platform/upstream/llvm.git
2 years ago[SystemZ] [z/OS] Add support for generating huge (1 MiB) stack frames in XPLINK64
Neumann Hon [Fri, 25 Feb 2022 07:37:05 +0000 (02:37 -0500)]
[SystemZ] [z/OS] Add support for generating huge (1 MiB) stack frames in XPLINK64

This patch extends support for generating huge stack frames on 64-bit XPLINK by implementing the ABI-mandated call to the stack extension routine.

Reviewed By: uweigand

Differential Revision: https://reviews.llvm.org/D120450

2 years ago[MLIR][Presburger] enable copy assignment operator for Simplex
Michel Weber [Fri, 25 Feb 2022 07:32:22 +0000 (07:32 +0000)]
[MLIR][Presburger] enable copy assignment operator for Simplex

This patch removes the `const` from `usingBigM` to enable the implicit copy assignment operator for Simplex.

Reviewed By: Groverkss

Differential Revision: https://reviews.llvm.org/D120542

2 years ago[clang-tidy] Fix `readability-non-const-parameter` for parameter referenced by an...
Sockke [Fri, 25 Feb 2022 06:44:26 +0000 (14:44 +0800)]
[clang-tidy] Fix `readability-non-const-parameter` for parameter referenced by an lvalue

The checker missed a check for a case when the parameter is referenced by an lvalue and this could cause build breakages.

Reviewed By: aaron.ballman

Differential Revision: https://reviews.llvm.org/D117090

2 years ago[RISCV] DAG Combine vcpop and vfirst with VL=0 to li imm
Chenbing Zheng [Fri, 25 Feb 2022 06:44:25 +0000 (14:44 +0800)]
[RISCV] DAG Combine vcpop and vfirst with VL=0 to li imm

vcpop and vfirst are still useful when VL=0.
vcpop equivalents to li 0 and vfirst equivalents to li -1,
since no mask elements are active.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D120302

2 years ago[RISCV] Update computeTargetABI from llc as well as clang
Zakk Chen [Fri, 25 Feb 2022 03:04:24 +0000 (19:04 -0800)]
[RISCV] Update computeTargetABI from llc as well as clang

Clang computes the default ABI if -mabi is empty
and encode it in LLVM IR module flag since D105555.
For correctness, llc need to give the same target-abi
(Options.MCOptions.ABIName) with ABI encoded in IR.
The getSubtargetImpl already has a check for them only if
Options.MCOptions.ABIName is not empty.

In order to get more robustness we could have a check for
explicit ABI, but now we have two different logic to
compute the default ABI.

The front-end ABI is defautl to the ilp32/ilp32e/lp64, and
ilp32d/lp64d when hardware support for extension D.
The backend ABI is default to the ilp32/ilp32e/lp64.

Reviewed by: asb, jrtc27

Differential Revision: https://reviews.llvm.org/D118333

2 years ago[AggressiveInstCombine] Fix `TruncInstCombine` (fix f84d732f)
Anton Afanasyev [Fri, 25 Feb 2022 05:04:11 +0000 (08:04 +0300)]
[AggressiveInstCombine] Fix `TruncInstCombine` (fix f84d732f)

Erase phi-nodes from `InstInfoMap` before erasing themselves

2 years ago[AggressiveInstCombine] Add `phi` nodes support to `TruncInstCombine`
Anton Afanasyev [Wed, 15 Sep 2021 06:19:53 +0000 (09:19 +0300)]
[AggressiveInstCombine] Add `phi` nodes support to `TruncInstCombine`

Expand `TruncInstCombine` to handle loops by adding `phi` nodes
to expression graph.

Reviewed by: RKSimon, lebedev.ri

(recommit of fixed f84d732f, reverted by 8ad6d5e after sanitizer breakage)

Differential Revision: https://reviews.llvm.org/D109817

2 years ago[CUDA][SPIRV] Assign global address space to CUDA kernel arguments
Shangwu Yao [Fri, 25 Feb 2022 04:51:43 +0000 (20:51 -0800)]
[CUDA][SPIRV] Assign global address space to CUDA kernel arguments

(resubmit https://reviews.llvm.org/D119207 after fixing the test for
some build settings)

This patch converts CUDA pointer kernel arguments with default address
space to CrossWorkGroup address space (__global in OpenCL). This is
because Generic or Function (OpenCL's private) is not supported as
storage class for kernel pointer types.

Differential revision: https://reviews.llvm.org/D120366

2 years ago[Driver] Support GCC detection for GCC compiled with --enable-version-specific-runtim...
Raúl Peñacoba [Fri, 25 Feb 2022 04:41:02 +0000 (04:41 +0000)]
[Driver] Support GCC detection for GCC compiled with --enable-version-specific-runtime-libs

GCC's compiled with --enable-version-specific-runtime-libs change the paths where includes and libs are found.
This patch adds support for these cases

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D118700

2 years ago[sanitizer] Don't collect unused info
Vitaly Buka [Fri, 25 Feb 2022 04:18:56 +0000 (20:18 -0800)]
[sanitizer] Don't collect unused info

2 years ago[ELF] Parallelize initializeLocalSymbols
Fangrui Song [Fri, 25 Feb 2022 04:05:59 +0000 (20:05 -0800)]
[ELF] Parallelize initializeLocalSymbols

ObjFile::parse combines symbol initialization and resolution. Many tasks
unrelated to symbol resolution can be postponed and parallelized. This patch
extracts local symbol initialization and parallelizes it.

Technically the new function initializeLocalSymbols can be merged into
ObjFile::postParse, but functions like getSrcMsg may access the
uninitialized (all nullptr) local part of InputFile::symbols.

Linking chrome: 1.02x as fast with glibc malloc, 1.04x as fast with mimalloc

Reviewed By: ikudrin

Differential Revision: https://reviews.llvm.org/D119909

2 years ago[SLP] Fix for the min/max intrinsic cost.
Vasileios Porpodas [Thu, 24 Feb 2022 02:04:36 +0000 (18:04 -0800)]
[SLP] Fix for the min/max intrinsic cost.

