platform/upstream/llvm.git
18 months agoSplit out `CodeGenTypes` from `CodeGen` for LLT/MVT
NAKAMURA Takumi [Mon, 17 Apr 2023 14:46:47 +0000 (23:46 +0900)]
Split out `CodeGenTypes` from `CodeGen` for LLT/MVT

This reduces dependencies on `llvm-tblgen` so much.

`CodeGenTypes` depends on `Support` at the moment.
Be careful to append deps on this, since Targets' tablegens
depend on this.

Depends on D149024

Differential Revision: https://reviews.llvm.org/D148769

18 months agoRestore CodeGen/MachineValueType.h from `Support`
NAKAMURA Takumi [Mon, 10 Apr 2023 15:05:24 +0000 (00:05 +0900)]
Restore CodeGen/MachineValueType.h from `Support`

This is rework of;

  - rG13e77db2df94 (r328395; MVT)

Since `LowLevelType.h` has been restored to `CodeGen`, `MachinveValueType.h`
can be restored as well.

Depends on D148767

Differential Revision: https://reviews.llvm.org/D149024

18 months agoRestore CodeGen/LowLevelType from `Support`
NAKAMURA Takumi [Mon, 10 Apr 2023 15:05:24 +0000 (00:05 +0900)]
Restore CodeGen/LowLevelType from `Support`

This is rework of;
  - D30046 (LLT)

Since I have introduced `llvm-min-tblgen` as D146352, `llvm-tblgen`
may depend on `CodeGen`.

`LowLevlType.h` originally belonged to `CodeGen`. Almost all userse are
still under `CodeGen` or `Target`. I think `CodeGen` is the right place
to put `LowLevelType.h`.

`MachineValueType.h` may be moved as well. (later, D149024)

I have made many modules depend on `CodeGen`. It is consistent but
inefficient. It will be split out later, D148769

Besides, I had to isolate MVT and LLT in modmap, since
`llvm::PredicateInfo` clashes between `TableGen/CodeGenSchedule.h`
and `Transforms/Utils/PredicateInfo.h`.
(I think better to introduce namespace llvm::TableGen)

Depends on D145937, D146352, and D148768.

Differential Revision: https://reviews.llvm.org/D148767

18 months ago[Bazel] Update for D148308 (riscv-sifive)
NAKAMURA Takumi [Tue, 2 May 2023 15:09:30 +0000 (00:09 +0900)]
[Bazel] Update for D148308 (riscv-sifive)

18 months agoRevert "[Clang][Sema] Fix comparison of constraint expressions"
Erich Keane [Tue, 2 May 2023 15:09:26 +0000 (08:09 -0700)]
Revert "[Clang][Sema] Fix comparison of constraint expressions"

This reverts commit e3b1083e00e62f5d157d15cb8c63a1c3dfdf12e2.

This was reverted because it breaks a number of libstdc++ examples, AND
required a workaround that causes hiding of legitimate bugs.

18 months agoRevert "[Clang][Sema] Add a temporary workaround in SemaConcept.cpp"
Erich Keane [Tue, 2 May 2023 15:09:01 +0000 (08:09 -0700)]
Revert "[Clang][Sema] Add a temporary workaround in SemaConcept.cpp"

This reverts commit ce861ec782ae3f41807b61e855512aaccf3c2149.

18 months ago[mlir][arith] Add narrowing patterns for `max*i` and `min*i`
Jakub Kuderski [Tue, 2 May 2023 14:48:01 +0000 (10:48 -0400)]
[mlir][arith] Add narrowing patterns for `max*i` and `min*i`

Reviewed By: antiagainst

Differential Revision: https://reviews.llvm.org/D149583

18 months ago[ThinLTO] Loosen up variable importing correctness checks
Teresa Johnson [Tue, 2 May 2023 02:04:00 +0000 (19:04 -0700)]
[ThinLTO] Loosen up variable importing correctness checks

After importing variables, we do some checking to ensure that variables
marked read or write only, which have been marked exported (e.g.
because a referencing function has been exported), are on at least one
module's imports list. This is because the read or write only variables
will be internalized, so we need a copy any any module that references
it.

This checking is overly conservative in the case of linkonce_odr or
other linkage types where there can already be a duplicate copy in
existence in the importing module, which therefore wouldn't need to
import it. Loosen up the checking for these linkage types.

Fixes https://github.com/llvm/llvm-project/issues/62468.

Differential Revision: https://reviews.llvm.org/D149630

18 months ago[mlir][arith] Add narrowing patterns for subi, divsi, and divui
Jakub Kuderski [Tue, 2 May 2023 14:41:05 +0000 (10:41 -0400)]
[mlir][arith] Add narrowing patterns for subi, divsi, and divui

Each of these ops is compatible with only one extension kind and
produces an extra result bit.

I checked these transformation in Alive2:
1. subi + extsi: https://alive2.llvm.org/ce/z/ipmZZA
2. divsi + extsi: https://alive2.llvm.org/ce/z/fAcqUv
3. divui + extui: https://alive2.llvm.org/ce/z/QZJpFp

Reviewed By: antiagainst

Differential Revision: https://reviews.llvm.org/D149531

18 months ago[mlir][SparseTensor][ArmSVE] Fix missing lli substitutions
Cullen Rhodes [Tue, 2 May 2023 14:30:02 +0000 (14:30 +0000)]
[mlir][SparseTensor][ArmSVE] Fix missing lli substitutions

The MLIR SVE integration tests are now enabled in the
clang-aarch64-full-2stage buildbot under emulation (QEMU) and two of the
sparse integration tests are failing [1]:

  * mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sorted_coo.mlir
  * mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_spmm.mlir

The reason for this is the SVE RUN lines use plain 'lli' rather than the
'%lli_host_or_aarch64_cmd' substitution that's necessary to run under
emulation. The CI doesn't support SVE so the tests will SIGILL unless
run under emulation.

