platform/upstream/llvm.git
18 months ago[CUDA] Allow targeting NVPTX directly without a host toolchain
Joseph Huber [Thu, 15 Dec 2022 20:49:09 +0000 (14:49 -0600)]
[CUDA] Allow targeting NVPTX directly without a host toolchain

Currently, the NVPTX compilation toolchain can only be invoked either
through CUDA or OpenMP via `--offload-device-only`. This is because we
cannot build a CUDA toolchain without an accompanying host toolchain for
the offloading. When using `--target=nvptx64-nvidia-cuda` this results
in generating calls to the GNU assembler and linker, leading to errors.

This patch abstracts the portions of the CUDA toolchain that are
independent of the host toolchain or offloading kind into a new base
class called `NVPTXToolChain`. We still need to read the host's triple
to build the CUDA installation, so if not present we just assume it will
match the host's system for now, or the user can provide the path
explicitly.

This should allow the compiler driver to create NVPTX device images
directly from C/C++ code.

Reviewed By: tra

Differential Revision: https://reviews.llvm.org/D140158

18 months ago[AMDGPU] Run autogen checks on test
Jeffrey Byrnes [Thu, 19 Jan 2023 00:12:18 +0000 (16:12 -0800)]
[AMDGPU] Run autogen checks on test

Change-Id: I46f2ced9ceac592c2a93a00631014a806d4b0693

18 months agoReland [pgo] Avoid introducing relocations by using private alias
Paul Kirth [Wed, 18 Jan 2023 21:54:21 +0000 (21:54 +0000)]
Reland [pgo] Avoid introducing relocations by using private alias

In many cases, we can use an alias to avoid a symbolic relocations,
instead of using the public, interposable symbol. When the instrumented
function is in a COMDAT, we can use a hidden alias, and still avoid
references to discarded sections.

Previous versions of this patch allowed the compiler to name the
generated alias, but that would only be valid when the functions were
local. Since the alias may be used across TUs we use a more
deterministic naming convention, and add a ".local" suffix to the alias
name just as we do for relative vtables aliases.

https://reviews.llvm.org/rG20894a478da224bdd69c91a22a5175b28bc08ed9
removed an incorrect assertion on Mach-O which caused assertion failures in LLD.

We addressed the link errors under ThinLTO + PGO + CFI by being more
selective about which comdat functions can be given aliases.
Specifically, we now do not emit an alias in the case of a comdat
function with hidden visibility, since the alias would have the same
linkage and visibility, giving no benefit over using the symbol
directly. This also prevents LowerTypeTest from incorrectly updating the
dangling alias after GlobalOpt replaces uses, and introducing a
duplicate symbol.

Reviewed By: phosek

Differential Revision: https://reviews.llvm.org/D137982

18 months ago[HWASAN] Remove FindHeapChunkByAddressFastLocked
Kirill Stoimenov [Wed, 18 Jan 2023 22:34:07 +0000 (22:34 +0000)]
[HWASAN] Remove FindHeapChunkByAddressFastLocked

Reviewed By: vitalybuka

Differential Revision: https://reviews.llvm.org/D142042

18 months ago[OpenMP] Fix inconsistent task state if hot team is not used
Shilei Tian [Wed, 18 Jan 2023 23:21:49 +0000 (18:21 -0500)]
[OpenMP] Fix inconsistent task state if hot team is not used

This patch fixes the inconsistent task state when hot team is not used.
When the primary thread executes `__kmp_join_call`, it calls `__kmp_free_team`,
where worker threads will get destroyed if not using hot team. The destroy of
worker threads also reset their task state. However, the primary thread's is not
reset. When the next parallel region is encountered, in `__kmp_task_team_sync`,
the task state of thread will be flipped. Since the state of primary thread is not
reset, it is still 1, but all the worker threads will be 0, this leads to the
inconsistent task state, causing those threads are using completely different
task team.

Fix #59190.

Reviewed By: tlwilmar

Differential Revision: https://reviews.llvm.org/D141979

18 months ago[Clang][NFC] Clang-format CUDA toolchain file
Joseph Huber [Wed, 18 Jan 2023 23:07:02 +0000 (17:07 -0600)]
[Clang][NFC] Clang-format CUDA toolchain file

Summary:
This file is not formatted, which makes further changes to it more
difficult. Format it.

18 months ago[-Wunsafe-buffer-usage] Group diagnostics by variable
Jan Korous [Wed, 18 Jan 2023 22:47:03 +0000 (14:47 -0800)]
[-Wunsafe-buffer-usage] Group diagnostics by variable

Differential Revision: https://reviews.llvm.org/D141356

18 months ago[BOLT][NFC] Replace ambiguous BinarySection::isReadOnly with isWritable
Amir Ayupov [Wed, 18 Jan 2023 22:21:28 +0000 (14:21 -0800)]
[BOLT][NFC] Replace ambiguous BinarySection::isReadOnly with isWritable

Address feedback in https://reviews.llvm.org/D102284#2755060

Reviewed By: yota9

Differential Revision: https://reviews.llvm.org/D141733

18 months ago[StatCacheFileSystem] Attempt to fix build
Fred Riss [Wed, 18 Jan 2023 22:46:55 +0000 (14:46 -0800)]
[StatCacheFileSystem] Attempt to fix build

Some bots are not happy with the way Error is returned here. Let's see if
std::moving it fixes this.

18 months ago[Clang] Give Clang the ability to use a shared stat cache
Fred Riss [Thu, 20 Oct 2022 00:29:36 +0000 (17:29 -0700)]
[Clang] Give Clang the ability to use a shared stat cache

Every Clang instance uses an internal FileSystemStatCache to avoid
stating the same content multiple times. However, different instances
of Clang will contend for filesystem access for their initial stats
during HeaderSearch or module validation.

On some workloads, the time spent in the kernel in these concurrent
stat calls has been measured to be over 20% of the overall compilation
time. This is extremly wassteful when most of the stat calls target
mostly immutable content like a SDK.

This commit introduces a new tool `clang-stat-cache` able to generate
an OnDiskHashmap containing the stat data for a given filesystem
hierarchy.

The driver part of this has been modeled after -ivfsoverlay given
the similarities with what it influences. It introduces a new
-ivfsstatcache driver option to instruct Clang to use a stat cache
generated by `clang-stat-cache`. These stat caches are inserted at
the bottom of the VFS stack (right above the real filesystem).

