platform/upstream/llvm.git
3 years agoRevert STRICT_FCMP nonan optimisation
Thomas Preud'homme [Tue, 9 Feb 2021 11:27:35 +0000 (11:27 +0000)]
Revert STRICT_FCMP nonan optimisation

Summary: This reverts commit b7b61a7b5bc63df0d84f3722a1dcfa375c35ba30 which fails on some of the builders: http://lab.llvm.org:8011/#/builders/14/builds/5806

Reviewers:

Subscribers:

3 years agoImprove STRICT_FSETCC codegen in absence of no NaN
Thomas Preud'homme [Thu, 19 Nov 2020 23:38:42 +0000 (23:38 +0000)]
Improve STRICT_FSETCC codegen in absence of no NaN

As for SETCC, use a less expensive condition code when generating
STRICT_FSETCC if the node is known not to have Nan.

Reviewed By: SjoerdMeijer

Differential Revision: https://reviews.llvm.org/D91972

3 years ago[NFC,Clang] Add more Asan Driver tests
Vitaly Buka [Tue, 9 Feb 2021 10:59:59 +0000 (02:59 -0800)]
[NFC,Clang] Add more Asan Driver tests

3 years ago[NFC,Clang] Add SanCov Driver tests
Vitaly Buka [Tue, 9 Feb 2021 10:56:57 +0000 (02:56 -0800)]
[NFC,Clang] Add SanCov Driver tests

3 years ago[NFC,Clang] Add LTO Driver MSan,KMsan tests
Vitaly Buka [Tue, 9 Feb 2021 10:55:33 +0000 (02:55 -0800)]
[NFC,Clang] Add LTO Driver MSan,KMsan tests

3 years ago[NFC,Clang] Add LTO Driver DFsan tests
Vitaly Buka [Tue, 9 Feb 2021 10:51:19 +0000 (02:51 -0800)]
[NFC,Clang] Add LTO Driver DFsan tests

3 years ago[NFC,Clang] Add LTO Driver Tsan tests
Vitaly Buka [Tue, 9 Feb 2021 10:49:51 +0000 (02:49 -0800)]
[NFC,Clang] Add LTO Driver Tsan tests

3 years ago[clangd] Fix false positive in local rename collision detetction
Kirill Bobyrev [Tue, 9 Feb 2021 10:46:22 +0000 (11:46 +0100)]
[clangd] Fix false positive in local rename collision detetction

Fixes https://github.com/clangd/clangd/issues/685

Reviewed By: hokein

Differential Revision: https://reviews.llvm.org/D96247

3 years ago[-Wcompletion-handler][NFC] Remove unexpected warnings on Windows
Valeriy Savchenko [Tue, 9 Feb 2021 10:48:24 +0000 (13:48 +0300)]
[-Wcompletion-handler][NFC] Remove unexpected warnings on Windows

3 years ago[clang][cli] Generate and round-trip CodeGen options
Jan Svoboda [Tue, 9 Feb 2021 10:42:01 +0000 (11:42 +0100)]
[clang][cli] Generate and round-trip CodeGen options

This patch implements generation of remaining codegen options and tests it by performing parse-generate-parse round trip.

Reviewed By: dexonsmith

Differential Revision: https://reviews.llvm.org/D96056

3 years ago[clang][codegen] Remember string used to create llvm::Regex for optimization remarks
Jan Svoboda [Tue, 9 Feb 2021 10:06:03 +0000 (11:06 +0100)]
[clang][codegen] Remember string used to create llvm::Regex for optimization remarks

Regular expression patterns passed through the command line are being used to create an instances of `llvm::Regex` and thrown away.

There is no API to serialize `Regex` back to the original pattern. This means we have no way to reconstruct the original pattern from command line. This is necessary for serializing `CompilerInvocation`.

This patch stores the original pattern string in `CodeGenOptions` alongside the `llvm::Regex` instance.

Reviewed By: dexonsmith, thegameg

Differential Revision: https://reviews.llvm.org/D96036

3 years agoRevert "[DebugInfo] Re-engineer a test to be stricter, add XFails"
Jeremy Morse [Tue, 9 Feb 2021 09:48:22 +0000 (09:48 +0000)]
Revert "[DebugInfo] Re-engineer a test to be stricter, add XFails"

This reverts commit e05c10380ce7c18d1a232dcc5baa7c10d8bd2bf6.

See parent commit, there's a bot which isn't captured in the XFail list,
reverting til I work out what it is.

3 years agoRevert "Follow up to e05c10380ce7: add aarch64 to test XFails"
Jeremy Morse [Tue, 9 Feb 2021 09:46:32 +0000 (09:46 +0000)]
Revert "Follow up to e05c10380ce7: add aarch64 to test XFails"

This reverts commit 4fd29e4fd3c5490b658f4b467d94aeed6b3d1ac3.

There's a report in D95617 that this is failing on what (I think?) is an
aarch64 bot, which should be covered by the XFail list... reverting this
follow-up and the base patch until I work out what's wrong here.

3 years ago[gn build] Port 40c261c41c4c
LLVM GN Syncbot [Tue, 9 Feb 2021 09:19:31 +0000 (09:19 +0000)]
[gn build] Port 40c261c41c4c

3 years ago[clang][cli] Generate and round-trip language options
Jan Svoboda [Tue, 9 Feb 2021 09:17:04 +0000 (10:17 +0100)]
[clang][cli] Generate and round-trip language options

This patch implements generation of remaining language options and tests it by performing parse-generate-parse round trip (on by default for assert builds, off otherwise).

This patch also correctly reports failures in `parseSanitizerKinds`, which is necessary for emitting diagnostics when an invalid sanitizer is passed to `-fsanitize=` during round-trip.

This patch also removes TableGen marshalling classes from two options:
* `fsanitize_blacklist` When parsing: it's first initialized via the generated code, but then also changed by manually written code, which is confusing.
* `fopenmp` When parsing: it's first initialized via generated code, but then conditionally changed by manually written code. This is also confusing. Moreover, we need to do some extra checks when generating it, which would be really cumbersome in TableGen. (Specifically, not emitting it when `-fopenmp-simd` was present.)