I should note the logs don't show a SIGILL, only the non-descript:

  FileCheck error: '<stdin>' is empty.

but I expect this is what's actually happening.

https://lab.llvm.org/buildbot/#/builders/179/builds/6051/steps/12/logs/stdio

18 months ago[libc] Fix some missing features from the hermetic test support
Joseph Huber [Tue, 2 May 2023 11:39:14 +0000 (06:39 -0500)]
[libc] Fix some missing features from the hermetic test support

This patch addresses some of the flags and features that are currently
missing from the hermetic test support. This mostly just fixes the
`add_libc_test` option failing to find a few dependencies or missing
arguments from the previous unit test support.

Reviewed By: lntue

Differential Revision: https://reviews.llvm.org/D149629

18 months ago[BOLT] Make sure Mach-O binaries are actually linked
Job Noorman [Tue, 2 May 2023 14:22:48 +0000 (16:22 +0200)]
[BOLT] Make sure Mach-O binaries are actually linked

Note that this issue is also solved by D147544.

Reviewed By: alexander-shaposhnikov

Differential Revision: https://reviews.llvm.org/D149244

18 months ago[BOLT] Make sure all section allocations have deterministic contents
Job Noorman [Tue, 2 May 2023 13:43:37 +0000 (15:43 +0200)]
[BOLT] Make sure all section allocations have deterministic contents

For empty sections, RuntimeDyld always allocates 1 byte but leaves it
uninitialized. This causes the contents of some output sections to be
non-deterministic.

Note that this issue is also solved by D147544.

Fixes #59008

Reviewed By: maksfb

Differential Revision: https://reviews.llvm.org/D149243

18 months ago[mlir][arith] Add narrowing patterns for `addi` and `muli`
Jakub Kuderski [Tue, 2 May 2023 14:09:51 +0000 (10:09 -0400)]
[mlir][arith] Add narrowing patterns for `addi` and `muli`

These two ops are handled in a very similar way -- the only difference
in the number result bits produced.

I checked these transformation with Alive2:
1.  addi + sext: https://alive2.llvm.org/ce/z/3NSs9T
2.  addi + zext: https://alive2.llvm.org/ce/z/t7XHOT
3.  muli + sext: https://alive2.llvm.org/ce/z/-7sfW9
4.  muli + zext: https://alive2.llvm.org/ce/z/h4yntF

Reviewed By: antiagainst

Differential Revision: https://reviews.llvm.org/D149530

18 months ago[CodeGen] One more use of MachineBasicBlock::phis. NFC.
Jay Foad [Tue, 2 May 2023 13:38:27 +0000 (14:38 +0100)]
[CodeGen] One more use of MachineBasicBlock::phis. NFC.

18 months ago[OpenMP] In libomptarget, assume alignment at powers of two
Joel E. Denny [Tue, 2 May 2023 13:44:58 +0000 (09:44 -0400)]
[OpenMP] In libomptarget, assume alignment at powers of two

This patch fixes a bug introduced by D142586, which landed as
434992c96ed1.  The fix was to only look for alignments that are powers
of 2.  See the new test case for details.

Reviewed By: jdoerfert, jhuber6

Differential Revision: https://reviews.llvm.org/D149490

18 months ago[DWARFLinker][NFC] Make interfaces to be compatible.
Alexey Lapshin [Fri, 31 Mar 2023 15:10:11 +0000 (17:10 +0200)]
[DWARFLinker][NFC] Make interfaces to be compatible.

This patch makes interface of AddressManager from DWARFLinker
to be compatible with AddressesMap from DWARFLinkerParallel.
This makes both linkers to be interchangeable.

Differential Revision: https://reviews.llvm.org/D147455

18 months ago[RISCV][RISCV][clang] Split out SiFive Vector C intrinsics from riscv_vector.td
4vtomat [Sat, 18 Mar 2023 12:15:24 +0000 (05:15 -0700)]
[RISCV][RISCV][clang] Split out SiFive Vector C intrinsics from riscv_vector.td

Since we don't always need the vendor extension to be in riscv_vector.td,
so it's better to make it be in separated header.

Depends on D148223 and D148680

Differential Revision: https://reviews.llvm.org/D148308

18 months ago[llvm][unittests] Silence warning on MSVC after 5b2423183cb3
Alexandre Ganea [Tue, 2 May 2023 12:49:35 +0000 (08:49 -0400)]
[llvm][unittests] Silence warning on MSVC after 5b2423183cb3

Differential revision: https://reviews.llvm.org/D149609

18 months ago[lldb] Add cstdio include to fix a595b931f1f91897317a4257df313bddfeb029a6
Dmitry Chernenkov [Tue, 2 May 2023 12:45:28 +0000 (12:45 +0000)]
[lldb] Add cstdio include to fix a595b931f1f91897317a4257df313bddfeb029a6

18 months ago[RISCV] Split out part of riscv_vector.td to riscv_vector_common.td
4vtomat [Wed, 19 Apr 2023 02:43:57 +0000 (19:43 -0700)]
[RISCV] Split out part of riscv_vector.td to riscv_vector_common.td

This makes other new targets able to reuse predefined classes
in their own *.td files.

Differential Revision: https://reviews.llvm.org/D148680

18 months ago[SiFive][RISCV][clang] Support C intrinsics for xsfvcp extension.
Nelson Chu [Wed, 18 May 2022 10:03:34 +0000 (03:03 -0700)]
[SiFive][RISCV][clang] Support C intrinsics for xsfvcp extension.

Depends on D147934 and D147935

Differential Revision: https://reviews.llvm.org/D148223

18 months ago[CodeGen] Make use of MachineBasicBlock::phis. NFC.
Jay Foad [Tue, 2 May 2023 12:29:51 +0000 (13:29 +0100)]
[CodeGen] Make use of MachineBasicBlock::phis. NFC.

