platform/upstream/llvm.git
2 years ago[AArch64] Use proper instruction mnemonics for FPRs
Bill Wendling [Fri, 20 May 2022 19:02:26 +0000 (12:02 -0700)]
[AArch64] Use proper instruction mnemonics for FPRs

The FPR128 regs need MOVIv2d_ns and SVE regs need DUP_ZI_D.

Differential Revision: https://reviews.llvm.org/D126083

2 years agoRecommit "[ConstantRange] Improve the implementation of binaryOr"
Alexander Shaposhnikov [Fri, 20 May 2022 18:39:21 +0000 (18:39 +0000)]
Recommit "[ConstantRange] Improve the implementation of binaryOr"

This recommits https://reviews.llvm.org/rG6990e7477d24ff585ae86549f5280f0be65422a6
as the problematic test has been updated updated in
https://reviews.llvm.org/rG3bd112c720dc614a59e3f34ebf9b45075037bfa0.

2 years agoAdd some diagnostics to diagnose bot-only failures for TestIgnoredExceptions.py
Jim Ingham [Fri, 20 May 2022 18:36:55 +0000 (11:36 -0700)]
Add some diagnostics to diagnose bot-only failures for TestIgnoredExceptions.py

The test for commit bff4673b41781ec5bff6b96b52cf321d2271726c is failing on the
GreenDragon bot but none of us can repro the failure locally.  Adding some logging
to the test failure to help diagnose the issue.

2 years agoAvoid uninitialized Diag.ID (which we pass but never read)
Sam McCall [Fri, 20 May 2022 18:29:47 +0000 (20:29 +0200)]
Avoid uninitialized Diag.ID (which we pass but never read)

2 years agoFix up fuzzing test on Windows.
Mitch Phillips [Fri, 20 May 2022 18:25:29 +0000 (11:25 -0700)]
Fix up fuzzing test on Windows.

3bd112c720dc fixed the fuzzing test on Linux, which, after
https://reviews.llvm.org/D125933, has one less branch. Turns out, on
Windows, that it still has the extra branch. I'm guessing that's because
exit() isn't known to be noreturn on Windows or something.

Either way, just make the test more tolerant.

2 years agoAdapt C++ std::string dataformatter for D125496
Adrian Prantl [Fri, 20 May 2022 18:24:24 +0000 (11:24 -0700)]
Adapt C++ std::string dataformatter for D125496

https://reviews.llvm.org/D125496 changed the layout of std::string
without updating the LLDB dataformatter. This patch adds code to
recognize the new format.

Differential Revision: https://reviews.llvm.org/D126080

2 years agoUpdate fuzzing test to comply with new optimisation.
Mitch Phillips [Fri, 20 May 2022 17:59:29 +0000 (10:59 -0700)]
Update fuzzing test to comply with new optimisation.

https://reviews.llvm.org/D125933 improved some of LLVM's handling of
binary ORs, which meant we have one less conditional branch, because the
'if (Size > 5 && Data[5] == 'R')' and 'if (bits == 63)' branches are now
correctly folded.

2 years agoRevert "[ConstantRange] Improve the implementation of binaryOr"
Douglas Yung [Fri, 20 May 2022 17:21:24 +0000 (10:21 -0700)]
Revert "[ConstantRange] Improve the implementation of binaryOr"

This reverts commit 6990e7477d24ff585ae86549f5280f0be65422a6.

This change was causing the test compiler-rt/test/fuzzer/merge_two_step.test to fail on
our internal bot as well as other build bots such as https://lab.llvm.org/buildbot/#/builders/179/builds/3712.

2 years ago[InstructionSimplify] Remove multiple 'break' after 'return'. NFC
Craig Topper [Fri, 20 May 2022 17:23:12 +0000 (10:23 -0700)]
[InstructionSimplify] Remove multiple 'break' after 'return'. NFC

2 years ago[RISCV] Add initial test coverage for LSR
Philip Reames [Fri, 20 May 2022 17:05:55 +0000 (10:05 -0700)]
[RISCV] Add initial test coverage for LSR

Establish the most basic possible test coverage for LSR transformation on RISCV.

Original patch by eopXD (D123458), modified by me to cleanup/simplify tests.

2 years ago[gn build] (manually) port 480dcdc8975d8 (llvm-ifs OptTable)
Nico Weber [Fri, 20 May 2022 17:13:37 +0000 (13:13 -0400)]
[gn build] (manually) port 480dcdc8975d8 (llvm-ifs OptTable)

2 years ago[mlir] Add missing NVGPU link dependency to VectorToGPU
Christopher Bate [Fri, 20 May 2022 16:44:31 +0000 (10:44 -0600)]
[mlir] Add missing NVGPU link dependency to VectorToGPU

The missing link dependency caused build failures in some
configurations.

2 years ago[lldb] Set correct register number for cpsr (GENERIC_REGNUM_FLAGS)
Jonas Devlieghere [Fri, 20 May 2022 16:27:44 +0000 (09:27 -0700)]
[lldb] Set correct register number for cpsr (GENERIC_REGNUM_FLAGS)

Report the correct register number (GENERIC_REGNUM_FLAGS) for cpsr. This
fixes TestLldbGdbServer.py on Apple Silicon.

Differential revision: https://reviews.llvm.org/D126076

2 years ago[TypePromotion] Fix another case for sext vs zext in promoted constant.
Craig Topper [Fri, 20 May 2022 04:19:12 +0000 (21:19 -0700)]
[TypePromotion] Fix another case for sext vs zext in promoted constant.

If the SafeWrap operation is a subtract, we negated the constant
to treat the subtract as an addition. The sext was based on the
operation being addition. So we really need to do (neg (sext (neg C)))
when promoting the constant. This is equivalent to (sext C) for
every value of C except the min signed value. For min signed value
we need to do (zext C) instead.

