platform/upstream/llvm.git
3 years ago[libc++][ranges] iterator.concept.sizedsentinel: sized_sentinel_for and disable_sized...
zoecarver [Mon, 26 Apr 2021 16:35:47 +0000 (09:35 -0700)]
[libc++][ranges] iterator.concept.sizedsentinel: sized_sentinel_for and disable_sized_sentinel_for.

Based on D100160.

Reviewed By: cjdb, ldionne, Quuxplusone, #libc, miscco

Differential Revision: https://reviews.llvm.org/D100587

3 years agoRevert "[clang][driver] Use the provided arch name for a Darwin target triple"
Alex Lorenz [Mon, 26 Apr 2021 21:56:56 +0000 (14:56 -0700)]
Revert "[clang][driver] Use the provided arch name for a Darwin target triple"

This reverts commit 6cc62043c8bf4daa27664a2e1674abbe8d0492c6.

This caused a test failure on a M1 mac CI job (https://reviews.llvm.org/D100807#2718006),
I will recommit this with a fix.

3 years ago[gn build] Port caf1294d9578
LLVM GN Syncbot [Mon, 26 Apr 2021 21:24:36 +0000 (21:24 +0000)]
[gn build] Port caf1294d9578

3 years ago[AMDGPU] Experiments show that the GCNRegBankReassign pass significantly impacts
Baptiste Saleil [Mon, 26 Apr 2021 19:48:12 +0000 (15:48 -0400)]
[AMDGPU] Experiments show that the GCNRegBankReassign pass significantly impacts
the compilation time and there is no case for which we see any improvement in
performance. This patch removes this pass and its associated test cases from
the tree.

Differential Revision: https://reviews.llvm.org/D101313

Change-Id: I0599169a7609c19a887f8d847a71e664030cc141

3 years agoImplementation for TargetTransformInfo::hasActiveVectorLength()
Vineet Kumar [Mon, 26 Apr 2021 19:50:12 +0000 (19:50 +0000)]
Implementation for TargetTransformInfo::hasActiveVectorLength()

This patch adds the missing implementation for
TargetTransformInfo::hasActiveVectorLength() without which using
hasActiveVectorLength() causes linker error.

Patch by Vineet Kumar!

Differential Revision: https://reviews.llvm.org/D100941

3 years agoSupport leak sanitizer in libcxx.
Sterling Augustine [Mon, 19 Apr 2021 17:56:25 +0000 (10:56 -0700)]
Support leak sanitizer in libcxx.

Support leak sanitizer in libcxx.

Simple addition for leak checking when running the libcxx testsuite.

Differential Revision: https://reviews.llvm.org/D100775

3 years agoDon't fail the shared_ptr test if libc++ has insufficient debug info.
Sterling Augustine [Fri, 16 Apr 2021 01:03:01 +0000 (18:03 -0700)]
Don't fail the shared_ptr test if libc++ has insufficient debug info.

Don't fail the shared_ptr test if libc++ has insufficient debug info.

This addresses https://bugs.llvm.org/show_bug.cgi?id=48937

Differential Revision: https://reviews.llvm.org/D100610

3 years ago[gn build] Port 8740360093b5
LLVM GN Syncbot [Mon, 26 Apr 2021 21:04:24 +0000 (21:04 +0000)]
[gn build] Port 8740360093b5

3 years ago[ORC] Add unit tests for parts of the Orc and LLJIT C APIs.
Lang Hames [Mon, 26 Apr 2021 20:55:26 +0000 (13:55 -0700)]
[ORC] Add unit tests for parts of the Orc and LLJIT C APIs.

Patch by Mats Larsen. Thanks Mats!

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D100506

3 years ago[ORC] C API updates.
Lang Hames [Mon, 26 Apr 2021 19:43:46 +0000 (12:43 -0700)]
[ORC] C API updates.

Adds support for creating custom MaterializationUnits in the C API with the new
LLVMOrcCreateCustomMaterializationUnit function.

Modifies ownership rules for LLVMOrcAbsoluteSymbols to make it consistent with
LLVMOrcCreateCustomMaterializationUnit. This is an ABI breaking change for any
clients of the LLVMOrcAbsoluteSymbols API.

Adds LLVMOrcLLJITGetObjLinkingLayer and LLVMOrcObjectLayerEmit functions to
allow clients to get a reference to an LLJIT instance's linking layer, then
emit an object file using it. This can be used to support construction of
custom materialization units in the common case where those units will
generate an object file that needs to be emitted to complete the
materialization.

3 years ago[ORC] Fix type name.
Lang Hames [Mon, 26 Apr 2021 19:30:57 +0000 (12:30 -0700)]
[ORC] Fix type name.

Rename JITTargetSymbolFlags to JITSymbolTargetFlags. This matches the convention
used for JITSymbolGenericFlags.

3 years ago[ADT] Remove StatisticBase and make NoopStatistic empty
Fangrui Song [Mon, 26 Apr 2021 20:39:35 +0000 (13:39 -0700)]
[ADT] Remove StatisticBase and make NoopStatistic empty

In LLVM_ENABLE_STATS=0 builds, `llvm::Statistic` maps to `llvm::NoopStatistic`
but has 3 unused pointers. GlobalOpt considers that the pointers can potentially
retain allocated objects, so GlobalOpt cannot optimize out the `NoopStatistic`
variables (see D69428 for more context), wasting 23KiB for stage 2 clang.

This patch makes `NoopStatistic` empty and thus reclaims the wasted space.  The
clang size is even smaller than applying D69428 (slightly smaller in both .bss and
.text).
```
# This means the D69428 optimization on clang is mostly nullified by this patch.
HEAD+D69428: size(.bss) = 0x0725a8
HEAD+D101211: size(.bss) = 0x072238

# bloaty - HEAD+D69428 vs HEAD+D101211
# With D101211, we also save a lot of string table space (.rodata).
    FILE SIZE        VM SIZE
 --------------  --------------
  -0.0%     -32  -0.0%     -24    .eh_frame
  -0.0%    -336  [ = ]       0    .symtab
  -0.0%    -360  [ = ]       0    .strtab
  [ = ]       0  -0.2%    -880    .bss
  -0.0% -2.11Ki  -0.0% -2.11Ki    .rodata
  -0.0% -2.89Ki  -0.0% -2.89Ki    .text
  -0.0% -5.71Ki  -0.0% -5.88Ki    TOTAL
```

Note: LoopFuse is a disabled pass. This patch adds `#if LLVM_ENABLE_STATS` so
`OptimizationRemarkMissed` is skipped in LLVM_ENABLE_STATS==0 builds.  If these
`OptimizationRemarkMissed` are useful and not noisy, we can replace
`llvm::Statistic` with `llvm::TrackingStatistic` in the future.

