platform/upstream/llvm.git
2 years ago[LoongArch] Remove useless empty pattern fields in some Pseudo instructions. NFC
wanglei [Tue, 1 Nov 2022 12:44:54 +0000 (20:44 +0800)]
[LoongArch] Remove useless empty pattern fields in some Pseudo instructions. NFC

2 years ago[sanitizer] Fix build error with current LoongArch Clang
Youling Tang [Tue, 1 Nov 2022 12:48:45 +0000 (20:48 +0800)]
[sanitizer] Fix build error with current LoongArch Clang

Fix the following build failures:
```
In file included from /home/loongson/llvm-work/llvm-project/compiler-rt/lib/sanitizer_common/sanit>
llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_syscall_linux_loongarch64.inc:27:23: error>
  register u64 a7 asm("a7") = nr;
                      ^
llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_syscall_linux_loongarch64.inc:28:23: error>
  register u64 a0 asm("a0");                      ^
```

The non-prefix `$` syntax in inline assembly is not supported in clang
yet (it is supported by gcc), add prefix `$` to solve the problem.

Differential Revision: https://reviews.llvm.org/D137145

2 years ago[asan][test] Fix build errors in loongarch64
Youling Tang [Tue, 1 Nov 2022 12:19:33 +0000 (20:19 +0800)]
[asan][test] Fix build errors in loongarch64

- Add get_bits_for_arch for loongarch64 detection, fix the following error,
```
$ make check-asan
-- sanitizer_common tests on "Linux" will run against "asan"
CMake Error at test/asan/CMakeLists.txt:22 (message):
  Unknown target architecture: loongarch64
Call Stack (most recent call first):
  test/asan/CMakeLists.txt:49 (get_bits_for_arch)
```

- Do not use __builtin_longjmp, fix the following error,
```
llvm-project/compiler-rt/lib/asan/tests/asan_test.cpp:635:3: error: __builtin_longjmp is not supported for the current target
  __builtin_longjmp((void**)buf, 1);
```

Differential Revision: https://reviews.llvm.org/D137012

2 years ago[builtins][LoongArch] Port __clear_cache to LoongArch Linux
Youling Tang [Tue, 1 Nov 2022 12:14:06 +0000 (20:14 +0800)]
[builtins][LoongArch] Port __clear_cache to LoongArch Linux

There are two failures in the current builtins,
Failed Tests (2):
  Builtins-loongarch64-linux :: clear_cache_test.c
  Builtins-loongarch64-linux :: enable_execute_stack_test.c

It is caused by __clear_cache not being implemented and triggering `abort`.

"__clear_cache" is implemented in the same way as "clear_cache" in gcc (
in gcc/config/loongarch/loongarch.md)

Differential Revision: https://reviews.llvm.org/D136921

2 years ago[compiler-rt][builtins] Support builtins for LoongArch
Youling Tang [Tue, 1 Nov 2022 12:08:06 +0000 (20:08 +0800)]
[compiler-rt][builtins] Support builtins for LoongArch

Initial builtins for LoongArch.
Add loongarch64 to ALL_CRT_SUPPORTED_ARCH list.
Support fe_getround and fe_raise_inexact in builtins.

Differential Revision: https://reviews.llvm.org/D136338

2 years ago[X86] combineConcatVectorOps - fold 512-bit concat(GF2P8AFFINEQB(x,y,c),GF2P8AFFINEQB...
Simon Pilgrim [Tue, 1 Nov 2022 12:06:34 +0000 (12:06 +0000)]
[X86] combineConcatVectorOps - fold 512-bit concat(GF2P8AFFINEQB(x,y,c),GF2P8AFFINEQB(z,w,c)) -> GF2P8AFFINEQB(concat(x,z),concat(y,w),c)

Now that D137036 has landed, we just need AVX512F support to generate 512-bit GF2P8AFFINEQB ops

2 years ago[GlobalISel] Compute debug location when merging stores more accurately
Anton Sidorenko [Fri, 28 Oct 2022 09:19:04 +0000 (12:19 +0300)]
[GlobalISel] Compute debug location when merging stores more accurately

Originaly the loop did almost nothing as the calculated location was
overwritten on the next iteration.

Differential Revision: https://reviews.llvm.org/D136937

2 years ago[AMDGPU] Fix RP tracker's live registers after processing a memory clause.
Valery Pykhtin [Mon, 31 Oct 2022 13:12:35 +0000 (14:12 +0100)]
[AMDGPU] Fix RP tracker's live registers after processing a memory clause.

It's incorrect to reuse live registers left from the first instruction in a clause after the clause as they don't contain in-clause defs.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D137081

2 years ago[GlobalISel] Precommit test for D136937
Anton Sidorenko [Mon, 31 Oct 2022 10:13:12 +0000 (13:13 +0300)]
[GlobalISel] Precommit test for D136937

2 years ago[AArch64][SVE2] Add the SVE2.1 contiguous stores to multiple consecutive vectors
David Sherwood [Thu, 20 Oct 2022 12:27:16 +0000 (12:27 +0000)]
[AArch64][SVE2] Add the SVE2.1 contiguous stores to multiple consecutive vectors

This patch adds the assembly/disassembly for the following instructions:

st1*   : Contiguous store of bytes to multiple consecutive vectors -
 (scalar + scalar) and (scalar + immediate)
stnt1* : Contiguous store non-temporal of bytes to multiple consecutive
 vectors - (scalar + scalar) and (scalar + immediate)

The reference can be found here:
https://developer.arm.com/documentation/ddi0602/2022-09

Differential Revision: https://reviews.llvm.org/D136686

2 years ago[AArch64][SVE2] Add the SVE2.1 contiguous load to multiple consecutive vector
David Sherwood [Wed, 19 Oct 2022 14:27:12 +0000 (14:27 +0000)]
[AArch64][SVE2] Add the SVE2.1 contiguous load to multiple consecutive vector

This patch adds the assembly/disassembly for the following instructions:

ld1*   : Contiguous load of bytes to multiple consecutive vectors -
         (scalar + scalar) and (scalar + immediate)
ldnt1* : Contiguous load non-temporal of bytes to multiple consecutive
         vectors - (scalar + scalar) and (scalar + immediate)

The reference can be found here:
https://developer.arm.com/documentation/ddi0602/2022-09

Differential Revision: https://reviews.llvm.org/D136680

2 years ago[JITLink] Revert template typename change after 2a3b257a935a (NFC)
Stefan Gränitz [Tue, 1 Nov 2022 09:13:43 +0000 (10:13 +0100)]
[JITLink] Revert template typename change after 2a3b257a935a (NFC)

As pointed out in the review https://reviews.llvm.org/D135523#C3696894NL515

2 years ago[LLDB][LoongArch] Add LoongArch ArchSpec and subtype detection
Tiezhu Yang [Tue, 1 Nov 2022 04:56:55 +0000 (12:56 +0800)]
[LLDB][LoongArch] Add LoongArch ArchSpec and subtype detection

Define LoongArch architecture subtypes, add the LoongArch ArchSpec bits,
and inspect the ELF header to detect the right subtype based on ELF class.