18 months agoRevert "[OpenMP] Make `libomptarget` link against `libomp`"
Shilei Tian [Tue, 2 May 2023 12:35:55 +0000 (08:35 -0400)]
Revert "[OpenMP] Make `libomptarget` link against `libomp`"

This reverts commit dc049a4ea681b1d0a4880bae3e19ae0ef40f6e80.

It causes issue of export target.

18 months ago[clang] Do not attempt to zero-extend _BitInt(1) when not required
Mariya Podchishchaeva [Tue, 2 May 2023 12:20:29 +0000 (08:20 -0400)]
[clang] Do not attempt to zero-extend _BitInt(1) when not required

`ConvertTypeForMem` doesn't return wider type for _BitInt unless it is
used in a bitfield, so no need to extend when trying to initialize a
global variable.

Fixes https://github.com/llvm/llvm-project/issues/62207

Reviewed By: erichkeane, shafik

Differential Revision: https://reviews.llvm.org/D149436

18 months ago[ARM] Don't allocate memory if free space in segmented stack is just enough
Zhiyao Ma [Tue, 2 May 2023 09:23:19 +0000 (10:23 +0100)]
[ARM] Don't allocate memory if free space in segmented stack is just enough

Assuming that the stack grows downwards, it is fine if the stack
pointer is exactly at the stacklet boundary. We should use
less-or-equal condition when deciding whether to skip new memory
allocation.

Differential Revision: https://reviews.llvm.org/D149315

18 months ago[Flang][OpenMP] Use fir.if instead of scf.if in lastprivate lowering
Kiran Chandramohan [Tue, 2 May 2023 11:29:42 +0000 (11:29 +0000)]
[Flang][OpenMP] Use fir.if instead of scf.if in lastprivate lowering

For finding the last iteration of a loop, or the last section an
if condition is generated. Using scf::if can cause some lowering
issues since the code contained inside it can have branches. Using
fir::if instead ensures that the fir::if is lowered into branches
along with any code contained inside that can generate branches.

Fixes #62458.

Reviewed By: NimishMishra

Differential Revision: https://reviews.llvm.org/D149547

18 months ago[lldb][test] TestCPP20Standard.py: make it a libc++ test
Michael Buch [Mon, 1 May 2023 17:09:54 +0000 (13:09 -0400)]
[lldb][test] TestCPP20Standard.py: make it a libc++ test

We just want to test whether the language switch works.
This is easier to control for libc++, since for bots building
the tests against libstdc++ we might not have the necessary
`<compare>` header available currently.

18 months ago[lldb][DWARFExpression] Fix DW_OP_div to use signed division
LU Hongyi [Mon, 1 May 2023 11:05:47 +0000 (07:05 -0400)]
[lldb][DWARFExpression] Fix DW_OP_div to use signed division

This patch resolves an issue where a value
is incorrectly displayed if it is represented
by DW_OP_div.

This issue is caused by lldb evaluating
operands of DW_OP_div as unsigned
and performed unintended unsigned
division.

This issue is resolved by creating two
temporary signed scalar and performing
signed division.

(Addresses GH#61727)

Differential Revision: https://reviews.llvm.org/D147370

18 months ago[clang][dataflow] Expose DataflowAnalysisContext from DataflowEnvironment.
Samira Bazuzi [Tue, 2 May 2023 00:08:30 +0000 (00:08 +0000)]
[clang][dataflow] Expose DataflowAnalysisContext from DataflowEnvironment.

This will eliminate the need for more pass-through APIs. Also replace pass-through usages with this exposure.

Reviewed By: ymandel, gribozavr2, xazax.hun

Differential Revision: https://reviews.llvm.org/D149464

18 months ago[EarlyCSE] Only combine metadata for load CSE
Nikita Popov [Tue, 2 May 2023 10:43:12 +0000 (12:43 +0200)]
[EarlyCSE] Only combine metadata for load CSE

There is no need to combine metadata if we're performing store to
load forwarding. In that case we would end up combining metadata
on an unrelated load instruction.

18 months ago[EarlyCSE] Add additional metadata preservation test (NFC)
Nikita Popov [Tue, 2 May 2023 10:37:52 +0000 (12:37 +0200)]
[EarlyCSE] Add additional metadata preservation test (NFC)

18 months ago[gn build] Port 852bf52cc957
LLVM GN Syncbot [Tue, 2 May 2023 10:31:09 +0000 (10:31 +0000)]
[gn build] Port 852bf52cc957

18 months ago[clang-tidy] Add check bugprone-multiple-new-in-one-expression.
Balázs Kéri [Tue, 2 May 2023 07:56:57 +0000 (09:56 +0200)]
[clang-tidy] Add check bugprone-multiple-new-in-one-expression.

Reviewed By: donat.nagy
Fixed test failures with previous commit.

Differential Revision: https://reviews.llvm.org/D138777

18 months ago[LCSSA] Remove unused ScalarEvolution argument (NFC)
Nikita Popov [Tue, 2 May 2023 10:13:53 +0000 (12:13 +0200)]
[LCSSA] Remove unused ScalarEvolution argument (NFC)

After D149435, LCSSA formation no longer needs access to
ScalarEvolution, so remove the argument from the utilities.

18 months ago[AArch64] Add sign bits handling for vector compare nodes
David Green [Tue, 2 May 2023 10:05:35 +0000 (11:05 +0100)]
[AArch64] Add sign bits handling for vector compare nodes

This adds ComputeNumSignBits for the NEON vector comparison nodes, which all
either return 0 or -1. Also adds sign_extend_inreg from VASHR+VSHL to show it
performing transforms.

Differential Revision: https://reviews.llvm.org/D148624

18 months agoFix MSVC "truncation of constant value" warnings. NFC.
Simon Pilgrim [Tue, 2 May 2023 09:42:13 +0000 (10:42 +0100)]
Fix MSVC "truncation of constant value" warnings. NFC.

18 months agoFix MSVC "not all control paths return a value" warnings. NFC.
Simon Pilgrim [Tue, 2 May 2023 09:30:29 +0000 (10:30 +0100)]
Fix MSVC "not all control paths return a value" warnings. NFC.