Reviewed By: lattner

Differential Revision: https://reviews.llvm.org/D101211

3 years ago[RISCV] Match splatted load to scalar load + splat. Form strided load during isel.
Craig Topper [Mon, 26 Apr 2021 20:23:22 +0000 (13:23 -0700)]
[RISCV] Match splatted load to scalar load + splat. Form strided load during isel.

This modifies my previous patch to push the strided load formation
to isel. This gives us opportunity to fold the splat into a .vx
operation first. Using a scalar register and a .vx operation reduces
vector register pressure which can be important for larger LMULs.

If we can't fold the splat into a .vx operation, then it can make
sense to use a strided load to free up the vector arithmetic
ALU to do actual arithmetic rather than tying it up with vmv.v.x.

Reviewed By: khchen

Differential Revision: https://reviews.llvm.org/D101138

3 years agoAdd LLVM_ATTRIBUTE_USED to fix problems which could be exposed by aggressive global...
Fangrui Song [Mon, 26 Apr 2021 20:31:37 +0000 (13:31 -0700)]
Add LLVM_ATTRIBUTE_USED to fix problems which could be exposed by aggressive global pointer variable removal

Note to BuryPointer.cpp:GraveYard. 'unused' cannot prevent (1) dead store
elimination and (2) removal of the global pointer variable (D69428) but 'used' can.

Discovered when comparing link maps between HEAD+D69428 and HEAD.

Reviewed By: lattner

Differential Revision: https://reviews.llvm.org/D101217

3 years ago[gcov] Set nounwind and respect module flags metadata "frame-pointer" & "uwtable...
Fangrui Song [Mon, 26 Apr 2021 20:30:20 +0000 (13:30 -0700)]
[gcov] Set nounwind and respect module flags metadata "frame-pointer" & "uwtable" for synthesized functions

This applies the D100251 mechanism to the gcov instrumentation pass.

With this patch, `-fno-omit-frame-pointer` in
`clang -fprofile-arcs -O1 -fno-omit-frame-pointer` will be respected for synthesized
`__llvm_gcov_writeout,__llvm_gcov_reset,__llvm_gcov_init` functions: the frame pointer
will be kept (note: on many targets -O1 eliminates the frame pointer by default).

`clang -fno-exceptions -fno-asynchronous-unwind-tables -g -fprofile-arcs` will
produce .debug_frame instead of .eh_frame.

Fix: https://github.com/ClangBuiltLinux/linux/issues/955

Reviewed By: nickdesaulniers

Differential Revision: https://reviews.llvm.org/D101129

3 years ago[compiler-rt][hwasan] Add definition for Symbolizer::SymbolizeFrame
Leonard Chan [Thu, 25 Mar 2021 22:24:12 +0000 (15:24 -0700)]
[compiler-rt][hwasan] Add definition for Symbolizer::SymbolizeFrame

This is undefined if SANITIZER_SYMBOLIZER_MARKUP is 1, which is the case for
Fuchsia, and will result in a undefined symbol error. This function is needed
by hwasan for online symbolization, but is not needed for us since we do
offline symbolization.

Differential Revision: https://reviews.llvm.org/D99386

3 years ago[libc++] [test] Fix a Python warning.
Arthur O'Dwyer [Mon, 26 Apr 2021 20:00:56 +0000 (16:00 -0400)]
[libc++] [test] Fix a Python warning.

params.py:106: SyntaxWarning: "is" with a literal. Did you mean "=="?

3 years ago[libc++] [test] Improve test_exceptions() in each string.modifiers test.
Arthur O'Dwyer [Fri, 16 Apr 2021 21:31:33 +0000 (17:31 -0400)]
[libc++] [test] Improve test_exceptions() in each string.modifiers test.

When checking the strong exception guarantee, also check that
iterators haven't been invalidated.

Reviewed as part of https://reviews.llvm.org/D98573

3 years ago[libc++] Remove the special logic for "noexcept iterators" in basic_string.
Arthur O'Dwyer [Fri, 16 Apr 2021 21:49:57 +0000 (17:49 -0400)]
[libc++] Remove the special logic for "noexcept iterators" in basic_string.

This reverts a large chunk of http://reviews.llvm.org/D15862 ,
and also fixes bugs in `insert`, `append`, and `assign`, which are now regression-tested.
(Thanks to Tim Song for pointing out the bug in `append`!)

Before this patch, we did a special dance in `append`, `assign`, and `insert`
(but not `replace`). All of these require the strong exception guarantee,
even when the user-provided InputIterator might have throwing operations.

The naive way to accomplish this is to construct a temporary string and
then append/assign/insert from the temporary; i.e., finish all the potentially
throwing and self-inspecting InputIterator operations *before* starting to
modify self. But this is slow, so we'd like to skip it when possible.

The old code (D15682) attempted to check that specific iterator operations
were nothrow: it assumed that if the iterator operations didn't throw, then
it was safe to iterate the input range multiple times and therefore it was
safe to use the fast-path non-naive version. This was wrong for two reasons:
(1) the old code checked the wrong operations (e.g. checked noexceptness of `==`,
but the code that ran used `!=`), and (2) the conversion of value_type to char
could still throw, or inspect the contents of self.

The new code is much simpler, although still much more complicated than it
really could be. We'll likely revisit this codepath at some point, but for now
this patch suffices to get it passing all the new regression tests.

The added tests all fail before this patch, and succeed afterward.
See https://quuxplusone.github.io/blog/2021/04/17/pathological-string-appends/

Differential Revision: https://reviews.llvm.org/D98573

3 years ago[libc++] __bit_iterator mustn't rely on deprecated SMF generation.
Arthur O'Dwyer [Sat, 24 Apr 2021 12:20:13 +0000 (08:20 -0400)]
[libc++] __bit_iterator mustn't rely on deprecated SMF generation.