Differential Revision: https://reviews.llvm.org/D136651

18 months ago[HWASAN] Fix Fuchsia link problem
Kirill Stoimenov [Wed, 18 Jan 2023 22:01:27 +0000 (22:01 +0000)]
[HWASAN] Fix Fuchsia link problem

Reviewed By: mysterymath

Differential Revision: https://reviews.llvm.org/D142057

18 months ago[BOLT][NFC] Simplify handleRelocation
Amir Ayupov [Wed, 18 Jan 2023 22:07:48 +0000 (14:07 -0800)]
[BOLT][NFC] Simplify handleRelocation

Reviewed By: rafauler

Differential Revision: https://reviews.llvm.org/D132089

18 months ago[-Wunsafe-buffer-usage] Use relevant source locations for warnings
Jan Korous [Tue, 10 Jan 2023 01:36:48 +0000 (17:36 -0800)]
[-Wunsafe-buffer-usage] Use relevant source locations for warnings

This way we highlight a particular unsafe subexpression by providing more accurate
source location than begin of an entire statement.

Differential Revision: https://reviews.llvm.org/D141340

18 months ago[clang][Sema] Fix uninitialized `SourceLocation` for types with multiple attributes...
Volodymyr Sapsai [Tue, 10 Jan 2023 20:41:11 +0000 (12:41 -0800)]
[clang][Sema] Fix uninitialized `SourceLocation` for types with multiple attributes and macros.

Some `TypeLoc`s are considered "sugar" and we go past them in
`GetTypeSourceInfoForDeclarator`. The problem is that we peel off only
the same kind of `TypeLoc` at the time which makes it impossible to
handle mixed sequences like
`AttributedTypeLoc - MacroQualifiedTypeLoc - AttributedTypeLoc - PointerTypeLoc`

In this situation, as shown in the added test, we don't get to
`PointerTypeLoc` and don't set its starLoc leaving it uninitialized.

Address FIXME and peel off "sugar" `TypeLoc`s regardless of their order.

rdar://102149264

Differential Revision: https://reviews.llvm.org/D141424

18 months ago[CMake] Allow setting the location of host tools with LLVM_NATIVE_TOOL_DIR
Martin Storsjö [Tue, 2 Aug 2022 11:53:36 +0000 (14:53 +0300)]
[CMake] Allow setting the location of host tools with LLVM_NATIVE_TOOL_DIR

This avoids having to specify the location of all individual tools.
In current builds, one may want to specify LLVM_TABLEGEN, CLANG_TABLEGEN,
LLDB_TABLEGEN, LLVM_CONFIG_PATH, CLANG_PSEUDO_GEN and
CLANG_TIDY_CONFUSABLE_CHARS_GEN; specifying just the base directory
containing all of them is much more convenient.

Factorize the code for setting up use of a tool that is used during
the build (which either is newly built in the same build, or
built in a separate nested cmake build - when cross compiling or
when e.g. optimized tablegen is requested - or used from an existing
prebuilt binary).

Differential Revision: https://reviews.llvm.org/D131052

18 months ago[BOLT][NFC] DataAggregator code cleanup
Amir Ayupov [Fri, 9 Dec 2022 23:09:39 +0000 (15:09 -0800)]
[BOLT][NFC] DataAggregator code cleanup

Reviewed By: #bolt, rafauler

Differential Revision: https://reviews.llvm.org/D139794

18 months ago[BOLT][NFC] Remove AArch64 override for evaluateBranch
Amir Ayupov [Wed, 18 Jan 2023 02:26:34 +0000 (18:26 -0800)]
[BOLT][NFC] Remove AArch64 override for evaluateBranch

Use MCInstrAnalysis implementation

Reviewed By: #bolt, rafauler, yota9

Differential Revision: https://reviews.llvm.org/D141983

18 months ago[SystemZ] Implement lowering of GET_ROUNDING
Tulio Magno Quites Machado Filho [Wed, 18 Jan 2023 19:26:24 +0000 (13:26 -0600)]
[SystemZ] Implement lowering of GET_ROUNDING

Add support for _FLT_ROUNDS_ in SystemZ.

Patch by Tulio Magno Quites Machado Filho.

Reviewed By: Ulrich Weigand

Differential Revision: https://reviews.llvm.org/D140988

18 months agoDo qProcessInfo-hint binary loading later in Process setup
Jason Molenda [Wed, 18 Jan 2023 20:29:52 +0000 (12:29 -0800)]
Do qProcessInfo-hint binary loading later in Process setup

The remote stub may give lldb hints about binaries to
be loaded, especially in a firmware type environment, and
relay those hints in the qProcessInfo response.  The
binary loading was done very early in Process setup, before
we had any threads, and this made it complicated for people
to write dSYM python scripts which need access to a thread.
Delay the binary loading until a bit later in the Process
startup.

Differential Revision: https://reviews.llvm.org/D141972
rdar://104235301

18 months ago[llvm][GenericUniformity] Hack around strict is_invocable() checks
Krzysztof Drewniak [Mon, 16 Jan 2023 18:39:21 +0000 (18:39 +0000)]
[llvm][GenericUniformity] Hack around strict is_invocable() checks

With recent (> 15, as far as I can tell, possibly > 16) clang, c++17,
and GNU's libstdc++ (versions 9 and 10 and maybe others), LLVM fails
to compile due to an is_invocable() check in unique_ptr::reset().

To resolve this issue, add a template argument to ImplDeleter to make
things work.

Differential Revision: https://reviews.llvm.org/D141865

18 months ago[RISCV] Use canonical move instruction in RISCVAsmPrinter::EmitHwasanMemaccessSymbols.
Craig Topper [Wed, 18 Jan 2023 19:38:38 +0000 (11:38 -0800)]
[RISCV] Use canonical move instruction in RISCVAsmPrinter::EmitHwasanMemaccessSymbols.

We were using an OR with X0 which is the canonical move for AArch64.
The canonical move for RISC-V is ADDI reg1, reg2, 0.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D142044

18 months ago[RISCV] Use TargetMachine MCSubtargetInfo in RISCVAsmPrinter::EmitHwasanMemaccessSymbols.
Craig Topper [Wed, 18 Jan 2023 19:38:17 +0000 (11:38 -0800)]
[RISCV] Use TargetMachine MCSubtargetInfo in RISCVAsmPrinter::EmitHwasanMemaccessSymbols.