Fixes PR55490.

Differential Revision: https://reviews.llvm.org/D125653

2 years ago[clang] Fixing arm-common, windows only and openmp header install targets
Qiongsi Wu [Fri, 20 May 2022 14:12:14 +0000 (10:12 -0400)]
[clang] Fixing arm-common, windows only and openmp header install targets

https://reviews.llvm.org/D123498 contains a few errors resulting in incorrect target contents or mismatched target/list names. This patch fixes all the known errors.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D126002

2 years ago[AMDGPU] Mark SMEM cache invalidations as not reading memory
Jay Foad [Thu, 19 May 2022 16:46:54 +0000 (17:46 +0100)]
[AMDGPU] Mark SMEM cache invalidations as not reading memory

This brings the MachineInstrs in line with the corresponding intrinsics
which have side effects but do not access memory. It also matches how
BUF cache invalidation instructions are defined.

The lit test changes are just because the machine scheduler previously
treated them like loads, and added an artificial scheduling edge from
them to the exit SU, which caused them to be scheduled earlier.

Differential Revision: https://reviews.llvm.org/D126074

2 years agoUse the public clang::Builtin API in the unit test
Dmitri Gribenko [Fri, 20 May 2022 15:45:45 +0000 (17:45 +0200)]
Use the public clang::Builtin API in the unit test

2 years agoAdjust BUILD files for [ifs] Switch to using OptTable
Dmitri Gribenko [Fri, 20 May 2022 16:08:29 +0000 (18:08 +0200)]
Adjust BUILD files for [ifs] Switch to using OptTable

2 years agoAdjust BUILD files for [MLIR][GPU] Add NvGpu mma.sync path to the VectorToGPU pass
Dmitri Gribenko [Fri, 20 May 2022 16:03:20 +0000 (18:03 +0200)]
Adjust BUILD files for [MLIR][GPU] Add NvGpu mma.sync path to the VectorToGPU pass

2 years ago[mlir][sparse] add support for complex zero/one building
Aart Bik [Fri, 20 May 2022 02:33:33 +0000 (19:33 -0700)]
[mlir][sparse] add support for complex zero/one building

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D126039

2 years ago[clangd][ObjC] Filter ObjC method completions on the remaining selector
David Goldman [Thu, 28 Apr 2022 19:13:21 +0000 (15:13 -0400)]
[clangd][ObjC] Filter ObjC method completions on the remaining selector

Previously, clangd would filter completions only on the first part of
the selector (first typed chunk) instead of all remaining selector
fragments (all typed chunks).

Differential Revision: https://reviews.llvm.org/D124637

2 years ago[MLIR][GPU] Add NvGpu mma.sync path to the VectorToGPU pass
Christopher Bate [Tue, 17 May 2022 23:54:29 +0000 (17:54 -0600)]
[MLIR][GPU] Add NvGpu mma.sync path to the VectorToGPU pass

This changes adds the option to lower to NvGpu dialect ops during the
VectorToGPU convsersion pass. Because this transformation reuses
existing VectorToGPU logic, a seperate VectorToNvGpu conversion pass is
not created. The option `use-nvgpu` is added to the VectorToGPU pass.
When this is true, the pass will attempt to convert slices rooted at
`vector.contract` operations into `nvgpu.mma.sync` ops, and
`vector.transfer_read` ops are converted to either `nvgpu.ldmatrix` or
one or more `vector.load` operations.  The specific data loaded will
depend on the thread id within a subgroup (warp). These index
calculations depend on data type and shape of the MMA op
according to the downstream PTX specification. The code for supporting
these details is separated into `NvGpuSupport.cpp|h`.

Differential Revision: https://reviews.llvm.org/D122940

2 years ago[ifs] Switch to using OptTable
Alex Brachet [Fri, 20 May 2022 15:28:18 +0000 (15:28 +0000)]
[ifs] Switch to using OptTable

Differential revision: https://reviews.llvm.org/D125658

2 years ago[X86] Add tests for FREM
Nabeel Omer [Fri, 20 May 2022 14:24:09 +0000 (15:24 +0100)]
[X86] Add tests for FREM

Introduces basic test coverage for frem on x86.

Split off from https://reviews.llvm.org/D125988

Differential Revision: https://reviews.llvm.org/D126055

2 years ago[lldb] Disable scripted_crashlog_json.test on Apple Silicon
Jonas Devlieghere [Fri, 20 May 2022 15:24:42 +0000 (08:24 -0700)]
[lldb] Disable scripted_crashlog_json.test on Apple Silicon

Disable scripted_crashlog_json.test on Apple Silicon until Ismail has
bandwidth to investigate.

rdar://93655633

2 years ago[AMDGPU] Handle mandatory literals in isOperandLegal
Jay Foad [Fri, 20 May 2022 12:23:53 +0000 (13:23 +0100)]
[AMDGPU] Handle mandatory literals in isOperandLegal

Extend SIInstrInfo::isOperandLegal to enforce a limit on the number of
literal operands for all VALU instructions, not just VOP3. In particular
it now handles VOP2 instructions with a mandatory literal operand like
V_FMAAK_F32.

Differential Revision: https://reviews.llvm.org/D126064

2 years ago[AMDGPU] Add verification for mandatory literals
Jay Foad [Fri, 20 May 2022 12:06:03 +0000 (13:06 +0100)]
[AMDGPU] Add verification for mandatory literals

Extend the literal operand checking in SIInstrInfo::verifyInstruction to
check VOP2 instructions like V_FMAAK_F32 which have a mandatory literal
operand. The rule is that src0 can also be a literal, but only if it is
the same literal value.