This allows us to turn -Wdeprecated-copy back on. We turned it off
in 3b71de41cc7c7 because Clang's implementation became more stringent
and started diagnosing the old code here.

Differential Revision: https://reviews.llvm.org/D101183

3 years ago[gn build] reformat all gn files
Nico Weber [Mon, 26 Apr 2021 20:04:23 +0000 (16:04 -0400)]
[gn build] reformat all gn files

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

3 years ago[gn build] (manually) port f8de9aaef2f4
Nico Weber [Mon, 26 Apr 2021 20:03:50 +0000 (16:03 -0400)]
[gn build] (manually) port f8de9aaef2f4

3 years ago[libc] Add differential quality and perf analysis targets for sinf and cosf.
Siva Chandra Reddy [Sat, 24 Apr 2021 06:21:12 +0000 (06:21 +0000)]
[libc] Add differential quality and perf analysis targets for sinf and cosf.

 Infrastructure needed for setting up the diff binaries has been added.
 Along the way, an exhaustive test for sinf and cosf have also been added.

Reviewed By: lntue

Differential Revision: https://reviews.llvm.org/D101276

3 years agoSupport DW_FORM_strx* in llvm-dwp.
Ali Tamur [Mon, 26 Apr 2021 19:09:10 +0000 (12:09 -0700)]
Support DW_FORM_strx* in llvm-dwp.

Currently llvm-dwp only handled DW_FORM_string and DW_FORM_GNU_str_index; with this patch it also starts to handle DW_FORM_strx[1-4]?

Reviewed By: dblaikie

Differential Revision: https://reviews.llvm.org/D75485

3 years ago[libc] Use different alignment for memcpy between ARM and x86.
Guillaume Chatelet [Mon, 26 Apr 2021 19:30:00 +0000 (19:30 +0000)]
[libc] Use different alignment for memcpy between ARM and x86.

Aligned copy used to be 'destination aligned' for x86 but this decision was reverted in D93457 where we noticed that it was better for ARM to be 'source aligned'.
More benchmarking confirmed that it can be up to 30% faster to align copy to destination for x86. This Patch offers both implementations and switches x86 back to destination aligned.
It also fixes alignment to 32 byte on x86.

Differential Revision: https://reviews.llvm.org/D101296

3 years ago[llvm-rc] Add a GNU windres-like frontend to llvm-rc
Martin Storsjö [Thu, 15 Apr 2021 11:38:27 +0000 (14:38 +0300)]
[llvm-rc] Add a GNU windres-like frontend to llvm-rc

This primarily parses a different set of options and invokes the same
resource compiler as llvm-rc normally. Additionally, it can convert
directly to an object file (which in MSVC style setups is done with the
separate cvtres tool, or by the linker).

(GNU windres also supports other conversions; from coff object file back
to .res, and from .res or object file back to .rc form; that's not yet
implemented.)

The other bigger complication lies in being able to imply or pass the
intended target triple, to let clang find the corresponding mingw sysroot
for finding include files, and for specifying the default output object
machine format.

It can be implied from the tool triple prefix, like
`<triple>-[llvm-]windres` or picked up from the windres option e.g.
`-F pe-x86-64`. In GNU windres, that option takes BFD style format names
such as pe-i386 or pe-x86-64. As libbfd in binutils doesn't support
Windows on ARM, there's no such canonical name for the ARM targets.
Therefore, as an LLVM specific extension, this option is extended to
allow passing full triples, too.

Differential Revision: https://reviews.llvm.org/D100756

3 years ago[Verifier] Tidy up the code a bit, NFC.
Chris Lattner [Sun, 25 Apr 2021 23:15:17 +0000 (16:15 -0700)]
[Verifier] Tidy up the code a bit, NFC.

This tidies up the code a bit:
 * Eliminate the ctx member, which doesn't need to be stored.
 * Rename verify(Operation) to make it more clear that it is
   doing more than verifyOperation and that the dominance check
   isn't being done multiple times.
 * Rename mayNotHaveTerminator which was confusing about whether
   it wasn't known whether it had a terminator, when it is really
   about whether it is legal to have a terminator.
 * Some minor optimizations: don't check for RegionKindInterface
   if there are no regions.  Don't do two passes over the
   operations in a block in OperationVerifier::verifyDominance when
   one will do.

The optimizations are actually a measurable (but minor) win in some
CIRCT cases.

Differential Revision: https://reviews.llvm.org/D101267

3 years agoRemove obsolete using declaration that was just kept around to help with API adoption.
Chris Lattner [Sun, 25 Apr 2021 23:57:45 +0000 (16:57 -0700)]
Remove obsolete using declaration that was just kept around to help with API adoption.

Differential Revision: https://reviews.llvm.org/D101269

3 years ago[clang][driver] Use the provided arch name for a Darwin target triple
Alex Lorenz [Mon, 26 Apr 2021 17:56:30 +0000 (10:56 -0700)]
[clang][driver] Use the provided arch name for a Darwin target triple

This ensures that the Darwin driver uses a consistent target triple
representation when the triple is printed out to the user.

Differential Revision: https://reviews.llvm.org/D100807

3 years ago[OpenMP][Documentation] Add FAQ entry for dynamically linked libraries
Joseph Huber [Mon, 26 Apr 2021 18:20:14 +0000 (14:20 -0400)]
[OpenMP][Documentation] Add FAQ entry for dynamically linked libraries

Summary:
Add an FAW entry detailing the support for using dynamically linked libraries
with OpenMP Offloading

3 years ago[AMDGPU][llvm-objdump] Fix memory leak in recent commit
Tim Renouf [Mon, 26 Apr 2021 17:49:33 +0000 (18:49 +0100)]
[AMDGPU][llvm-objdump] Fix memory leak in recent commit

Hopefully stops sanitizer fail in AMDGPU llvm-objdump test.