Here is a simple test:

```
[loongson@linux ~]$ cat hello.c

int main()
{
printf("Hello, World!\n");
return 0;
}
[loongson@linux ~]$ clang hello.c -g -o hello
```

Without this patch:

```
[loongson@linux ~]$ llvm-project/llvm/build/bin/lldb hello
(lldb) target create "hello"
error: '/home/loongson/hello' doesn't contain any 'host' platform architectures: unknown
```

With this patch:

```
[loongson@linux ~]$ llvm-project/llvm/build/bin/lldb hello
(lldb) target create "hello"
Current executable set to '/home/loongson/hello' (loongarch64).
(lldb) run
Process 735167 launched: '/home/loongson/hello' (loongarch64)
Hello, World!
Process 735167 exited with status = 0 (0x00000000)
(lldb) quit
[loongson@linux ~]$ llvm-project/llvm/build/bin/llvm-lit llvm-project/lldb/test/Shell/ObjectFile/ELF/loongarch-arch.yaml
llvm-lit: /home/loongson/llvm-project/llvm/utils/lit/lit/llvm/config.py:456: note: using clang: /home/loongson/llvm-project/llvm/build/bin/clang
-- Testing: 1 tests, 1 workers --
PASS: lldb-shell :: ObjectFile/ELF/loongarch-arch.yaml (1 of 1)

Testing Time: 0.09s
  Passed: 1
```

Reviewed By: SixWeining, xen0n, DavidSpickett

Differential Revision: https://reviews.llvm.org/D137057

2 years ago[IR] Take operand bundles into account for call argument readonly/writeonly
Nikita Popov [Thu, 27 Oct 2022 10:45:08 +0000 (12:45 +0200)]
[IR] Take operand bundles into account for call argument readonly/writeonly

We currently only take operand bundle effects into account when
querying the function-level memory attributes. However, I believe
that we also need to do the same for parameter attributes. For
example, a call with deopt bundle to a function with readnone
parameter attribute cannot treat that parameter as readnone,
because the deopt bundle may read it.

Differential Revision: https://reviews.llvm.org/D136834

2 years agoAMDGPU: Add testcase from issue 58639
Matt Arsenault [Tue, 1 Nov 2022 07:40:46 +0000 (00:40 -0700)]
AMDGPU: Add testcase from issue 58639

This was fixed in bf789b1957efd2482e1dbd164d91a6612a450fe3

2 years ago[VP][RISCV] Add vp.rint and RISC-V support.
Yeting Kuo [Mon, 24 Oct 2022 15:40:48 +0000 (23:40 +0800)]
[VP][RISCV] Add vp.rint and RISC-V support.

FRINT uses dynamic rounding mode instead of static rounding mode. The patch
rename VFCVT_X_F_VL to VFCVT_RM_X_F_VL for static rounding mode uses and added
new ISDNode VFCVT_X_F_VL directly selected to PseudoVFCVT_X_F_V.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D136662

2 years agoAMDGPU/GlobalISel: Add missing run line for gfx7 packed operations
Matt Arsenault [Sat, 29 Oct 2022 18:32:04 +0000 (11:32 -0700)]
AMDGPU/GlobalISel: Add missing run line for gfx7 packed operations

We were only testing targets with legal 16-bit operations.

2 years agoAtomicExpand: Use InstSimplifyFolder
Matt Arsenault [Tue, 20 Sep 2022 22:17:25 +0000 (18:17 -0400)]
AtomicExpand: Use InstSimplifyFolder

Automatically cleanup operations if we know the atomic has higher
alignment.

2 years ago[lit] Add pyproject.toml and fix build w/ modern setuptools backend
Michał Górny [Fri, 28 Oct 2022 20:13:35 +0000 (22:13 +0200)]
[lit] Add pyproject.toml and fix build w/ modern setuptools backend

Add a `pyproject.toml` file that provides build system information
for PEP 517-compliant builders.  While all the commonly used builders
provide fallback to running `setup.py` for backwards compatibility, this
ensures the best forward compatibility.  It also provides a reliable way
of specifying the minimum required setuptools version.  Effectively, it
will make it possible to remove `setup.py` in favor of purely
declarative configuration in the future, or even switch to a different
build system.

Update `setup.py` to explicitly add the current directory to `sys.path`
for importing `lit`.  This is necessary, as the modern setuptools
backend does not guarantee that the current directory is present there.

Differential Revision: https://reviews.llvm.org/D136976

2 years ago[MLIR] Fix bug in the method constructing semi affine expression from flattened form
Arnab Dutta [Mon, 31 Oct 2022 11:49:51 +0000 (17:19 +0530)]
[MLIR] Fix bug in the method constructing semi affine expression from flattened form

Set proper offset to the second element of the index pair, so that
we do not have same index values for more than one local expression.

Reviewed By: springerm

Differential Revision: https://reviews.llvm.org/D137062

2 years ago[lldb][Test] Fix TestFrameFormatNameWithArgs.test on Windows/Linux
Michael Buch [Tue, 1 Nov 2022 05:56:15 +0000 (22:56 -0700)]
[lldb][Test] Fix TestFrameFormatNameWithArgs.test on Windows/Linux

* Windows doesn't support setting these breakpoints by basename
* On Linux std::function arguments aren't formatted as such

2 years ago[IR][Instruction] Annotate more methods with LLVM_READONLY
Alexander Shaposhnikov [Tue, 1 Nov 2022 05:55:28 +0000 (05:55 +0000)]
[IR][Instruction] Annotate more methods with LLVM_READONLY

This diff annotates more methods of the class Instruction with
LLVM_READONLY (similarly to isAssociative/isCommutative).

