platform/upstream/llvm.git
17 months ago[libc++] Remove availability markup for std::format
Louis Dionne [Thu, 16 Mar 2023 17:09:44 +0000 (13:09 -0400)]
[libc++] Remove availability markup for std::format

std::format is currently experimental, so there is technically no
deployment target requirement for it (since the only symbols required
for it are in `libc++experimental.a`).

However, some parts of std::format depend indirectly on the floating
point std::to_chars implementation, which does have deployment target
requirements.

This patch removes all the availability format for std::format and
updates the XFAILs in the tests to properly explain why they fail
on old deployment targets, when they do. It also changes a couple
of tests to avoid depending on floating-point std::to_chars when
it isn't fundamental to the test.

Finally, some tests are marked as XFAIL but I added a comment saying

   TODO FMT This test should not require std::to_chars(floating-point)

These tests do not fundamentally depend on floating-point std::to_chars,
however they end up failing because calling std::format even without a
floating-point argument to format will end up requiring floating-point
std::to_chars. I believe this is an implementation artifact that could
be avoided in all cases where we know the format string at compile-time.
In the tests, I added the TODO comment only to the places where we could
do better and actually avoid relying on floating-point std::to_chars
because we know the format string at compile-time.

Differential Revision: https://reviews.llvm.org/D134598

17 months ago[test] Add tools/llvm-dwarfdump/ARM/lit.local.cfg after D143513
Fangrui Song [Wed, 22 Mar 2023 19:56:13 +0000 (12:56 -0700)]
[test] Add tools/llvm-dwarfdump/ARM/lit.local.cfg after D143513

17 months ago[JITLink] Introduce target flags for Symbol and prepare ObjectLinkingLayer to account...
Stefan Gränitz [Wed, 22 Mar 2023 19:21:59 +0000 (20:21 +0100)]
[JITLink] Introduce target flags for Symbol and prepare ObjectLinkingLayer to account for them

AArch32 branch offsets explicitly encode the target instruction subset (Arm/Thumb) in their least significant bit. We want this bit set (or clear) in addreses we hand out, but the addresses in the LinkGraph should be the real/physical addresses.

This patch allows ELFLinkGraphBuilder's to set target-specific flags in jitlink::Symbol and prepares ObjectLinkingLayer to account for them.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D146641

17 months ago[gn build] Port e655d8a54880
LLVM GN Syncbot [Wed, 22 Mar 2023 19:30:55 +0000 (19:30 +0000)]
[gn build] Port e655d8a54880

17 months agoPrecommit tests for D146327
Jeff Byrnes [Wed, 22 Mar 2023 19:19:00 +0000 (12:19 -0700)]
Precommit tests for D146327

17 months ago[LFTR] Simplify integer case for genLoopLimit [nfc-ish]
Philip Reames [Wed, 22 Mar 2023 19:07:08 +0000 (12:07 -0700)]
[LFTR] Simplify integer case for genLoopLimit [nfc-ish]

The integer case in genLoopLimit reduces down to a special case for narrowing the bitwidth of the limit, and then performing the same expansion we would for a pointer IV.

Differential Revision: https://reviews.llvm.org/D146638

17 months ago[libc++] Granularize __mutex_base
Nikolas Klauser [Thu, 9 Mar 2023 00:14:03 +0000 (01:14 +0100)]
[libc++] Granularize __mutex_base

This also updates the moved code to the current style. (i.e. `_VSTD` -> `std`, `_LIBCPP_INLINE_VISIBILITY` -> `_LIBCPP_HIDE_FROM_ABI`, clang-format).

Reviewed By: Mordante, #libc, EricWF

Spies: arichardson, libcxx-commits, mikhail.ramalho

Differential Revision: https://reviews.llvm.org/D146228

17 months ago[SCEV] Infer no-self-wrap via constant ranges
Philip Reames [Wed, 22 Mar 2023 16:51:58 +0000 (09:51 -0700)]
[SCEV] Infer no-self-wrap via constant ranges

Without this, pointer IVs in loops with small constant trip counts couldn't be proven no-self-wrap. This came up in a new LSR transform, but may also benefit other SCEV consumers as well.