AMDGPUAsmParser::validateConstantBusLimitations already handles this
correctly.

Differential Revision: https://reviews.llvm.org/D126063

2 years agoFix an unused variable warning in no-asserts build mode
Dmitri Gribenko [Fri, 20 May 2022 15:11:18 +0000 (17:11 +0200)]
Fix an unused variable warning in no-asserts build mode

2 years ago[libc++] Use _LIBCPP_ASSERT by default for _PSTL_ASSERTions
Nikolas Klauser [Thu, 19 May 2022 11:10:37 +0000 (13:10 +0200)]
[libc++] Use _LIBCPP_ASSERT by default for _PSTL_ASSERTions

Reviewed By: ldionne, #libc

Spies: jwakely, rodgert, libcxx-commits

Differential Revision: https://reviews.llvm.org/D125634

2 years agoRevert "[libc] Apply no-builtin everywhere, remove unnecessary flags"
Guillaume Chatelet [Fri, 20 May 2022 14:37:17 +0000 (14:37 +0000)]
Revert "[libc] Apply no-builtin everywhere, remove unnecessary flags"

This reverts commit 94d6dd90576637fa0eb2c40ca92320ad4c1a6942.

2 years ago[clangd] Provide links to clang-tidy and include-cleaner diagnostic docs
Sam McCall [Fri, 20 May 2022 13:09:53 +0000 (15:09 +0200)]
[clangd] Provide links to clang-tidy and include-cleaner diagnostic docs

LSP supports Diagnostic.codeInformation since 3.16.
In VSCode, this turns the code (e.g. "unused-includes" or "bugprone-foo") into
a clickable link that opens the docs in a web browser.

Differential Revision: https://reviews.llvm.org/D126065

2 years ago[clang-format] Don't break lines after pragma region
Tobias Hieta [Fri, 20 May 2022 14:11:02 +0000 (16:11 +0200)]
[clang-format] Don't break lines after pragma region

We have autogenerated pragma regions in our code
which where awkwardly broken up like this:

```
#pragma region foo(bar : hello)
```
becomes

```
#pragma region foo(bar \
                   : hello)
```

This fixes the problem by adding region as a keyword
and handling it the same way as pragma mark

Reviewed By: curdeius

Differential Revision: https://reviews.llvm.org/D125961

2 years ago[analyzer][NFC] Remove the unused LocAsInteger::getPersistentLoc()
Balazs Benics [Fri, 20 May 2022 14:06:46 +0000 (16:06 +0200)]
[analyzer][NFC] Remove the unused LocAsInteger::getPersistentLoc()

Reviewed By: martong

Differential Revision: https://reviews.llvm.org/D125920

2 years ago[clang] Honor __attribute__((no_builtin("foo"))) on functions
Stephen Long [Fri, 20 May 2022 13:41:37 +0000 (06:41 -0700)]
[clang] Honor __attribute__((no_builtin("foo"))) on functions

Support for `__attribute__((no_builtin("foo")))` was added in https://reviews.llvm.org/D68028,
but builtins were still being used even when the attribute was placed on a function.

Reviewed By: hans

Differential Revision: https://reviews.llvm.org/D124701

2 years ago[libc++] Improve error messages for disabled modes
Louis Dionne [Wed, 18 May 2022 17:17:14 +0000 (13:17 -0400)]
[libc++] Improve error messages for disabled modes

We should not surface CMake-level options like LIBCXX_ENABLE_FILESYSTEM
to our users, since they don't know what it means. Instead, use a slightly
more general wording.

Also, add an error in <ios> to improve the quality of errors for people
trying to use <iostream> when localization is disabled.

Differential Revision: https://reviews.llvm.org/D125910

2 years ago[libunwind] Introduce a cmake-bridge.cfg.in file to reduce test config duplication
Louis Dionne [Thu, 19 May 2022 15:16:37 +0000 (11:16 -0400)]
[libunwind] Introduce a cmake-bridge.cfg.in file to reduce test config duplication

Differential Revision: https://reviews.llvm.org/D125981

2 years ago[libc++abi] Add missing XFAIL on test
Louis Dionne [Fri, 20 May 2022 13:35:44 +0000 (09:35 -0400)]
[libc++abi] Add missing XFAIL on test

2 years agoFix test from df2a4e to work with 32 bit windows targets.
Erich Keane [Fri, 20 May 2022 13:20:13 +0000 (06:20 -0700)]
Fix test from df2a4e to work with 32 bit windows targets.

32 Bit windows includes attribute 'thiscall' on member functions as a
default calling convention.  This test was not written in a way that
works with that, so added wildcards so it is tolerant of it.

2 years ago[DebugInfo][NFC] Add instr-ref documentation, migration guide
J. Ryan Stinnett [Mon, 9 May 2022 14:45:56 +0000 (15:45 +0100)]
[DebugInfo][NFC] Add instr-ref documentation, migration guide

This used to be D102158, but all the code it describes got re-written, so I
figured I'd take another shot at documenting the new instruction referencing
variable locations, this time from a higher level. Happily there's no longer any
need to describe LiveDebugValues in any detail seeing how it's all SSA-based
now.

Probably the most important part is the explanation of what targets need to do
to support instruction referencing. The list is small, mostly because there's
nothing especially complicated that targets need to do: just instrument their
target-specific optimisations and implement the stack spill/restore recognition
target hooks.

This is a small amount of text (which is a virtue), I'm extremely happy to
expand on anything.

Differential Revision: https://reviews.llvm.org/D113586

Co-authored-by: Jeremy Morse <jeremy.morse@sony.com>
2 years ago[InstCombine] add use check to fold of bitwise logic with cast ops
Sanjay Patel [Fri, 20 May 2022 13:07:21 +0000 (09:07 -0400)]
[InstCombine] add use check to fold of bitwise logic with cast ops

This was shown as a potential regression in D126040.