Reviewed By: dexonsmith

Differential Revision: https://reviews.llvm.org/D95793

3 years ago[ELF] Drop Android specific workaround -m aarch64_elf64_le_vec
Fangrui Song [Tue, 9 Feb 2021 08:43:10 +0000 (00:43 -0800)]
[ELF] Drop Android specific workaround -m aarch64_elf64_le_vec

`extern const bfd_target aarch64_elf64_le_vec;` is a variable in BFD.
It was somehow misused as an emulation by Android.

```
% aarch64-linux-gnu-ld -m aarch64_elf64_le_vec a.o
aarch64-linux-gnu-ld: unrecognised emulation mode: aarch64_elf64_le_vec
Supported emulations: aarch64linux aarch64elf aarch64elf32 aarch64elf32b aarch64elfb armelf armelfb aarch64linuxb aarch64linux32 aarch64linux32b armelfb_linux_eabi armelf_linux_eabi
```

Acked by Stephen Hines, who removed the flag from Android a while back.

3 years ago[-Wcompletion-handler] Support checks with builtins
Valeriy Savchenko [Mon, 8 Feb 2021 15:47:21 +0000 (18:47 +0300)]
[-Wcompletion-handler] Support checks with builtins

It is very common to check callbacks and completion handlers for null.
This patch supports such checks using built-in functions:
  * __builtin_expect
  * __builtin_expect_with_probablity
  * __builtin_unpredictable

rdar://73455388

Differential Revision: https://reviews.llvm.org/D96268

3 years ago[ELF] Rewriting the path of sample profile file for --reproduce response.txt
Hongtao Yu [Sat, 6 Feb 2021 01:56:45 +0000 (17:56 -0800)]
[ELF] Rewriting the path of sample profile file for --reproduce response.txt

Rewritting the path of the sample profile file in response.txt to be relative to the repro tar.

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D96193

3 years ago[NFC] [Coroutine] Remove Unused Variables
Chuanqi Xu [Tue, 9 Feb 2021 07:52:11 +0000 (15:52 +0800)]
[NFC] [Coroutine] Remove Unused Variables

3 years ago[RISCV] Use whole register load/store for generic load/store.
Hsiangkai Wang [Tue, 9 Feb 2021 06:43:10 +0000 (14:43 +0800)]
[RISCV] Use whole register load/store for generic load/store.

In vector v0.10, there are whole vector register load/store
instructions. I suggest to use the whole register load/store
instructions for generic load/store for scalable vector types. It could
save up vset{i}vl{i} for these load/store.

For fractional LMUL, I keep to use vle{eew}.v/vse{eew}.v instructions to
load/store partial vector registers.

Differential Revision: https://reviews.llvm.org/D95853

3 years ago[Docs] Fix Typo
Zakk Chen [Tue, 9 Feb 2021 07:43:32 +0000 (23:43 -0800)]
[Docs] Fix Typo

3 years ago[MLIR][AVX512] Add integration test for vp2intersect
Matthias Springer [Tue, 9 Feb 2021 07:43:04 +0000 (16:43 +0900)]
[MLIR][AVX512] Add integration test for vp2intersect

Differential Revision: https://reviews.llvm.org/D96306

3 years ago[test] Drop redundant REQUIRES: x86-registered-target
Fangrui Song [Tue, 9 Feb 2021 07:36:37 +0000 (23:36 -0800)]
[test] Drop redundant REQUIRES: x86-registered-target

3 years ago[test] Add REQUIRES: x86-registered-target to DebugInfo/Symbolize/ELF llvm-mc tests
Fangrui Song [Tue, 9 Feb 2021 07:34:41 +0000 (23:34 -0800)]
[test] Add REQUIRES: x86-registered-target to DebugInfo/Symbolize/ELF llvm-mc tests

3 years agoMark 4 tests added in 6d766c8bf9df as requiring an x86 backend as they fail when...
Douglas Yung [Tue, 9 Feb 2021 06:47:14 +0000 (22:47 -0800)]
Mark 4 tests added in 6d766c8bf9df as requiring an x86 backend as they fail when it is not present.

This should fix buildbot failures like http://lab.llvm.org:8011/#/builders/107/builds/4469

3 years ago[Transforms] Use range-based for loops (NFC)
Kazu Hirata [Tue, 9 Feb 2021 06:33:53 +0000 (22:33 -0800)]
[Transforms] Use range-based for loops (NFC)

3 years ago[TableGen] Use ListSeparator (NFC)
Kazu Hirata [Tue, 9 Feb 2021 06:33:51 +0000 (22:33 -0800)]
[TableGen] Use ListSeparator (NFC)

3 years ago[Transforms/Utils] Drop unnecessary const from a return type (NFC)
Kazu Hirata [Tue, 9 Feb 2021 06:33:49 +0000 (22:33 -0800)]
[Transforms/Utils] Drop unnecessary const from a return type (NFC)

Identified with const-return-type.

3 years agoReturn "[Test] Add failing test for PR49087"
Max Kazantsev [Tue, 9 Feb 2021 04:00:12 +0000 (11:00 +0700)]
Return "[Test] Add failing test for PR49087"

Another attempt, this time with tripple fix.

3 years ago[MLIR] Add C API for navigating up the IR tree
George [Tue, 9 Feb 2021 03:54:19 +0000 (19:54 -0800)]
[MLIR] Add C API for navigating up the IR tree

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D96301

3 years ago[lld][WebAssembly] Fix typo in function name
Sam Clegg [Tue, 9 Feb 2021 01:12:23 +0000 (17:12 -0800)]
[lld][WebAssembly] Fix typo in function name

addOptionalGlobalSymbols should be addOptionalGlobalSymbol.

Also, remove unnecessary additional argument to make the signature match
the sibling function: addOptionalDataSymbol.