The STI object in the class is updated each time runOnMachineFunction
is called. That means STI will be based on the last function processed.
If that function has different attributes than the rest of the module
that could theoretically result in odd behavior.

Use the copy in TargetMachine instead. That will use the global
CPU and mattrs.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D142041

18 months ago[RISCV] Make it explicit that attributes use the MCSubtargetInfo from TargetMachine...
Craig Topper [Wed, 18 Jan 2023 19:31:29 +0000 (11:31 -0800)]
[RISCV] Make it explicit that attributes use the MCSubtargetInfo from TargetMachine. NFC

The MCSTI variable is initialized to TM.getMCSubtargetInfo(), but is
re-assigned in every call to runOnMachineFunction. emitAttributes is
called before any call to runOnMachineFunction, but it's not
immediately obvious.

This patch removes the MCSTI variable, and instead queries
TM.getMCSubtargetInfo() at the time emitAttributes is called.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D142037

18 months ago[OpenMP] RecordReplay saves bitcode when JIT-ing
Giorgis Georgakoudis [Wed, 14 Dec 2022 06:57:02 +0000 (22:57 -0800)]
[OpenMP] RecordReplay saves bitcode when JIT-ing

This patch enables to store bitcode images when JIT is enabled for the record-and-replay functionality (see https://reviews.llvm.org/D138931). Credits to @jdoerfert for refactoring the code.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D141986

18 months ago[hwasan] add pattern for short tag in symbolizer
Johan Erlandsson [Wed, 18 Jan 2023 19:10:55 +0000 (11:10 -0800)]
[hwasan] add pattern for short tag in symbolizer

Lines with 'record_addr:' are not processed when failing to decode
access tag. The regular format is: %02x/%02x but for a short tag it's:
%02x/%02x(%02x). Now it will handle both cases.

Tested-by: Ivar Henckel <ivar.henckel@sony.com>
Reviewed By: fmayer

Differential Revision: https://reviews.llvm.org/D141906

18 months ago[libTooling] Rename `getRangeForEdit` as `getFileRangeForEdit`
Eric Li [Thu, 12 Jan 2023 22:29:17 +0000 (17:29 -0500)]
[libTooling] Rename `getRangeForEdit` as `getFileRangeForEdit`

With the addition of `getFileRange`, we rename `getRangeForEdit` as
`getFileRangeForEdit` for consistency in the API.

Depends on D141634

Differential Revision: https://reviews.llvm.org/D141636

18 months ago[InstSimplify] reduce "mul nsw i1" to "false"
Sanjay Patel [Wed, 18 Jan 2023 18:35:12 +0000 (13:35 -0500)]
[InstSimplify] reduce "mul nsw i1" to "false"

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

Follow-up for:
68c197f07eeae71

18 months ago[OpenMP] Disable building `libomptarget` on 32-bit systems
Shilei Tian [Wed, 18 Jan 2023 18:39:58 +0000 (13:39 -0500)]
[OpenMP] Disable building `libomptarget` on 32-bit systems

There are plenty of assumptions in `libomptarget` and the device runtime
about the pointer size or `size_t`, etc. 32-bit systems are not supported. There
is no point to refine whole things to make it portable. This patch simply disables
building on 32-bit systems.

Fix #60121.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D142023

18 months ago[InstSimplify] add no-wrap parameters to simplifyMul and add more tests; NFC
Sanjay Patel [Wed, 18 Jan 2023 18:21:32 +0000 (13:21 -0500)]
[InstSimplify] add no-wrap parameters to simplifyMul and add more tests; NFC

This gives mul the same capabilities as add/sub.
A potential improvement with nsw was noted in:
1720ec6da040729f17

18 months ago[clang][Darwin] Try to guess the SDK root with xcrun when unspecified
Caleb Zulawski [Wed, 18 Jan 2023 18:14:15 +0000 (18:14 +0000)]
[clang][Darwin] Try to guess the SDK root with xcrun when unspecified

Picking up from https://reviews.llvm.org/D109460

Reviewed By: ldionne

Differential Revision: https://reviews.llvm.org/D136315

18 months ago[lldb][Language] List supported languages in expr error text
Michael Buch [Wed, 18 Jan 2023 16:42:20 +0000 (16:42 +0000)]
[lldb][Language] List supported languages in expr error text

Before:
```
(lldb) expr --language abc -- 1 + 1
error: unknown language type: 'abc' for expression
```

After:
```
(lldb) expr --language abc -- 1 + 1
error: unknown language type: 'abc' for expression. List of supported languages:
  c++
  objective-c++
  c++03
  c++11
  c++14
  objc++
```

We choose to only list the languages which `expr` will actually
accept instead of all the language constants defined in `Language.cpp`
since that's what the user will most likely need.

Differential Revision: https://reviews.llvm.org/D142034

18 months ago[NFC] Release Note for D138995
Tobias Hieta [Wed, 18 Jan 2023 18:17:25 +0000 (19:17 +0100)]
[NFC] Release Note for D138995

18 months ago[OpenMP][libomp] Switch Intel topology type values: module, tile
Jonathan Peyton [Wed, 18 Jan 2023 18:05:01 +0000 (12:05 -0600)]
[OpenMP][libomp] Switch Intel topology type values: module, tile

According to Software Developer Manual, modules should be value 3 and
tile should be value 4.

18 months agoReduce input arguments for `getPackingInfoFromConsumer` (NFC)
Lorenzo Chelini [Thu, 12 Jan 2023 14:22:59 +0000 (15:22 +0100)]
Reduce input arguments for `getPackingInfoFromConsumer` (NFC)

Pass to `getPackingInfoFromConsumer` the tensor.pack op instead of all
the arguments derived from it. Additionally, remove the padding from the
struct, as we currently don't handle propagation when `tensor.pack`
requires padding. We will add back the field when we will need it.