2 years ago[InstCombine] add casted bitwise logic tests to show missing use check; NFC
Sanjay Patel [Fri, 20 May 2022 13:01:41 +0000 (09:01 -0400)]
[InstCombine] add casted bitwise logic tests to show missing use check; NFC

While here, update the auto-generated checks to also check
the match the function parameters - there was a potential
miscompile that would go unnoticed with the more lenient
check lines.

2 years ago[SLP]Do not emit extract elements for insertelements users, replace with shuffles...
Alexey Bataev [Wed, 4 Aug 2021 17:58:37 +0000 (10:58 -0700)]
[SLP]Do not emit extract elements for insertelements users, replace with shuffles directly.

SLP vectorizer emits extracts for externally used vectorized scalars and
estimates the cost for each such extract. But in many cases these
scalars are input for insertelement instructions, forming buildvector,
and instead of extractelement/insertelement pair we can emit/cost
estimate shuffle(s) cost and generate series of shuffles, which can be
further optimized.

Tested using test-suite (+SPEC2017), the tests passed, SLP was able to
generate/vectorize more instructions in many cases and it allowed to reduce
number of re-vectorization attempts (where we could try to vectorize
buildector insertelements again and again).

Differential Revision: https://reviews.llvm.org/D107966

2 years ago[SLP][NFC]Use SmallPtrSet to avoid n*m complexity, NFC.
Alexey Bataev [Fri, 20 May 2022 12:31:03 +0000 (05:31 -0700)]
[SLP][NFC]Use SmallPtrSet to avoid n*m complexity, NFC.

2 years ago[clangd] Support UnresolvedUsingTypeLoc AST node in FindTarget.
Haojian Wu [Mon, 16 May 2022 12:46:09 +0000 (14:46 +0200)]
[clangd] Support UnresolvedUsingTypeLoc AST node in FindTarget.

to make features like hover, go-to-def work when the cursor is on the
UnresolvedUsingTypeLoc.

Differential Revision: https://reviews.llvm.org/D125684

2 years ago[AArch64] Generate AND in place of CSEL for predicated CTTZ
Rahul Anand R [Fri, 20 May 2022 12:41:32 +0000 (13:41 +0100)]
[AArch64] Generate AND in place of CSEL for predicated CTTZ

This patch implements a for a target specific optimization that replaces
the cmp and csel from cttz with an and mask.

Recommitted with a fix for truncated value sizes.

Differential Revision: https://reviews.llvm.org/D123782

2 years ago[RISCV] Fix logic for determining RVV stack padding
Fraser Cormack [Thu, 19 May 2022 10:03:36 +0000 (11:03 +0100)]
[RISCV] Fix logic for determining RVV stack padding

We must add padding when using SP or BP to access stack objects.
Checking whether we're missing FP is not sufficient as stack realignment
uses SP too. The test in D125962 explains the specific issue in more
detail.

Split from D125787.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D125964

2 years ago[RISCV] Add a test showing overlapping stack offsets with RVV
Fraser Cormack [Thu, 19 May 2022 09:58:17 +0000 (10:58 +0100)]
[RISCV] Add a test showing overlapping stack offsets with RVV

This test (and its forthcoming fix) was split off from D125787. It shows
that the logic we use to determine when we need to add extra RVV padding
is insufficient.

In this example, we may have a situation involving dynamic stack
alignment -- but no variable-sized objects -- where we have no FP but
must still use SP to index objects. In this case we also need the
extra RVV padding, otherwise objects may overlap. Specifically, the test
shows that the RVV vector object may clobber the lowest callee-save.

    |------------------------------| -- <-- Incoming SP
    | 4-byte callee-save (ra)      |
    |------------------------------| -- <-- SP + VLENB*2 + 60
    | 4-byte callee-save (s0)      |
    |------------------------------| -- <-- SP + VLENB*2 + 56  --
    | 4-byte callee-save (s9)      |                            |
    |------------------------------| -- <-- SP + VLENB*2 + 52   | RVV object(!!)
    | VLENB*2 RVV object           |                            |
    |------------------------------| -- <-- SP + 56            --
    | 4-byte local object          |
    |------------------------------| -- <-- SP + 32
    | Dead area                    |
    |------------------------------| -- <-- InSP - 2*VLENB - 64
    | Possibly-zero realignment    |
    |------------------------------| -- <-- SP (realigned to 32)

This diagram should help show that when SP==InSP -- e.g., when the incoming SP
is 32-byte aligned, subtracting 2*VLENB+64 may keep it that way -- the RVV
object clobbers the spill of s9.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D125962

2 years agoMinutes for pauth sync-ups have moved to Discourse.
Kristof Beyls [Fri, 20 May 2022 12:00:53 +0000 (14:00 +0200)]
Minutes for pauth sync-ups have moved to Discourse.

2 years ago[AMDGPU] Add a test case for an SIFoldOperands bug
Jay Foad [Fri, 20 May 2022 11:19:59 +0000 (12:19 +0100)]
[AMDGPU] Add a test case for an SIFoldOperands bug

2 years agotsan: add lock free stack pattern test
Alexey Katranov [Fri, 20 May 2022 09:03:54 +0000 (11:03 +0200)]
tsan: add lock free stack pattern test

Add a set of tests that iterate over possible combinations of
memory orders for lock free stack implementation.

Reviewed By: dvyukov

Differential Revision: https://reviews.llvm.org/D110552

2 years ago[X86][AMX][NFC] Refactor X86LowerAMXCast.cpp
Luo, Yuanke [Fri, 20 May 2022 10:26:13 +0000 (18:26 +0800)]
[X86][AMX][NFC] Refactor X86LowerAMXCast.cpp

Change static function to X86LowerAMXCast member function.