The min/max intrinsic cost is currently too low because in the cost calculation
we subtract the cost of the vector compare as we will not emit it.
For the cost of the vector compare we are currently passing BAD_ICMP_PREDICATE
which returns 3, the worst case cost.
I think we should be passing VecPred instead, since we know the predicates of
the compare instr.

I think this is related to commit b3b993a7ad817 which introduced the predicate
argument to getCmpSelInstrCost().
https://reviews.llvm.org/rGb3b993a7ad817c3c5801341fa78f34332900eb83

Differential Revision: https://reviews.llvm.org/D120439

2 years ago[SLP][NFC] Test for a follow-up fix of the the vector min/max instrinsic cost calcula...
Vasileios Porpodas [Thu, 24 Feb 2022 02:03:20 +0000 (18:03 -0800)]
[SLP][NFC] Test for a follow-up fix of the the vector min/max instrinsic cost calculation.

The code in this test should not have been vectorized.
It looks worse than the scalar code.

Differential Revision: https://reviews.llvm.org/D120438

2 years ago[RISCV] Add schedule class for Zbt extension
lian wang [Wed, 16 Feb 2022 01:59:50 +0000 (01:59 +0000)]
[RISCV] Add schedule class for Zbt extension

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D119808

2 years ago[sanitizer] Enable trace pc guard coverage test on PPC64/s390x/MIPS
Xiaodong Liu [Fri, 25 Feb 2022 01:42:19 +0000 (09:42 +0800)]
[sanitizer] Enable trace pc guard coverage test on PPC64/s390x/MIPS

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D120472

2 years ago[AMDGPU] Extend pre-emit peephole to redundantly masked VCC
Carl Ritson [Fri, 25 Feb 2022 00:42:55 +0000 (09:42 +0900)]
[AMDGPU] Extend pre-emit peephole to redundantly masked VCC

Extend pre-emit peephole for S_CBRANCH_VCC[N]Z to eliminate
redundant S_AND operations against EXEC for V_CMP results in VCC.
These occur after after register allocation when VCC has been
selected as the comparison destination.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D120202

2 years ago[ADT, CSSPGO] Specify set comparer
Harald van Dijk [Fri, 25 Feb 2022 00:58:50 +0000 (00:58 +0000)]
[ADT, CSSPGO] Specify set comparer

In _GLIBCXX_DEBUG builds, potentially implicitly enabled by
LLVM_ENABLE_EXPENSIVE_CHECKS, std::set<A, B>::iterator and
std::set<A, C>::iterator are distinct types that are not
interconvertible. This change aligns the iterator types with the set
types.

Reviewed By: hoy

Differential Revision: https://reviews.llvm.org/D119798

2 years ago[libc++] Make sure calls to std::move are always qualified
Corentin Jabot [Fri, 25 Feb 2022 00:54:00 +0000 (19:54 -0500)]
[libc++] Make sure calls to std::move are always qualified

This fixes instances of the newly added `-Wunqualified-std-cast-call`.

(Commit 7853371146 removed unqualified `move` from the tests,
but these unqualified `move`s remained undetected in the actual headers.)

Differential Revision: https://reviews.llvm.org/D120509

2 years ago[OpaquePtr][AArch64] Use load/store value type instead of pointer type for ldnt1...
Arthur Eubanks [Fri, 25 Feb 2022 00:54:18 +0000 (16:54 -0800)]
[OpaquePtr][AArch64] Use load/store value type instead of pointer type for ldnt1/stnt1 alignment

2 years ago[Debuginfod] Add BUILD_ID syntax to llvm-symbolizer.
Daniel Thornburgh [Tue, 15 Feb 2022 22:21:14 +0000 (22:21 +0000)]
[Debuginfod] Add BUILD_ID syntax to llvm-symbolizer.

This adds a BUILD_ID prefix to the llvm-symbolizer stdin and argument
syntax. The prefix causes the given binary name to be interpreted as a
build ID instead of an object file path. The semantics are analagous to
the behavior of --obj and --build-id.

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D119901

2 years ago[RISCV] Change rvv version to 1.0 and remove ratify notice
Qihan Cai [Thu, 24 Feb 2022 22:13:51 +0000 (09:13 +1100)]
[RISCV] Change rvv version to 1.0 and remove ratify notice

This patch changes the version of V extension from 0.1 to 1.0 in RISCVInstrInfoVPseudos.td, RISCVInstrInfoVSDPatterns.td,  RISCVInstrInfoVVLPatterns.td, RISCVInstrInfoV.td

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D120525

2 years ago[Driver][test] Make linux-ld.c work with CLANG_DEFAULT_PIE_ON_LINUX=on
Fangrui Song [Fri, 25 Feb 2022 00:34:05 +0000 (00:34 +0000)]
[Driver][test] Make linux-ld.c work with CLANG_DEFAULT_PIE_ON_LINUX=on

2 years ago[Symbolize] LRU cache binaries in llvm-symbolizer.
Daniel Thornburgh [Tue, 8 Feb 2022 22:07:14 +0000 (22:07 +0000)]
[Symbolize] LRU cache binaries in llvm-symbolizer.

This change adds a simple LRU cache to the Symbolize class to put a cap
on llvm-symbolizer memory usage. Previously, the Symbolizer's virtual
memory footprint would grow without bound as additional binaries were
referenced.

I'm putting this out there early for an informal review, since there may be
a dramatically different/better way to go about this. I still need to
figure out a good default constant for the memory cap and benchmark the
implementation against a large symbolization workload. Right now I've
pegged max memory usage at zero for testing purposes, which evicts the whole
cache every time.