Test plan: ninja check-all

Differential revision: https://reviews.llvm.org/D137053

2 years ago[SLP] Extend reordering data of tree entry to support PHI nodes
skc7 [Wed, 26 Oct 2022 04:33:13 +0000 (04:33 +0000)]
[SLP] Extend reordering data of tree entry to support PHI nodes

Reviewed By: ABataev

Differential Revision: https://reviews.llvm.org/D136757

2 years agoRevert "[compiler-rt] Switch from llvm-config to find_package(LLVM)"
Michał Górny [Tue, 1 Nov 2022 04:49:24 +0000 (05:49 +0100)]
Revert "[compiler-rt] Switch from llvm-config to find_package(LLVM)"

This reverts commit 37acf9bdd4686397a6d37000e8d3fe35f456a860.  It broke
a buildbot.

2 years ago[AAPointerInfo] refactor how offsets and Access objects are tracked
Sameer Sahasrabuddhe [Tue, 1 Nov 2022 04:26:58 +0000 (09:56 +0530)]
[AAPointerInfo] refactor how offsets and Access objects are tracked

AAPointerInfo now maintains a list of all Access objects that it owns, along
with the following maps:

- OffsetBins: OffsetAndSize -> { Access }
- InstTupleMap: RemoteI x LocalI -> Access

A RemoteI is any instruction that accesses memory. RemoteI is different from
LocalI if and only if LocalI is a call; then RemoteI is some instruction in the
callgraph starting from LocalI.

Motivation: When AAPointerInfo recomputes the offset for an instruction, it sets
the value to Unknown if the new offset is not the same as the old offset. The
instruction must now be moved from its current bin to the bin corresponding to
the new offset. This happens for example, when:

- A PHINode has operands that result in different offsets.
- The same remote inst is reachable from the same local inst via different paths
  in the callgraph:

```
               A (local inst)
               |
               B
              / \
             C1  C2
              \ /
               D (remote inst)

```
This fixes a bug where a store is incorrectly eliminated in a lit test.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D136526

2 years agollvm-reduce: Write bitcode temporary files for bitcode inputs
Matt Arsenault [Sat, 29 Oct 2022 01:50:19 +0000 (18:50 -0700)]
llvm-reduce: Write bitcode temporary files for bitcode inputs

Most tools accept .ll or .bc inputs interchangably, but some don't.
Default to writing temporary files that match the input. This
will also aid reducing deserialization bugs.

2 years agollvm-reduce: Drop guessing output format based on file extension
Matt Arsenault [Sat, 29 Oct 2022 01:24:21 +0000 (18:24 -0700)]
llvm-reduce: Drop guessing output format based on file extension

2 years agollvm-reduce: Support emitting bitcode for final result
Matt Arsenault [Wed, 26 Oct 2022 20:05:46 +0000 (13:05 -0700)]
llvm-reduce: Support emitting bitcode for final result

Previously, this unconditionally emitted text IR. I ran
into a bug that manifested in broken disassembly, so the
desired output was the bitcode format. If the input format
was binary bitcode, the requested output file ends in .bc,
or an explicit -output-bitcode option was used, emit bitcode.

2 years ago[compiler-rt] Switch from llvm-config to find_package(LLVM)
Michał Górny [Sat, 29 Oct 2022 17:03:37 +0000 (19:03 +0200)]
[compiler-rt] Switch from llvm-config to find_package(LLVM)

Replace the use of the deprecated `llvm-config` tool with LLVM's CMake
files for detecting LLVM in standalone builds.

Differential Revision: https://reviews.llvm.org/D137024

2 years agoFix missing C++ mode comment
Matt Arsenault [Sat, 29 Oct 2022 00:17:00 +0000 (17:17 -0700)]
Fix missing C++ mode comment

2 years agollvm-reduce: Report file opening errors
Matt Arsenault [Mon, 24 Oct 2022 16:08:54 +0000 (09:08 -0700)]
llvm-reduce: Report file opening errors

This was also trying to write the bitcode to the failed file
on failure, which asserts. Also, consistently use
ToolOutputFile, instead of one path manually removing
the temp file.

2 years agoAtomicExpand: Don't create unused instructions for some atomicrmw
Matt Arsenault [Thu, 29 Sep 2022 20:21:49 +0000 (16:21 -0400)]
AtomicExpand: Don't create unused instructions for some atomicrmw

This wasn't used by every atomicrmw expansion.

2 years agoHexagon: Register null TargetStreamer
Matt Arsenault [Sat, 29 Oct 2022 19:34:03 +0000 (12:34 -0700)]
Hexagon: Register null TargetStreamer

Unlike just about every target, I don't
see an obvious null dereference this fixes.

2 years agoHexagon: Add missing C++ mode comment
Matt Arsenault [Sat, 29 Oct 2022 19:34:42 +0000 (12:34 -0700)]
Hexagon: Add missing C++ mode comment

2 years agoNVPTX: Register null target streamer
Matt Arsenault [Sat, 29 Oct 2022 19:38:55 +0000 (12:38 -0700)]
NVPTX: Register null target streamer

2 years agoX86: Implement null target streamer
Matt Arsenault [Fri, 28 Oct 2022 20:42:37 +0000 (13:42 -0700)]
X86: Implement null target streamer

There should no need for null checks in the AsmPrinter

2 years agoAArch64: Register null MCTargetStreamer
Matt Arsenault [Fri, 28 Oct 2022 17:32:15 +0000 (10:32 -0700)]
AArch64: Register null MCTargetStreamer

Fixes crashes with clang -emit-codegen-only

2 years agoRevert "[InstCombine] fold mul with decremented "shl -1" factor"
Florian Mayer [Tue, 1 Nov 2022 00:39:05 +0000 (17:39 -0700)]
Revert "[InstCombine] fold mul with decremented "shl -1" factor"

This reverts commit bc886e9b587b9d009f49b12eaaa9ebc1c71905a1.