Change-Id: I7331151d1cb65292bd06b6ae283349fe7231cf6b

3 years ago[AST] Fix DeclarationNameInfo introspection
Stephen Kelly [Mon, 26 Apr 2021 17:28:50 +0000 (18:28 +0100)]
[AST] Fix DeclarationNameInfo introspection

Some AST classes return `const DeclarationNameInfo &` instead of
returning by value (eg CXXDependentScopeMemberExpr).

3 years ago[RuntimeDyld] Fixed buffer overflows with absolute symbols
Moritz Sichert [Fri, 30 Oct 2020 10:35:12 +0000 (11:35 +0100)]
[RuntimeDyld] Fixed buffer overflows with absolute symbols

Differential Revision: https://reviews.llvm.org/D95596

3 years agoPreserve the lexical order for global variables during llvm-link merge
Jin Lin [Sat, 24 Apr 2021 05:37:08 +0000 (22:37 -0700)]
Preserve the lexical order for global variables during llvm-link merge

The order of global variables is generated in the order of recursively materializing variables if the global variable has the attribute of hasLocalLinkage or hasLinkOnceLinkage during the module merging. In practice, it is often the exact reverse of source order. This new order may cause performance regression.

The change is to preserve the original lexical order for global variables.

Reviewed By: jdoerfert, dexonsmith

Differential Revision: https://reviews.llvm.org/D94202

3 years ago[AMDGPU] Fix autogenerated wwm-reserved-spill.ll
Sebastian Neubauer [Mon, 26 Apr 2021 17:04:46 +0000 (19:04 +0200)]
[AMDGPU] Fix autogenerated wwm-reserved-spill.ll

Due to a bug in update_llc_test_checks.py, the test is wrongly
coalesced between run lines. Remove common check prefix to fix that.
NFC.

3 years agoXFAIL thinlto testcases on AIX
Wael Yehia [Mon, 26 Apr 2021 14:18:51 +0000 (10:18 -0400)]
XFAIL thinlto testcases on AIX

In D100350, we switched off thinLTO on the AIX OS. So, XFAIL tests that
invoke clang with -flto=thin on AIX.

3 years ago[lldb] Skip TestPointerToMemberTypeDependingOnParentSize on Windows and GCC
Raphael Isemann [Mon, 26 Apr 2021 16:52:07 +0000 (18:52 +0200)]
[lldb] Skip TestPointerToMemberTypeDependingOnParentSize on Windows and GCC

The test added in D100977 is failing to compile on these platforms. This seems
to be caused by GCC, MSVC and Clang@Windows rejecting the code because
`ToLayout` isn't complete when pointer_to_member_member is declared (even though
that seems to be valid code).

This also reverts the test changes in the lazy-loading test from D100977 as
that failed for the same reason.

3 years ago[Lexer] Allow LLLexer to be used as an API
William S. Moses [Wed, 21 Apr 2021 01:59:45 +0000 (21:59 -0400)]
[Lexer] Allow LLLexer to be used as an API

Explose LLVM Lexer for usage externally as an API

Differential Revision: https://reviews.llvm.org/D100920

3 years ago[MLIR] Remove empty shape operands from `cstr_broadcastable` ops
Frederik Gossen [Mon, 26 Apr 2021 16:33:36 +0000 (18:33 +0200)]
[MLIR] Remove empty shape operands from `cstr_broadcastable` ops

Differential Revision: https://reviews.llvm.org/D101170

3 years ago[lldb] Don't use ::fork or ::vfork on watchOS or tvOS
Jonas Devlieghere [Mon, 26 Apr 2021 16:29:37 +0000 (09:29 -0700)]
[lldb] Don't use ::fork or ::vfork on watchOS or tvOS

Update lldb-server to not use fork or vfork on watchOS and tvOS as these
functions are explicitly marked unavailable there.

llvm-project/lldb/test/API/tools/lldb-server/main.cpp:304:11:
error: 'fork' is unavailable: not available on watchOS
      if (fork() == 0)
          ^
WatchSimulator6.2.sdk/usr/include/unistd.h:447:8: note: 'fork' has been
explicitly marked unavailable here
pid_t    fork(void) __WATCHOS_PROHIBITED __TVOS_PROHIBITED;
         ^
llvm-project/lldb/test/API/tools/lldb-server/main.cpp:307:11:
error: 'vfork' is unavailable: not available on watchOS
      if (vfork() == 0)
          ^
WatchSimulator6.2.sdk/usr/include/unistd.h:602:8: note: 'vfork' has been
explicitly marked unavailable here
pid_t    vfork(void) __WATCHOS_PROHIBITED __TVOS_PROHIBITED;
         ^

3 years ago[flang] Check for attributes specific to dummy arguments
Peter Steinfeld [Fri, 23 Apr 2021 02:29:14 +0000 (19:29 -0700)]
[flang] Check for attributes specific to dummy arguments

We were not checking that attributes that are supposed to be specific to
dummy arguments were not being used for local entities.  I added the checks
along with tests for them.

After implementing these new checks, I found that one of the tests in
separate-mp02.f90 was erroneous, and I fixed it.

Differential Revision: https://reviews.llvm.org/D101126

3 years ago[Doc] Add SYCLSupport.rst to index toctree.
Alexey Bader [Mon, 26 Apr 2021 13:15:03 +0000 (16:15 +0300)]
[Doc] Add SYCLSupport.rst to index toctree.

3 years ago[CSSPGO] Explicitly disallow Binary and Compact Binary profile format for CSSPGO
Wenlei He [Sat, 24 Apr 2021 17:37:55 +0000 (10:37 -0700)]
[CSSPGO] Explicitly disallow Binary and Compact Binary profile format for CSSPGO

CSSPGO only supports text and extended binary profile now. Raw binary does not have the metadata section CSSPGO needs, and Compact binary profile needs special handling for GUID based context names, which is not yet implemented.

Disasslow these two format for CSSPGO profile writing to avoid silently generating invalid profiles.

Differential Revision: https://reviews.llvm.org/D101300

3 years ago[SimplifyCFG] Preserve metadata when unconditionalizing branches (constant condition).
Michael Kruse [Mon, 26 Apr 2021 15:52:06 +0000 (10:52 -0500)]
[SimplifyCFG] Preserve metadata when unconditionalizing branches (constant condition).

When replacing a conditional branch by an unconditional one because the condition is a constant, transfer the metadata to the new branch instruction.