Unfortunately, it looks like StringRefs in the returned DI objects can
directly refer to the contents of binaries. Accordingly, the cache
pruning must be explicitly requested by the caller, as the caller must
guarantee that none of the returned objects will be used afterwards.

For llvm-symbolizer this a light burden; symbolization occurs
line-by-line, and the returned objects are discarded after each.

Implementation wise, there are a number of nested caches that depend
on one another. I've implemented a simple Evictor callback system to
allow derived caches to register eviction actions to occur when the
underlying binaries are evicted.

Reviewed By: dblaikie

Differential Revision: https://reviews.llvm.org/D119784

2 years ago[clang][NFC] Move all avr CodeGen tests to avr specific directory
Ben Shi [Thu, 24 Feb 2022 11:23:05 +0000 (11:23 +0000)]
[clang][NFC] Move all avr CodeGen tests to avr specific directory

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D120468

2 years ago[Driver] Default CLANG_DEFAULT_PIE_ON_LINUX to ON
Fangrui Song [Fri, 25 Feb 2022 00:22:12 +0000 (00:22 +0000)]
[Driver] Default CLANG_DEFAULT_PIE_ON_LINUX to ON

Default the option introduced in D113372 to ON to match all(?) major Linux
distros. This matches GCC and improves consistency with Android and linux-musl
which always default to PIE.
Note: CLANG_DEFAULT_PIE_ON_LINUX will be removed in the future.

Reviewed By: thesamesam

Differential Revision: https://reviews.llvm.org/D120305

2 years ago[asan][test] asan_prelink_test.cpp: use -fno-pie -no-pie
Fangrui Song [Fri, 25 Feb 2022 00:09:18 +0000 (16:09 -0800)]
[asan][test] asan_prelink_test.cpp: use -fno-pie -no-pie

prelink (will be removed by glibc 2.37) does not support PIE.

2 years ago[Driver][test] Remove soon irrelevant pie tests
Fangrui Song [Fri, 25 Feb 2022 00:04:07 +0000 (00:04 +0000)]
[Driver][test] Remove soon irrelevant pie tests

CLANG_DEFAULT_PIE_ON_LINUX=on will soon become the default. The purpose
of these tests has gone.

2 years ago[Driver][test] Make hexagon-toolchain-elf.c work with CLANG_DEFAULT_PIE_ON_LINUX=on
Fangrui Song [Thu, 24 Feb 2022 23:53:38 +0000 (23:53 +0000)]
[Driver][test] Make hexagon-toolchain-elf.c work with CLANG_DEFAULT_PIE_ON_LINUX=on

2 years ago.mailmap: remove stray space in comment
Nico Weber [Thu, 24 Feb 2022 23:50:08 +0000 (18:50 -0500)]
.mailmap: remove stray space in comment

2 years ago[clang][test] Make mips-vector-return.c work with CLANG_DEFAULT_PIE_ON_LINUX=on
Fangrui Song [Thu, 24 Feb 2022 23:25:35 +0000 (15:25 -0800)]
[clang][test] Make mips-vector-return.c work with CLANG_DEFAULT_PIE_ON_LINUX=on

2 years ago[RISCV] Add 'i64' to some isel so tablegen will remove them for RV32. NFC
Craig Topper [Thu, 24 Feb 2022 23:09:20 +0000 (15:09 -0800)]
[RISCV] Add 'i64' to some isel so tablegen will remove them for RV32. NFC

Saves a 100 bytes or so from the isel table.

2 years agoClang `unused-but-set-variable` warnings should not apply to `__attribute__((objc_pre...
Michael Wyman [Wed, 23 Feb 2022 18:47:16 +0000 (10:47 -0800)]
Clang `unused-but-set-variable` warnings should not apply to `__attribute__((objc_precise_lifetime))` Objective-C pointers

The `objc_precise_lifetime` attribute is applied to Objective-C pointers to ensure the optimizer does not prematurely release an object under Automatic Reference Counting (ARC). It is a common enough pattern to assign values to these variables but not reference them otherwise, and annotating them with `__unused` is not really correct as they are being used to ensure an object's lifetime.

Differential Revision: https://reviews.llvm.org/D120372

2 years ago[ELF] Update comment. NFC
Fangrui Song [Thu, 24 Feb 2022 22:09:00 +0000 (14:09 -0800)]
[ELF] Update comment. NFC

2 years ago[ELF] Simplify resolveDefined and resolveCommon
Fangrui Song [Thu, 24 Feb 2022 22:08:06 +0000 (14:08 -0800)]
[ELF] Simplify resolveDefined and resolveCommon

This is NFC for valid input (COMMON symbols cannot be weak or versioned).

2 years agoRevert "Encode address offsets of basic blocks relative to the end of the previous...
Rahman Lavaee [Thu, 24 Feb 2022 20:29:08 +0000 (12:29 -0800)]
Revert "Encode address offsets of basic blocks relative to the end of the previous basic blocks."

This reverts commit 029283c1c0d8d06fbf000f5682c56b8595a1101f.
The code in `ELFFile::decodeBBAddrMap` was not changed in the submitted patch.

Differential Revision: https://reviews.llvm.org/D120457

2 years ago[lld/MachO] Fix +asserts build after recent change
Reid Kleckner [Thu, 24 Feb 2022 21:12:48 +0000 (13:12 -0800)]
[lld/MachO] Fix +asserts build after recent change

2 years ago[AArch64][x86] add tests for bitwise logic + shifts; NFC
Sanjay Patel [Thu, 24 Feb 2022 20:42:30 +0000 (15:42 -0500)]
[AArch64][x86] add tests for bitwise logic + shifts; NFC

2 years ago[RISCV] Add tests for (neg (abs X)) where the abs has an additional user.
Craig Topper [Thu, 24 Feb 2022 20:36:33 +0000 (12:36 -0800)]
[RISCV] Add tests for (neg (abs X)) where the abs has an additional user.