Differential Revision: https://reviews.llvm.org/D96305

3 years ago[CUDA][HIP] Add -fuse-cuid
Yaxun (Sam) Liu [Tue, 19 Jan 2021 22:35:23 +0000 (17:35 -0500)]
[CUDA][HIP] Add -fuse-cuid

This patch added a distinct CUID for each input file, which is represented by InputAction.
clang initially creates an InputAction for each input file for the host compilation. In CUDA/HIP action
builder, each InputAction is given a CUID and cloned for each GPU arch, and the CUID is also cloned. In this way,
we guarantee the corresponding device and host compilation for the same file shared the
same CUID. On the other hand, different compilation units have different CUID.

-fuse-cuid=random|hash|none is added to control the method to generate CUID. The default
is hash. -cuid=X is also added to specify CUID explicitly, which overrides -fuse-cuid.

Reviewed by: Artem Belevich

Differential Revision: https://reviews.llvm.org/D95007

3 years ago[lldb] Fix crash in FormatEntity for mangled-name
Dave Lee [Tue, 9 Feb 2021 01:33:53 +0000 (17:33 -0800)]
[lldb] Fix crash in FormatEntity for mangled-name

Check a `Block` pointer before dereferencing.

Using `function.mangled-name` led to a crash for a frame where the symbol
context had no block info. In my case, the frame's function was a system frame.

Differential Revision: https://reviews.llvm.org/D96307

3 years agoRevert "[CostModel] Remove VF from IntrinsicCostAttributes"
Jinsong Ji [Tue, 9 Feb 2021 02:12:54 +0000 (02:12 +0000)]
Revert "[CostModel] Remove VF from IntrinsicCostAttributes"

This reverts commit 502a67dd7f23901834e05071ab253889f671b5d9.

This expose a failure in test-suite build on PowerPC,
revert to unblock buildbot first,
Dave will re-commit in https://reviews.llvm.org/D96287.

Thanks Dave.

3 years agoPR48606: The lifetime of a constexpr heap allocation always started
Richard Smith [Tue, 9 Feb 2021 01:58:05 +0000 (17:58 -0800)]
PR48606: The lifetime of a constexpr heap allocation always started
during the same evaluation.

It looks like the only case for which this matters is determining
whether mutable subobjects of a heap allocation can be modified during
constant evaluation.

3 years agoPR48587: is_constant_evaluated() should not evaluate to true during a
Richard Smith [Tue, 9 Feb 2021 01:32:52 +0000 (17:32 -0800)]
PR48587: is_constant_evaluated() should not evaluate to true during a
variable's destruction if it didn't do so during construction.

The standard doesn't give any guidance as to what to do here, but this
approach seems reasonable and conservative, and has been proposed to the
standard committee.

3 years ago[llvm-objcopy][test] Stablize build-id-link-dir.test
Fangrui Song [Tue, 9 Feb 2021 01:22:22 +0000 (17:22 -0800)]
[llvm-objcopy][test] Stablize build-id-link-dir.test

3 years ago[gn build] Port 87104faac433
LLVM GN Syncbot [Tue, 9 Feb 2021 01:14:44 +0000 (01:14 +0000)]
[gn build] Port 87104faac433

3 years ago[lld-macho] Add ARM64 target arch
Greg McGary [Sat, 26 Sep 2020 20:00:22 +0000 (13:00 -0700)]
[lld-macho] Add ARM64 target arch

This is an initial base commit for ARM64 target arch support. I don't represent that it complete or bug-free, but wish to put it out for review now that some basic things like branch target & load/store address relocs are working.

I can add more tests to this base commit, or add them in follow-up commits.

It is not entirely clear whether I use the "ARM64" (Apple) or "AArch64" (non-Apple) naming convention. Guidance is appreciated.

Differential Revision: https://reviews.llvm.org/D88629

3 years ago[MC][WebAssembly] Fix provisional values for data alias relocations
Sam Clegg [Mon, 8 Feb 2021 02:51:42 +0000 (18:51 -0800)]
[MC][WebAssembly] Fix provisional values for data alias relocations

When calculating the symbol offsets to write as provisitonal values
in object files we are only interested in the offset of the symbol
itself.  For aliases this offset already includes the offset of the
base symbol.

The testin question was added back in https://reviews.llvm.org/D87407
but I believe the expectations here were incorrect.   sym_a lives
at offset 4 and sym_b lives 4 bytes into that (should be 8).

The addresses of the 3 symbosl in this object file are:

foo  : 0
sym_a: 4
sym_b: 8

Differential Revision: https://reviews.llvm.org/D96234

3 years ago[TableGen] Use return value from EmitVBRValue instead of calling GetVBRSize on the...
Craig Topper [Tue, 9 Feb 2021 00:25:18 +0000 (16:25 -0800)]
[TableGen] Use return value from EmitVBRValue instead of calling GetVBRSize on the same value. Consistently use unsigned for child sizes. NFCI

getSize and setSize both use unsigned. So size_t doesn't
increase range here and might get truncated if passed to
setSize.

Also not sure why EmitVBRValue was returning uint64_t, but used
an unsigned to supply the value.

3 years ago[MLIR][NFC] Fix std.copysign op documentation
Uday Bondhugula [Sun, 7 Feb 2021 06:30:22 +0000 (12:00 +0530)]
[MLIR][NFC] Fix std.copysign op documentation

Fix std.copysign op documentation. NFC.

Differential Revision: https://reviews.llvm.org/D96217

3 years agoFix failure in cuda-external-tools.cu
Yaxun (Sam) Liu [Tue, 9 Feb 2021 00:27:17 +0000 (19:27 -0500)]
Fix failure in cuda-external-tools.cu

-fgpu-rdc is output in different order

3 years ago[SPARC] Recognize and handle the %lm(sym) operator
LemonBoy [Tue, 9 Feb 2021 00:23:38 +0000 (19:23 -0500)]
[SPARC] Recognize and handle the %lm(sym) operator

Reviewed By: joerg

Differential Revision: https://reviews.llvm.org/D77737

3 years agoRevert "Renovate CMake files in the `llvm-exegesis` tool."
Jameson Nash [Tue, 9 Feb 2021 00:11:08 +0000 (19:11 -0500)]
Revert "Renovate CMake files in the `llvm-exegesis` tool."