18 months ago[clang-repl] Add a test coverage for nested out-of-line dtor disambiguation.
Vassil Vassilev [Tue, 2 May 2023 09:29:37 +0000 (09:29 +0000)]
[clang-repl] Add a test coverage for nested out-of-line dtor disambiguation.

This came up as a review comment in https://reviews.llvm.org/D148425

18 months ago[AMDGPU] Make use of new tablegen operator !range. NFC.
Jay Foad [Tue, 2 May 2023 09:29:17 +0000 (10:29 +0100)]
[AMDGPU] Make use of new tablegen operator !range. NFC.

18 months ago[gn build] Port 7b7a6b641afd
LLVM GN Syncbot [Tue, 2 May 2023 09:28:16 +0000 (09:28 +0000)]
[gn build] Port 7b7a6b641afd

18 months agoRevert "[clang-tidy] Add check bugprone-multiple-new-in-one-expression."
Balázs Kéri [Tue, 2 May 2023 09:23:31 +0000 (11:23 +0200)]
Revert "[clang-tidy] Add check bugprone-multiple-new-in-one-expression."

This reverts commit 1aa36da15369678d94add0f64809b11f95795efd.

18 months ago[PGO] Avoid potential const_cast UB (NFC)
Christian Ulmann [Tue, 2 May 2023 09:15:29 +0000 (09:15 +0000)]
[PGO] Avoid potential const_cast UB (NFC)

This commit removes potential UB in the PGO instrumentation passes that
was caused by casting away constness and then potentially modifying the
object.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D148903

18 months ago[gn build] Port 1aa36da15369
LLVM GN Syncbot [Tue, 2 May 2023 09:11:37 +0000 (09:11 +0000)]
[gn build] Port 1aa36da15369

18 months ago[mlir][llvm] Add trap intrinsics to the dialect
Victor Perez [Mon, 1 May 2023 08:34:32 +0000 (09:34 +0100)]
[mlir][llvm] Add trap intrinsics to the dialect

Define llvm.trap, llvm.debugtrap, and llvm.ubsantrap intrinsics in the
llvm dialect.

Signed-off-by: Victor Perez <victor.perez@codeplay.com>
Differential Revision: https://reviews.llvm.org/D149574

18 months ago[SCEV] Reuse Accum variable when handling GEP flags
Nikita Popov [Tue, 2 May 2023 09:00:27 +0000 (11:00 +0200)]
[SCEV] Reuse Accum variable when handling GEP flags

The GEP minus the base pointer (which is the pre-inc addrec) is
exactly the Accum value that was already calculated.

18 months ago[clang-tidy] Add check bugprone-multiple-new-in-one-expression.
Balázs Kéri [Tue, 2 May 2023 07:56:57 +0000 (09:56 +0200)]
[clang-tidy] Add check bugprone-multiple-new-in-one-expression.

Reviewed By: donat.nagy

Differential Revision: https://reviews.llvm.org/D138777

18 months agoTypo fix
Hans Wennborg [Tue, 2 May 2023 08:31:37 +0000 (10:31 +0200)]
Typo fix

18 months agoadd_tablegen: Quick fix to reflect LLVM_TABLEGEN to llvm-min-tblgen
NAKAMURA Takumi [Tue, 2 May 2023 08:02:06 +0000 (17:02 +0900)]
add_tablegen: Quick fix to reflect LLVM_TABLEGEN to llvm-min-tblgen

`sanitizer-x86_64-linux-android` uses LLVM_TABLEGEN.

18 months ago[LCSSA] Don't invalidate SCEV
Nikita Popov [Thu, 27 Apr 2023 10:23:11 +0000 (12:23 +0200)]
[LCSSA] Don't invalidate SCEV

LCSSA currently invalidates SCEV for all instructions for which
LCSSA phi nodes are created. This used to be necessary, because
SCEV has historically tried to maintain LCSSA form as well. As
such, some SCEV uses of the value would have to become uses of
the phi node instead.

However, nowadays SCEV itself no longer maintains LCSSA form.
The SCEV of on LCSSA phi node will be the same as the SCEV of its
argument. LCSSA is instead maintained in the SCEVExpander.

As such, I believe it is no longer necessary to perform any SCEV
invalidation during LCSSA construction.

After this patch the ScalarEvolution argument to the LCSSA utilities
is no longer necessary -- I'll remove it in a followup NFC patch to
keep this patch more concise.

Differential Revision: https://reviews.llvm.org/D149435

18 months ago[LLD][MachO] Do not run Mach-O tests on big-endian hosts
Ulrich Weigand [Tue, 2 May 2023 07:33:16 +0000 (09:33 +0200)]
[LLD][MachO] Do not run Mach-O tests on big-endian hosts

Currently, most Mach-O tests fail when executed on a big-endian host. This is
because the Mach-O back-end does not perform the necessary byte swaps when
accessing the (little-endian) binary file format.

For now, simply consider all Mach-O tests unsupported on big-endian hosts,
to enable running the test suite at all on such hosts.

Reviewed by: oontvoo

Differential Revision: https://reviews.llvm.org/D149270

18 months ago[lldb] Remove unused will_modify argument (NFC)
Jonas Devlieghere [Tue, 2 May 2023 07:17:52 +0000 (00:17 -0700)]
[lldb] Remove unused will_modify argument (NFC)

Various OptionValue related classes are passing around will_modify but
the value is never used. This patch simplifies the interfaces by
removing the redundant argument.

18 months ago[mlir][Bazel] Adjust BUILD file for changes in 5e118f933b6590cecd7f1afb30845a1594bc4a5d
Adrian Kuegel [Tue, 2 May 2023 06:37:44 +0000 (08:37 +0200)]
[mlir][Bazel] Adjust BUILD file for changes in 5e118f933b6590cecd7f1afb30845a1594bc4a5d

18 months ago[mlir][mem2reg] Follow-up adjustments (NFC).
Théo Degioanni [Tue, 2 May 2023 06:06:12 +0000 (06:06 +0000)]
[mlir][mem2reg] Follow-up adjustments (NFC).