2 years ago[ELF][test] Remove invalid weak COMMON tests
Fangrui Song [Thu, 24 Feb 2022 20:54:16 +0000 (12:54 -0800)]
[ELF][test] Remove invalid weak COMMON tests

GNU as reports `Error: symbol `foo' can not be both weak and common`,
though LLVM integrated assembler does not report an error yet.

2 years ago[clang][dataflow] Add limits to size of modeled data structures in environment.
Yitzhak Mandelbaum [Thu, 24 Feb 2022 20:02:00 +0000 (20:02 +0000)]
[clang][dataflow] Add limits to size of modeled data structures in environment.

Adds two new parameters to control the size of data structures modeled in the environment: # of values and depth of data structure.  The environment already prevents creation of recursive data structures, but that was insufficient in practice. Very large structs still ground the analysis to a halt.  These new parameters allow tuning the size more effectively.

In this patch, the parameters are set as internal constants. We leave to a future patch to make these proper model parameters.

Differential Revision: https://reviews.llvm.org/D120510

2 years ago[lld-macho] Implement -why_live (without perf overhead)
Jez Ng [Thu, 24 Feb 2022 20:39:59 +0000 (15:39 -0500)]
[lld-macho] Implement -why_live (without perf overhead)

This was based off @thakis' draft in {D103517}. I employed templates to ensure
the support for `-why_live` wouldn't slow down the regular non-why-live code
path.

No stat sig perf difference on my 3.2 GHz 16-Core Intel Xeon W:

             base           diff           difference (95% CI)
  sys_time   1.195 ± 0.015  1.199 ± 0.022  [  -0.4% ..   +1.0%]
  user_time  3.716 ± 0.022  3.701 ± 0.025  [  -0.7% ..   -0.1%]
  wall_time  4.606 ± 0.034  4.597 ± 0.046  [  -0.6% ..   +0.2%]
  samples    44             37

Reviewed By: #lld-macho, thakis

Differential Revision: https://reviews.llvm.org/D120377

2 years agoFix typo in file name; NFC
Aaron Ballman [Thu, 24 Feb 2022 20:41:25 +0000 (15:41 -0500)]
Fix typo in file name; NFC

2 years agoAdd -Wno-strict-prototypes to C tests; NFC
Aaron Ballman [Thu, 24 Feb 2022 20:30:30 +0000 (15:30 -0500)]
Add -Wno-strict-prototypes to C tests; NFC

This patch adds -Wno-strict-prototypes to all of the test cases that
use functions without prototypes, but not as the primary concern of the
test. e.g., attributes testing whether they can/cannot be applied to a
function without a prototype, etc.

This is done in preparation for enabling -Wstrict-prototypes by
default.

2 years agoUse functions with prototypes when appropriate; NFC
Aaron Ballman [Thu, 24 Feb 2022 20:27:07 +0000 (15:27 -0500)]
Use functions with prototypes when appropriate; NFC

A significant number of our tests in C accidentally use functions
without prototypes. This patch converts the function signatures to have
a prototype for the situations where the test is not specific to K&R C
declarations. e.g.,

  void func();

becomes

  void func(void);

This is the final batch of tests being updated to add prototypes,
hopefully.

2 years ago[flang] Simple array assignment lowering
Valentin Clement [Thu, 24 Feb 2022 20:09:40 +0000 (21:09 +0100)]
[flang] Simple array assignment lowering

This patch handles lowering of simple array assignment.

```
a(:) = 10
```

or

```
a(1) = 1
```

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: PeteSteinfeld, schweitz

Differential Revision: https://reviews.llvm.org/D120501

Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: V Donaldson <vdonaldson@nvidia.com>
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
2 years ago[AMDGPU] Divergence-driven instruction selection for bitreverse
Jay Foad [Mon, 14 Feb 2022 09:51:56 +0000 (09:51 +0000)]
[AMDGPU] Divergence-driven instruction selection for bitreverse

Differential Revision: https://reviews.llvm.org/D119702

2 years ago[ELF] Simplify --fortran-common. NFC
Fangrui Song [Thu, 24 Feb 2022 20:21:40 +0000 (12:21 -0800)]
[ELF] Simplify --fortran-common. NFC

2 years ago[ELF] De-template Symbol::resolveLazy. NFC
Fangrui Song [Thu, 24 Feb 2022 20:20:05 +0000 (12:20 -0800)]
[ELF] De-template Symbol::resolveLazy. NFC

2 years ago[flang] add semantics tests for sync team
Damian Rouson [Fri, 12 Nov 2021 00:24:01 +0000 (16:24 -0800)]
[flang] add semantics tests for sync team

Test a range of acceptable forms of SYNC TEAM statements,
including combinations with and without the stat-variable
and errmsg-variable present.  Also test that several invalid
forms of SYNC TEAM call generate the correct error messages.

Differential Revision: https://reviews.llvm.org/D120099

2 years agoRevert "[OPENMP]Fix PR50347: Mapping of global scope deep object fails."
Alexey Bataev [Thu, 24 Feb 2022 20:04:39 +0000 (12:04 -0800)]
Revert "[OPENMP]Fix PR50347: Mapping of global scope deep object fails."

This reverts commit 638938117aeae5518d6cacd066ffd9830ef4fc9a. Need to
fix reported fail https://lab.llvm.org/buildbot/#/builders/193/builds/7496

2 years ago[OPENMP]Fix PR50347: Mapping of global scope deep object fails.
Alexey Bataev [Thu, 1 Jul 2021 11:16:56 +0000 (04:16 -0700)]
[OPENMP]Fix PR50347: Mapping of global scope deep object fails.

Changed the we handle llvm::Constants in sizes arrays. ConstExprs and
GlobalValues cannot be used as initializers, need to put them at the
runtime, otherwise there wight be the compilation errors.