This reverts commit 549a1e2e59508d4aeaf9a93912b479798954bd5e.

I see some buildbot failures, so reverting while I look into them.

3 years agoMake sure a module file with errors produced via '-fallow-pcm-with-compiler-errors...
Argyrios Kyrtzidis [Mon, 8 Feb 2021 08:24:31 +0000 (00:24 -0800)]
Make sure a module file with errors produced via '-fallow-pcm-with-compiler-errors' can be loaded when using implicit modules

A module with errors would be marked as out-of-date, then the `compilerModule` action would produce it, but due to the error it would be treated as failure and the resulting PCM would not get used.

rdar://74087062

Differential Revision: https://reviews.llvm.org/D96246

3 years ago[CUDA][HIP] Pass -fgpu-rdc to host clang -cc1
Yaxun (Sam) Liu [Mon, 8 Feb 2021 23:51:31 +0000 (18:51 -0500)]
[CUDA][HIP] Pass -fgpu-rdc to host clang -cc1

Currently -fgpu-rdc is not passed to host clang -cc1.
This causes issue because -fgpu-rdc affects shadow
variable linkage in host compilation.

Reviewed by: Artem Belevich

Differential Revision: https://reviews.llvm.org/D96105

3 years ago[flang][fir] Add OpaqueAttr.
Eric Schweitz [Mon, 8 Feb 2021 21:47:16 +0000 (13:47 -0800)]
[flang][fir] Add OpaqueAttr.

Add the opaque attribute class used in flang.

https://github.com/flang-compiler/f18-llvm-project/pull/402

Differential Revision: https://reviews.llvm.org/D96293

3 years agoRenovate CMake file for the `llvm-cfi-verify` tool
Jameson Nash [Mon, 8 Feb 2021 23:20:16 +0000 (18:20 -0500)]
Renovate CMake file for the `llvm-cfi-verify` tool

Hopefully this is the non-problematic part from https://reviews.llvm.org/rL342148, which later got reverted in r342336 (b09a8c9bd9b819741b38071a7ccd95042ef2643a) due to problems with the llvm-exegesis part of the change. That part would also still be desirable, but currently appears not to be possible (https://reviews.llvm.org/D81922).

I think this should replace https://reviews.llvm.org/D44650, per Keno's comment there.

Reviewed By: hctim

Differential Revision: https://reviews.llvm.org/D90969

3 years agoRenovate CMake files in the `llvm-exegesis` tool.
Jameson Nash [Mon, 8 Feb 2021 23:03:23 +0000 (18:03 -0500)]
Renovate CMake files in the `llvm-exegesis` tool.

This attempts to move all tools over to using `add_llvm_library` for
better consistency. After doing this, I noticed it ended up as nearly a
reimplementation of https://reviews.llvm.org/rL342148, which later got
reverted in r342336 (b09a8c9bd9b819741b38071a7ccd95042ef2643a).

With ccache and ninja on a large core machine (40), I haven't run into
build errors, so I'm hopeful it's better now, though it doesn't seem to
be any different / new.

Reviewed By: stephenneuendorffer

Differential Revision: https://reviews.llvm.org/D90970

3 years ago[RISCV] Initial support of LoopVectorizer for RISC-V Vector.
Hsiangkai Wang [Wed, 27 Jan 2021 07:00:46 +0000 (15:00 +0800)]
[RISCV] Initial support of LoopVectorizer for RISC-V Vector.

Define an option -riscv-vector-bits-max to specify the maximum vector
bits for vectorizer. Loop vectorizer will use the value to check if it
is safe to use the whole vector registers to vectorize the loop.

It is not the optimum solution for loop vectorizing for scalable vector.
It assumed the whole vector registers will be used to vectorize the code.
If it is possible, we should configure vl to do vectorize instead of
using whole vector registers.

We only consider LMUL = 1 in this patch.

This patch just an initial work for loop vectorizer for RISC-V Vector.

Differential Revision: https://reviews.llvm.org/D95659

3 years agoGlobalISel: Use correct calling convention in handleAssignments
Matt Arsenault [Mon, 8 Feb 2021 21:53:42 +0000 (16:53 -0500)]
GlobalISel: Use correct calling convention in handleAssignments

This was using the calling convention of the calling function, not the
callee. Avoids regressions in a future patch.

3 years agoAMDGPU: Stop adding stack passed wide arguments to call conv handler
Matt Arsenault [Sun, 7 Feb 2021 17:12:09 +0000 (12:12 -0500)]
AMDGPU: Stop adding stack passed wide arguments to call conv handler

The generated calling convention code shouldn't see these types since
we split large types into 32-bit chunks before the calling convention
code is triggered.

GlobalISel ends up directly calls the generated CC code before
checking for the register count breakdown. Arguably this difference is
a bug, but this was dead code for the DAG anyway.

3 years agoAMDGPU/GlobalISel: Remove dead check prefixes
Matt Arsenault [Sun, 7 Feb 2021 21:15:08 +0000 (16:15 -0500)]
AMDGPU/GlobalISel: Remove dead check prefixes

3 years ago[NVPTX][NewPM] Re-enable NVVMReflectPass
Arthur Eubanks [Mon, 8 Feb 2021 21:52:14 +0000 (13:52 -0800)]
[NVPTX][NewPM] Re-enable NVVMReflectPass

Disabled alongside NVVMIntrRangePass in https://reviews.llvm.org/D96166,
but turns out NVVMIntrRangePass was the issue.