Broke MSAN bootstrap buildbots with Assertion `RangeAfterCopy % ExtraScale == 0 && "Extra instruction requires immediate to be aligned"' failed.

2 years ago[NFC] use has_value instead of hasValue
Florian Mayer [Mon, 31 Oct 2022 23:34:17 +0000 (16:34 -0700)]
[NFC] use has_value instead of hasValue

2 years ago[gn build] Port 627465cf9fa7
LLVM GN Syncbot [Mon, 31 Oct 2022 23:37:46 +0000 (23:37 +0000)]
[gn build] Port 627465cf9fa7

2 years ago[libc++][PMR] Deprecate the implementation in std::experimental
Nikolas Klauser [Wed, 19 Oct 2022 09:06:12 +0000 (11:06 +0200)]
[libc++][PMR] Deprecate the implementation in std::experimental

This commit deprecates <experimental/memory_resource> since we now ship the non-experimental
version of it. Per the libc++ policy [1], we are deprecating the experimental feature in
upcoming LLVM 16 and will remove it entirely in LLVM 18.

[1]: https://libcxx.llvm.org/DesignDocs/ExperimentalFeatures.html#id4

Reviewed By: ldionne, #libc

Spies: EricWF, libcxx-commits

Differential Revision: https://reviews.llvm.org/D136245

2 years ago[libc++][PMR] Move the pmr::memory_resource destructor into the dylib
Nikolas Klauser [Tue, 18 Oct 2022 19:40:49 +0000 (21:40 +0200)]
[libc++][PMR] Move the pmr::memory_resource destructor into the dylib

This avoids emitting the VTable of `pmr::memory_resource` in every TU.

Reviewed By: ldionne

Spies: EricWF, nemanjai, libcxx-commits

Differential Revision: https://reviews.llvm.org/D136196

2 years ago[libc++][PMR] Mark completed papers and issues as such
Nikolas Klauser [Tue, 18 Oct 2022 19:33:32 +0000 (21:33 +0200)]
[libc++][PMR] Mark completed papers and issues as such

Reviewed By: ldionne, #libc

Spies: krytarowski, Quuxplusone, tschuett, libcxx-commits

Differential Revision: https://reviews.llvm.org/D136194

2 years ago[libc++][PMR] Add attributes
Nikolas Klauser [Tue, 18 Oct 2022 18:40:13 +0000 (20:40 +0200)]
[libc++][PMR] Add attributes

This allows the compiler to do more optimizations.

Reviewed By: ldionne, #libc

Spies: libcxx-commits, krytarowski

Differential Revision: https://reviews.llvm.org/D136191

2 years ago[libc++] Move preferred_name declarations into the forward declaring headers and...
Nikolas Klauser [Wed, 12 Oct 2022 23:03:58 +0000 (01:03 +0200)]
[libc++] Move preferred_name declarations into the forward declaring headers and add pmr preferred names

We currently define the preferred names in multiple places. `basic_string` and `basic_string_view` also have a lot of aliases, which makes the declarations quite long. So let's only add the preferred names in forward-declaring headers to make the implementation more readable and have all the preferred names in one place.

Reviewed By: ldionne

Spies: EricWF, krytarowski, libcxx-commits

Differential Revision: https://reviews.llvm.org/D135824

2 years ago[SPIRV] support the enqueue_kernel builtin function
Ilia Diachkov [Sat, 29 Oct 2022 00:07:02 +0000 (03:07 +0300)]
[SPIRV] support the enqueue_kernel builtin function

The patch adds support of the enqueue_kernel builtin function.
It is implemented in the same way as in the SPIRV translator.
2 LIT tests are added to show the improvement.

Differential Revision: https://reviews.llvm.org/D137018

2 years ago[Bazel] Add llvm:Support to Bazel deps of SPIRVCommonConversion
Anlun Xu [Mon, 31 Oct 2022 22:26:58 +0000 (15:26 -0700)]
[Bazel] Add llvm:Support to Bazel deps of SPIRVCommonConversion

Differential Revision: https://reviews.llvm.org/D137119

2 years ago[JumpThreading] Put a limit on the PHI nodes when duplicating a BB.
Usman Nadeem [Mon, 31 Oct 2022 22:08:11 +0000 (15:08 -0700)]
[JumpThreading] Put a limit on the PHI nodes when duplicating a BB.

Do not duplicate a BB if it has a lot of PHI nodes.
If a threadable chain is too long then the number of duplicated PHI nodes
can add up, leading to a substantial increase in compile time when rewriting
the SSA.

Fixes https://github.com/llvm/llvm-project/issues/58203
Differential Revision: https://reviews.llvm.org/D136716

The threshold of 76 in this patch is reasonably high and reduces the compile
time of cldwat2m_macro.f90 in SPEC2017/cam4 from 80+min to <2min.

Change-Id: I153c89a8e0d89b206a5193dc1b908c67e320717e

2 years ago[RISCV] Fix crash a vector add has a 4x sext and zext operand.
Craig Topper [Mon, 31 Oct 2022 19:37:51 +0000 (12:37 -0700)]
[RISCV] Fix crash a vector add has a 4x sext and zext operand.

We can narrow one of the extends and keep the other original by
using a vwaddu.wv or vwadd.wv.

We were previously forgetting to keep the original operand and
instead took the source of its extend. This resulted in a type
mismatch that later failed with an impossible physical register copy.

To fix this I've refactored some code to maintain information about
whether the source needs to be extended at all for longer so we could
use it in materialize.

Differential Revision: https://reviews.llvm.org/D137106

2 years ago[libc++] Add test for checking progress on P0533R9
Evgeny Shulgin [Sat, 22 Oct 2022 20:17:56 +0000 (20:17 +0000)]
[libc++] Add test for checking progress on P0533R9

This test checks progress on P0533R9 (https://wg21.link/p0533r9).
Whenever a desired function becomes constexpr, the programmer won't forget
to switch `ASSERT_NOT_CONSTEXPR_CXX23` to `ASSERT_CONSTEXPR_CXX23` and
eventually to change the paper's implementation status. The test also works
as a reference list of unimplemented functions.