Differential Revision: https://reviews.llvm.org/D105297

2 years ago[CMake][WinMsvc] Replace MSVC_BASE/WINSDK_BASE with LLVM_WINSYSROOT
Yuanfang Chen [Thu, 24 Feb 2022 19:46:37 +0000 (11:46 -0800)]
[CMake][WinMsvc] Replace MSVC_BASE/WINSDK_BASE with LLVM_WINSYSROOT

- Using LLVM_WINSYSROOT would pick up DIA SDK path automatically,
  otherwise llvm-pdbutil has no DIA support.
- Add MSVC_VER to specify VC tools version.
- Make MSVC_VER/WINSDK_VER optional. If not specified, use the highest
  version number like the driver does.

Reviewed By: thakis

Differential Revision: https://reviews.llvm.org/D117852

2 years ago[DAG] Attempt to fold bswap(shl(x,c)) -> zext(bswap(trunc(shl(x,c-bw/2))))
Simon Pilgrim [Thu, 24 Feb 2022 19:18:37 +0000 (19:18 +0000)]
[DAG] Attempt to fold bswap(shl(x,c)) -> zext(bswap(trunc(shl(x,c-bw/2))))

If the shl is at least half the bitwidth (i.e. the lower half of the bswap source is zero), then we can reduce the shift and perform the bswap at half the bitwidth and just zero extend.

Based off PR51391 + PR53867

Differential Revision: https://reviews.llvm.org/D120192

2 years ago[AArch64] Regenerate dp1.ll test, NFC
David Green [Thu, 24 Feb 2022 19:33:45 +0000 (19:33 +0000)]
[AArch64] Regenerate dp1.ll test, NFC

The old check lines were not showing enough congtext to show issues.
Regenerate the test with theua auto-check lines to be clearer.

2 years ago[ELF] Set config->exportDynamic to true if config->shared. NFC
Fangrui Song [Thu, 24 Feb 2022 19:31:58 +0000 (11:31 -0800)]
[ELF] Set config->exportDynamic to true if config->shared. NFC

2 years ago[AMDGPU] Extend SILoadStoreOptimizer to handle global stores
Stanislav Mekhanoshin [Tue, 22 Feb 2022 20:26:41 +0000 (12:26 -0800)]
[AMDGPU] Extend SILoadStoreOptimizer to handle global stores

TODO: merge flat load/stores.
TODO: merge flat with global promoting to flat.

Differential Revision: https://reviews.llvm.org/D120346

2 years ago[flang] add semantics test for sync memory
Damian Rouson [Fri, 12 Nov 2021 00:22:38 +0000 (16:22 -0800)]
[flang] add semantics test for sync memory

Test a range of acceptable forms of SYNC MEMORY statements,
including combinations with and without the stat-variable
and errmsg-variable present.  Also test that several invalid
forms of SYNC MEMORY call generate the correct error messages.

Differential Revision: https://reviews.llvm.org/D120097

2 years ago[AArch64] Async unwind - Always place the first LDP at the end when ReverseCSRRestore...
Momchil Velikov [Thu, 24 Feb 2022 18:36:37 +0000 (18:36 +0000)]
[AArch64] Async unwind - Always place the first LDP at the end when ReverseCSRRestoreSeq is true

This patch is in preparation for the async unwind CFI.

Put the first `LDP` the end, so that the load-store optimizer can run
and merge the `LDP` and the `ADD` into a post-index `LDP`.

Do this always and as early as at the time of the initial creation of
the CSR restore instructions, even if that `LDP` is not guaranteed to
be mergeable with a subsequent `SP` increment.

This greatly simplifies the CFI generation for prologue, as otherwise
we have to take extra steps to ensure reordering does not cross CFI
instructions.

Reviewed By: danielkiss

Differential Revision: https://reviews.llvm.org/D112328

2 years ago[AMDGPU] Fix combined MMO in load-store merge
Stanislav Mekhanoshin [Wed, 23 Feb 2022 00:33:06 +0000 (16:33 -0800)]
[AMDGPU] Fix combined MMO in load-store merge

Loads and stores can be out of order in the SILoadStoreOptimizer.
When combining MachineMemOperands of two instructions operands are
sent in the IR order into the combineKnownAdjacentMMOs. At the
moment it picks the first operand and just replaces its offset and
size. This essentially loses alignment information and may generally
result in an incorrect base pointer to be used.

Use a base pointer in memory addresses order instead and only adjust
size.

Differential Revision: https://reviews.llvm.org/D120370

2 years ago[X86] Introduce x86-cmov-converter-force-all
Amir Ayupov [Wed, 16 Feb 2022 01:27:31 +0000 (17:27 -0800)]
[X86] Introduce x86-cmov-converter-force-all

Introduce an option to expand all CMOV groups into hammocks, matching GCC's
`-fno-if-conversion2` flag. The motivation is to leave CMOV conversion
opportunities to a binary optimizer that can make the decision based on branch
misprediction rate (available e.g. in Intel's LBR).

Reviewed By: MaskRay, skan

Differential Revision: https://reviews.llvm.org/D119777

2 years ago[tblgen] Compress CompositeSequences to 1/8th of its size. NFCI.
Benjamin Kramer [Thu, 24 Feb 2022 18:30:50 +0000 (19:30 +0100)]
[tblgen] Compress CompositeSequences to 1/8th of its size. NFCI.

2 years ago[RISCV] Add Zbb RUN lines to neg-abs.ll.
Craig Topper [Thu, 24 Feb 2022 18:15:07 +0000 (10:15 -0800)]
[RISCV] Add Zbb RUN lines to neg-abs.ll.