Reviewed By: hanchung

Differential Revision: https://reviews.llvm.org/D141837

18 months ago[OpenMP][libomp] Fix stats-gathering for new MSVC sections API
Jonathan Peyton [Mon, 12 Dec 2022 19:25:51 +0000 (13:25 -0600)]
[OpenMP][libomp] Fix stats-gathering for new MSVC sections API

Differential Revision: https://reviews.llvm.org/D139867

18 months ago[flang] Do not perform INTERNAL_CHECK for deallocation of unlimited polymorphic with...
Valentin Clement [Wed, 18 Jan 2023 17:52:44 +0000 (18:52 +0100)]
[flang] Do not perform INTERNAL_CHECK for deallocation of unlimited polymorphic with intrinsic type spec

When an unlimited polymorphic descriptor is establish for an intrinsic
type spec, the `PointerNullifyIntrinsic` or `AllocatableInitIntrinsic` runtime
function is called. These functions do not initialize an addendum with a derivedType.
When the deallocation on this descriptor is performed, the runtime should not
crash if the addendum is not present. This patch updates `PointerDeallocatePolymorphic`
and `AllocatableDeallocatePolymorphic` for this use case.

Depends on D141996

Reviewed By: jeanPerier, PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D142010

18 months ago[flang] Allocate polymorphic with SOURCE= using ApplyMold
Valentin Clement [Wed, 18 Jan 2023 17:51:18 +0000 (18:51 +0100)]
[flang] Allocate polymorphic with SOURCE= using ApplyMold

Source allocation is similar to mold allocation + assignment. Use
ApplyMold runtime entry point for polymorphic source allocation.
It could be generalized for other source allocation.

Reviewed By: jeanPerier, PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D141996

18 months ago[Clang] Update the help message for `--offload-arch`
Joseph Huber [Wed, 18 Jan 2023 17:40:40 +0000 (11:40 -0600)]
[Clang] Update the help message for `--offload-arch`

Summary:
This works for `OpenMP` and supports the `native` option. It should be
better documented.

18 months ago[clang][Interp][NFC] Simplify InterpFrame constructor
Timm Bäder [Wed, 12 Oct 2022 07:07:27 +0000 (09:07 +0200)]
[clang][Interp][NFC] Simplify InterpFrame constructor

Try to decrease the block depth here a bit.

18 months ago[mlir][tosa] Prefer tosa.transpose composition canonicalization to reshape
Rob Suderman [Wed, 18 Jan 2023 16:59:02 +0000 (08:59 -0800)]
[mlir][tosa] Prefer tosa.transpose composition canonicalization to reshape

It is preferred to merge tosa.transpose operations together rather than convert
one to a tosa.reshape. This is to leverage the tosa.transpose -> tosa.transpose
merging canonicalization.

Reviewed By: AviadCo

Differential Revision: https://reviews.llvm.org/D141434

18 months ago[mlir][tosa] Set explicit benefit for tosa.reshape to linalg
Rob Suderman [Wed, 18 Jan 2023 16:37:05 +0000 (08:37 -0800)]
[mlir][tosa] Set explicit benefit for tosa.reshape to linalg

The patterns used to lower tosa.reshape to linalg are order
dependent which varies depending on platform. Setting the
benefit appropriately guarantees compilation executes the
rewriters in the correct ordering.

Reviewed By: benvanik

Differential Revision: https://reviews.llvm.org/D141981

18 months ago[Libomptarget] Only build GPU tests if a GPU is found on the system
Joseph Huber [Wed, 18 Jan 2023 14:08:06 +0000 (08:08 -0600)]
[Libomptarget] Only build GPU tests if a GPU is found on the system

    Currently we build tests as long as the libraries are found on the
    machine. This doesn't necessarily mean there is a GPU to use though.
    This patch changes it to where we only will build the tests if we found
    a compatible GPU via `nvptx-arch` or `amdgpu-arch`.

    The only downside to this I could see if someone were to build LLVM on a
    home node of a cluster and then wished to run the tests after switching
    to a compute node. For this I think we should allow it to be overridden.
    I think that's better than allowing us to run tests that will fail by
    default.

Reviewed By: tianshilei1992

Differential Revision: https://reviews.llvm.org/D142018

18 months ago[libc++][format] Adds new test macros.
Mark de Wever [Sat, 17 Dec 2022 11:58:40 +0000 (12:58 +0100)]
[libc++][format] Adds new test macros.

These macros make it easier to log additional information. This is
useful for formatting tests. It also properly disables additional
information when locales are disabled in libc++.

Reviewed By: #libc, ldionne

Differential Revision: https://reviews.llvm.org/D140651

18 months ago[clang][Interp] Track initialization state of local variables
Timm Bäder [Mon, 26 Dec 2022 08:29:04 +0000 (09:29 +0100)]
[clang][Interp] Track initialization state of local variables

Use an InlineDescriptor per local variable to track whether locals
have been initialized or not. This way we can support uninitialized
local variables in constexpr functions.

Differential Revision: https://reviews.llvm.org/D135750

18 months ago[DAGCombiner] `combineShuffleOfSplatVal()`: don't assert that shuffle is non-undef
Roman Lebedev [Wed, 18 Jan 2023 15:41:22 +0000 (18:41 +0300)]
[DAGCombiner] `combineShuffleOfSplatVal()`: don't assert that shuffle is non-undef

As per the test case from Steven Johnson in https://reviews.llvm.org/rGf8d9097168b7#1165311
we can indeed encounter such shuffles, that produce all-undef after folding,
before something else manages to optimize them away.

18 months ago[Assignment Tracking] Replace metadata number with variable capture in tests
OCHyams [Wed, 18 Jan 2023 15:33:46 +0000 (15:33 +0000)]
[Assignment Tracking] Replace metadata number with variable capture in tests

18 months ago[flang] Add implementation of move_alloc to the runtime
David Truby [Mon, 9 Jan 2023 15:13:58 +0000 (15:13 +0000)]
[flang] Add implementation of move_alloc to the runtime

This patch adds a move_alloc implementation to the flang runtime.
Most of the checks required by the standard for move_alloc are
done by semenatic analysis; these checks are not replicated here.

Differential Revision: https://reviews.llvm.org/D141286

18 months ago[InstCombine] restrict no-wrap propagation for i1/i2 to avoid miscompiles
Sanjay Patel [Wed, 18 Jan 2023 15:26:18 +0000 (10:26 -0500)]
[InstCombine] restrict no-wrap propagation for i1/i2 to avoid miscompiles

This transform was added with 68c197f07eeae71b9b7,
and the post-commit review noted the potential
for miscompiles at narrow bitwidths.

I'm not sure how to expose the i1 nuw bug because we
already simplify that, but other cases show that
there are missing transforms to add in follow-up
patches.