Part of fix for llvm.org/PR50060

Reviewed By: lebedev.ri

Differential Revision: https://reviews.llvm.org/D101141

3 years ago[MLIR][Shape] Ensure to preserve op type of `shape.broadcast`
Frederik Gossen [Mon, 26 Apr 2021 15:52:37 +0000 (17:52 +0200)]
[MLIR][Shape] Ensure to preserve op type of `shape.broadcast`

Ensure to preserve the correct type during when folding and canonicalization.
`shape.broadcast` of of a single operand can only be folded away if the argument
type is correct.

Differential Revision: https://reviews.llvm.org/D101158

3 years ago[libc++][NFC] Remove non-ASCII from comment.
Mark de Wever [Mon, 26 Apr 2021 15:53:20 +0000 (17:53 +0200)]
[libc++][NFC] Remove non-ASCII from comment.

Seems our check-generated-output doesn't work as intended. Will
investigate it further.

3 years ago[AMDGPU] Use MapVector for WWMReservedRegs
Sebastian Neubauer [Mon, 26 Apr 2021 15:43:00 +0000 (17:43 +0200)]
[AMDGPU] Use MapVector for WWMReservedRegs

Use MapVector instead of SmallDenseMap because it has a deterministic
iteration order.

Differential Revision: https://reviews.llvm.org/D101299

3 years ago[dexter] Update failing regression test
OCHyams [Mon, 26 Apr 2021 15:33:10 +0000 (16:33 +0100)]
[dexter] Update failing regression test

Commit 7deb970ef added REQUIRES and UNSUPPORTED lines to a test which is
sensitive to line number changes. Update the line number reference in the test.

3 years ago[SYCL][Doc] Add design document for SYCL mode
Alexey Bader [Wed, 24 Mar 2021 13:25:20 +0000 (16:25 +0300)]
[SYCL][Doc] Add design document for SYCL mode

Initial version of the document covers address space handling

Differential Revision: https://reviews.llvm.org/D99488

3 years ago[X86] Rename multiclass ATOMIC_LOAD_BINOP -> ATOMIC_RMW_BINOP. NFCI.
Simon Pilgrim [Mon, 26 Apr 2021 14:10:20 +0000 (15:10 +0100)]
[X86] Rename multiclass ATOMIC_LOAD_BINOP -> ATOMIC_RMW_BINOP. NFCI.

Noticed while triaging the rG2149aa73f640c96 regressions - the LXADD ops are load+store RMW instructions, not just loads.

3 years ago[lld-macho] Add option --error-limit=N
Greg McGary [Mon, 26 Apr 2021 00:28:49 +0000 (17:28 -0700)]
[lld-macho] Add option --error-limit=N

Add option to limit (or remove limits) on the number of errors printed before exiting. This option exists in the other lld ports: COFF & ELF.

Differential Revision: https://reviews.llvm.org/D101274

3 years agoRevert "Only ignore -Wdeprecated-copy if the used compiler supports the warning"
Dávid Bolvanský [Mon, 26 Apr 2021 14:08:18 +0000 (16:08 +0200)]
Revert "Only ignore -Wdeprecated-copy if the used compiler supports the warning"

This reverts commit b125c034af89e32d5258d593080ff0c4810776a0.

Some bots say " missing binary operator before token "(" "

3 years ago[clang][XCore] Define __xcore__ for XCore target.
Nigel Perks [Wed, 21 Apr 2021 17:41:19 +0000 (18:41 +0100)]
[clang][XCore] Define __xcore__ for XCore target.

The headers shipped with the XMOS XCore compiler expect __xcore__ to be defined.
The __XS1B__ macro, already defined, is for the default subtarget.

No other targets affected.

3 years ago[TableGen] Remove a TODO comment that is no longer relevant [NFC]
Paul C. Anagnostopoulos [Mon, 26 Apr 2021 14:00:19 +0000 (10:00 -0400)]
[TableGen] Remove a TODO comment that is no longer relevant [NFC]

3 years agoOnly ignore -Wdeprecated-copy if the used compiler supports the warning
Dávid Bolvanský [Mon, 26 Apr 2021 13:44:52 +0000 (15:44 +0200)]
Only ignore -Wdeprecated-copy if the used compiler supports the warning

This is needed after
  https://reviews.llvm.org/rG9658d045926545e62cc3f963fe611d7c5d0c9d98
which introduced code that at least didn't compile clean with clang 8.

Reviewed By: xbolva00

Differential Revision: https://reviews.llvm.org/D101288

3 years ago[TableGen] Change assertion information from a tuple to a struct [NFC]
Paul C. Anagnostopoulos [Tue, 20 Apr 2021 14:00:24 +0000 (10:00 -0400)]
[TableGen] Change assertion information from a tuple to a struct [NFC]

Differential Revision: https://reviews.llvm.org/D100854

3 years ago[SYCL] Implement SYCL address space attributes handling
Alexey Bader [Fri, 20 Dec 2019 09:05:42 +0000 (12:05 +0300)]
[SYCL] Implement SYCL address space attributes handling

Default address space (applies when no explicit address space was
specified) maps to generic (4) address space.

Added SYCL named address spaces `sycl_global`, `sycl_local` and
`sycl_private` defined as sub-sets of the default address space.

Static variables without address space now reside in global address
space when compile for SPIR target, unless they have an explicit address
space qualifier in source code.

Differential Revision: https://reviews.llvm.org/D89909

3 years ago[InstCombine] C - ctpop(a) - > ctpop(~a)) if C is bitwidth (PR50104)
Dávid Bolvanský [Mon, 26 Apr 2021 12:22:14 +0000 (14:22 +0200)]
[InstCombine] C - ctpop(a) - > ctpop(~a)) if C is bitwidth (PR50104)

Proof: https://alive2.llvm.org/ce/z/mncA9K
Solves https://bugs.llvm.org/show_bug.cgi?id=50104

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D101257

3 years agoAArch64: support atomics in GISel
Tim Northover [Thu, 22 Apr 2021 12:30:30 +0000 (13:30 +0100)]
AArch64: support atomics in GISel

3 years ago[lldb] Use forward type in pointer-to-member
Emre Kultursay [Mon, 26 Apr 2021 13:20:33 +0000 (15:20 +0200)]
[lldb] Use forward type in pointer-to-member

This change is similar in spirit to the change at:
https://reviews.llvm.org/rG34c697c85e9d0af11a72ac4df5578aac94a627b3

It fixes the problem where the layout of a type was being accessed
while its base classes were not populated yet; which caused an
incorrect layout to be produced and cached.

