for performance reasons. This helps to
fix a performance issue observed in the bug above, in which debug values
would be salvaged hundreds of times, producing expressions with over
1000 elements and causing the compiler to hang. Limiting the size of
debug values that we will produce to 128 largely fixes this issue.

Reviewed By: dblaikie, jmorse

Differential Revision: https://reviews.llvm.org/D110332

3 years ago[mlir][sparse] implement sparse tensor init operation
Aart Bik [Thu, 14 Oct 2021 01:33:53 +0000 (18:33 -0700)]
[mlir][sparse] implement sparse tensor init operation

Next step towards supporting sparse tensors outputs.
Also some minor refactoring of enum constants as well
as replacing tensor arguments with proper buffer arguments
(latter is required for more general sizes arguments for
the sparse_tensor.init operation, as well as more general
spares_tensor.convert operations later)

Reviewed By: wrengr

Differential Revision: https://reviews.llvm.org/D111771

3 years ago[AIX] Enable int128 in 64 bit mode
Jinsong Ji [Fri, 15 Oct 2021 15:29:55 +0000 (15:29 +0000)]
[AIX] Enable int128 in 64 bit mode

This patch remove the override in AIX target,
so the int128 is enabled in 64 bit mode or with ForceEnableInt128.

Reviewed By: lkail

Differential Revision: https://reviews.llvm.org/D111078

3 years ago[IR] Fix a few incorrect paths in file header comments. NFC
Craig Topper [Fri, 15 Oct 2021 16:14:04 +0000 (09:14 -0700)]
[IR] Fix a few incorrect paths in file header comments. NFC

3 years ago[clang] Capture Framework when HeaderSearch is resolved via headermap
Cyndy Ishida [Fri, 15 Oct 2021 15:12:54 +0000 (08:12 -0700)]
[clang] Capture Framework when HeaderSearch is resolved via headermap

When building frameworks, headermaps responsible for mapping angle-included headers to their source file location are passed via
`-I` and not `-index-header-map`. Also, `-index-header-map` is only used for indexing purposes and not during most builds.
This patch holds on to the framework's name in HeaderFileInfo as this is retrieveable for cases outside of IndexHeaderMaps and
still represents the framework that is being built.

resolves: rdar://84046893

Reviewed By: jansvoboda11

Differential Revision: https://reviews.llvm.org/D111468

3 years ago[ARM] Don't use TARGET_HEADER_BUILTIN in arm_mve_builtins.inc or arm_cde_builtins.inc
Craig Topper [Fri, 15 Oct 2021 16:01:02 +0000 (09:01 -0700)]
[ARM] Don't use TARGET_HEADER_BUILTIN in arm_mve_builtins.inc or arm_cde_builtins.inc

The attributes string doesn't include 'f' or 'h'. I don't think
any code looks at the header name without those.

Reviewed By: simon_tatham

Differential Revision: https://reviews.llvm.org/D111755

3 years ago[mlir][vector] NFC - Refactor and extract a helper StructuredGenerator class
Nicolas Vasilache [Fri, 15 Oct 2021 15:56:58 +0000 (15:56 +0000)]
[mlir][vector] NFC - Refactor and extract a helper StructuredGenerator class

Differential Revision: https://reviews.llvm.org/D111893

3 years ago[libc++] LWG3266: delete the to_chars(bool) overload.
Konstantin Varlamov [Fri, 15 Oct 2021 15:40:42 +0000 (17:40 +0200)]
[libc++] LWG3266: delete the to_chars(bool) overload.

This PR only updates the synopsis in `<charconv>` -- the current
implementation already [deletes](https://github.com/llvm/llvm-project/blob/e9e6266c704df43e2c52308e1b653dccefa89e04/libcxx/include/charconv#L108)
the overload and has a [test](https://github.com/llvm/llvm-project/blob/main/libcxx/test/std/utilities/charconv/charconv.to.chars/integral.bool.fail.cpp)
for it (and this has been the case from the first [commit](https://reviews.llvm.org/D41458)
where `<charconv>` was added).

Reviewed By: #libc, Mordante

Differential Revision: https://reviews.llvm.org/D111845

3 years ago[OpenMP][Tools][NFC] Make an Archer test more robust
Joachim Protze [Fri, 15 Oct 2021 15:16:07 +0000 (17:16 +0200)]
[OpenMP][Tools][NFC] Make an Archer test more robust

The execution order of the tasks is not fixed, so there is no ordering
for the write accesses. Enforce the ordering that is expected in the check.

3 years ago[libc++][doc] Use issue labels.
Mark de Wever [Fri, 8 Oct 2021 19:01:55 +0000 (21:01 +0200)]
[libc++][doc] Use issue labels.