2 years ago[AArch64] Async unwind - helper functions to decide on CFI emission
Momchil Velikov [Thu, 24 Feb 2022 17:36:19 +0000 (17:36 +0000)]
[AArch64] Async unwind - helper functions to decide on CFI emission

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D112327

2 years ago[analyzer] Just use default capture after 7fd60ee6e0a87957a718297a4a42d9881fc561e3
Fangrui Song [Thu, 24 Feb 2022 18:06:11 +0000 (10:06 -0800)]
[analyzer] Just use default capture after 7fd60ee6e0a87957a718297a4a42d9881fc561e3

2 years ago[mlir][memref] Add transformation to do loop multi-buffering
Thomas Raoux [Wed, 16 Feb 2022 07:49:32 +0000 (23:49 -0800)]
[mlir][memref] Add transformation to do loop multi-buffering

This transformation is useful to break dependency between consecutive loop
iterations by increasing the size of a temporary buffer. This is usually
combined with heavy software pipelining.

Differential Revision: https://reviews.llvm.org/D119406

2 years ago[RISCV] Update some tests to use floating point ABI where it makes sense.
Craig Topper [Thu, 24 Feb 2022 17:23:14 +0000 (09:23 -0800)]
[RISCV] Update some tests to use floating point ABI where it makes sense.

Trying to reduce the diffs from D118333 for cases where it makes
more sense to use an FP ABI.

Reviewed By: asb, kito-cheng

Differential Revision: https://reviews.llvm.org/D120447

2 years ago[AArch64] Async unwind - do not schedule frame setup/destroy
Momchil Velikov [Thu, 24 Feb 2022 16:08:14 +0000 (16:08 +0000)]
[AArch64] Async unwind  - do not schedule frame setup/destroy

The PostRA scheduler can reorder non-CFI instructions in a way that
makes the unwind info not instruction precise.

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D112326

2 years ago[RISCV] Fold (sext_inreg (fmv_x_anyexth X), i16) -> (fmv_x_signexth X).
Craig Topper [Thu, 24 Feb 2022 17:18:48 +0000 (09:18 -0800)]
[RISCV] Fold (sext_inreg (fmv_x_anyexth X), i16) -> (fmv_x_signexth X).

Add a new ISD opcode to represent the sign extending behavior of
vmv.x.h. Keep the previous anyext opcode to allow the existing
(fmv_x_anyexth (fmv_h_x X)) combine to keep working without needing
to generate a sign extend.

For fmv.x.w we are able to match the sext_inreg in an isel pattern,
but a 16-bit sext_inreg is lowered to a shift pair before isel. This
seemed like a larger match than we should do in isel.

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D118974

2 years ago[flang] Lower allocatable assignment for scalar
Valentin Clement [Thu, 24 Feb 2022 17:11:41 +0000 (18:11 +0100)]
[flang] Lower allocatable assignment for scalar

Add lowering for simple assignement on allocatable
scalars.

This patch is part of the upstreaming effort from fir-dev branch.

Depends on D120483

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D120488

Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
2 years ago[clang][dataflow] Update StructValue child when assigning a value
Stanislav Gatev [Wed, 23 Feb 2022 15:39:26 +0000 (15:39 +0000)]
[clang][dataflow] Update StructValue child when assigning a value

When assigning a value to a storage location of a struct member we
need to also update the value in the corresponding `StructValue`.

This is part of the implementation of the dataflow analysis framework.
See "[RFC] A dataflow analysis framework for Clang AST" on cfe-dev.

Reviewed-by: ymandel, xazax.hun
Differential Revision: https://reviews.llvm.org/D120414

2 years ago[clang-tidy] Remove opencl-c.h inclusion from tests
Sven van Haastregt [Thu, 24 Feb 2022 16:28:52 +0000 (16:28 +0000)]
[clang-tidy] Remove opencl-c.h inclusion from tests

After D120254 some clang-tidy tests started failing on release builds.

clang-tidy has been using the `-fdeclare-opencl-builtins` functionality
since this became the default in clang, so there is no need to include
`opencl-c.h`.

Differential Revision: https://reviews.llvm.org/D120470

2 years ago[SDAG] remove shift that is redundant with part of funnel shift
Sanjay Patel [Thu, 24 Feb 2022 16:09:57 +0000 (11:09 -0500)]
[SDAG] remove shift that is redundant with part of funnel shift

This is the SDAG translation of D120253 :
https://alive2.llvm.org/ce/z/qHpmNn

The SDAG nodes can have different operand types than the result value.
We can see an example of that with AArch64 - the funnel shift amount
is an i64 rather than i32.

We may need to make that match even more flexible to handle
post-legalization nodes, but I have not stepped into that yet.

Differential Revision: https://reviews.llvm.org/D120264

2 years ago[flang] Handle allocatable dummy arguments
Valentin Clement [Thu, 24 Feb 2022 16:16:02 +0000 (17:16 +0100)]
[flang] Handle allocatable dummy arguments

This patch handles allocatable dummy argument lowering
in function and subroutines.

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: schweitz

Differential Revision: https://reviews.llvm.org/D120483

Co-authored-by: Jean Perier <jperier@nvidia.com>
2 years agoDisable Mailgun click tracking
Anton Korobeynikov [Thu, 24 Feb 2022 16:03:21 +0000 (19:03 +0300)]
Disable Mailgun click tracking

2 years agoRemove useless RUN lines in the middle of the file and pipe to FileCheck; NFC
Aaron Ballman [Thu, 24 Feb 2022 16:00:40 +0000 (11:00 -0500)]
Remove useless RUN lines in the middle of the file and pipe to FileCheck; NFC

2 years ago[OpenMP] Make section variable external to prevent collisions
Joseph Huber [Thu, 24 Feb 2022 15:24:39 +0000 (10:24 -0500)]
[OpenMP] Make section variable external to prevent collisions

Summary:
We use a section to embed offloading code into the host for later
linking. This is normally unique to the translation unit as it is thrown
away during linking. However, if the user performs a relocatable link
the sections will be merged and we won't be able to access the files
stored inside. This patch changes the section variables to have external
linkage and a name defined by the section name, so if two sections are
combined during linking we get an error.