Ajustments to the MLIR mem2reg changes.

Reviewed By: gysit, kuhar

Differential Revision: https://reviews.llvm.org/D149431

18 months agoIntroduce MLIR Op Properties
Mehdi Amini [Sun, 26 Feb 2023 15:46:01 +0000 (10:46 -0500)]
Introduce MLIR Op Properties

This new features enabled to dedicate custom storage inline within operations.
This storage can be used as an alternative to attributes to store data that is
specific to an operation. Attribute can also be stored inside the properties
storage if desired, but any kind of data can be present as well. This offers
a way to store and mutate data without uniquing in the Context like Attribute.
See the OpPropertiesTest.cpp for an example where a struct with a
std::vector<> is attached to an operation and mutated in-place:

struct TestProperties {
  int a = -1;
  float b = -1.;
  std::vector<int64_t> array = {-33};
};

More complex scheme (including reference-counting) are also possible.

The only constraint to enable storing a C++ object as "properties" on an
operation is to implement three functions:

- convert from the candidate object to an Attribute
- convert from the Attribute to the candidate object
- hash the object

Optional the parsing and printing can also be customized with 2 extra
functions.

A new options is introduced to ODS to allow dialects to specify:

  let usePropertiesForAttributes = 1;

When set to true, the inherent attributes for all the ops in this dialect
will be using properties instead of being stored alongside discardable
attributes.
The TestDialect showcases this feature.

Another change is that we introduce new APIs on the Operation class
to access separately the inherent attributes from the discardable ones.
We envision deprecating and removing the `getAttr()`, `getAttrsDictionary()`,
and other similar method which don't make the distinction explicit, leading
to an entirely separate namespace for discardable attributes.

Recommit d572cd1b067f after fixing python bindings build.

Differential Revision: https://reviews.llvm.org/D141742

18 months ago[SelectionDAG] Use int64_t to store the integer power of llvm.powi
Shengchen Kan [Tue, 2 May 2023 05:49:15 +0000 (13:49 +0800)]
[SelectionDAG] Use int64_t to store the integer power of llvm.powi

https://llvm.org/docs/LangRef.html#llvm-powi-intrinsic
The max length of the integer power of `llvm.powi` intrinsic is 32, and
the value can be negative. If we use `int32_t` to store this value, `-Val`
will underflow when it is `INT32_MIN`

The issue was reported in D149033.

18 months ago[Coverity] Fix unchecked return value, NFC
Phoebe Wang [Mon, 1 May 2023 14:24:24 +0000 (22:24 +0800)]
[Coverity] Fix unchecked return value, NFC

The `ReversePredicate` should have made sure the reverse predicate is
supported by target, but the check comes from early function and might
be invalid by any mistake. So it's better to double confirm it here.

Differential Revision: https://reviews.llvm.org/D149586

18 months ago[MC] Remove dead code, NFC
Phoebe Wang [Mon, 1 May 2023 13:59:01 +0000 (21:59 +0800)]
[MC] Remove dead code, NFC

This code was added in 408b5e660300f9, but neither the return value nor
`Res` was used in the following code.

Differential Revision: https://reviews.llvm.org/D149584

18 months ago[RISCV] Add Smaia and Ssaia extensions support
4vtomat [Wed, 12 Apr 2023 02:49:14 +0000 (19:49 -0700)]
[RISCV] Add Smaia and Ssaia extensions support

This patch implements 1.0-RC3:
https://github.com/riscv/riscv-aia/releases/download/1.0-RC3/riscv-interrupts-1.0-RC3.pdf

Differential Revision: https://reviews.llvm.org/D148066

18 months agoRevert "[SelectionDAG] Use logic right shift to avoid loop hang"
Shengchen Kan [Tue, 2 May 2023 05:13:29 +0000 (13:13 +0800)]
Revert "[SelectionDAG] Use logic right shift to avoid loop hang"

This reverts commit b73229e55543b4ba2b293adcb8b7d6025f01f7d9.

It caused LIT failure on non-X86 targets.

18 months ago[RISCV] Add missing constraints for vwsll
4vtomat [Tue, 25 Apr 2023 06:31:49 +0000 (23:31 -0700)]
[RISCV] Add missing constraints for vwsll

Add missing early clobber and widen constraints for vector crypto instruction: vwsll

Differential Revision: https://reviews.llvm.org/D149127

18 months agoadd_tablegen: Remove LLVM_ENABLE_OBJLIB
NAKAMURA Takumi [Tue, 2 May 2023 04:57:35 +0000 (13:57 +0900)]
add_tablegen: Remove LLVM_ENABLE_OBJLIB

- It doesn't work if `ALL_FILES` has `$<TARGET_OBJECTS>` with Makefile generator.
- It was a micro-optimization in the ancient age. (introduced in rGe6bc093b42ff)

18 months ago[SelectionDAG] Use logic right shift to avoid loop hang
Shengchen Kan [Tue, 2 May 2023 03:46:57 +0000 (11:46 +0800)]
[SelectionDAG] Use logic right shift to avoid loop hang

Issue was reported in D149033, `Val` can be negative value and
arithmetic right shift always keeps the sign bit.

BTW, the redundant code `Val = -Val` is removed by this patch.

18 months ago[lldb] Refactor OptionValueProperties to return a std::optional (NFC)
Jonas Devlieghere [Tue, 2 May 2023 04:04:24 +0000 (21:04 -0700)]
[lldb] Refactor OptionValueProperties to return a std::optional (NFC)

Similar to fdbe7c7faa54, refactor OptionValueProperties to return a
std::optional instead of taking a fail value. This allows the caller to
handle situations where there's no value, instead of being unable to
distinguish between the absence of a value and the value happening the
match the fail value. When a fail value is required,
std::optional::value_or() provides the same functionality.