Reviewed By: tra

Differential Revision: https://reviews.llvm.org/D96291

3 years ago[flang][NFC] Add comment.
Eric Schweitz [Mon, 8 Feb 2021 21:49:16 +0000 (13:49 -0800)]
[flang][NFC] Add comment.

3 years ago[test] Fix unused check prefixes
Fangrui Song [Mon, 8 Feb 2021 21:44:29 +0000 (13:44 -0800)]
[test] Fix unused check prefixes

3 years ago[FileCheck] Default --allow-unused-prefixes to false
Fangrui Song [Mon, 8 Feb 2021 21:37:03 +0000 (13:37 -0800)]
[FileCheck] Default --allow-unused-prefixes to false

Link: https://lists.llvm.org/pipermail/llvm-dev/2020-October/146162.html
If a downstream project using lit needs time for transition,
add the following to `lit.local.cfg`:

```
from lit.llvm.subst import ToolSubst

fc = ToolSubst('FileCheck', unresolved='fatal')
config.substitutions.insert(0, (fc.regex, 'FileCheck --allow-unused-prefixes'))
```

Differential Revision: https://reviews.llvm.org/D95849

3 years ago[Verifier] Allow DW_TAG_class_type/DW_TAG_union_type to have no filename
Fangrui Song [Mon, 8 Feb 2021 21:31:05 +0000 (13:31 -0800)]
[Verifier] Allow DW_TAG_class_type/DW_TAG_union_type to have no filename

`clang/lib/CodeGen/CGOpenMPRuntime.cpp` synthesized union
(`distinct !DICompositeType(tag: DW_TAG_union_type, name: "kmp_cmplrdata_t", size: 64, elements: <0x62b690>)`)
does not have meaningful filename/line number.

D94735 dropped the previously arbitrary and untested filename/line from the union and caused a verifier error here.

This fixes `check-libarcher` failures.

Differential Revision: https://reviews.llvm.org/D96212

3 years ago[ARM] One-off identity shuffle
David Green [Mon, 8 Feb 2021 21:24:32 +0000 (21:24 +0000)]
[ARM] One-off identity shuffle

A One-Off Identity mask is a shuffle that is mostly an identity mask
from as single source but contains a single element out-of-place, either
from a different vector or from another position in the same vector. As
opposed to lowering this via a ARMISD::BUILD_VECTOR we can generate an
extract/insert pair directly. Under ARM with individually accessible
lane elements this often becomes a simple lane move.

This also alters the LowerVECTOR_SHUFFLEUsingMovs code to use v4f32 (not
v4i32), a more natural type for lane moves.

Differential Revision: https://reviews.llvm.org/D95551

3 years ago[SimpleLoopUnswitch] Don't non-trivially unswitch loops that are unsafe to clone
Arthur Eubanks [Mon, 8 Feb 2021 20:52:20 +0000 (12:52 -0800)]
[SimpleLoopUnswitch] Don't non-trivially unswitch loops that are unsafe to clone

Non-trivial unswitching can clone loops.

The legacy -loop-unswitch pass also checks for this.

Fixes PR49085.

Reviewed By: asbirlea

Differential Revision: https://reviews.llvm.org/D96288

3 years ago[gn build] reformat all gn files
Nico Weber [Mon, 8 Feb 2021 21:10:59 +0000 (16:10 -0500)]
[gn build] reformat all gn files

$ git ls-files '*.gn' '*.gni' | xargs llvm/utils/gn/gn.py format

3 years ago[CMake] [MinGW] Enable use of LLVM_USE_SANITIZER in a MinGW environment
Markus Böck [Mon, 8 Feb 2021 21:04:04 +0000 (23:04 +0200)]
[CMake] [MinGW] Enable use of LLVM_USE_SANITIZER in a MinGW environment

Currently using LLVM_USE_SANITIZER with a MinGW target leads to a fatal
configuration error due to an unsupported platform. MinGW targets on
clang however implement a few sanitizers, currently ASAN and UBSAN.

This patch enables LLVM_USE_SANITIZER in a MinGW environment as well.

Differential Revision: https://reviews.llvm.org/D95750

3 years ago[AArch64][GlobalISel] Support the 'returned' parameter attribute.
Amara Emerson [Wed, 3 Feb 2021 18:33:48 +0000 (10:33 -0800)]
[AArch64][GlobalISel] Support the 'returned' parameter attribute.

On AArch64 (which seems to be the only target that supports it), this
attribute allows codegen to avoid saving/restoring the value in x0
across a call.

Gives a 0.1% geomean -Os code size improvement on CTMark.

Differential Revision: https://reviews.llvm.org/D96099

3 years agoRevert "[Test] Add failing test for PR49087"
Nico Weber [Mon, 8 Feb 2021 20:46:29 +0000 (15:46 -0500)]
Revert "[Test] Add failing test for PR49087"

This reverts commit 0fc1738eb75d613b9e16143b83e7cb80512e84eb.
The test passes (unexpectedly, due to the XFAIL: *) when x86 isn't
the default triple (such as on an arm machine).

3 years ago[AArch64] Use '//' as comment string for MSVC assembly
Martin Storsjö [Mon, 8 Feb 2021 13:24:42 +0000 (15:24 +0200)]
[AArch64] Use '//' as comment string for MSVC assembly

As the actual MSVC toolset doesn't use the GAS-style assembly that
Clang/LLVM produces and consumes, there's no reference for what
string to use for e.g. comments when building with a MSVC triple.

This frees up the use of semicolon as separator string, just like
was done for GNU targets in 23413195649d0cf6f3860ae8b5fb115b35032075.
(Previously, both the separator and comment strings were set to
the same, a semicolon.)

Compiler-rt extensively uses separator chars in its assembly,
and that assembly should be buildable with clang-cl for MSVC too.