Reviewed By: philnik, #libc

Differential Revision: https://reviews.llvm.org/D136538

2 years ago[gn build] Port 84ea6b6f78df
LLVM GN Syncbot [Mon, 31 Oct 2022 21:45:27 +0000 (21:45 +0000)]
[gn build] Port 84ea6b6f78df

2 years ago[gn build] Port 0d01300aacf6
LLVM GN Syncbot [Mon, 31 Oct 2022 21:45:26 +0000 (21:45 +0000)]
[gn build] Port 0d01300aacf6

2 years ago[lldb] Emit diagnostic events in the diagnostic dump
Jonas Devlieghere [Mon, 31 Oct 2022 17:16:32 +0000 (10:16 -0700)]
[lldb] Emit diagnostic events in the diagnostic dump

Connect the diagnostic events with the diagnostic infrastructure.

 - Emit existing diagnostic events (warnings and errors) to the
   diagnostic log.
 - Introduce a new diagnostic event (info) that's used exclusively for
   diagnostic logging and does not get broadcast.

Differential revision: https://reviews.llvm.org/D136648

2 years ago[lldb] Add a "diagnostics dump" command
Jonas Devlieghere [Mon, 31 Oct 2022 17:11:12 +0000 (10:11 -0700)]
[lldb] Add a "diagnostics dump" command

Add a "diagnostics dump" command to, as the name implies, dump the
diagnostics to disk. The goal of this command is to let the user
generate the diagnostics in case of an issue that doesn't cause the
debugger to crash.

This command is also critical for testing, where we don't want to cause
a crash to emit the diagnostics.

Differential revision: https://reviews.llvm.org/D135622

2 years ago[lldb] Add diagnostics
Jonas Devlieghere [Fri, 28 Oct 2022 21:26:11 +0000 (14:26 -0700)]
[lldb] Add diagnostics

Around this time last year, I said on the mailing list [1] that I wanted
to to transform the reproducers into something that resembles a
sysdiagnose on Apple platforms: a collection of files containing a
variety of information to help diagnose bugs or troubleshoot issues.

This patch adds that framework. Based on lessons learned from the
reproducers, I've intentionally tried to keep it small and simple.
Different parts of LLDB can register callbacks (this is necessary for
layering purposes) that will get called when the diagnostics should be
generated.

[1] https://lists.llvm.org/pipermail/lldb-dev/2021-September/017045.html

Differential revision: https://reviews.llvm.org/D134991

2 years ago[InstCombine] adjust branch on logical-and fold
Sanjay Patel [Mon, 31 Oct 2022 21:32:34 +0000 (17:32 -0400)]
[InstCombine] adjust branch on logical-and fold

The transform was just added with:
115d2f69a515cd756fa51
...but as noted in post-commit feedback, it was
confusingly coded. Now, we create the final
expected canonicalized form directly and put
an extra use check on the match, so we should
not ever end up with more instructions.

2 years ago[InstCombine] add test for branch on logical-and; NFC
Sanjay Patel [Mon, 31 Oct 2022 21:29:03 +0000 (17:29 -0400)]
[InstCombine] add test for branch on logical-and; NFC

More coverage for the change 115d2f69a515cd756fa51 as
suggested in post-commit feedback.

Given that the transform is canonicalized with another
'not' op, we should adjust the use checks and create
that form directly.

2 years ago[Hexagon] Add LLVM codegen testcases for V6_[add|sub]carryo intrinsics
Krzysztof Parzyszek [Mon, 31 Oct 2022 21:06:39 +0000 (14:06 -0700)]
[Hexagon] Add LLVM codegen testcases for V6_[add|sub]carryo intrinsics

2 years ago[libc][math] Add place-holder implementation for pow function.
Tue Ly [Mon, 31 Oct 2022 20:04:41 +0000 (16:04 -0400)]
[libc][math] Add place-holder implementation for pow function.

Add place-holder implementation for pow function to unblock libc demo
examples.

Reviewed By: michaelrj

Differential Revision: https://reviews.llvm.org/D137109

2 years ago[libc][math] Add place-holder implementation for asin to unblock demo examples.
Tue Ly [Mon, 31 Oct 2022 19:42:00 +0000 (15:42 -0400)]
[libc][math] Add place-holder implementation for asin to unblock demo examples.

Add a place-holder implementation for asin to unblock libc demo
examples.

Reviewed By: michaelrj

Differential Revision: https://reviews.llvm.org/D137105

2 years ago[flang] Rework line of code to dodge clang 16 warning
Peter Klausler [Mon, 31 Oct 2022 21:17:02 +0000 (14:17 -0700)]
[flang] Rework line of code to dodge clang 16 warning

Recode a non-short-circuiting conjunction of two Boolean function calls
into separate statements to avoid a warning from clang 16.

2 years ago[mlir][arith][spirv] Hard fail in `-convert-arith-to-spirv`
Jakub Kuderski [Mon, 31 Oct 2022 21:00:45 +0000 (17:00 -0400)]
[mlir][arith][spirv] Hard fail in `-convert-arith-to-spirv`

Turn legalization failures into hard failures to make sure that we do
not miss conversion pattern application failures.

Add a message on type conversion failure.

Move unsupported cases into a separate test file.

Reviewed By: antiagainst

Differential Revision: https://reviews.llvm.org/D137102

2 years ago[wasm-ld] Allow importing/exporting the output module's memory with arbitrary names
Dan Gohman [Tue, 4 Oct 2022 19:50:10 +0000 (12:50 -0700)]
[wasm-ld] Allow importing/exporting the output module's memory with arbitrary names

This adds an `--export-memory` option to wasm-ld which allows passing
a name to give to the exported memory, and extends `--import-memory` to
allow passing a <module>,<name> pair specifying where the memory should
be imported from.

This is based on https://reviews.llvm.org/D131376, with the main
difference being that it only supports exporting memory by one name.