Differential Revision: https://reviews.llvm.org/D126058

2 years ago[AMDGPU][MC][GFX8+] Correct SMEM offset parsing
Dmitry Preobrazhensky [Fri, 20 May 2022 10:58:54 +0000 (13:58 +0300)]
[AMDGPU][MC][GFX8+] Correct SMEM offset parsing

Differential Revision: https://reviews.llvm.org/D125907

2 years ago[mlir] do not elide dialect prefix for ops with dots in the name
Alex Zinenko [Thu, 19 May 2022 14:13:51 +0000 (16:13 +0200)]
[mlir] do not elide dialect prefix for ops with dots in the name

For the hypothetical "a.b.c" op printed within a region that declares "a" as
the default dialect, MLIR would currently elide the "a." prefix and only print
"b.c". However, this becomes ambiguous while parsing as "b.c" may be exist as
the "c" op in the "b" dialect. If it does not, the parsing currently fails. Do
not elide the default dialect if the op name contains further dots to avoid the
ambiguity.

See https://discourse.llvm.org/t/dropping-dialect-prefix-for-ops-with-multiple-dots-in-the-name/62562

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D125975

2 years ago[ARM] Cost modelling for MVE vector fptoi_sat
David Green [Fri, 20 May 2022 10:00:34 +0000 (11:00 +0100)]
[ARM] Cost modelling for MVE vector fptoi_sat

Building on top of D125665, this adds MVE costs for fptosi.sat and
fptoui.sat, providing MVE is available and the types are legal.

Differential Revision: https://reviews.llvm.org/D125666

2 years ago[AArch64][SME]Tied up ZA operand for accumulate instructions
Caroline Concatto [Thu, 12 May 2022 12:19:50 +0000 (13:19 +0100)]
[AArch64][SME]Tied up ZA operand for accumulate instructions

This patch updates SMEInstrFormats.td to tie up ZA operand for instructions
that accumulate their results into ZA or use part of ZA as input.

Depends on: D125534

Differential Revision: https://reviews.llvm.org/D125537

2 years ago[AArch64][SME][NFC] Add implicit operands for SME instructions in the disassembly.
Caroline Concatto [Fri, 13 May 2022 10:04:08 +0000 (11:04 +0100)]
[AArch64][SME][NFC] Add implicit operands for SME instructions in the disassembly.

This patch simplifies the switch statement in getInstruction to add
implicit operands (register ZA and Immediate  equal to zero)
in the SME operands when disassembly.

The register ZA and the zero immediate  can be added by checking the operand
in MCInstDesc.

Differential Revision: https://reviews.llvm.org/D125534

2 years ago[flang] Fix use-associated false-positive error
Daniil Dudkin [Fri, 20 May 2022 09:11:36 +0000 (12:11 +0300)]
[flang] Fix use-associated false-positive error

For the program provided as the test case flang fired the following
error:

    error: Semantic errors in main.f90
    error: 'foo' is not a procedure

This change fixes the error by postponing handling of `UseErrorDetails`
from `CharacterizeProcedure` to a later stage.

Reviewed By: kiranchandramohan

Differential Revision: https://reviews.llvm.org/D125791

2 years ago[LV] Do not LoopSimplify/LCSSA after generating main vector loop.
Florian Hahn [Fri, 20 May 2022 08:58:40 +0000 (09:58 +0100)]
[LV] Do not LoopSimplify/LCSSA after generating main vector loop.

At the moment LV runs LoopSimplify and reconstructs LCSSA form after
generating the main vector loop and before generating the epilogue
vector loop.

In practice, this adds a new exit block for the scalar loop because the
middle block now also branches to the original exit block of the scalar
loop. It also requires adding a new LCSSA phi in the newly created exit
block.

This complicates things when modeling exit values in VPlan, because we
would need to update the VPlan for the epilogue loop to update the newly
created LCSSA phi node.

But none of that should be necessary, as all analysis requiring
loop-simplify form is already done at this point and LCSSA form of the
original loop is not broken.

Reviewed By: bmahjour

Differential Revision: https://reviews.llvm.org/D125810

2 years ago[NFC][test] Fix the line num of expected-error for CSKY at builtin-alloca-with-align.c
Zi Xuan Wu [Fri, 20 May 2022 08:54:23 +0000 (16:54 +0800)]
[NFC][test] Fix the line num of expected-error for CSKY at builtin-alloca-with-align.c

2 years ago[AArch64] Fix the generation of BE Nops
David Green [Fri, 20 May 2022 08:31:00 +0000 (09:31 +0100)]
[AArch64] Fix the generation of BE Nops

Big endian Nops were being generated as d5 03 20 1f   fnmadd  s21, s30,
s0, s0, getting the bytes of the NOP in the wrong order. This switches
the bytes to not be dependant on the endianness.

Differential Revision: https://reviews.llvm.org/D125980

2 years ago[amdgpu] Add amdgpu_kernel calling conv attribute to clang
Jon Chesterfield [Fri, 20 May 2022 07:50:36 +0000 (08:50 +0100)]
[amdgpu] Add amdgpu_kernel calling conv attribute to clang

Allows emitting define amdgpu_kernel void @func() IR from C or C++.

This replaces the current workflow which is to write a stub in opencl that
calls an external C function implemented in C++ combined through llvm-link.

Calling the resulting function still requires a manual implementation of the
ABI from the host side. The primary application is for more rapid debugging
of the amdgpu backend by permuting a C or C++ test file instead of manually
updating an IR file.

Implementation closely follows D54425. Non-amd reviewers from there.