2 years ago[InstCombine] try harder to preserve 'nsz' in fneg-of-select transform
Sanjay Patel [Thu, 24 Feb 2022 15:20:18 +0000 (10:20 -0500)]
[InstCombine] try harder to preserve 'nsz' in fneg-of-select transform

The corner case where 'nsz' needs to be removed is very narrow
as discussed here:
https://reviews.llvm.org/rG3cdd05e519dd

If the select condition is not undef, there's no problem with
propagating 'nsz':
https://alive2.llvm.org/ce/z/4GWJdq

2 years ago[InstCombine] add test for fneg of select with FMF; NFC
Sanjay Patel [Thu, 24 Feb 2022 15:19:23 +0000 (10:19 -0500)]
[InstCombine] add test for fneg of select with FMF; NFC

2 years ago[MIRParser] Diagnose too large align values in MachineMemOperands
Jay Foad [Wed, 23 Feb 2022 11:35:55 +0000 (11:35 +0000)]
[MIRParser] Diagnose too large align values in MachineMemOperands

When parsing MachineMemOperands, MIRParser treated the "align" keyword
the same as "basealign". Really "basealign" should specify the
alignment of the MachinePointerInfo base value, and "align" should
specify the alignment of that base value plus the offset.

This worked OK when the specified alignment was no larger than the
alignment of the offset, but in cases like this it just caused
confusion:

    STW killed %18, 4, %stack.1.ap2.i.i :: (store (s32) into %stack.1.ap2.i.i + 4, align 8)

MIRPrinter would never have printed this, with an offset of 4 but an
align of 8, so it must have been written by hand. MIRParser would
interpret "align 8" as "basealign 8", but I think it is better to give
an error and force the user to write "basealign 8" if that is what they
really meant.

Differential Revision: https://reviews.llvm.org/D120400

Change-Id: I7eeeefc55c2df3554ba8d89f8809a2f45ada32d8

2 years ago[mlir][emitc] Add a variable op
Marius Brehler [Thu, 17 Feb 2022 09:57:09 +0000 (09:57 +0000)]
[mlir][emitc] Add a variable op

This adds a variable op, emitted as C/C++ locale variable, which can be
used if the `emitc.constant` op is not sufficient.

As an example, the canonicalization pass would transform
```mlir
%0 = "emitc.constant"() {value = 0 : i32} : () -> i32
%1 = "emitc.constant"() {value = 0 : i32} : () -> i32
%2 = emitc.apply "&"(%0) : (i32) -> !emitc.ptr<i32>
%3 = emitc.apply "&"(%1) : (i32) -> !emitc.ptr<i32>
emitc.call "write"(%2, %3) : (!emitc.ptr<i32>, !emitc.ptr<i32>) -> ()
```
into
```mlir
%0 = "emitc.constant"() {value = 0 : i32} : () -> i32
%1 = emitc.apply "&"(%0) : (i32) -> !emitc.ptr<i32>
%2 = emitc.apply "&"(%0) : (i32) -> !emitc.ptr<i32>
emitc.call "write"(%1, %2) : (!emitc.ptr<i32>, !emitc.ptr<i32>) -> ()
```
resulting in pointer aliasing, as %1 and %2 point to the same address.
In such a case, the `emitc.variable` operation can be used instead.

Reviewed By: jpienaar

Differential Revision: https://reviews.llvm.org/D120098

2 years ago[clang] Warn on unqualified calls to std::move and std::forward
Corentin Jabot [Thu, 24 Feb 2022 15:00:44 +0000 (07:00 -0800)]
[clang] Warn on unqualified calls to std::move and std::forward

This adds a diagnostic when an unqualified call is resolved
to std::move or std::forward.

This follows some C++ committee discussions where some
people where concerns that this might be an usual anti pattern
particularly britle worth warning about - both because move
is a common name and because these functions accept any values.

This warns inconditionnally of whether the current context is in
std:: or not, as implementations probably want to always qualify
these calls too, to avoid triggering adl accidentally.

Differential Revision: https://reviews.llvm.org/D119670

2 years ago[OpenCL] Handle TypeExtensions in OpenCLBuiltinFileEmitter
Sven van Haastregt [Thu, 24 Feb 2022 15:17:24 +0000 (15:17 +0000)]
[OpenCL] Handle TypeExtensions in OpenCLBuiltinFileEmitter

Until now, any types that had TypeExtensions attached to them were not
guarded with those extensions.  Extend the OpenCLBuiltinFileEmitter
such that all required extensions are emitted for the types of a
builtin function.

The `clang-tblgen -gen-clang-opencl-builtin-tests` emitter will now
produce e.g.:

  #if defined(cl_khr_fp16) && defined(cl_khr_fp64)
  half8 test11802_convert_half8_rtp(double8 arg1) {
    return convert_half8_rtp(arg1);
  }
  #endif // TypeExtension

Differential Revision: https://reviews.llvm.org/D120262

2 years ago[AArch64] Simplify and extend tests added in 0c5ea01b20623e40.
Florian Hahn [Thu, 24 Feb 2022 14:51:43 +0000 (14:51 +0000)]
[AArch64] Simplify and extend tests added in 0c5ea01b20623e40.

2 years ago[X86] LowerRotate - enable v8i16 ROTL/ROTR on all pre-SSE41 targets
Simon Pilgrim [Thu, 24 Feb 2022 14:14:08 +0000 (14:14 +0000)]
[X86] LowerRotate - enable v8i16 ROTL/ROTR on all pre-SSE41 targets

We're still better off expanding this once we have PMOVZX

2 years ago[X86] SimplifyDemandedVectorEltsForTargetNode - add X86ISD::ANDNP handling
Simon Pilgrim [Thu, 24 Feb 2022 13:51:51 +0000 (13:51 +0000)]
[X86] SimplifyDemandedVectorEltsForTargetNode - add X86ISD::ANDNP handling

2 years ago[mlir][OpenMP][NFC] Remove unused binary operator enum
Shraiysh Vaishay [Thu, 24 Feb 2022 13:30:40 +0000 (19:00 +0530)]
[mlir][OpenMP][NFC] Remove unused binary operator enum

This patch removes binary operator enum which was introduced with `omp.atomic.update`. Now the update operation handles update in a region so this is no longer required.