Differential Revision: https://reviews.llvm.org/D135898

2 years ago[InstCombine] Precommit test for D136582; NFC
chenglin.bi [Mon, 31 Oct 2022 20:45:59 +0000 (04:45 +0800)]
[InstCombine] Precommit test for D136582; NFC

2 years ago[Hexagon] Add builtins and intrinsics for V6_v[add|sub]carryo
Krzysztof Parzyszek [Mon, 31 Oct 2022 19:28:25 +0000 (12:28 -0700)]
[Hexagon] Add builtins and intrinsics for V6_v[add|sub]carryo

2 years ago[InstCombine] Allow simplify demanded transformations on scalable vectors
Philip Reames [Mon, 31 Oct 2022 20:34:28 +0000 (13:34 -0700)]
[InstCombine] Allow simplify demanded transformations on scalable vectors

Differential Revision: https://reviews.llvm.org/D136475

2 years ago[lit] Fix typo in REQUIRES clause
Paul Robinson [Mon, 31 Oct 2022 20:14:56 +0000 (13:14 -0700)]
[lit] Fix typo in REQUIRES clause

2 years ago[AliasAnalysis] Introduce getModRefInfoMask() as a generalization of pointsToConstant...
Patrick Walton [Mon, 24 Oct 2022 22:31:35 +0000 (15:31 -0700)]
[AliasAnalysis] Introduce getModRefInfoMask() as a generalization of pointsToConstantMemory().

The pointsToConstantMemory() method returns true only if the memory pointed to
by the memory location is globally invariant. However, the LLVM memory model
also has the semantic notion of *locally-invariant*: memory that is known to be
invariant for the life of the SSA value representing that pointer. The most
common example of this is a pointer argument that is marked readonly noalias,
which the Rust compiler frequently emits.

It'd be desirable for LLVM to treat locally-invariant memory the same way as
globally-invariant memory when it's safe to do so. This patch implements that,
by introducing the concept of a *ModRefInfo mask*. A ModRefInfo mask is a bound
on the Mod/Ref behavior of an instruction that writes to a memory location,
based on the knowledge that the memory is globally-constant memory (in which
case the mask is NoModRef) or locally-constant memory (in which case the mask
is Ref). ModRefInfo values for an instruction can be combined with the
ModRefInfo mask by simply using the & operator. Where appropriate, this patch
has modified uses of pointsToConstantMemory() to instead examine the mask.

The most notable optimization change I noticed with this patch is that now
redundant loads from readonly noalias pointers can be eliminated across calls,
even when the pointer is captured. Internally, before this patch,
AliasAnalysis was assigning Ref to reads from constant memory; now AA can
assign NoModRef, which is a tighter bound.

Differential Revision: https://reviews.llvm.org/D136659

2 years ago[lit] Remove undefined keyword from a Driver test
Paul Robinson [Mon, 31 Oct 2022 19:58:53 +0000 (12:58 -0700)]
[lit] Remove undefined keyword from a Driver test

`clang-driver` hasn't been a valid keyword for a long time.
The test fails when the keyword is removed, so added XFAIL: *
to reflect the true status of the test (XFAIL not UNSUPPORTED)
and filed issue #58711 to get it fixed.

2 years ago[InstCombine] canonicalize branch with logical-and-not condition
Sanjay Patel [Mon, 31 Oct 2022 19:49:03 +0000 (15:49 -0400)]
[InstCombine] canonicalize branch with logical-and-not condition

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

In the motivating case from issue #58313,
this allows forming a duplicate 'not' op
which then gets CSE'd and simplifyCFG'd
and combined into the expected 'xor'.

2 years ago[InstCombine] add tests for branch on logical and/or; NFC
Sanjay Patel [Mon, 31 Oct 2022 18:00:05 +0000 (14:00 -0400)]
[InstCombine] add tests for branch on logical and/or; NFC

2 years ago[test] Add coverage for sign bits and known non-zero for scalable vectors
Philip Reames [Mon, 31 Oct 2022 18:54:37 +0000 (11:54 -0700)]
[test] Add coverage for sign bits and known non-zero for scalable vectors

2 years ago[flang] Fix pointer definition semantic checking via refactoring
Peter Klausler [Mon, 17 Oct 2022 19:51:00 +0000 (12:51 -0700)]
[flang] Fix pointer definition semantic checking via refactoring

The infrastructure in semantics that is used to check that the
left-hand sides of normal assignment statements are really definable
variables was not being used to check whether the LHSs of pointer assignments
are modifiable, and so most cases of unmodifiable pointers are left
undiagnosed.  Rework the semantics checking for pointer assignments,
NULLIFY statements, pointer dummy arguments, &c. so that cases of
unmodifiable pointers are properly caught.  This has been done
by extracting all the various definability checking code that has
been implemented for different contexts in Fortran into one new
facility.

The new consolidated definability checking code returns messages
meant to be attached as "because: " explanations to context-dependent
errors like "left-hand side of assignment is not definable".
These new error message texts and their attached explanations
affect many existing tests, which have been updated.  The testing
infrastructure was extended by another patch to properly compare
warnings and explanatory messages, which had been ignored until
recently.

Differential Revision: https://reviews.llvm.org/D136979

2 years ago[mlir][scf] Add scf-to-cf lowering for `scf.index_switch`
Jeff Niu [Thu, 27 Oct 2022 20:43:41 +0000 (13:43 -0700)]
[mlir][scf] Add scf-to-cf lowering for `scf.index_switch`

This patch adds lowering from `scf.index_switch` to `cf.switch.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D136883

2 years ago[test] Refresh an autogened test to pickup naming change
Philip Reames [Mon, 31 Oct 2022 18:36:14 +0000 (11:36 -0700)]
[test] Refresh an autogened test to pickup naming change

2 years agoRevert "[clang] ASTImporter: Fix importing of va_list types and declarations"
Matheus Izvekov [Mon, 31 Oct 2022 18:34:23 +0000 (19:34 +0100)]
Revert "[clang] ASTImporter: Fix importing of va_list types and declarations"

This reverts commit 5f820c0f55cd9d4f4520cc92eae0a1c4afe05a54.

Apparently it breaks aarch64 buildbots.
https://lab.llvm.org/buildbot#builders/188/builds/21591

2 years ago[flang] Be more precise about CHARACTER known length discrepancies
Peter Klausler [Thu, 13 Oct 2022 22:22:55 +0000 (15:22 -0700)]
[flang] Be more precise about CHARACTER known length discrepancies

Many intrinsic functions in Fortran require that two or more of their
arguments have types that agree in the values of all of their type
parameters, while others only require the same type category and kind
type parameters but not lengths, including the important case of
CHARACTER.  The intrinsic procedure tables need to be adjusted in
some cases so that discrepancies in character lengths that are known
at compilation time can be diagnosed as errors where they should be,
as in for example MOVE_ALLOC().