18 months ago[flang] Move TODO to the source allocation part
Valentin Clement [Wed, 18 Jan 2023 15:31:19 +0000 (16:31 +0100)]
[flang] Move TODO to the source allocation part

18 months ago[InstCombine] reduce code duplication in visitSub(); NFC
Sanjay Patel [Wed, 18 Jan 2023 15:13:20 +0000 (10:13 -0500)]
[InstCombine] reduce code duplication in visitSub(); NFC

18 months ago[InstCombine][InstSimplify] add tests for i1/i2 mul with no-wrap; NFC
Sanjay Patel [Wed, 18 Jan 2023 15:03:36 +0000 (10:03 -0500)]
[InstCombine][InstSimplify] add tests for i1/i2 mul with no-wrap; NFC

A bug was introduced with 68c197f07eeae71 as noted in the
post-commit review comments, and there are potentially
missed smaller transforms/simplifications because no-wrap
multiply with only 1 or 2 bits eliminates some potential
results.

18 months ago[StatepointLowering] Move statepoint correctness checks to Verifier. NFC.
Denis Antrushin [Tue, 17 Jan 2023 11:01:42 +0000 (18:01 +0700)]
[StatepointLowering] Move statepoint correctness checks to Verifier. NFC.

Since D140504, GCStrategy is available for use in opt.
Now we can move statepoint correctness checks from
StatepointLowering.cpp to Verifier.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D141948

18 months ago[VPlan] Disconnect VPRegionBlock from successors in graph iterator(NFCI)
Florian Hahn [Wed, 18 Jan 2023 15:02:40 +0000 (15:02 +0000)]
[VPlan] Disconnect VPRegionBlock from successors in graph iterator(NFCI)

This updates the VPAllSuccessorsIterator to not connect the
VPRegionBlock itself to its successors. The successors are connected to
the exit block of the region. At the moment, this doesn't change any
exisint functionality.

But the new schema ensures the following property when used for
VPDominatorTree:

1. Entry & exit blocks of regions dominate the successors of the region.

This allows for convenient checking of dominance between defs and uses
that are not defined in the same region. I will share a follow-up patch
to use it for the VPDominatorTree soon.

Depends on D140500.

Reviewed By: Ayal

Differential Revision: https://reviews.llvm.org/D140511

18 months ago[mlir][vector] Limit vector.mask op to single vector result
Matthias Springer [Wed, 18 Jan 2023 14:55:13 +0000 (15:55 +0100)]
[mlir][vector] Limit vector.mask op to single vector result

Multiple vector results require multiple masks and passthru values.

Depends On: D141683

Differential Revision: https://reviews.llvm.org/D141786

18 months ago[Regex] Check two chars in step back optimization (NFC)
Nikita Popov [Wed, 18 Jan 2023 14:21:27 +0000 (15:21 +0100)]
[Regex] Check two chars in step back optimization (NFC)

When stepping back, and there is a following fixed character, also
try to check whether another following fixed character matches.

For our tests the next fixed character is often " ", which occurs
pretty frequently, so checking a second character is worthwhile in
practice.

This drops FileCheck runtime for the vloxseg.c test from 25s to
17s for me.

18 months agoRevert "[ConstraintElim] Enable pass by default."
Florian Hahn [Wed, 18 Jan 2023 14:25:00 +0000 (14:25 +0000)]
Revert "[ConstraintElim] Enable pass by default."

This reverts commit fb13dcf3431cd83911fe56899d2fade808dc5b8d.

A large compile-time regression for code generated by sanitizers has
been reported. Revert while I investigate the issue. Details and
reproducers are available here: https://reviews.llvm.org/D135915

18 months ago[VPlan] Replace VPInterleaveRecipe::classof with VP_CLASSOF_IMPL. (NFC)
Florian Hahn [Wed, 18 Jan 2023 14:23:22 +0000 (14:23 +0000)]
[VPlan] Replace VPInterleaveRecipe::classof with VP_CLASSOF_IMPL. (NFC)

18 months ago[mlir][linalg][bufferize][NFC] Utilize DestinationStyleOpInterface helper
Matthias Springer [Wed, 18 Jan 2023 14:19:23 +0000 (15:19 +0100)]
[mlir][linalg][bufferize][NFC] Utilize DestinationStyleOpInterface helper

`DstBufferizableOpInterfaceExternalModel` provides suitable default implementations for Linalg ops.

Differential Revision: https://reviews.llvm.org/D141921

18 months agoHarden -slice-analysis-test MLIR testing pass to check for some invariants
Mehdi Amini [Wed, 18 Jan 2023 14:14:33 +0000 (14:14 +0000)]
Harden -slice-analysis-test MLIR testing pass to check for some invariants

The pass would assert on some input IR that it didn't expect. Instead
signal a pass failure and properly interupt the flow.

Fixes #60022

18 months ago[mlir] GreedyPatternRewriteDriver: Fix termination criteria in OpPatternRewriteDriver
Matthias Springer [Wed, 18 Jan 2023 14:10:14 +0000 (15:10 +0100)]
[mlir] GreedyPatternRewriteDriver: Fix termination criteria in OpPatternRewriteDriver

This driver should iterate until convergence or until the specified op was erased. However, it used to stop when any op was erased.

Differential Revision: https://reviews.llvm.org/D141921

18 months ago[llvm-objdump] Fix reporting error for processing target features
Elena Lepilkina [Wed, 18 Jan 2023 11:09:03 +0000 (14:09 +0300)]
[llvm-objdump] Fix reporting error for processing target features

Differential Revision: https://reviews.llvm.org/D142003

18 months agoDiagnose extensions in 'offsetof'
Aaron Ballman [Wed, 18 Jan 2023 13:49:45 +0000 (08:49 -0500)]
Diagnose extensions in 'offsetof'

https://www.open-std.org/jtc1/sc22/wg14/www/docs/n2350.htm made very
clear that it is an UB having type definitions with in offsetof.
Clang supports defining a type as the first argument as a conforming
extension due to how many projects use the construct in C99 and earlier
to calculate the alignment of a type. GCC also supports defining a type
as the first argument.

This adds extension warnings and documentation for the functionality
Clang explicitly supports.