Reviewed By: yaxunl

Differential Revision: https://reviews.llvm.org/D125970

2 years agoMake CompoundStmtBitfields::NumStmts not a bit-field
Serge Pavlov [Sun, 15 May 2022 15:35:46 +0000 (22:35 +0700)]
Make CompoundStmtBitfields::NumStmts not a bit-field

Number of statements in CompoundStmt is kept in a bit-field of the common
part of Stmt. The field has 24 bits for the number. To allocate a new
bit field (as attempted in https://reviews.llvm.org/D123952), this
number must be reduced, maximal number of statements in a compound
statement becomes smaller. It can result in compilation errors of some
programs.

With this change the number of statements is kept in a field of type
'unsigned int' rather than in bit-field. To make room in CompoundStmtBitfields
LBraceLoc is moved to fields of CompoundStmt.

Differential Revision: https://reviews.llvm.org/D125635

2 years ago[flang][OpenMP] Fix the types of worksharing-loop variables
Peixin-Qiao [Fri, 20 May 2022 07:16:03 +0000 (15:16 +0800)]
[flang][OpenMP] Fix the types of worksharing-loop variables

The types of lower bound, upper bound, and step are converted into the
type of the loop variable if necessary. OpenMP runtime requires 32-bit
or 64-bit loop variables. OpenMP loop iteration variable cannot have
more than 64 bits size and will be narrowed.

This patch is part of upstreaming code from the fir-dev branch of
https://github.com/flang-compiler/f18-llvm-project. (#1256)

Co-authored-by: kiranchandramohan <kiranchandramohan@gmail.com>
Reviewed By: kiranchandramohan, shraiysh

Differential Revision: https://reviews.llvm.org/D125740

2 years ago[RISCV] Add VL patterns for vector widening floating-point fused multiply-add instruc...
jacquesguan [Wed, 27 Apr 2022 07:20:31 +0000 (07:20 +0000)]
[RISCV] Add VL patterns for vector widening floating-point fused multiply-add instructions.

This patch adds VL patterns for vector widening floating-point fused multiply-add instructions to support fixed length vector type.

Differential Revision: https://reviews.llvm.org/D124505

2 years ago[RISCV][NFC] Remove `*=` operator for LMULType
eopXD [Fri, 20 May 2022 06:23:34 +0000 (23:23 -0700)]
[RISCV][NFC] Remove `*=` operator for LMULType

LMULType always manipulate on Log2LMUL, let all manipulations go
through LMULType::MulLog2LMUL.

Reviewed By: khchen

Differential Revision: https://reviews.llvm.org/D126042

2 years ago[MIR] Provide location of extra instruction operand when diagnosing it.
Ivan Kosarev [Fri, 20 May 2022 04:55:36 +0000 (05:55 +0100)]
[MIR] Provide location of extra instruction operand when diagnosing it.

Also resolves misspelled FileCheck directives caught with D125604.

Reviewed By: foad

Differential Revision: https://reviews.llvm.org/D125965

2 years ago[lldb] Update test_software_breakpoint_set_and_remove_work for AS
Jonas Devlieghere [Thu, 19 May 2022 20:38:18 +0000 (13:38 -0700)]
[lldb] Update test_software_breakpoint_set_and_remove_work for AS

On Apple Silicon the platform arch is arm64 rather than AArch64.

2 years ago[lldb] Fix 'ptsname_r' is only available on macOS 10.13.4 or newer
Jonas Devlieghere [Thu, 19 May 2022 20:07:20 +0000 (13:07 -0700)]
[lldb] Fix 'ptsname_r' is only available on macOS 10.13.4 or newer

A deployment target less than 10.13.4 causes an error saying that
'ptsname_r' is only available on macOS 10.13.4 or newer. The current
logic only checks if the symbol is available and doesn't account for the
deployment target. This patch fixes that by adding an availability
check.

Differential revision: https://reviews.llvm.org/D125995

2 years ago[RISCV] Add test showing codegen for unaligned loads and stores of scalar types
Philip Reames [Fri, 20 May 2022 03:59:26 +0000 (20:59 -0700)]
[RISCV] Add test showing codegen for unaligned loads and stores of scalar types

2 years ago[ASan] Add sleep_before_init flag
Julian Lettner [Fri, 20 May 2022 00:46:42 +0000 (17:46 -0700)]
[ASan] Add sleep_before_init flag

Also do a little bit of refactoring instead of just copy&paste.

Differential Revision: https://reviews.llvm.org/D126037

2 years ago[InstCombine] [NFC] Use a pattern matcher for ExtractElementInst
Chenbing Zheng [Fri, 20 May 2022 01:56:33 +0000 (09:56 +0800)]
[InstCombine] [NFC] Use a pattern matcher for ExtractElementInst

Reviewed By: RKSimon, rampitec

Differential Revision: https://reviews.llvm.org/D125857

2 years ago[lit] Fix setup of sanitizer environment
Vitaly Buka [Fri, 20 May 2022 02:22:02 +0000 (19:22 -0700)]
[lit] Fix setup of sanitizer environment

Not all options were propageted into tests.

Reviewed By: ychen

Differential Revision: https://reviews.llvm.org/D122869

2 years ago[mlir][Arithmetic] fold overlapping negf.
jacquesguan [Thu, 19 May 2022 09:25:22 +0000 (09:25 +0000)]
[mlir][Arithmetic] fold overlapping negf.

This patch folds negf(negf(x)) to x.

Differential Revision: https://reviews.llvm.org/D125955

2 years ago[AArch64] Add support for -fzero-call-used-regs
Bill Wendling [Thu, 19 May 2022 23:57:40 +0000 (16:57 -0700)]
[AArch64] Add support for -fzero-call-used-regs

Support the "-fzero-call-used-regs" option on AArch64. This involves much less
specialized code than the X86 version. Most of the checks can be done with
TableGen.