Differential Revision: https://reviews.llvm.org/D137032

2 years agoAddress post commit style comment from 087bb0f
Philip Reames [Mon, 31 Oct 2022 18:15:48 +0000 (11:15 -0700)]
Address post commit style comment from 087bb0f

2 years ago[lit] Remove undefined keyword from an MCJIT test
Paul Robinson [Mon, 31 Oct 2022 17:51:29 +0000 (10:51 -0700)]
[lit] Remove undefined keyword from an MCJIT test

Changed 'XFAIL: mcjit-ia32' to 'XFAIL: i686 && windows' because the
'mcjit-ia32 feature isn't defined anywhere, or used anywhere else.

I found this as part of other work to clean up lit keywords, so
I figured I would go ahead and fix it.  Verified by hacking my
lit.site.cfg.py so host_triple = target_triple = "i686-pc-windows-msvc"
and the test correctly reported XFAIL, while "i686-pc-linux" passes.

2 years agoRevert "[lit] Define keyword used by MCJIT test"
Paul Robinson [Mon, 31 Oct 2022 17:44:49 +0000 (10:44 -0700)]
Revert "[lit] Define keyword used by MCJIT test"

This reverts commit 6d859266803e2a9060c4e8770f92cc2c7bd05a3b.

Causing an XPASS on 32-bit Gentoo, as reported by mgorny.

2 years ago[utils] Remove ancient scripts using llvm-config --src-root
Fangrui Song [Mon, 31 Oct 2022 17:45:00 +0000 (10:45 -0700)]
[utils] Remove ancient scripts using llvm-config --src-root

The last effective changes for the 4 scripts were more than 10 years ago.
They do grep and lines of code counting which have mature modern replacements
(rg, tokei, etc).

Reviewed By: mgorny

Differential Revision: https://reviews.llvm.org/D136969

2 years ago[X86][Driver] Remove stale FIXME. NFC
Craig Topper [Mon, 31 Oct 2022 17:32:22 +0000 (10:32 -0700)]
[X86][Driver] Remove stale FIXME. NFC

We use getHostCPUFeatures in x86::getX86TargetFeatures so I think
this FIXME is taken care of already.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D137054

2 years ago[flang] Always diagnose incompatible function results when associating function pointers
Peter Klausler [Wed, 12 Oct 2022 19:59:48 +0000 (12:59 -0700)]
[flang] Always diagnose incompatible function results when associating function pointers

There are some exceptional cases where the compiler permits association of
procedure pointers or dummy arguments with slightly incompatible procedure
targets, but they should not override any incompatibilty of function
result types.

(Includes a second fix to resolve the original motivating test failure, in
which a COMPLEX intrinsic function was getting its result kind divided by
two due to an implicit C++ conversion of the kind to a "*kind" parse tree
node, and those legacy "COMPLEX*size" type designators' values are twice
the type kind value.)

Differential Revision: https://reviews.llvm.org/D136964

2 years ago[OpenMP] [OMPT] [2/8] Implemented a connector for communication of OMPT callbacks...
Dhruva Chakrabarti [Fri, 17 Jun 2022 23:05:39 +0000 (16:05 -0700)]
[OpenMP] [OMPT] [2/8] Implemented a connector for communication of OMPT callbacks between libraries.

This is part of a set of patches implementing OMPT target callback support and has been split out of the originally submitted https://reviews.llvm.org/D113728. The overall design can be found in https://rice.app.box.com/s/pf3gix2hs4d4o1aatwir1set05xmjljc

The purpose of this patch is to provide a way to register tool-provided callbacks into libomp when libomptarget is loaded.

Introduced a cmake variable LIBOMPTARGET_OMPT_SUPPORT that can be used to control OMPT target support. It follows host OMPT support, controlled by LIBOMP_HAVE_OMPT_SUPPORT.

Added a connector that can be used to communicate between OMPT implementations in libomp and libomptarget or libomptarget and a plugin.

Added a global constructor in libomptarget that uses the connector to force registration of tool-provided callbacks in libomp. A pair of init and fini functions are provided to libomp as part of the connect process which will be used to register the tool-provided callbacks in libomptarget.

Depends on D123429

Patch from John Mellor-Crummey <johnmc@rice.edu>
(With contributions from Dhruva Chakrabarti <Dhruva.Chakrabarti@amd.com>)

Reviewed By: dreachem

Differential Revision: https://reviews.llvm.org/D123572

2 years ago[X86] Enable EVEX GFNI instructions without avx512bw.
Craig Topper [Mon, 31 Oct 2022 17:04:36 +0000 (10:04 -0700)]
[X86] Enable EVEX GFNI instructions without avx512bw.

We only really need avx512bw for masking 256 or 512 bit GFNI
instructions due to the need for v32i1 or v64i1.

I wanted to enable 128-bit intrinsics with avx512vl, but the
__builtin_ia32_selectb_128 used in the header file requires avx512bw.
The codegen test for the same is also not using a masked instruction
because vselect with v16i1 mask and v16i8 is not legal so is expanded
before isel. To fix these issues we need a mask specific builtin and a
mask specific ISD opcode.

Fixes PR58687.

Reviewed By: pengfei

Differential Revision: https://reviews.llvm.org/D137036

2 years ago[flang] Add atomic_and to list of intrinsics
Katherine Rasmussen [Fri, 28 Oct 2022 00:55:18 +0000 (17:55 -0700)]
[flang] Add atomic_and to list of intrinsics

Add the atomic subroutine, atomic_and, to the list of
intrinsic subroutines, add its last dummy argument to a check
for coindexed-object, and update test.

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D136907

2 years ago[clang] Instantiate alias templates with sugar
Matheus Izvekov [Sun, 23 Oct 2022 14:57:12 +0000 (16:57 +0200)]
[clang] Instantiate alias templates with sugar

This makes use of the changes introduced in D134604, in order to
instantiate alias templates witn a final sugared substitution.