Differential Revision: https://reviews.llvm.org/D146596

17 months ago[clang-tidy][NFC] Move avoid-underscore-in-googletest-name to google folder
Carlos Galvez [Wed, 22 Mar 2023 18:38:10 +0000 (18:38 +0000)]
[clang-tidy][NFC] Move avoid-underscore-in-googletest-name to google folder

Since the check belongs to the google module, it makes sense
that the corresponding test also belongs to the google module.

Differential Revision: https://reviews.llvm.org/D146653

17 months ago[CodeGen] Fix type of MachineRegisterInfo::RegAllocHints. NFC.
Jay Foad [Wed, 22 Mar 2023 16:56:05 +0000 (16:56 +0000)]
[CodeGen] Fix type of MachineRegisterInfo::RegAllocHints. NFC.

The first member of the pair should be unsigned instead of Register
because it is the hint type, 0 for simple (target independent) hints and
other values for target dependent hints.

Differential Revision: https://reviews.llvm.org/D146646

17 months ago[Driver][test] Fix avr-ld.c for -DCLANG_DEFAULT_LINKER=lld after D145646
Fangrui Song [Wed, 22 Mar 2023 18:36:02 +0000 (11:36 -0700)]
[Driver][test] Fix avr-ld.c for -DCLANG_DEFAULT_LINKER=lld after D145646

17 months ago[lldb] Update some uses of Python2 API in typemaps.
Jorge Gorbe Moya [Tue, 21 Mar 2023 22:30:32 +0000 (15:30 -0700)]
[lldb] Update some uses of Python2 API in typemaps.

Python 3 doesn't have a distinction between PyInt and PyLong, it's all
PyLong now.

This also fixes a bug in SetNumberFromObject. This used to crash LLDB:
```
lldb -o "script data=lldb.SBData(); data.SetDataFromUInt64Array([2**63])"
```

The problem happened in the PyInt path:
```
  if (PyInt_Check(obj))
      number = static_cast<T>(PyInt_AsLong(obj));
```
when obj doesn't fit in a signed long, `PyInt_AsLong` would fail with
"OverflowError: Python int too large to convert to C long".

The existing long path does the right thing, as it will call
`PyLong_AsUnsignedLongLong` for uint64_t.

Differential Revision: https://reviews.llvm.org/D146590

17 months ago[clang-format] Annotate noexcept, explicit specifiers as containing expressions
Emilia Dreamer [Wed, 22 Mar 2023 18:26:38 +0000 (20:26 +0200)]
[clang-format] Annotate noexcept, explicit specifiers as containing expressions

The noexcept specifier and explicit specifier can optionally include a
boolean expression to make these specifiers apply conditionally,
however, clang-format didn't set the context for the parenthesized
content of these specifiers, meaning they inherited the parent context,
which usually isn't an expressions, leading to misannotated binary
operators.

This patch applies expression context to the content of these
specifiers, making them similar to the static_assert keyword.

Fixes https://github.com/llvm/llvm-project/issues/44543

Reviewed By: owenpan, MyDeveloperDay

Differential Revision: https://reviews.llvm.org/D146284

17 months agoSymbolFile: ensure that we have a value before invoking `getBitWidth`
Saleem Abdulrasool [Tue, 21 Mar 2023 16:03:54 +0000 (12:03 -0400)]
SymbolFile: ensure that we have a value before invoking `getBitWidth`

Ensure that the variant returned by `member->getValue()` has a value and
is not `Empty`.  Failure to do so will trigger an assertion failure in
`llvm::pdb::Variant::getBitWidth()`.  This can occur when the `static`
member is a forward declaration.

Differential Revision: https://reviews.llvm.org/D146536
Reviewed By: sgraenitz

17 months ago[TSan] Avoid deadlock between ReportRace() and dlopen() interceptor
Julian Lettner [Tue, 21 Mar 2023 23:20:22 +0000 (16:20 -0700)]
[TSan] Avoid deadlock between ReportRace() and dlopen() interceptor

This change prevents rare deadlocks observed for specific macOS/iOS GUI
applications which issue many `dlopen()` calls from multiple different
threads at startup and where TSan finds and reports a race during
startup.  Providing a reliable test for this has been deemed infeasible.