During the review of D111166 I had a private discussion with @ldionne to
avoid the duplication of the C++2b issues in the Ranges and Format
status pages. The main reason for duplicating them is to make it easier to
find them. The title of the paper may not always make it clear to which
project the paper belongs.

This commit removes all LWG-issues from the Ranges and Format status page
and adds labels for these issue in the C++20/C++23 issues list.

A quick scan revealed there are some issues that are missing a label since
they weren't on the ranges issue list. These can be labelled in a separate
commit. In that commit I'll also look for issues for the spaceship operator
and chrono.

Reviewed By: Quuxplusone, ldionne, #libc

Differential Revision: https://reviews.llvm.org/D111458

3 years ago[llvm-readelf] Make -W an alias of --wide
gbreynoo [Fri, 15 Oct 2021 15:27:53 +0000 (16:27 +0100)]
[llvm-readelf] Make -W an alias of --wide

Currently -W and --wide are treated as two options as they are only
included for gnu readelf compatibility and ignored. This change makes -W
an alias of --wide to be consistent with other option aliases.

Differential Revision: https://reviews.llvm.org/D111731

3 years ago[VectorCombine] add tests for shuffle of binops; NFC
Sanjay Patel [Fri, 15 Oct 2021 12:37:33 +0000 (08:37 -0400)]
[VectorCombine] add tests for shuffle of binops; NFC

3 years ago[lldb] Add test for struct merging in scratch ASTContext
Raphael Isemann [Fri, 15 Oct 2021 13:01:27 +0000 (15:01 +0200)]
[lldb] Add test for struct merging in scratch ASTContext

3 years ago[AArch64] Fix failing test target-invalid-cpu-note.c
Tomas Matheson [Fri, 15 Oct 2021 15:09:54 +0000 (16:09 +0100)]
[AArch64] Fix failing test target-invalid-cpu-note.c

3 years ago[Driver][NetBSD] Use Triple reference instead of ToolChain.getTriple().
Frederic Cambus [Fri, 15 Oct 2021 14:36:19 +0000 (16:36 +0200)]
[Driver][NetBSD] Use Triple reference instead of ToolChain.getTriple().

Differential Revision: https://reviews.llvm.org/D111805

3 years ago[AMDGPU] Avoid redundant calls to numBits in AMDGPUCodeGenPrepare::replaceMulWithMul24().
Abinav Puthan Purayil [Fri, 15 Oct 2021 03:59:58 +0000 (09:29 +0530)]
[AMDGPU] Avoid redundant calls to numBits in AMDGPUCodeGenPrepare::replaceMulWithMul24().

The isU24() and isI24() calls numBits to make its decision. This change
replaces them with the internal numBits call so that we can use its
result for the > 32 bit width cases.

Differential Revision: https://reviews.llvm.org/D111864

3 years ago[Polly] Remove checkIslAstExprInt and use RAII instead of manually freeing Expr....
Max Fan [Fri, 15 Oct 2021 13:45:28 +0000 (08:45 -0500)]
[Polly] Remove checkIslAstExprInt and use RAII instead of manually freeing Expr. NFC.

Polly is trying to move towards using isl::ast_expr / isl-noexceptions.h
(which implements RAII) where possible instead of manually managing memory.
checkIslAstExprInt manually frees Expr, so it has been removed to be
more idiomatic and consistent.

Reviewed By: Meinersbur

Differential Revision: https://reviews.llvm.org/D111769

3 years ago[lldb] [test] Add TestGDBServerTargetXML tests for x86 duplicate subregs
Michał Górny [Fri, 15 Oct 2021 14:03:44 +0000 (16:03 +0200)]
[lldb] [test] Add TestGDBServerTargetXML tests for x86 duplicate subregs

3 years agoFix a crash on an invalid templated UDL declaration
Aaron Ballman [Fri, 15 Oct 2021 13:56:54 +0000 (09:56 -0400)]
Fix a crash on an invalid templated UDL declaration

We were missing a null pointer check that a template parameter existed
at all.

3 years ago[libc] Memory function benchmarks: rename MemcmpConfiguration in MemcmpOrBcmpConfigur...
Guillaume Chatelet [Fri, 15 Oct 2021 09:26:12 +0000 (09:26 +0000)]
[libc] Memory function benchmarks: rename MemcmpConfiguration in MemcmpOrBcmpConfiguration

This will help make sense of the double use of the same type in https://reviews.llvm.org/D111622#inline-1065560.

Differential Revision: https://reviews.llvm.org/D111868

3 years ago[X86] Enable promotion of i16 popcnt (PR52056)
Dávid Bolvanský [Fri, 15 Oct 2021 13:37:11 +0000 (15:37 +0200)]
[X86] Enable promotion of i16 popcnt (PR52056)

Solves https://bugs.llvm.org/show_bug.cgi?id=52056

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D111507