This fixes PR50054

Reviewed By: teemperor

Differential Revision: https://reviews.llvm.org/D100977

3 years ago[AMDGPU][llvm-objdump] Add lit.local.cfg missing from recent commit
Tim Renouf [Mon, 26 Apr 2021 13:06:56 +0000 (14:06 +0100)]
[AMDGPU][llvm-objdump] Add lit.local.cfg missing from recent commit

Stops llvm-objdump tests failing when AMDGPU target is not supported.

Change-Id: Ic4ae443958c41c303ff6bee0966e5f21ab7a1851

3 years ago[MC][AMDGPU][llvm-objdump] Synthesized local labels in disassembly
Tim Renouf [Fri, 23 Apr 2021 07:59:24 +0000 (08:59 +0100)]
[MC][AMDGPU][llvm-objdump] Synthesized local labels in disassembly

1. Add an accessor function to MCSymbolizer to retrieve addresses
   referenced by a symbolizable operand, but not resolved to a symbol.
   That way, the caller can synthesize labels at those addresses and
   then retry disassembling the section.

2. Implement that in AMDGPU -- a failed symbol lookup results in the
   address being added to a vector returned by the new function.

3. Use that in llvm-objdump when using MCSymbolizer (which only happens
   on AMDGPU) and SymbolizeOperands is on.

Differential Revision: https://reviews.llvm.org/D101145

Change-Id: I19087c3bbfece64bad5a56ee88bcc9110d83989e

3 years ago[IndVars] avoid 'tmp' names in test file with auto-generated CHECK lines; NFC
Sanjay Patel [Mon, 26 Apr 2021 12:30:27 +0000 (08:30 -0400)]
[IndVars] avoid 'tmp' names in test file with auto-generated CHECK lines; NFC

The script warns that this could be buggy when updating.

3 years ago[LSR][DebugInfo] Don't unnecessarily drop DebugLocs
Yuanbo Li [Mon, 26 Apr 2021 12:07:18 +0000 (13:07 +0100)]
[LSR][DebugInfo] Don't unnecessarily drop DebugLocs

When transforming a loop terminating condition into a "max" comparison,
the DebugLoc from the old condition should be set on the newly created
comparison. They are the same operation, just optimized. Fixes PR48067.

Differential Revision: https://reviews.llvm.org/D98218

3 years ago[AArch64][SVE] Add missing patterns for scalar versions of SQSHL/UQSHL
Bradley Smith [Thu, 22 Apr 2021 11:15:43 +0000 (12:15 +0100)]
[AArch64][SVE] Add missing patterns for scalar versions of SQSHL/UQSHL

Differential Revision: https://reviews.llvm.org/D101058

3 years ago[InstCombine] Added testcases for PR50104, NFC
Dávid Bolvanský [Sun, 25 Apr 2021 14:33:57 +0000 (16:33 +0200)]
[InstCombine] Added testcases for PR50104, NFC

3 years ago[ARM] Expand VMOVRRD simplification pattern
David Green [Mon, 26 Apr 2021 11:27:38 +0000 (12:27 +0100)]
[ARM] Expand VMOVRRD simplification pattern

This expands the VMOVRRD(extract(..(build_vector(a, b, c, d)))) pattern,
to also handle insert_vectors. Providing we can find the correct insert,
this helps further simplify patterns by removing the redundant VMOVRRD.

Differential Revision: https://reviews.llvm.org/D100245

3 years ago[ARM] Additional soft float BE test. NFC
David Green [Mon, 26 Apr 2021 09:06:28 +0000 (10:06 +0100)]
[ARM] Additional soft float BE test. NFC

3 years ago[NFC] Add scalable vectorisation tests for int/FP <> int/FP conversions
David Sherwood [Tue, 6 Apr 2021 12:23:44 +0000 (13:23 +0100)]
[NFC] Add scalable vectorisation tests for int/FP <> int/FP conversions

We can already vectorize loops that involve int<>int, fp<>fp, int<>fp
and fp<>int conversions, however we didn't previously have any tests
for them. This patch adds some tests for each conversion type.

Differential Revision: https://reviews.llvm.org/D99951

3 years ago[AArch64] Add AArch64TTIImpl::getMaskedMemoryOpCost function
David Sherwood [Fri, 16 Apr 2021 15:08:38 +0000 (16:08 +0100)]
[AArch64] Add AArch64TTIImpl::getMaskedMemoryOpCost function

When vectorising for AArch64 targets if you specify the SVE attribute
we automatically then treat masked loads and stores as legal. Also,
since we have no cost model for masked memory ops we believe it's
cheap to use the masked load/store intrinsics even for fixed width
vectors. This can lead to poor code quality as the intrinsics will
currently be scalarised in the backend. This patch adds a basic
cost model that marks fixed-width masked memory ops as significantly
more expensive than for scalable vectors.

Tests for the cost model are added here:

  Transforms/LoopVectorize/AArch64/masked-op-cost.ll

Differential Revision: https://reviews.llvm.org/D100745

3 years ago[VPlan] Make blocksOnly work properly with ranges over const pointers.
Florian Hahn [Mon, 26 Apr 2021 09:16:25 +0000 (10:16 +0100)]
[VPlan] Make blocksOnly work properly with ranges over const pointers.

When iterating over const blocks, the base type in the lambdas needs
to use const VPBlockBase *, otherwise it cannot be used with input
iterators over const VPBlockBase.

Also adjust the type of the input iterator range to const &, as it
does not take ownership of the input range.

3 years ago[flang][driver] Fine-tune `-fdebug-dump-symbols`
Andrzej Warzynski [Fri, 16 Apr 2021 14:07:09 +0000 (14:07 +0000)]
[flang][driver] Fine-tune `-fdebug-dump-symbols`

When generating output for `-fdebug-dump-symbols`, make sure that
BuildRuntimeDerivedTypeTables is also run. This change is needed in
order to make the implementation of `-fdebug-dump-symbols` in
`flang-new` consistent with `f18`. It also allows us to port more tests
to use the new driver whenever it is enabled.