Although I've only observed this deadlock on Apple platforms,
conceptually the cause is not confined to Apple code so the fix lives in
platform-independent code.

Deadlock scenario:
```
Thread 2                    | Thread 4
ReportRace()                |
Lock internal TSan mutexes  |
  &ctx->slot_mtx            |
                            | dlopen() interceptor
                            | OnLibraryLoaded()
                            | MemoryMappingLayout::DumpListOfModules()
                            | calls dyld API, which takes internal lock
                            | lock() interceptor
                            | TSan tries to take internal mutexes again
                            |   &ctx->slot_mtx
call into symbolizer        |
MemoryMappingLayout::DumpListOfModules()
calls dyld API, which hangs on trying to take lock
```
Resulting in:
* Thread 2 has internal TSan mutex, blocked on dyld lock
* Thread 4 has dyld lock, blocked on internal TSan mutex

The fix prevents this situation by not intercepting any of the calls
originating from `MemoryMappingLayout::DumpListOfModules()`.

Stack traces for deadlock between ReportRace() and dlopen() interceptor:
```
thread #2, queue = 'com.apple.root.default-qos'
  frame #0: libsystem_kernel.dylib
  frame #1: libclang_rt.tsan_osx_dynamic.dylib`::wrap_os_unfair_lock_lock_with_options(lock=<unavailable>, options=<unavailable>) at tsan_interceptors_mac.cpp:306:3
  frame #2: dyld`dyld4::RuntimeLocks::withLoadersReadLock(this=0x000000016f21b1e0, work=0x00000001814523c0) block_pointer) at DyldRuntimeState.cpp:227:28
  frame #3: dyld`dyld4::APIs::_dyld_get_image_header(this=0x0000000101012a20, imageIndex=614) at DyldAPIs.cpp:240:11
  frame #4: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::MemoryMappingLayout::CurrentImageHeader(this=<unavailable>) at sanitizer_procmaps_mac.cpp:391:35
  frame #5: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::MemoryMappingLayout::Next(this=0x000000016f2a2800, segment=0x000000016f2a2738) at sanitizer_procmaps_mac.cpp:397:51
  frame #6: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::MemoryMappingLayout::DumpListOfModules(this=0x000000016f2a2800, modules=0x00000001011000a0) at sanitizer_procmaps_mac.cpp:460:10
  frame #7: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::ListOfModules::init(this=0x00000001011000a0) at sanitizer_mac.cpp:610:18
  frame #8: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::Symbolizer::FindModuleForAddress(unsigned long) [inlined] __sanitizer::Symbolizer::RefreshModules(this=0x0000000101100078) at sanitizer_symbolizer_libcdep.cpp:185:12
  frame #9: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::Symbolizer::FindModuleForAddress(this=0x0000000101100078, address=6465454512) at sanitizer_symbolizer_libcdep.cpp:204:5
  frame #10: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::Symbolizer::SymbolizePC(this=0x0000000101100078, addr=6465454512) at sanitizer_symbolizer_libcdep.cpp:88:15
  frame #11: libclang_rt.tsan_osx_dynamic.dylib`__tsan::SymbolizeCode(addr=6465454512) at tsan_symbolize.cpp:106:35
  frame #12: libclang_rt.tsan_osx_dynamic.dylib`__tsan::SymbolizeStack(trace=StackTrace @ 0x0000600002d66d00) at tsan_rtl_report.cpp:112:28
  frame #13: libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedReportBase::AddMemoryAccess(this=0x000000016f2a2a90, addr=4381057136, external_tag=<unavailable>, s=<unavailable>, tid=<unavailable>, stack=<unavailable>, mset=0x00000001012fc310) at tsan_rtl_report.cpp:190:16
  frame #14: libclang_rt.tsan_osx_dynamic.dylib`__tsan::ReportRace(thr=0x00000001012fc000, shadow_mem=0x000008020a4340e0, cur=<unavailable>, old=<unavailable>, typ0=1) at tsan_rtl_report.cpp:795:9
  frame #15: libclang_rt.tsan_osx_dynamic.dylib`__tsan::DoReportRace(thr=0x00000001012fc000, shadow_mem=0x000008020a4340e0, cur=Shadow @ x22, old=Shadow @ 0x0000600002d6b4f0, typ=1) at tsan_rtl_access.cpp:166:3
  frame #16: libclang_rt.tsan_osx_dynamic.dylib`::__tsan_read8(void *) at tsan_rtl_access.cpp:220:5