Fixes #57065
Reverts the revert of 39da55e8f548a11f7dadefa73ea73d809a5f1729

Co-authored-by: Yingchi Long <i@lyc.dev>
Co-authored-by: Aaron Ballman <aaron@aaronballman.com>
Differential Revision: https://reviews.llvm.org/D133574

18 months ago[flang] Add TODO for allocate array with scalar source
Peixin Qiao [Wed, 18 Jan 2023 13:49:22 +0000 (21:49 +0800)]
[flang] Add TODO for allocate array with scalar source

This scenario is not supported yet.

Reviewed By: clementval, jeanPerier, sscalpone

Differential Revision: https://reviews.llvm.org/D141932

18 months ago[NFC][WebAssembly] Add tests
Samuel Parker [Wed, 18 Jan 2023 13:30:53 +0000 (13:30 +0000)]
[NFC][WebAssembly] Add tests

Add more variations to fpclamptosat.

18 months ago[InstCombine] add tests for diff-of-squares; NFC
Sanjay Patel [Tue, 17 Jan 2023 20:21:14 +0000 (15:21 -0500)]
[InstCombine] add tests for diff-of-squares; NFC

Ideally, these negative tests would have been added
before 68c197f07eea to make sure the specific
operand matching is working as expected, but I did
not remember to include them.

18 months ago[InstCombine] simplify code for matching shift-logic-shift pattern; NFC
Sanjay Patel [Tue, 17 Jan 2023 20:15:48 +0000 (15:15 -0500)]
[InstCombine] simplify code for matching shift-logic-shift pattern; NFC

We can match and capture in one statement. Also, make the
code more closely resemble the description comment by using
the constant name of an operand value.

18 months ago[include-mapping] Fix gen_std.py test
Viktoriia Bakalova [Tue, 17 Jan 2023 16:38:14 +0000 (16:38 +0000)]
[include-mapping] Fix gen_std.py test

Differential Revision: https://reviews.llvm.org/D141944

18 months ago[gn build] Port f10d452db841
LLVM GN Syncbot [Wed, 18 Jan 2023 12:50:04 +0000 (12:50 +0000)]
[gn build] Port f10d452db841

18 months agoReland "[JITLink] Add an initial implementation of JITLink for ELF/LoongArch"
wanglei [Tue, 17 Jan 2023 07:10:00 +0000 (15:10 +0800)]
Reland "[JITLink] Add an initial implementation of JITLink for ELF/LoongArch"

This implementation supports basic relocation types and adds EHFrame,
Got/Plt handling passes.
This patch also enables JIT support for LoongArch64.

With this patch, I successfully run hello.ll and simple_throw.ll
(which is generated from test-suite/SingleSource/Regression/C++/EH/simple_throw.cpp)
using the `lli` command with options `--jit-kind=orc --jit-linker=jitlink`.

Note: `hasJIT` property of LoongArch32 remains false as there is no
validation environment.

New changes: Since LoongArch does not support RuntimeDyld, JITLink is set
by default.  Add a null-terminator to eh-frame sections. This should fix
the test failure on LoongArch bot.
(https://lab.llvm.org/staging/#/builders/236/builds/896)

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D141036

18 months ago[bazel] Fix build after c33b9395b1
Benjamin Kramer [Wed, 18 Jan 2023 12:31:35 +0000 (13:31 +0100)]
[bazel] Fix build after c33b9395b1

18 months ago[BitcodeReader] Allow reading pointer types from old IR
Sebastian Neubauer [Wed, 18 Jan 2023 12:19:59 +0000 (13:19 +0100)]
[BitcodeReader] Allow reading pointer types from old IR

When opaque pointers are enabled and old IR with typed pointers is read,
the BitcodeReader automatically upgrades all typed pointers to opaque
pointers. This is a lossy conversion, i.e. when a function argument is a
pointer and unused, it’s impossible to reconstruct the original type
behind the pointer.

There are cases where the type information of pointers is needed. One is
reading DXIL, which is bitcode of old LLVM IR and makes a lot of use of
pointers in function signatures.
We’d like to keep using up-to-date llvm to read in and process DXIL, so
in the face of opaque pointers, we need some way to access the type
information of pointers from the read bitcode.

This patch allows extracting type information by supplying functions to
parseBitcodeFile that get called for each function signature or metadata
value. The function can access the type information via the reader’s
type IDs and the getTypeByID and getContainedTypeID functions.
The tests exemplarily shows how type info from pointers can be stored in
metadata for use after the BitcodeReader finished.

Differential Revision: https://reviews.llvm.org/D127728

18 months ago[VPlan] Replace VPBranchOnMaskSC::classof with VP_CLASSOF_IMPL. (NFC)
Florian Hahn [Wed, 18 Jan 2023 12:14:57 +0000 (12:14 +0000)]
[VPlan] Replace VPBranchOnMaskSC::classof with VP_CLASSOF_IMPL. (NFC)

18 months ago[RISCV][ISelLowering] Fix select lowering issue
Dmitry Bushev [Tue, 17 Jan 2023 15:19:38 +0000 (18:19 +0300)]
[RISCV][ISelLowering] Fix select lowering issue

Fix bug that leads to some pseudo instructions not being lowered.

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D141395

18 months agoRemove an unused variable, NFC
Haojian Wu [Wed, 18 Jan 2023 12:08:21 +0000 (13:08 +0100)]
Remove an unused variable, NFC

18 months ago[DAG] combineInsertEltToShuffle - split off mergeInsertEltWithShuffle fold. NFC.
Simon Pilgrim [Wed, 18 Jan 2023 11:21:06 +0000 (11:21 +0000)]
[DAG] combineInsertEltToShuffle - split off mergeInsertEltWithShuffle fold. NFC.

combineInsertEltToShuffle was performing 2 very different folds in the same call, merging "(insert_vector_elt (vector_shuffle X, Y), (extract_vector_elt X, N), IdxC) --> (vector_shuffle X, Y)" and "(insert_vector_elt V, (bitcast X from vector type), IdxC) --> bitcast(shuffle (bitcast V), (extended X), Mask)"

The folds are currently still attempted in the same order as before (just as 2 seperate calls) so there should be no change in behaviour.

First step towards some adjustments to mergeInsertEltWithShuffle for D127115.