18 months agoRevert "[SimpleLoopUnswitch] unswitch selects"
Vitaly Buka [Tue, 2 May 2023 04:37:41 +0000 (21:37 -0700)]
Revert "[SimpleLoopUnswitch] unswitch selects"
Revert "Don't loop unswitch vector selects"

Breaks msan. Details in D138526.

This reverts commit bf089732775520624cb4983bfed6c341e1b4c405.
This reverts commit e479ed90b591c18873fda68c12946b9d08cbe02f.

18 months ago[lldb] Remove TestExternalEditor.test
Jonas Devlieghere [Tue, 2 May 2023 04:25:12 +0000 (21:25 -0700)]
[lldb] Remove TestExternalEditor.test

Remove TestExternalEditor.test as the code to open the transcript in an
external editor is conditional on lldb running in a graphical session.
As a result this test passes locally but not on the bots.

18 months ago[lldb] Fix warning: extra ';' outside of a function is incompatible with C++98
Jonas Devlieghere [Tue, 2 May 2023 04:23:47 +0000 (21:23 -0700)]
[lldb] Fix warning: extra ';' outside of a function is incompatible with C++98

Fix warning, remove trailing whitespace and fix formatting.

18 months ago[lldb] Fix bug introduced by fdbe7c7faa54
Jonas Devlieghere [Tue, 2 May 2023 04:22:40 +0000 (21:22 -0700)]
[lldb] Fix bug introduced by fdbe7c7faa54

I didn't account for the scenario where the string was set but was
empty. This commit restores the old behavior and fixes
TestMemoryFind.py.

18 months ago[clang][Interp][NFC] Make OptionScope final
Timm Bäder [Sun, 30 Apr 2023 17:37:03 +0000 (19:37 +0200)]
[clang][Interp][NFC] Make OptionScope final

18 months ago[clang][Interp][NFC] Add documentation for SetLocal
Timm Bäder [Sun, 30 Apr 2023 14:02:46 +0000 (16:02 +0200)]
[clang][Interp][NFC] Add documentation for SetLocal

18 months ago[lldb] Refactor OptionValue to return a std::optional (NFC)
Jonas Devlieghere [Tue, 2 May 2023 03:34:51 +0000 (20:34 -0700)]
[lldb] Refactor OptionValue to return a std::optional (NFC)

Refactor OptionValue to return a std::optional instead of taking a fail
value. This allows the caller to handle situations where there's no
value, instead of being unable to distinguish between the absence of a
value and the value happening the match the fail value. When a fail
value is required, std::optional::value_or() provides the same
functionality.

18 months agoAdd deps on LLVMTableGenCommon even if it is actually unused.
NAKAMURA Takumi [Tue, 2 May 2023 03:43:16 +0000 (12:43 +0900)]
Add deps on LLVMTableGenCommon even if it is actually unused.

18 months agollvm-tblgen: Split out `obj.LLVMTableGenCommon`
NAKAMURA Takumi [Tue, 2 May 2023 03:19:27 +0000 (12:19 +0900)]
llvm-tblgen: Split out `obj.LLVMTableGenCommon`

`$<TARGET_OBJECTS:llvm-min-tblgen>` was too lazy.
It has `rc.res` in the list with MS toolchain.

Fixup for D146352

18 months ago[mlir][tosa] Typo in transpose canonicalization message
Kai Sasaki [Tue, 2 May 2023 02:47:26 +0000 (11:47 +0900)]
[mlir][tosa] Typo in transpose canonicalization message

18 months agoIntroduce `llvm-min-tblgen` to build public header files
NAKAMURA Takumi [Fri, 17 Mar 2023 20:04:22 +0000 (05:04 +0900)]
Introduce `llvm-min-tblgen` to build public header files

`llvm-min-tblgen` is capable of building `llvm/include/llvm`;

- `-gen-attrs`
- `-gen-directive-*`
- `-gen-intrinsics-*`
- `-gen-riscv-target-def`

`llvm-min-tblgen` is built and used only when `llvm-tblgen` is built in-tree.
This is not installed.

`llvm-tblgen` is built with complete set and may be installed.
`check-llvm` uses not `llvm-min-tblgen` but `llvm-tblgen`.

`LLVM_TABLEGEN_PROJECT` overrides the definition of `tablegen(project)`.
`LLVM_HEADERS` is used as the overridden prefix for LLVM header generators.

If `EXPORT` is not specified in `add_tablegen`, its tablegen is treated as internal.

Let `llvm-tblgen` depend on `intrinsics_gen`

Depends on D149072

Differential Revision: https://reviews.llvm.org/D146352

18 months ago[CMake] add_tablegen: Rename host tablegen to `${target}-host"
NAKAMURA Takumi [Thu, 27 Apr 2023 13:15:46 +0000 (22:15 +0900)]
[CMake] add_tablegen: Rename host tablegen to `${target}-host"

Since `llvm-min-tblgen` has been introduced, `llvm-min-tblgen-host`
would make more sense than `LLVM_HEADERS-tablegen-host`.

Depends on D146352

Differential Revision: https://reviews.llvm.org/D149343

18 months ago[RISCV] Use vslide1down idiom for generic build_vector
Philip Reames [Tue, 2 May 2023 01:47:28 +0000 (18:47 -0700)]
[RISCV] Use vslide1down idiom for generic build_vector

We had previously been going through the stack.

A couple small notes:
    We have the vslide1down idiom in a few other places. As a post patch, I plan to try to common the code a bit.
    VF=2 case is still going through the splat + insert path. Picking the optimal sequence for this seems to be a bit fiddly (due to constant mat costs), so I restricted this to cases which would have previously hit the stack.
    I'm only handling integer vectors for the moment. Mostly because I don't see the existing vfslide1down ISD nodes being in place. This will be an obvious followup.
    One of the test diffs does expose a missing combine - a build_vector with a prefix coming from a vector extract sequence. The code after this is arguably worse (due to domain crossing vs stack store), but I think this is a narrow enough case to be non-blocking for now. Let me know if you disagree.