Differential Revision: https://reviews.llvm.org/D96259

3 years agoDebugInfo/Symbolize: Allow STT_NOTYPE/STT_GNU_IFUNC symbols for .symtab symbolization
Fangrui Song [Mon, 8 Feb 2021 20:29:11 +0000 (12:29 -0800)]
DebugInfo/Symbolize: Allow STT_NOTYPE/STT_GNU_IFUNC symbols for .symtab symbolization

In assembly files, omitting `.type foo,@function` is common. Such functions have
type `STT_NOTYPE` and llvm-symbolizer reports `??` for them.

An ifunc symbol usually has an associated resolver symbol which is defined at
the same address. Returning either one is fine for symbolization. The resolver
symbol may not end up in the symbol table if (object file) `.L` is used (linked
image) .symtab is stripped while .dynsym is retained.

This patch allows ELF STT_NOTYPE/STT_GNU_IFUNC symbols for .symtab symbolization.

I have left TODO in the test files for an unimplemented STT_FILE heuristic.

Differential Revision: https://reviews.llvm.org/D95916

3 years ago[scudo/standalone] Use .arch_extension memtag, not mte
Roland McGrath [Thu, 4 Feb 2021 03:28:29 +0000 (19:28 -0800)]
[scudo/standalone] Use .arch_extension memtag, not mte

GNU binutils accepts only `.arch_extension memtag` while Clang
accepts either that or `.arch_extension mte` to mean the same thing.

Reviewed By: pcc

Differential Revision: https://reviews.llvm.org/D95996

3 years ago[flang][NFC] Update comments.
Eric Schweitz [Mon, 8 Feb 2021 20:10:43 +0000 (12:10 -0800)]
[flang][NFC] Update comments.

3 years ago[libomptarget][amdgcn] Fix language linkage post D95300, drop use of assert
Jon Chesterfield [Mon, 8 Feb 2021 20:07:51 +0000 (20:07 +0000)]
[libomptarget][amdgcn] Fix language linkage post D95300, drop use of assert

3 years ago[dfsan] Refactor visitCallBase
Jianzhou Zhao [Fri, 5 Feb 2021 21:50:15 +0000 (21:50 +0000)]
[dfsan] Refactor visitCallBase

To simplify the review of https://reviews.llvm.org/D95835.

Reviewed-by: morehouse
Differential Revision: https://reviews.llvm.org/D96177

3 years ago[lld-macho] Try to make ubsan happy
Jez Ng [Mon, 8 Feb 2021 19:50:13 +0000 (14:50 -0500)]
[lld-macho] Try to make ubsan happy

Summary: We should avoid passing a null pointer to memcpy.

3 years ago[RISCV] Use _COMMUTABLE fma pseudos for fixed vectors.
Craig Topper [Mon, 8 Feb 2021 19:25:50 +0000 (11:25 -0800)]
[RISCV] Use _COMMUTABLE fma pseudos for fixed vectors.

This matches what we do in the VLMAX SDNode patterns.

3 years ago[GWP-ASan] Add aligned allocations.
Mitch Phillips [Mon, 8 Feb 2021 18:46:45 +0000 (10:46 -0800)]
[GWP-ASan] Add aligned allocations.

Adds a new allocation API to GWP-ASan that handles size+alignment
restrictions.

Reviewed By: cryptoad, eugenis

Differential Revision: https://reviews.llvm.org/D94830

3 years agoRevert "[Utils] Add a switch controlling prefix warnings in UpdateTestChecks"
Mircea Trofin [Mon, 8 Feb 2021 19:21:56 +0000 (11:21 -0800)]
Revert "[Utils] Add a switch controlling prefix warnings in UpdateTestChecks"

This reverts commit 87f8a08ce36e5bc72f11129d2cf36b5848f86f63.

3 years ago[RISCV] Add support for splat fixed length build_vectors using RVV.
Craig Topper [Mon, 8 Feb 2021 18:59:38 +0000 (10:59 -0800)]
[RISCV] Add support for splat fixed length build_vectors using RVV.

Building on the fixed vector support from D95705

I've added ISD nodes for vmv.v.x and vfmv.v.f and switched to
lowering the intrinsics to it. This allows us to share the same
isel patterns for both.

This doesn't handle splats of i64 on RV32 yet. The build_vector
gets converted to a vXi32 build_vector+bitcast during type
legalization. Not sure the best way to handle this at the moment.

Differential Revision: https://reviews.llvm.org/D96108

3 years ago[RISCV] Add support for fixed vector FMA.
Craig Topper [Mon, 8 Feb 2021 18:45:45 +0000 (10:45 -0800)]
[RISCV] Add support for fixed vector FMA.

Follow up to D95705. Does not include the commuting support from D95800.

Differential Revision: https://reviews.llvm.org/D96103

3 years ago[mlir][Linalg] Fix padding related bugs.
Nicolas Vasilache [Mon, 8 Feb 2021 15:28:29 +0000 (15:28 +0000)]
[mlir][Linalg] Fix padding related bugs.

This revision fixes the fact that the padding transformation did not have enough information to set the proper type for the padding value.
Additionally, the verifier for Yield in the presence of PadTensorOp is fixed to properly report incorrect number of results or operands. Previously, the error would be silently ignored which made the core issue difficult to debug.

Differential Revision: https://reviews.llvm.org/D96264

3 years ago[lld-macho] Emit LSDA info in compact unwind
Jez Ng [Mon, 8 Feb 2021 18:47:34 +0000 (13:47 -0500)]
[lld-macho] Emit LSDA info in compact unwind

The LSDA pointers are encoded as offsets from the image base,
and arranged in one big contiguous array. Each second-level page records
the offset within that LSDA array which corresponds to the LSDA for its
first CU entry.

Reviewed By: clayborg

Differential Revision: https://reviews.llvm.org/D95810

3 years ago[lld-macho] Emit personalities in compact unwind
Jez Ng [Mon, 8 Feb 2021 18:47:33 +0000 (13:47 -0500)]
[lld-macho] Emit personalities in compact unwind

Note that there is a triple indirection involved with
personalities and compact unwind:

1. Two bits of each CU encoding are used as an offset into the
   personality array.
2. Each entry of the personality array is an offset from the image base.
   The resulting address (after adding the image base) should point within the
   GOT.
3. The corresponding GOT entry contains the actual pointer to the
   personality function.