18 months ago[DAG][ARM][AArch64] Transform max(a,b) - min(a,b) -> abd(a,b)
David Green [Wed, 18 Jan 2023 11:44:26 +0000 (11:44 +0000)]
[DAG][ARM][AArch64] Transform max(a,b) - min(a,b) -> abd(a,b)

This adds both signed and unsigned transforms for
max(a, b) - min(a, b) -> abd(a, b).

unsigned: https://alive2.llvm.org/ce/z/RF4jGQ
signed: https://alive2.llvm.org/ce/z/Cjr2zE

Fixes: #59894

Differential Revision: https://reviews.llvm.org/D141706

18 months agoReland "nullptr returned from ActOnTag() is not a valid result"
Haojian Wu [Wed, 18 Jan 2023 11:18:26 +0000 (12:18 +0100)]
Reland "nullptr returned from ActOnTag() is not a valid result"

The commit was reverted in 346e1c43a11b8af5a818dac321f83f043862c1ec as
part of the f1f0a0d8e8fdd2e534d9423b2e64c6b8aaa53aee revert.

18 months agoAArch64: allocate small fixed args properly in varargs functions.
Tim Northover [Tue, 10 Jan 2023 11:52:12 +0000 (11:52 +0000)]
AArch64: allocate small fixed args properly in varargs functions.

On Darwin, function arguments occupy their real size when passed on the stack
(e.g. an i16 only consumes 2 bytes). This means that, even for fixed args in
varargs calls we need to keep track of the original type being passed before
any DAG/GISel promotions. Existing logic only applied this fix to the
non-varargs case leading to mismatch between caller & callee in those
situations.

On Linux & Windows these arguments always occupy a 64-bit slot anyway so
there's no special handling needed.

18 months agoReapply [AArch64] fold subs ugt/ult to ands when the second operand is mask/pow2
chenglin.bi [Wed, 18 Jan 2023 11:23:58 +0000 (19:23 +0800)]
Reapply [AArch64] fold subs ugt/ult to ands when the second operand is mask/pow2

Origianl patch made a mistake that ugt is reverse cc should be ule.
And ule < C will be generalize to ult < C + 1. So the new patch add support for ult < Pow2 case.

https://alive2.llvm.org/ce/z/naBw5A

Reviewed By: samtebbs, chapuni

Differential Revision: https://reviews.llvm.org/D141829

18 months ago[libc++][ranges] Remove a leftover include version control markers in Cxx20Papers.csv
Haojian Wu [Wed, 18 Jan 2023 11:04:38 +0000 (12:04 +0100)]
[libc++][ranges] Remove a leftover include version control markers in Cxx20Papers.csv

18 months ago[RISCV] Generalize performFP_TO_INTCombine to vectors
Luke Lau [Thu, 12 Jan 2023 12:20:07 +0000 (12:20 +0000)]
[RISCV] Generalize performFP_TO_INTCombine to vectors

Like in the scalar domain, combine calls to (fp_to_int (ftrunc X)) on
scalable and fixed-length vectors into a single vfcvt instruction.
For truncating rounds, the static vfcvt.rtz rounding mode is used.
Otherwise use the VFCVT_RM_ variants to set the rounding mode
dynamically.
Closes https://github.com/llvm/llvm-project/issues/56737

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D141599

18 months ago[RISCV][NFC] Add test cases for rounding vfcvt
Luke Lau [Thu, 12 Jan 2023 12:14:44 +0000 (12:14 +0000)]
[RISCV][NFC] Add test cases for rounding vfcvt

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D141600

18 months ago[LLVMContext] Remove hasSetOpaquePointersValue() API (NFC)
Nikita Popov [Wed, 18 Jan 2023 10:04:12 +0000 (11:04 +0100)]
[LLVMContext] Remove hasSetOpaquePointersValue() API (NFC)

Now that we no longer perform any kind of typed pointer guessing,
we no longer need this API.

18 months ago[DAG][AArch64][ARM] Combine abd(sub(x, y)) to abd if the sub is nsw
David Green [Wed, 18 Jan 2023 10:10:52 +0000 (10:10 +0000)]
[DAG][AArch64][ARM] Combine abd(sub(x, y)) to abd if the sub is nsw

This implements the fold (abs (sub nsw x, y)) -> abds(x, y). Providing
the sub is nsw this appears to be valid without the extensions that are
usually used for abds. https://alive2.llvm.org/ce/z/XHVaB3. The
equivalent abdu combine seems to not be valid.

Differential Revision: https://reviews.llvm.org/D141665

18 months ago[LTO] Always use opaque pointers mode from config
Nikita Popov [Wed, 18 Jan 2023 10:01:16 +0000 (11:01 +0100)]
[LTO] Always use opaque pointers mode from config

Always use the configured opaque pointers mode, don't check whether
it has already been set through a different means.

Removes the last user of the hasSetOpaquePointersValue() API.

18 months ago[GlobalISel] Detect splats built with G_CONCAT_VECTORS
Diana Picus [Tue, 17 Jan 2023 08:08:34 +0000 (09:08 +0100)]
[GlobalISel] Detect splats built with G_CONCAT_VECTORS

Add support to the MI matching of vector splats for patterns that
consist of `G_CONCAT_VECTORS` of smaller splats with the same constant
value. With this, we would consider the following pseudo-MIR to be a splat:

```
%0 = G_[F]CONSTANT [...]
%1 = G_BUILD_VECTOR %0, %0, ..., %0
%2 = G_CONCAT_VECTORS %1, %1, ..., %1
```

Since it uses recursion for matching splats, it could match pretty
complicated patterns with all sorts of combinations of `G_BUILD_VECTOR`
and `G_CONCAT_VECTORS` (e.g. a `G_CONCAT_VECTORS` with
a `G_BUILD_VECTOR_TRUNC` and another `G_CONCAT_VECTORS` as operands),
and it should also look through copies etc.

This should make it easier to match complex immediates for certain
instructions on AMDGPU, where for instance a <8 x s16> will be split
before instruction selection into a `G_CONCAT_VECTORS` of <2 x s16>
splats.

Differential Revision: https://reviews.llvm.org/D141902

18 months agoFix typo. NFC
Diana Picus [Mon, 16 Jan 2023 14:03:12 +0000 (15:03 +0100)]
Fix typo. NFC

18 months ago[mlir][tblgen] Emit deprecation warning if `kEmitRawAttributes` is used
Markus Böck [Thu, 12 Jan 2023 17:07:57 +0000 (18:07 +0100)]
[mlir][tblgen] Emit deprecation warning if `kEmitRawAttributes` is used

As discussed in https://reviews.llvm.org/D140886, emitting a warning if the old API is used may be beneficial to encourage migration to the new fold API.
This reuse the existing `Deprecated` infrastructure within TableGen, and simply marks the `def` for `kEmitRawAttributesFolder` causing a use of it in a record (even if set within a base class) to emit a warning.