This comes at no additional relevant cost.
Since we don't track / unique them in specializations, we wouldn't be
able to resugar them later anyway.

Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Differential Revision: https://reviews.llvm.org/D136565

2 years ago[clang] Instantiate NTTPs and template default arguments with sugar
Matheus Izvekov [Sun, 23 Oct 2022 11:41:57 +0000 (13:41 +0200)]
[clang] Instantiate NTTPs and template default arguments with sugar

This makes use of the changes introduced in D134604, in order to
instantiate non-type template parameters and default template arguments
with a final sugared substitution.

This comes at no additional relevant cost.
Since we don't track / unique them in specializations, we wouldn't be
able to resugar them later anyway.

Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Differential Revision: https://reviews.llvm.org/D136564

2 years ago[clang] ASTImporter: Fix importing of va_list types and declarations
Matheus Izvekov [Sat, 29 Oct 2022 18:36:31 +0000 (20:36 +0200)]
[clang] ASTImporter: Fix importing of va_list types and declarations

This fixes a problem where __va_list_tag was not correctly imported,
possibly leading to multiple definitions with different types.

This adds __va_list_tag to it's proper scope, so that the ASTImporter
can find it.

Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Differential Revision: https://reviews.llvm.org/D136886

2 years ago[clang] Fix handling of unexpanded packs in template argument expressions.
Matheus Izvekov [Fri, 28 Oct 2022 20:22:39 +0000 (22:22 +0200)]
[clang] Fix handling of unexpanded packs in template argument expressions.

Closes #58679.

Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Differential Revision: https://reviews.llvm.org/D136977

2 years ago[clang] Improve error recovery for pack expansion of expressions
Matheus Izvekov [Fri, 28 Oct 2022 16:23:02 +0000 (18:23 +0200)]
[clang] Improve error recovery for pack expansion of expressions

Closes #58673.

Signed-off-by: Matheus Izvekov <mizvekov@gmail.com>
Differential Revision: https://reviews.llvm.org/D136962

2 years ago[libc] Implement getopt
Alex Brachet [Mon, 31 Oct 2022 16:55:53 +0000 (16:55 +0000)]
[libc] Implement getopt

Differential Revision: https://reviews.llvm.org/D133487

2 years ago[SLP]Fix a crash in the analysis of the compatible cmp operands.
Alexey Bataev [Mon, 31 Oct 2022 15:54:49 +0000 (08:54 -0700)]
[SLP]Fix a crash in the analysis of the compatible cmp operands.

We can skip the analysis of the operands opcodes, can compare directly
them in some cases.

2 years ago[AAPointerInfo] check for Unknown offsets in callee
Sameer Sahasrabuddhe [Mon, 31 Oct 2022 16:45:56 +0000 (22:15 +0530)]
[AAPointerInfo] check for Unknown offsets in callee

When translating offset info from the callee at a call site, first check if the
offset is Unknown. Any offset in the caller should be added only if the callee
offset is valid.

Differential Revision: https://reviews.llvm.org/D137011

2 years ago[gn build] Port aee2a35ac4ab
LLVM GN Syncbot [Mon, 31 Oct 2022 15:53:48 +0000 (15:53 +0000)]
[gn build] Port aee2a35ac4ab

2 years ago[LoopFuse] Ensure loops are in loop simplified form under new PM
Mengxuan Cai [Mon, 31 Oct 2022 15:45:44 +0000 (11:45 -0400)]
[LoopFuse] Ensure loops are in loop simplified form under new PM

Loop Fusion (Function Pass) requires loops in simplified form. With
legacy-pm, loop-simplify pass is added as a dependency for loop-fusion.
But the new pass manager does not always ensure this format. This patch
tries to invoke simplifyLoop() on loops that are not in simplified form
only for new PM.

Reviewed By: aeubanks

Differential Revision: https://reviews.llvm.org/D136781

2 years ago[X86] Add AVX-NE-CONVERT instructions.
Freddy Ye [Mon, 31 Oct 2022 15:01:31 +0000 (23:01 +0800)]
[X86] Add AVX-NE-CONVERT instructions.

For more details about these instructions, please refer to the latest ISE document: https://www.intel.com/content/www/us/en/develop/download/intel-architecture-instruction-set-extensions-programming-reference.html

Reviewed By: pengfei

Differential Revision: https://reviews.llvm.org/D135930

2 years ago[JITLink][AArch46] Update aarch64 pointer jump stub sequence, add utilities.
Lang Hames [Mon, 31 Oct 2022 05:14:30 +0000 (22:14 -0700)]
[JITLink][AArch46] Update aarch64 pointer jump stub sequence, add utilities.

Updates the aarch64 pointer jump stub sequence to:

ADRP x16, <ptr>@page21
LDR  x16, [x16, <ptr>@pageoff12]
BR   x16

from:

LDR  x16, <ptr>@ldrimm19
BR   x16

The old sequence can only reference pointers within +/-1Mb, which may not be
enough for large object files. The new sequence can reach pointers within
+/-4Gb. (A future pre-fixup-pass could apply range-based optimizations to
turn this into an ldr-br-nop sequence.)

Also adds createPointerJumpStubBlock and createAnonymousPointerJumpStub
utilities along the same lines as their x86-64 counterparts.

The PLTTableManager is updated to use the new utility functions.

2 years ago[flang] Fix code generation for polymorphic deallocation
Valentin Clement [Mon, 31 Oct 2022 15:33:45 +0000 (16:33 +0100)]
[flang] Fix code generation for polymorphic deallocation

This patch fix issues found during call to deallocate.
fir.class can be handled the same way as fir.box in fir.store operation
code generation. In MutableBox::createNewFirBox, the fir.class is
also already a boxed entity.

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D137079

2 years ago[Support] Do not run test on AIX
Serge Pavlov [Mon, 31 Oct 2022 12:09:32 +0000 (19:09 +0700)]
[Support] Do not run test on AIX

A part of the unit test CommandLineTest/BadResponseFile, added in
the commit fd3d7a9f8cbb (Handle errors in expansion of response files)
fails on AIX. The test checks if an error is issued when `file` in
`@file` is a path to directory. This change excludes that check on
AIX platform.

Differential Revision: https://reviews.llvm.org/D136090