  frame #17: libclang_rt.tsan_osx_dynamic.dylib`::__tsan_read8(void *) [inlined] __tsan::MemoryAccess(thr=0x00000001012fc000, pc=<unavailable>, addr=<unavailable>, size=8, typ=1) at tsan_rtl_access.cpp:442:3
  frame #18: libclang_rt.tsan_osx_dynamic.dylib`::__tsan_read8(addr=<unavailable>) at tsan_interface.inc:34:3
  <call into TSan from from instrumented code>

thread #4, queue = 'com.apple.dock.fullscreen'
  frame #0:  libsystem_kernel.dylib
  frame #1:  libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::FutexWait(p=<unavailable>, cmp=<unavailable>) at sanitizer_mac.cpp:540:3
  frame #2:  libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::Semaphore::Wait(this=<unavailable>) at sanitizer_mutex.cpp:35:7
  frame #3:  libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::Mutex::Lock(this=0x0000000102992a80) at sanitizer_mutex.h:196:18
  frame #4:  libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedInterceptor::~ScopedInterceptor() [inlined] __sanitizer::GenericScopedLock<__sanitizer::Mutex>::GenericScopedLock(this=<unavailable>, mu=0x0000000102992a80) at sanitizer_mutex.h:383:10
  frame #5:  libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedInterceptor::~ScopedInterceptor() [inlined] __sanitizer::GenericScopedLock<__sanitizer::Mutex>::GenericScopedLock(this=<unavailable>, mu=0x0000000102992a80) at sanitizer_mutex.h:382:77
  frame #6:  libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedInterceptor::~ScopedInterceptor() at tsan_rtl.h:708:10
  frame #7:  libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedInterceptor::~ScopedInterceptor() [inlined] __tsan::TryTraceFunc(thr=0x000000010f084000, pc=0) at tsan_rtl.h:751:7
  frame #8:  libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedInterceptor::~ScopedInterceptor() [inlined] __tsan::FuncExit(thr=0x000000010f084000) at tsan_rtl.h:798:7
  frame #9:  libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedInterceptor::~ScopedInterceptor(this=0x000000016f3ba280) at tsan_interceptors_posix.cpp:300:5
  frame #10: libclang_rt.tsan_osx_dynamic.dylib`__tsan::ScopedInterceptor::~ScopedInterceptor(this=<unavailable>) at tsan_interceptors_posix.cpp:293:41
  frame #11: libclang_rt.tsan_osx_dynamic.dylib`::wrap_os_unfair_lock_lock_with_options(lock=0x000000016f21b1e8, options=OS_UNFAIR_LOCK_NONE) at tsan_interceptors_mac.cpp:310:1
  frame #12: dyld`dyld4::RuntimeLocks::withLoadersReadLock(this=0x000000016f21b1e0, work=0x00000001814525d4) block_pointer) at DyldRuntimeState.cpp:227:28
  frame #13: dyld`dyld4::APIs::_dyld_get_image_vmaddr_slide(this=0x0000000101012a20, imageIndex=412) at DyldAPIs.cpp:273:11
  frame #14: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::MemoryMappingLayout::Next(__sanitizer::MemoryMappedSegment*) at sanitizer_procmaps_mac.cpp:286:17
  frame #15: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::MemoryMappingLayout::Next(this=0x000000016f3ba560, segment=0x000000016f3ba498) at sanitizer_procmaps_mac.cpp:432:15
  frame #16: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::MemoryMappingLayout::DumpListOfModules(this=0x000000016f3ba560, modules=0x000000016f3ba618) at sanitizer_procmaps_mac.cpp:460:10
  frame #17: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::ListOfModules::init(this=0x000000016f3ba618) at sanitizer_mac.cpp:610:18
  frame #18: libclang_rt.tsan_osx_dynamic.dylib`__sanitizer::LibIgnore::OnLibraryLoaded(this=0x0000000101f3aa40, name="<some library>") at sanitizer_libignore.cpp:54:11
  frame #19: libclang_rt.tsan_osx_dynamic.dylib`::wrap_dlopen(filename="<some library>", flag=<unavailable>) at sanitizer_common_interceptors.inc:6466:3
  <library code>
```

rdar://106766395

Differential Revision: https://reviews.llvm.org/D146593

17 months agoRevert "Remove the LINK_COMPONENTS entry from lldb-instr CMakery"
Heejin Ahn [Thu, 2 Mar 2023 20:04:27 +0000 (12:04 -0800)]
Revert "Remove the LINK_COMPONENTS entry from lldb-instr CMakery"