Differential Revision: https://reviews.llvm.org/D100649

3 years ago[Clang][Coroutine][DebugInfo] Relax test ordering requirement
Jeremy Morse [Mon, 26 Apr 2021 08:48:36 +0000 (09:48 +0100)]
[Clang][Coroutine][DebugInfo] Relax test ordering requirement

The test added in D97533 (and modified by this patch) has some overly
strict printed metadata ordering requirements, specifically the
interleaving of DILocalVariable nodes and DILocation nodes. Slight changes
in metadata emission can easily break this unfortunately.

This patch stops after clang codegen rather than allowing the coro splitter
to run,  and reduces the need for ordering: it picks out the
DILocalVariable nodes being sought, in any order (CHECK-DAG), and doesn't
examine any DILocations. The implicit CHECK-NOT is what's important: the
test seeks to ensure a duplicate set of DILocalVariables aren't emitted in
the same scope.

Differential Revision: https://reviews.llvm.org/D100298

3 years ago[clang-format] fix indent in alignChainedConditionals
Krasimir Georgiev [Thu, 22 Apr 2021 08:56:15 +0000 (10:56 +0200)]
[clang-format] fix indent in alignChainedConditionals

Clang-format was indenting the lines following the `?` in the added test
case by +5 instead of +4. This only happens in a very specific
situation, where the `?` is followed by a multiline block comment, as in
the example. This fix addresses this without regressing any of the
existing tests.

Differential Revision: https://reviews.llvm.org/D101033

3 years ago[ARM] Ensure loop invariant active.lane.mask operands
David Green [Mon, 26 Apr 2021 09:04:33 +0000 (10:04 +0100)]
[ARM] Ensure loop invariant active.lane.mask operands

CGP can move instructions like a ptrtoint into a loop, but the
MVETailPredication when converting them will currently assume invariant
trip counts. This tries to ensure the operands are loop invariant, and
bails if not.

Differential Revision: https://reviews.llvm.org/D100550

3 years ago[RISCV] Add REQUIRES: riscv-registered-target for RISC-V test cases.
Hsiangkai Wang [Mon, 26 Apr 2021 09:01:49 +0000 (17:01 +0800)]
[RISCV] Add REQUIRES: riscv-registered-target for RISC-V test cases.

3 years ago[llvm-dwarfdump] Fix split-dwarf bug in stats for inlined var loc cov
Djordje Todorovic [Mon, 26 Apr 2021 08:14:12 +0000 (01:14 -0700)]
[llvm-dwarfdump] Fix split-dwarf bug in stats for inlined var loc cov

Initial (D96045) patch didn't handle split dwarf cases,
so this fixes that bug.

In addition, before applying this patch, we had a slowdown
that happened after the D96045. With this patch,
the slowdown will be fixed as well.

Differential Revision: https://reviews.llvm.org/D100951

3 years ago[clang][deps] NFC: Fix typo
Jan Svoboda [Mon, 26 Apr 2021 08:55:24 +0000 (10:55 +0200)]
[clang][deps] NFC: Fix typo

3 years ago[clang][deps] Only generate absolute paths when asked to
Jan Svoboda [Thu, 22 Apr 2021 11:49:25 +0000 (13:49 +0200)]
[clang][deps] Only generate absolute paths when asked to

Add option to `clang-scan-deps` to enable/disable generation of command-line arguments with absolute paths. This is essentially a revert of D100533, but with improved naming and added test.

Reviewed By: dexonsmith

Differential Revision: https://reviews.llvm.org/D101051

3 years ago[OpenCL] Add inc/dec/cmpxchg C++ legacy atomics with generic
Sven van Haastregt [Mon, 26 Apr 2021 08:46:11 +0000 (09:46 +0100)]
[OpenCL] Add inc/dec/cmpxchg C++ legacy atomics with generic

Mirror the remaining C++ for OpenCL specific builtins from opencl-c.h
to the TableGen builtin functions.

Fixes PR50041 (part 2).

3 years ago[RISCV] Implement the vwcvt{u}.x.x.v/vncvt.x.x.w builtin.
Hsiangkai Wang [Tue, 20 Apr 2021 07:31:13 +0000 (15:31 +0800)]
[RISCV] Implement the vwcvt{u}.x.x.v/vncvt.x.x.w builtin.

Differential Revision: https://reviews.llvm.org/D100824

3 years ago[RISCV] Implement the pseudo compare builtin.
Hsiangkai Wang [Tue, 20 Apr 2021 06:50:42 +0000 (14:50 +0800)]
[RISCV] Implement the pseudo compare builtin.

Differential Revision: https://reviews.llvm.org/D100823

3 years ago[RISCV] Implement the vfabs.v/vfneg.v builtin.
Hsiangkai Wang [Tue, 20 Apr 2021 05:17:40 +0000 (13:17 +0800)]
[RISCV] Implement the vfabs.v/vfneg.v builtin.

Differential Revision: https://reviews.llvm.org/D100822

3 years ago[RISCV] Implement the vmmv.m/vmnot.m builtin.
Hsiangkai Wang [Tue, 20 Apr 2021 03:12:53 +0000 (11:12 +0800)]
[RISCV] Implement the vmmv.m/vmnot.m builtin.

Differential Revision: https://reviews.llvm.org/D100821

3 years ago[RISCV] Cleanup setOperationAction calls for INTRINSIC_WO_CHAIN/INTRINSIC_W_CHAIN
Craig Topper [Mon, 26 Apr 2021 06:36:28 +0000 (23:36 -0700)]
[RISCV] Cleanup setOperationAction calls for INTRINSIC_WO_CHAIN/INTRINSIC_W_CHAIN

We have several extensions that need i32 to be Custom for
INTRINSIC_WO_CHAIN with RV64 so enable it for all RV64.

For V extension, make i32 Custom for RV64 and i64 Custom for RV32.
When the i32 or i64 is legal, the operation action doesn't matter.
LegalizeDAG checks MVT::Other rather than the real type.