To further complicate things, when the personality function is in the
object file (as opposed to a dylib), its references in
`__compact_unwind` may refer to it via a section + offset relocation
instead of a symbol relocation. Since our GOT implementation can only
create entries for symbols, we have to create a synthetic symbol at the
given section offset.

Reviewed By: clayborg

Differential Revision: https://reviews.llvm.org/D95809

3 years ago[RISCV] Add initial support for converting fixed vectors to scalable vectors during...
Craig Topper [Mon, 8 Feb 2021 18:32:32 +0000 (10:32 -0800)]
[RISCV] Add initial support for converting fixed vectors to scalable vectors during lowering to use RVV instructions.

This is an alternative to D95563.

This is modeled after a similar feature for AArch64's SVE that uses
predicated scalable vector instructions.a

Rather than use predication, this patch uses an explicit VL operand.
I've limited it to always use LMUL=1 for now, but we can improve this
in the future.

This requires a bunch of new ISD opcodes to carry the VL operand.
I think we can probably lower intrinsics to these ISD opcodes to
cut down on the size of the isel table. Which is why I've added
patterns for all integer/float types and not just LMUL=1.

I'm only testing one vector width right now, but the width is
programmable via the command line.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D95705

3 years ago[ELF] Inspect -EL & -EB for OUTPUT_FORMAT(default, big, little)
Fangrui Song [Mon, 8 Feb 2021 18:34:57 +0000 (10:34 -0800)]
[ELF] Inspect -EL & -EB for OUTPUT_FORMAT(default, big, little)

Choose big if -EB is specified, little if -EL is specified, or default if neither is specified.
The new behavior matches GNU ld.

Fixes: https://github.com/ClangBuiltLinux/linux/issues/1025

Differential Revision: https://reviews.llvm.org/D96214

3 years ago[mlir] Drop deprecated syntax for LLVM dialect types
Alex Zinenko [Mon, 8 Feb 2021 17:22:23 +0000 (18:22 +0100)]
[mlir] Drop deprecated syntax for LLVM dialect types

After the LLVM dialect types were ported to use built-in types, the parser kept
supporting the old syntax for LLVM dialect types to produce built-in types for
compatibility. Drop this support.

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D96275

3 years ago[ConstraintElimination] Decompose a few more GEP indices.
Florian Hahn [Mon, 8 Feb 2021 17:28:34 +0000 (17:28 +0000)]
[ConstraintElimination] Decompose a few more GEP indices.

This patch adds handling for zero-extended GEP indices.

3 years ago[RISCV] Make scalable vector FMA commutable for register allocation.
Craig Topper [Mon, 8 Feb 2021 17:56:47 +0000 (09:56 -0800)]
[RISCV] Make scalable vector FMA commutable for register allocation.

This adds support for commuting operands and converting between
vfmadd and vfmacc to avoid register copies.

To avoid messing up intrinsic behavior, I've added new pseudo
instructions that have the isCommutable flag set. These pseudos also
force a tail agnostic policy. The intrinsic version still use
the tail undisturbed policy.

For best results it looks like we need to start with fmadd and only
pick fmacc if its beneficial. MachineCSE commutes without contraining
the operands and then commutes back if it didn't help with CSE. So
I've made sure that when the operand choice isn't constrained, we
will keep fmadd for MachineCSE and when it does the second commute,
we get back the original instruction.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D95800

3 years ago[Sanitizer] Fix failing sanitizer tests
Julian Lettner [Sat, 6 Feb 2021 05:31:22 +0000 (21:31 -0800)]
[Sanitizer] Fix failing sanitizer tests

The new pass manager was enabled by default [1].

The commit message states the following relevant differences:
  * The inliner works slightly differently
  * -O1 does some amount of inlining

These tests are affected because they specify `-O1` and then check the
reported stack trace.

[1] https://reviews.llvm.org/D95380

Differential Revision: https://reviews.llvm.org/D96198

3 years ago[RISCV] Use SplatPat/SplatPat_simm5 to handle PseudoVMV_V_X_/PseudoVMV_V_I_ selection...
Craig Topper [Mon, 8 Feb 2021 17:48:25 +0000 (09:48 -0800)]
[RISCV] Use SplatPat/SplatPat_simm5 to handle PseudoVMV_V_X_/PseudoVMV_V_I_ selection as well.

This ensures that we'll match immediates consistently regardless
of whether we match them as a standalone splat or as part of
another operation.

While I was there I added complexities to the simm5/uimm5 patterns so
we didn't have to assume that the 1 on the non-immediate was lower
than what tablegen inferred.

I had to make a minor tweak to tablegen to fix one place that
didn't expect to see a ComplexPattern that wasn't a "leaf".

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D96199

3 years ago[AMDGPU] Use named unified buffer format constant. NFC.
Jay Foad [Mon, 8 Feb 2021 17:34:27 +0000 (17:34 +0000)]
[AMDGPU] Use named unified buffer format constant. NFC.

3 years ago[lldb] [Process/FreeBSDRemote] Introduce mips64 support
Michał Górny [Thu, 28 Jan 2021 11:23:54 +0000 (12:23 +0100)]
[lldb] [Process/FreeBSDRemote] Introduce mips64 support

Introduce mips64 support to match the legacy FreeBSD plugin. Similarly
to the legacy plugin, the code does not support FPU registers at the
moment.  The support for them will be submitted separately as it
requires changes to the register context shared by both plugins.

This also includes software single-stepping support that is moved from
the Linux plugin into a common Utility class.  The FreeBSD code also
starts explicitly ignoring EINVAL from PT_CLEARSTEP since this is easier
to implement than checking whether hardware single-stepping were used.