Differential Revision: https://reviews.llvm.org/D149263

18 months ago[RISCV][InsertVSETVLI] Avoid VL toggles for extractelement patterns
Philip Reames [Tue, 2 May 2023 00:49:35 +0000 (17:49 -0700)]
[RISCV][InsertVSETVLI] Avoid VL toggles for extractelement patterns

Noticed this while looking at some SLP output. If we have an extractelement, we're probably using a slidedown into an destination with no contents. Given this, we can allow the slideup to use a larger VL and clobber tail elements of the destination vector. Doing this allows us to avoid vsetvli toggles in many fixed length vector examples.

Differential Revision: https://reviews.llvm.org/D148834

18 months ago[ELF][test] Improve AVR tests
Ben Shi [Mon, 1 May 2023 08:33:40 +0000 (16:33 +0800)]
[ELF][test] Improve AVR tests

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D149318

18 months ago[clang][Driver] Fix crash with unsupported architectures in MinGW and CrossWindows
KOMATA Manabu [Tue, 2 May 2023 01:34:09 +0000 (18:34 -0700)]
[clang][Driver] Fix crash with unsupported architectures in MinGW and CrossWindows

MinGW and CrossWindows crash if the target is an unsupported target architecture.
Changed it to emit an error message.

Fixes https://github.com/llvm/llvm-project/issues/59545

Reviewed By: MaskRay, mstorsjo

Differential Revision: https://reviews.llvm.org/D148944

18 months ago[Driver] Drop incorrect hint from err_target_unknown_triple
Fangrui Song [Tue, 2 May 2023 01:22:49 +0000 (18:22 -0700)]
[Driver] Drop incorrect hint from err_target_unknown_triple

For an unknown target triple, the ", please use -triple or -arch" hint of the
diagnostic is almost always wrong.
For cc1, the error is due to a specified but unknown -triple.
For driver, -triple and -arch are not driver options.

18 months ago[PowerPC] Implement DFP add and sub instructions.
Stefan Pintilie [Mon, 1 May 2023 17:11:47 +0000 (13:11 -0400)]
[PowerPC] Implement DFP add and sub instructions.

Add the following Decimal Floating Point (DFP) instructions for PowerPC.
dadd, daddq, dsub, dsubq

In order to add these instructions a new register class for a pair
of floating point registers is added.

This patch is only to allow the user to specify the instructions in
assembly. There is no scheduling or patterns for the instructions.

Reviewed By: amyk

Differential Revision: https://reviews.llvm.org/D148597

18 months ago[libc++][PSTL] Simplify pstl_config.h
Nikolas Klauser [Fri, 28 Apr 2023 23:26:09 +0000 (16:26 -0700)]
[libc++][PSTL] Simplify pstl_config.h

This is done by removing configurations we don't support.

Reviewed By: ldionne, #libc

Spies: libcxx-commits

Differential Revision: https://reviews.llvm.org/D149499

18 months ago[gn build] Port a50e54fbeb48
LLVM GN Syncbot [Mon, 1 May 2023 23:27:58 +0000 (23:27 +0000)]
[gn build] Port a50e54fbeb48

18 months ago[CUDA] Temporarily undefine __noinline__ when including bits/shared_ptr_base.h
Artem Belevich [Thu, 27 Apr 2023 18:10:37 +0000 (11:10 -0700)]
[CUDA] Temporarily undefine __noinline__ when including bits/shared_ptr_base.h

This avoid CUDA compilation errors caused by CUDA headers defining __noinline__
and conflicting with __noinline__ use in libstdc++.

Differential Revision: https://reviews.llvm.org/D149364

18 months ago[flang][openacc] Remove unused variable
Valentin Clement [Mon, 1 May 2023 23:23:46 +0000 (16:23 -0700)]
[flang][openacc] Remove unused variable

Fix buildbot failure after landing D149385

18 months agoRe-land "[lldb] Make the libcxx unique_ptr prettyprinter support custom deleters."
Jorge Gorbe Moya [Mon, 1 May 2023 22:57:53 +0000 (15:57 -0700)]
Re-land "[lldb] Make the libcxx unique_ptr prettyprinter support custom deleters."

This reverts commit 45351120105a7257ccb1e38ec1b1f8a452269da2.

18 months ago[lldb] Change ObjectValueDictionary to use a StringMap
Alex Langford [Fri, 28 Apr 2023 20:23:55 +0000 (13:23 -0700)]
[lldb] Change ObjectValueDictionary to use a StringMap

llvm has a structure for maps where the key's type is a string. Using
that also means that the keys for OptionValueDictionary don't stick
around forever in ConstString's StringPool (even after they are gone).

The only thing we lose here is ordering: iterating over the map where the keys
are ConstStrings guarantees that we iterate in alphabetical order.
StringMap makes no guarantees about the ordering when you iterate over
the entire map.

Differential Revision: https://reviews.llvm.org/D149482

18 months ago[flang][openacc] Add basic support for derived type component in data operand
Valentin Clement [Mon, 1 May 2023 23:15:34 +0000 (16:15 -0700)]
[flang][openacc] Add basic support for derived type component in data operand

Add lowering support for derived type component appearing
in a data clause to the newly added data operand operations.