Reviewed By: nickdesaulniers, MaskRay

Differential Revision: https://reviews.llvm.org/D124836

2 years ago[mlir][sparse] Adding x-macros for OverheadType
wren romano [Thu, 19 May 2022 23:00:39 +0000 (16:00 -0700)]
[mlir][sparse] Adding x-macros for OverheadType

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D126026

2 years ago[Serialization] Delta encode locations in expansion sloc entries
Sam McCall [Thu, 19 May 2022 08:01:44 +0000 (10:01 +0200)]
[Serialization] Delta encode locations in expansion sloc entries

This is a 1.9% reduction in PCH size in my measurements.

In abbreviated records, VBR6 seems to be slightl better than VBR8 for locations
that may be delta-encoded (i.e. not the first)

Differential Revision: https://reviews.llvm.org/D125952

2 years ago[bazel][libc] Fix bazel build
Alex Brachet [Thu, 19 May 2022 22:57:59 +0000 (22:57 +0000)]
[bazel][libc] Fix bazel build

Differential revision: https://reviews.llvm.org/D126028

2 years ago[mlir][sparse] Factored out a "FATAL" macro for unrecoverable assertion failure
wren romano [Thu, 19 May 2022 22:01:23 +0000 (15:01 -0700)]
[mlir][sparse] Factored out a "FATAL" macro for unrecoverable assertion failure

Depends On D126019

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D126022

2 years ago[TableGen] Add generation of argument register lists
Bill Wendling [Thu, 19 May 2022 20:45:43 +0000 (13:45 -0700)]
[TableGen] Add generation of argument register lists

There are cases, like with -fzero-call-used-regs,  where we need to know
which registers can be used by a certain calling convention. This change
generates a list of registers used by each calling convention defined in
*CallingConv.td.

Calling conventions that use registers conditioned on Swift have those
registers placed in a separate list. This allows us to be flexible about
whether to use the Swift registers or not.

Reviewed By: nickdesaulniers

Differential Revision: https://reviews.llvm.org/D125421

2 years agoRevert "[TableGen] Add generation of argument register lists"
Bill Wendling [Thu, 19 May 2022 22:16:08 +0000 (15:16 -0700)]
Revert "[TableGen] Add generation of argument register lists"

There are build bot failures.

This reverts commit 3fa1b6557d08a148ef853c2a761f1c43e09fef5e.

2 years ago[mlir][sparse] Simplifying closure
wren romano [Thu, 19 May 2022 21:04:35 +0000 (14:04 -0700)]
[mlir][sparse] Simplifying closure

By closing over the `rank` itself rather than `this`, we save a method call on each iteration.  A minor optimization, but one that adds up.

Depends On D126016

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D126019

2 years ago[mlir][sparse] Using the name "dimSizes" more consistently
wren romano [Thu, 19 May 2022 20:47:36 +0000 (13:47 -0700)]
[mlir][sparse] Using the name "dimSizes" more consistently

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D126016

2 years ago[TableGen] Add generation of argument register lists
Bill Wendling [Thu, 19 May 2022 20:45:43 +0000 (13:45 -0700)]
[TableGen] Add generation of argument register lists

There are cases, like with -fzero-call-used-regs,  where we need to know
which registers can be used by a certain calling convention. This change
generates a list of registers used by each calling convention defined in
*CallingConv.td.

Calling conventions that use registers conditioned on Swift have those
registers placed in a separate list. This allows us to be flexible about
whether to use the Swift registers or not.

Reviewed By: nickdesaulniers

Differential Revision: https://reviews.llvm.org/D125421

2 years ago[DirectX backend] When cleanup module flags only remove unused flags.
python3kgae [Tue, 17 May 2022 23:57:18 +0000 (16:57 -0700)]
[DirectX backend] When cleanup module flags only remove unused flags.

Only remove dx.valver from module flags when cleanup module flags in DXILTranslateMetadataPass.

Reviewed By: beanz

Differential Revision: https://reviews.llvm.org/D125842

2 years ago[libc] Add strlcat
Alex Brachet [Thu, 19 May 2022 21:47:28 +0000 (21:47 +0000)]
[libc] Add strlcat

Differential Revision: https://reviews.llvm.org/D125978

2 years ago[lldb/test] Fix PExpect.launch issue when disabling color support
Med Ismail Bennani [Thu, 19 May 2022 21:47:04 +0000 (14:47 -0700)]
[lldb/test] Fix PExpect.launch issue when disabling color support

This patch should fix a bug in PExpect.launch that happened when color
support is not enabled.

In that case, we need to add the `--no-use-colors` flag to lldb's launch
argument list. However, previously, each character to the string was
appended separately to the `args` list. This patch solves that by adding
the whole string to the list.

This should fix the TestIOHandlerResize failure on GreenDragon.

Differential Revision: https://reviews.llvm.org/D126021

Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
2 years agoRevert "[RISCV] Add test cases showing failure to remove mask on rotate amounts."
Craig Topper [Thu, 19 May 2022 21:33:02 +0000 (14:33 -0700)]
Revert "[RISCV] Add test cases showing failure to remove mask on rotate amounts."

This reverts commit e2f410feeab27a8bb2c015fc02bb8527702e401f.

This exposes a pre-existing bug in type legalization that is failing
expensive checks.

2 years agoRevert "[RISCV] Use selectShiftMaskXLen ComplexPattern for isel of rotates."
Craig Topper [Thu, 19 May 2022 21:32:15 +0000 (14:32 -0700)]
Revert "[RISCV] Use selectShiftMaskXLen ComplexPattern for isel of rotates."