Differential Revision: https://reviews.llvm.org/D95802

3 years ago[LLDB] Fix `Wunused-result` warning
Frederik Gossen [Mon, 8 Feb 2021 17:08:47 +0000 (18:08 +0100)]
[LLDB] Fix `Wunused-result` warning

3 years ago[ELF] Support aarch64_be
Fangrui Song [Mon, 8 Feb 2021 16:55:28 +0000 (08:55 -0800)]
[ELF] Support aarch64_be

This patch adds

* Big-endian values for `R_AARCH64_{ABS,PREL}{16,32,64}` and `R_AARCH64_PLT32`
* aarch64elfb & aarch64linuxb BFD emulations
* elf64-bigaarch64 output format (bfdname)

Link: https://github.com/ClangBuiltLinux/linux/issues/1288
Differential Revision: https://reviews.llvm.org/D96188

3 years ago[llvm-objdump] Support PLT decoding for aarch64_be
Fangrui Song [Mon, 8 Feb 2021 16:50:25 +0000 (08:50 -0800)]
[llvm-objdump] Support PLT decoding for aarch64_be

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D96211

3 years ago[flang][fir] Update FIR's character type.
Eric Schweitz [Fri, 5 Feb 2021 22:40:10 +0000 (14:40 -0800)]
[flang][fir] Update FIR's character type.

Upstream the changes made to the !fir.char type.

https://github.com/flang-compiler/f18-llvm-project/pull/269
https://github.com/flang-compiler/f18-llvm-project/pull/557

Author: Eric Schweitz, Jean Perier

Differention Revision: https://reviews.llvm.org/D96183

3 years ago[FE] Manipulate the first byte of guard variable type in both load and store operation
Xiangling Liao [Mon, 8 Feb 2021 16:14:23 +0000 (11:14 -0500)]
[FE] Manipulate the first byte of guard variable type in both load and store operation

As Itanium ABI[http://itanium-cxx-abi.github.io/cxx-abi/abi.html#once-ctor]
points out:

"The size of the guard variable is 64 bits. The first byte (i.e. the byte at
the address of the full variable) shall contain the value 0 prior to
initialization of the associated variable, and 1 after initialization is complete."

Differential Revision: https://reviews.llvm.org/D95822

3 years ago[OpenCL] Fix pipe type printing in arg info metadata
Anastasia Stulova [Mon, 8 Feb 2021 16:01:44 +0000 (16:01 +0000)]
[OpenCL] Fix pipe type printing in arg info metadata

Pipe element type spelling for arg info metadata
should follow the same behavior as normal type spelling.

We should only use the canonical type spelling in the
base type field.

This patch also removed duplication in type handling.

Tags: #clang

Differential Revision: https://reviews.llvm.org/D96151

3 years ago[AArch64AsmParser] Fix type-limits warning for VectorIndex.
Sander de Smalen [Mon, 8 Feb 2021 14:33:05 +0000 (14:33 +0000)]
[AArch64AsmParser] Fix type-limits warning for VectorIndex.

Making VectorIndex an `int` instead of `unsigned`, silences the warning:
  comparison of unsigned expression in ‘>= 0’ is always true

in:
  template <int Min, int Max>
  DiagnosticPredicate isVectorIndex() const {
    ...
    if (VectorIndex.Val >= Min && VectorIndex.Val <= Max)
      return DiagnosticPredicateTy::Match;
    ...
  }

when Min is 0.

3 years ago[libc++] Add a wait step in the BuildKite pipeline to shield macOS builders
Louis Dionne [Mon, 8 Feb 2021 15:04:38 +0000 (10:04 -0500)]
[libc++] Add a wait step in the BuildKite pipeline to shield macOS builders

We don't have many of those and they are rather slow, so we'd rather not run
those jobs if we know other jobs in the pipeline failed anyway.

3 years agoAArch64: use a constpool for blockaddress(...) on MachO
Tim Northover [Mon, 8 Feb 2021 14:47:14 +0000 (14:47 +0000)]
AArch64: use a constpool for blockaddress(...) on MachO

More MachO madness for everyone. MachO relocations are only 32-bits, which
means the ARM64_RELOC_ADDEND one only actually has 24 (signed) bits for the
actual addend. This is a problem when calculating the address of a basic block;
because it has no symbol of its own, the sequence

adrp x0, Ltmp0@PAGE
add x0, x0, x0 Ltmp0@PAGEOFF

is represented by relocation with an addend that contains the offset from the
function start to Ltmp, and so the largest function where this is guaranteed to
work is 8MB. That's not quite big enough that we can call it user error (IMO).

So this patch puts the any blockaddress into a constant-pool, where the addend
is instead stored in the (x)word being relocated, which is obviously big enough
for any function.

3 years agoIntroduce -print-changed=[diff | diff-quiet] which show changes in patch-like format
Jamie Schmeiser [Mon, 8 Feb 2021 15:09:34 +0000 (10:09 -0500)]
Introduce -print-changed=[diff | diff-quiet] which show changes in patch-like format
Summary:
Introduce base classes that hold a textual represent of the IR
based on basic blocks and a base class for comparing this
representation.  A new change printer is introduced that uses these
classes to save and compare representations of the IR before and after
each pass.  It only reports when changes are made by a pass (similar to
-print-changed) except that the changes are shown in a patch-like format
with those lines that are removed shown in red prefixed with '-' and those
added shown in green with '+'.  This functionality was introduced in my
tutorial at the 2020 virtual developer's meeting.

Author: Jamie Schmeiser <schmeise@ca.ibm.com>
Reviewed By: aeubanks (Arthur Eubanks)
Differential Revision: https://reviews.llvm.org/D91890

3 years ago[DAG] visitVECTOR_SHUFFLE - move shuffle legality check into MergeInnerShuffle lamda...
Simon Pilgrim [Mon, 8 Feb 2021 13:46:31 +0000 (13:46 +0000)]
[DAG] visitVECTOR_SHUFFLE - move shuffle legality check into MergeInnerShuffle lamda. NFCI.

This is going to be necessary for a future reuse of MergeInnerShuffle