Error message as printed in the terminal:
```
Included from C:/llvm-project/mlir/python/mlir/dialects/TensorOps.td:13:
Included from C:/llvm-project/mlir/include\mlir/Dialect/Tensor/IR/TensorOps.td:12:
C:/llvm-project/mlir/include\mlir/Dialect/Tensor/IR/TensorBase.td:14:5: warning: Using deprecated def `kEmitRawAttributesFolder`
def Tensor_Dialect : Dialect {
    ^
note: 'useFoldAPI' of 'kEmitRawAttributesFolder' (default) has been deprecated and is pending removal. Please switch to 'kEmitFoldAdaptorFolder'. See https://discourse.llvm.org/t/psa-new-improved-fold-method-signature-has-landed-please-update-your-downstream-projects/67618
```

Differential Revision: https://reviews.llvm.org/D141604

18 months ago[gn build] Port 48c2172e51b2
LLVM GN Syncbot [Wed, 18 Jan 2023 09:05:51 +0000 (09:05 +0000)]
[gn build] Port 48c2172e51b2

18 months ago[CSKY] Fix errors caused by change of compressInst
wangpc [Wed, 18 Jan 2023 08:49:32 +0000 (16:49 +0800)]
[CSKY] Fix errors caused by change of compressInst

We don't need MCContext since D141951.

Reviewed By: zixuan-wu

Differential Revision: https://reviews.llvm.org/D141995

18 months ago[Release] Produce mlir tarball
Nikita Popov [Tue, 17 Jan 2023 11:59:54 +0000 (12:59 +0100)]
[Release] Produce mlir tarball

MLIR supports standalone builds, so I think it makes sense to also
produce a release tarball for the MLIR subproject.

Differential Revision: https://reviews.llvm.org/D141919

18 months ago[AsmParser] Remove typed pointer auto-detection
Nikita Popov [Wed, 14 Dec 2022 12:41:10 +0000 (13:41 +0100)]
[AsmParser] Remove typed pointer auto-detection

IR is now always parsed in opaque pointer mode, unless
-opaque-pointers=0 is explicitly given. There is no automatic
detection of typed pointers anymore.

The -opaque-pointers=0 option is added to any remaining IR tests
that haven't been migrated yet.

Differential Revision: https://reviews.llvm.org/D141912

18 months agoRevert "[JITLink] Add an initial implementation of JITLink for ELF/LoongArch"
wanglei [Wed, 18 Jan 2023 08:50:32 +0000 (16:50 +0800)]
Revert "[JITLink] Add an initial implementation of JITLink for ELF/LoongArch"

This reverts commit 28b6f38d0a841451af7b4893368543382dd5e630.

This breaks on LoongArch64 bot.
https://lab.llvm.org/staging/#/builders/236/builds/896

18 months agoRevert "[BitcodeReader] Allow reading pointer types from old IR"
Nikita Popov [Wed, 18 Jan 2023 08:48:31 +0000 (09:48 +0100)]
Revert "[BitcodeReader] Allow reading pointer types from old IR"

This reverts commit b56df190b01335506ce30a4559d880da76d1a181.

The unit tests are implemented in a way that requires support for
writing typed pointer bitcode, which is going away soon. Please
rewrite it in a way that not have requirement, e.g. by shipping
pre-compiled bitcode, as we do for integration tests.

18 months ago[GVN][NFC] Add pre-commit tests for non-local load cases (D141680)
Sergey Kachkov [Mon, 16 Jan 2023 14:43:35 +0000 (17:43 +0300)]
[GVN][NFC] Add pre-commit tests for non-local load cases (D141680)

Add test cases when load is non-local for select dependency, but can be
found in extended BB (chain of blocks with single predecessor). Check
that type of found load is the same.

18 months ago[CodeGen] Prevent overlapping subregs in getCoveringSubRegIndexes
Pierre van Houtryve [Thu, 12 Jan 2023 08:52:33 +0000 (03:52 -0500)]
[CodeGen] Prevent overlapping subregs in getCoveringSubRegIndexes

If `getCoveringSubRegIndexes` returns a set of subregister indexes where some subregisters overlap others, it can create unsatisfiable copy bundles that eventually cause VirtRegRewriter to error out due to "cycles in copy bundle".

We can simply prevent this by making the algorithm skip over subregisters indexes that would cause an overlap with already-covered lanes.

Note that in the case of AMDGPU, this problem is caused by the lack of subregisters indexes for 13/14/15-register tuples. We have everything up until 12, then we have 16 and 32 but nothing between 12 and 16.
This means that the best candidate to do the least amount of copies when splitting a 29-register tuple was to copy (e.g.) 0-15 and 14-29, causing an overlap.
With this change, getCoveringSubRegIndexes will now prefer using something like 0-15, 16-28 and 1

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D141576

18 months ago[AMDGPU] Precommit test for D141576
Pierre van Houtryve [Tue, 17 Jan 2023 08:50:29 +0000 (03:50 -0500)]
[AMDGPU] Precommit test for D141576

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D141903

18 months ago[LowerIFunc] Remove typed pointer test (NFC)
Nikita Popov [Wed, 18 Jan 2023 08:45:25 +0000 (09:45 +0100)]
[LowerIFunc] Remove typed pointer test (NFC)

18 months ago[HWASAN] Add leak sanitizer flag support
Kirill Stoimenov [Wed, 18 Jan 2023 07:50:27 +0000 (23:50 -0800)]
[HWASAN] Add leak sanitizer flag support

Reviewed By: vitalybuka

Differential Revision: https://reviews.llvm.org/D141710

18 months ago[HWASA] Restore LsanMetadata from D141642 Diff5
Vitaly Buka [Wed, 18 Jan 2023 07:40:30 +0000 (23:40 -0800)]
[HWASA] Restore LsanMetadata from D141642 Diff5

Unlike asan, we can't use pointer arithmetics to get from user ptr to
metadata. Asan does not use CombinedAllocator::GetMetadata and store
metadata next to the user data.