Reviewed By: kiranchandramohan

Differential Revision: https://reviews.llvm.org/D120458

2 years ago[Symbolizer] Move ctor/dtor into .cpp file
Benjamin Kramer [Thu, 24 Feb 2022 13:19:19 +0000 (14:19 +0100)]
[Symbolizer] Move ctor/dtor into .cpp file

On some standard library configurations these have a dependency on the
complete type of SymbolizableModule. They also do a lot of
copying/freeing so no point in inlining them.

2 years ago[mlir] Document creation of Python bindings for a dialect
Alex Zinenko [Tue, 11 Jan 2022 10:56:21 +0000 (11:56 +0100)]
[mlir] Document creation of Python bindings for a dialect

Documentation exists about the details of the API but is missing a
description of the overall structure per dialect.

Reviewed By: shabalin

Differential Revision: https://reviews.llvm.org/D117002

2 years ago[mlir][linalg] Cast back to the original type after making linalg.generic outputs...
Benjamin Kramer [Thu, 24 Feb 2022 12:34:04 +0000 (13:34 +0100)]
[mlir][linalg] Cast back to the original type after making linalg.generic outputs more static

This codepath was entirely untested.

Differential Revision: https://reviews.llvm.org/D120473

2 years ago[NFC][SROA] Update tests for D113520
Roman Lebedev [Thu, 24 Feb 2022 05:47:14 +0000 (08:47 +0300)]
[NFC][SROA] Update tests for D113520

2 years agoCleanup includes: ProfileData
serge-sans-paille [Wed, 23 Feb 2022 21:18:20 +0000 (22:18 +0100)]
Cleanup includes: ProfileData

Estimation of the impact on preprocessor output:

before: 1067349756
after: 1065940348

Discourse thread: https://discourse.llvm.org/t/include-what-you-use-include-cleanup
Differential Revision: https://reviews.llvm.org/D120434

2 years agoCleanup include: DebugInfo/Symbolize
serge-sans-paille [Wed, 23 Feb 2022 21:16:39 +0000 (22:16 +0100)]
Cleanup include: DebugInfo/Symbolize

Estimation of the impact on preprocessor output
after: 1067349756
before:1067487786

Discourse thread: https://discourse.llvm.org/t/include-what-you-use-include-cleanup
Differential Revision: https://reviews.llvm.org/D120433

2 years ago[AMDGPU] Fix permissions on test files
Jay Foad [Thu, 24 Feb 2022 12:17:34 +0000 (12:17 +0000)]
[AMDGPU] Fix permissions on test files

2 years ago[AArch64] Add vector select test showing redundant operations.
Florian Hahn [Thu, 24 Feb 2022 12:12:21 +0000 (12:12 +0000)]
[AArch64] Add vector select test showing redundant operations.

The tests show sub-optimal lowering of extend/cmp/select chains starting
with v16i8 vectors.

2 years ago[NFC][RISCV] Reuse ISD::NodeType in float extension
Shao-Ce SUN [Thu, 24 Feb 2022 03:14:43 +0000 (11:14 +0800)]
[NFC][RISCV] Reuse ISD::NodeType in float extension

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D120412

2 years ago[OpenCL] opencl-c.h: remove arg names for image builtins
Sven van Haastregt [Thu, 24 Feb 2022 11:52:32 +0000 (11:52 +0000)]
[OpenCL] opencl-c.h: remove arg names for image builtins

This simplifies completeness comparisons against OpenCLBuiltins.td and
also makes the header no longer "claim" the identifiers "image",
"image_array", "coord", "sampler", "sample", "gradientX", "gradientY",
"lod", and "color".

Continues the direction set out in D119560.

2 years ago[lldb] Fix macos build for D120425
Pavel Labath [Thu, 24 Feb 2022 11:47:17 +0000 (12:47 +0100)]
[lldb] Fix macos build for D120425

2 years ago[mlir][LLVM] Allow scalable vectors in ShuffleVectorOp
Javier Setoain [Thu, 27 Jan 2022 13:36:16 +0000 (13:36 +0000)]
[mlir][LLVM] Allow scalable vectors in ShuffleVectorOp

The current implementation of ShuffleVectorOp assumes all vectors are
scalable. LLVM IR allows shufflevector operations on scalable vectors,
and the current translation between LLVM Dialect and LLVM IR does the
rigth thing when the shuffle mask is all zeroes. This is required to
do a splat operation on a scalable vector, but it doesn't make sense
for scalable vectors outside of that operation, i.e.: with non-all zero
masks.

Differential Revision: https://reviews.llvm.org/D118371

2 years ago[X86] LowerShiftByScalarVariable - use getSplatSourceVector for vXi8 shift expansion
Simon Pilgrim [Thu, 24 Feb 2022 11:12:52 +0000 (11:12 +0000)]
[X86] LowerShiftByScalarVariable - use getSplatSourceVector for vXi8 shift expansion

Using getSplatValue causes poor codegen due to not always being able to remove the EXTRACT_VECTOR_ELT created inside getSplatValue.

The vXi16 shifts/rotates are still showing occasional regressions but vXi8 is a definite improvement.

2 years agoRevert "[RISCV] add the MC layer support of Zfinx extension"
Nikita Popov [Thu, 24 Feb 2022 11:12:05 +0000 (12:12 +0100)]
Revert "[RISCV] add the MC layer support of Zfinx extension"

This reverts commit 7798ecca9c3db42241169d31fea4fb820ed01830.

As reported in https://reviews.llvm.org/D93298#3331641 and
following, this causes assertion failures with inline assembly.