3 years ago[RISCV] Optimize addition with immediate
Ben Shi [Mon, 26 Apr 2021 05:26:17 +0000 (13:26 +0800)]
[RISCV] Optimize addition with immediate

Reviewed by: craig.topper

Differential Revision: https://reviews.llvm.org/D101244

3 years ago[RISCV] Teach DAG combine what bits Zbp instructions demanded from their inputs.
Craig Topper [Mon, 26 Apr 2021 03:17:13 +0000 (20:17 -0700)]
[RISCV] Teach DAG combine what bits Zbp instructions demanded from their inputs.

This teaches DAG combine that shift amount operands for grev, gorc
shfl, unshfl only read a few bits.

This also teaches DAG combine that grevw, gorcw, shflw, unshflw,
bcompressw, bdecompressw only consume the lower 32 bits of their
inputs.

In the future we can teach SimplifyDemandedBits to also propagate
demanded bits of the output to the inputs in some cases.

3 years ago[RISCV] [1/2] Add IR intrinsic for Zbe extension
Levy Hsu [Mon, 26 Apr 2021 01:49:40 +0000 (18:49 -0700)]
[RISCV] [1/2] Add IR intrinsic for Zbe extension

RV32/64:
bcompress
bdecompress

RV64 ONLY:
bcompressw
bdecompressw

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D101143

3 years ago[DebugInfo][AIX] Set target debugger-tune default to dbx
Jinsong Ji [Sun, 25 Apr 2021 21:59:24 +0000 (21:59 +0000)]
[DebugInfo][AIX] Set target debugger-tune default to dbx

https://reviews.llvm.org/D99400 set clang DefaultDebuggerTuning for AIX
to dbx. However, we still need to update the target default so that llc
and other tools will get the same default debuggertuning, and avoid
passing extra options in LTO.

Reviewed By: #powerpc, shchenz, dblaikie

Differential Revision: https://reviews.llvm.org/D101197

3 years ago[ORC] Avoid invalidating iterators in EHFrameRegistrationPlugin.
Lang Hames [Sun, 25 Apr 2021 23:47:56 +0000 (16:47 -0700)]
[ORC] Avoid invalidating iterators in EHFrameRegistrationPlugin.

In EHFrameRegistrationPlugin::notifyTransferringResources if SrcKey had
eh-frames associated but DstKey did not we would create a new entry for DskKey,
invalidating the iterator for SrcKey in the process. This commit fixes that by
removing SrcKey first in this case.

3 years ago[clangd] Rename HeuristicResolver::resolveCallExpr() to resolveTypeOfCallExpr()
Nathan Ridge [Mon, 19 Apr 2021 06:57:16 +0000 (02:57 -0400)]
[clangd] Rename HeuristicResolver::resolveCallExpr() to resolveTypeOfCallExpr()

Differential Revision: https://reviews.llvm.org/D100741

3 years ago[clangd] Omit parameter hint for setter functions
Nathan Ridge [Sun, 18 Apr 2021 19:26:04 +0000 (15:26 -0400)]
[clangd] Omit parameter hint for setter functions

Differential Revision: https://reviews.llvm.org/D100731

3 years ago[clangd] Omit parameter hint if parameter name comment is present
Nathan Ridge [Sun, 18 Apr 2021 07:15:15 +0000 (03:15 -0400)]
[clangd] Omit parameter hint if parameter name comment is present

Differential Revision: https://reviews.llvm.org/D100715

3 years ago[scudo][NFC] Fix cast warning
Vitaly Buka [Sun, 25 Apr 2021 07:22:26 +0000 (00:22 -0700)]
[scudo][NFC] Fix cast warning

3 years ago[scudo] Mark ARM64 as supported platform
Vitaly Buka [Sun, 25 Apr 2021 07:21:34 +0000 (00:21 -0700)]
[scudo] Mark ARM64 as supported platform

3 years ago[sanitizer] Use COMPILER_RT_EMULATOR with gtests
Vitaly Buka [Thu, 22 Apr 2021 18:16:21 +0000 (11:16 -0700)]
[sanitizer] Use COMPILER_RT_EMULATOR with gtests

Reviewed By: morehouse

Differential Revision: https://reviews.llvm.org/D100998

3 years ago[NFC][X86][AVX2] Add baseline CodeGen/CostModel tests for interleaved loads/stores...
Roman Lebedev [Sun, 25 Apr 2021 18:35:05 +0000 (21:35 +0300)]
[NFC][X86][AVX2] Add baseline CodeGen/CostModel tests for interleaved loads/stores of i16 w/ strides 2/3/4

`X86TTIImpl::getInterleavedMemoryOpCostAVX2()` currently contains data
only for a handful of tuples. For now, at least add tests for a few more.

I'm guessing that we care how well the patterns codegen since
we use their presumed cost for vectorization decisions,
so i've added codegen tests too.

There's one really easy caveat for these codegen tests:
for interleaved load tests, we really have to ensure that the
deinterleaved vectors are escaped separately. Similarly for stores.

3 years ago[clangd] Use HeuristicResolver to produce a better semantic token for name referring...
Nathan Ridge [Mon, 22 Mar 2021 06:43:41 +0000 (02:43 -0400)]
[clangd] Use HeuristicResolver to produce a better semantic token for name referring to UnresolvedUsingValueDecl

Fixes https://github.com/clangd/clangd/issues/686

Differential Revision: https://reviews.llvm.org/D99056

3 years ago[clangd] Produce semantic token for name referring to UnresolvedUsingValueDecl
Nathan Ridge [Mon, 22 Mar 2021 06:13:53 +0000 (02:13 -0400)]
[clangd] Produce semantic token for name referring to UnresolvedUsingValueDecl

For now, use the token kind Unknown. We may be able to improve on this
using HeuristicResolver.

Differential Revision: https://reviews.llvm.org/D99052

3 years ago[googlemock] Fix warnings about superfluous semicolons after 68ff493dfc950c05c102e09e...
Martin Storsjö [Sun, 25 Apr 2021 20:41:37 +0000 (23:41 +0300)]
[googlemock] Fix warnings about superfluous semicolons after 68ff493dfc950c05c102e09e14a16d06253ffa16. NFC.