This reverts commit 86f7d7074a0129955aa2f5c82fe8c383eb17a35a.

The test cases added for this exposed an pre-existing bug that is failing
the expensive checks bot. Reverting so I can revert that patch.

2 years ago[ConstantRange] Improve the implementation of binaryOr
Alexander Shaposhnikov [Thu, 19 May 2022 20:25:47 +0000 (20:25 +0000)]
[ConstantRange] Improve the implementation of binaryOr

This diff adjusts binaryOr to take advantage of the analysis
based on KnownBits.

Differential revision: https://reviews.llvm.org/D125933

Test plan:
1/ ninja check-llvm
2/ ninja check-llvm-unit

2 years ago[bazel] Add lib/Basic/BuiltinTargetFeatures.h to clang:basic `hdrs`.
Jorge Gorbe Moya [Thu, 19 May 2022 21:17:36 +0000 (14:17 -0700)]
[bazel] Add lib/Basic/BuiltinTargetFeatures.h to clang:basic `hdrs`.

This header is included by
clang/unittests/CodeGen/CheckTargetFeaturesTest.cpp
so it needs to be exposed here to make it visible.

2 years ago[Flang][OpenMP] Upstream the lowering of the parallel do combined construct
Kiran Chandramohan [Thu, 19 May 2022 20:23:04 +0000 (20:23 +0000)]
[Flang][OpenMP] Upstream the lowering of the parallel do combined construct

When parallel is used in a combined construct, then use a separate
function to create the parallel operation. It handles the parallel
specific clauses and leaves the rest for handling at the inner
operations.

Reviewed By: peixin, shraiysh

Differential Revision: https://reviews.llvm.org/D125465

Co-authored-by: Sourabh Singh Tomar <SourabhSingh.Tomar@amd.com>
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Valentin Clement <clementval@gmail.com>
Co-authored-by: Nimish Mishra <neelam.nimish@gmail.com>
2 years agoHandle instrumentation of scalar single-precision (_ss) intrinsics
Nicolas Capens [Thu, 19 May 2022 20:46:49 +0000 (13:46 -0700)]
Handle instrumentation of scalar single-precision (_ss) intrinsics

Instrumentation of scalar double-precision intrinsics such as
x86_sse41_round_sd was already handled by https://reviews.llvm.org/D82398,
but not their single-precision counterparts.

https://issuetracker.google.com/172238865

Reviewed By: vitalybuka

Differential Revision: https://reviews.llvm.org/D124871

2 years ago[mlir][sparse] fix unsigned comparison bug in assert
Aart Bik [Thu, 19 May 2022 19:23:52 +0000 (12:23 -0700)]
[mlir][sparse] fix unsigned comparison bug in assert

Reviewed By: bixia, wrengr

Differential Revision: https://reviews.llvm.org/D126007

2 years ago[AMDGPU] Mark s_get_waveid_in_workgroup as not reading memory
Jay Foad [Thu, 19 May 2022 15:52:41 +0000 (16:52 +0100)]
[AMDGPU] Mark s_get_waveid_in_workgroup as not reading memory

It is already marked as having side effects, at least in MIR. It does
not interact with anything else that is modelled as a memory access
either in IR or MachineIR.

Differential Revision: https://reviews.llvm.org/D125985

2 years ago[AMDGPU] Mark s_getreg as having side effects instead of reading memory
Jay Foad [Thu, 19 May 2022 12:43:43 +0000 (13:43 +0100)]
[AMDGPU] Mark s_getreg as having side effects instead of reading memory

s_getreg does not interact with anything else that is modelled as a
memory access either in IR or MachineIR.

Differential Revision: https://reviews.llvm.org/D125968

2 years ago[mlir] Remove unused properties from the standalone example's lit configuration
Stella Stamenova [Thu, 19 May 2022 19:51:37 +0000 (12:51 -0700)]
[mlir] Remove unused properties from the standalone example's lit configuration

Since these are unused, I've removed them from the configuration, so that it can be easier to read and follow.

Reviewed By: stellaraccident

Differential Revision: https://reviews.llvm.org/D125132

2 years ago[Clang][[OpenMP5.1] Initial parser/sema for default(private) clause
Jennifer Yu [Tue, 17 May 2022 21:17:32 +0000 (14:17 -0700)]
[Clang][[OpenMP5.1] Initial parser/sema for default(private) clause

This implements the default(private) clause as defined in OMP5.1

Differential Revision: https://reviews.llvm.org/D125912

2 years ago[LV] Drop wrap flags for reductions using VP def-use chain.
Florian Hahn [Thu, 19 May 2022 19:36:46 +0000 (20:36 +0100)]
[LV] Drop wrap flags for reductions using VP def-use chain.

Update clearReductionWrapFlags to use the VPlan def-use chain from the
reduction phi recipe to drop reduction wrap flags.

This addresses an existing FIXME and fixes a crash when instructions in
the reduction chain are not used and have been removed before VPlan
codegeneration.

Fixes #55540.

2 years ago[gn build] (manually) port 505ddb6b7450 (remove Unit/lit.site.cfg.py)
Nico Weber [Thu, 19 May 2022 19:18:56 +0000 (15:18 -0400)]
[gn build] (manually) port 505ddb6b7450 (remove Unit/lit.site.cfg.py)

2 years ago[lld][test] Delete empty Unit test directory
Keith Smiley [Thu, 12 May 2022 00:12:03 +0000 (17:12 -0700)]
[lld][test] Delete empty Unit test directory

This became empty when we removed the legacy macho lld. This results in
a warning when running `check-lld`. We can revert this in the future if
we want unit tests.

Differential Revision: https://reviews.llvm.org/D125436