Depends on D149297

Reviewed By: razvanlupusoru, jeanPerier

Differential Revision: https://reviews.llvm.org/D149385

18 months ago[mlir][sparse] relax constraints on tensor.cast with pre-rewriting
Aart Bik [Mon, 1 May 2023 19:56:26 +0000 (12:56 -0700)]
[mlir][sparse] relax constraints on tensor.cast with pre-rewriting

Reviewed By: wrengr

Differential Revision: https://reviews.llvm.org/D149489

18 months ago[OpenMP] Make `libomptarget` link against `libomp`
Shilei Tian [Mon, 1 May 2023 23:01:33 +0000 (19:01 -0400)]
[OpenMP] Make `libomptarget` link against `libomp`

In `libomptarget` we use a couple of functions from `libomp`, but we didn't link
`libomptarget` against `libomp`. That will not work on some platforms such
as macOS. A linker error will be encountered because those symbols are not resolved
at link time when building `libomptarget`. This patch simply makes `libomptarget`
link agains `libomp`, makes it a "user" of `libomp`. I think this will not break
the policies between `libomp` and `libomptarget`.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D149617

18 months agoRevert "Introduce MLIR Op Properties"
Mehdi Amini [Mon, 1 May 2023 22:55:58 +0000 (15:55 -0700)]
Revert "Introduce MLIR Op Properties"

This reverts commit d572cd1b067f1177a981a4711bf2e501eaa8117b.

Some bots are broken and investigation is needed before relanding.

18 months ago[OpenMP] Handle function calls from `libomp` to `libomptarget` correctly
Shilei Tian [Mon, 1 May 2023 22:25:11 +0000 (18:25 -0400)]
[OpenMP] Handle function calls from `libomp` to `libomptarget` correctly

D132005 introduced function calls from `libomp` to `libomptarget` if offloading
is enabled. However, the external function declaration may not always work. For
example, it causes a link error on macOS. Currently it is guarded properly by
a macro, but in order to get OpenMP target offloading working on macOS, it has
to be handled correctly. This patch applies the same idea of how we support
target memory extension by using function pointer indirect call for that function.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D149557

18 months agoIntroduce MLIR Op Properties
Mehdi Amini [Sun, 26 Feb 2023 15:46:01 +0000 (10:46 -0500)]
Introduce MLIR Op Properties

This new features enabled to dedicate custom storage inline within operations.
This storage can be used as an alternative to attributes to store data that is
specific to an operation. Attribute can also be stored inside the properties
storage if desired, but any kind of data can be present as well. This offers
a way to store and mutate data without uniquing in the Context like Attribute.
See the OpPropertiesTest.cpp for an example where a struct with a
std::vector<> is attached to an operation and mutated in-place:

struct TestProperties {
  int a = -1;
  float b = -1.;
  std::vector<int64_t> array = {-33};
};

More complex scheme (including reference-counting) are also possible.

The only constraint to enable storing a C++ object as "properties" on an
operation is to implement three functions:

- convert from the candidate object to an Attribute
- convert from the Attribute to the candidate object
- hash the object

Optional the parsing and printing can also be customized with 2 extra
functions.

A new options is introduced to ODS to allow dialects to specify:

  let usePropertiesForAttributes = 1;

When set to true, the inherent attributes for all the ops in this dialect
will be using properties instead of being stored alongside discardable
attributes.
The TestDialect showcases this feature.

Another change is that we introduce new APIs on the Operation class
to access separately the inherent attributes from the discardable ones.
We envision deprecating and removing the `getAttr()`, `getAttrsDictionary()`,
and other similar method which don't make the distinction explicit, leading
to an entirely separate namespace for discardable attributes.

Differential Revision: https://reviews.llvm.org/D141742

18 months ago[RISCV] Don't store CPUKind in CPUInfo. NFC
Craig Topper [Mon, 1 May 2023 22:26:17 +0000 (15:26 -0700)]
[RISCV] Don't store CPUKind in CPUInfo. NFC

This field is never used today. If you have a pointer to the row
you can find the CPUKind by subtracting the pointer from the start
of the array.

18 months ago[RISCV] Remove INVALID from the list of CPUs in RISCVTargetParser. NFC
Craig Topper [Mon, 1 May 2023 22:26:09 +0000 (15:26 -0700)]
[RISCV] Remove INVALID from the list of CPUs in RISCVTargetParser. NFC

This value is never used outside and is only used as a sentinel
internally which we can solve with other means.

18 months ago[RISCV] Remove or simplify some StringSwitches in RISCVTargetParser.cpp. NFC
Craig Topper [Mon, 1 May 2023 22:26:03 +0000 (15:26 -0700)]
[RISCV] Remove or simplify some StringSwitches in RISCVTargetParser.cpp. NFC

We can iterate over the RISCVCPUInfo table instead of using a
separate StringSwitch.

18 months agoRecommit "[ValueTracking] Use knownbits interface for determining if `div`/`rem`...
Noah Goldstein [Mon, 1 May 2023 20:37:23 +0000 (15:37 -0500)]
Recommit "[ValueTracking] Use knownbits interface for determining if `div`/`rem` are safe to speculate" (2nd Try)

Add `poison` checks that where missing.

Reviewed By: nikic

18 months ago[InstCombine] Fix bug in `FoldOpIntoSelect` where we would incorrectly fold `undef...
Noah Goldstein [Mon, 1 May 2023 20:37:11 +0000 (15:37 -0500)]
[InstCombine] Fix bug in `FoldOpIntoSelect` where we would incorrectly fold `undef` as constant

D146349 Introduced the ability to use the information from the
`select` condition to deduce constants as we folded a binop into
select. I.e if the `select` cond was `icmp eq %A, 10`, then in the
true-arm of `select`, we would be able to replace usage of `A` with
`10`.

This is broken for vectors that contain `undef` elements. I.e with
`icmp eq %A, <10, undef>`, subsituting `<10, undef>` for `A` can
result in creating a more undefined result than we otherwise would
have.

We fix the issue with simply checking if the candidate constant for
substituting may contain `undef` elements and don't do it in that
case.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D149592

18 months ago[libc++][PSTL] Fix clang-tidy
Nikolas Klauser [Fri, 28 Apr 2023 18:02:46 +0000 (11:02 -0700)]
[libc++][PSTL] Fix clang-tidy

Reviewed By: ldionne, #libc

Spies: libcxx-commits, miyuki, carlosgalvezp

Differential Revision: https://reviews.llvm.org/D149500