This reverts commit e12a950d90f88aeddaa97d6e7c8fd0bfedc42f73.

D142241 broke `-sBUILD_SHARED_LIBS=ON` build. After investigations in
https://github.com/llvm/llvm-project/issues/60314, the issue that
prompted D142441 now seems gone.

Fixes https://github.com/llvm/llvm-project/issues/60314.

Reviewed By: JDevlieghere

Differential Revision: https://reviews.llvm.org/D145181

17 months ago[RISCV] Remove AnyReg RegisterClass used by .insn instructions. Use custom operand...
Craig Topper [Wed, 22 Mar 2023 17:24:57 +0000 (10:24 -0700)]
[RISCV] Remove AnyReg RegisterClass used by .insn instructions. Use custom operand instead.

The fake register class interferes too much with the autogenerated
register class tables. Especially the fake spill size.

I'm working on .insn support for compressed instructions and adding
AnyRegC broke CodeGen.

17 months ago[CMake] Build runtimes for riscv64-unknown-fuchsia
Petr Hosek [Wed, 22 Mar 2023 04:57:08 +0000 (04:57 +0000)]
[CMake] Build runtimes for riscv64-unknown-fuchsia

This is necessary to have a complete RISC-V toolchain for Fuchsia.

Differential Revision: https://reviews.llvm.org/D146608

17 months ago[gn build] Port 700cd99061ed
LLVM GN Syncbot [Wed, 22 Mar 2023 17:21:25 +0000 (17:21 +0000)]
[gn build] Port 700cd99061ed

17 months agoRestore "[MemProf] Context disambiguation cloning pass [patch 1a/3]"
Teresa Johnson [Wed, 22 Mar 2023 17:01:45 +0000 (10:01 -0700)]
Restore "[MemProf] Context disambiguation cloning pass [patch 1a/3]"

This restores commit d6ad4f01c3dafcab335bca66dac6e36d9eac8421, which was
reverted in commit 883dbb9c86be87593a58ef10b070b3a0564c7fee, along with
a fix for gcc 12.2 build errors in the original commit.

Support for building, printing, and displaying CallsiteContextGraph
which represents the MemProf metadata contexts. Uses CRTP to enable
support for both IR (regular LTO) and summary (ThinLTO). This patch
includes the support for building it in regular LTO mode (from
memprof and callsite metadata), and the next patch will add the
handling for building it from ThinLTO summaries.

Also includes support for dumping the graph to text and to dot files.

Follow-on patches will contain the support for cloning on the graph and
in the IR.

The graph represents the call contexts in all memprof metadata on
allocation calls, with nodes for the allocations themselves, as well as
for the calls in each context. The graph is initially built from the
allocation memprof metadata (or summary) MIBs. It is then updated to
match calls with callsite metadata onto the nodes, updating it to
reflect any inlining performed on those calls.

Each MIB (representing an allocation's call context with allocation
behavior) is assigned a unique context id during the graph build. The
edges and nodes in the graph are decorated with the context ids they
carry. This is used to correctly update the graph when cloning is
performed so that we can uniquify the context for a single (possibly
cloned) allocation.

Differential Revision: https://reviews.llvm.org/D140908

17 months ago[RISCV] Add FallbackRegAltNameIndex to ABIRegAltName.
Craig Topper [Wed, 22 Mar 2023 17:07:18 +0000 (10:07 -0700)]
[RISCV] Add FallbackRegAltNameIndex to ABIRegAltName.

Remove now redundant fake ABI names from vector registers.

This also fixes a crash that occurs if you use fflags as an instruction
operand in the assembly and use -debug. It's not a valid register
for any instruction since this wouldn't be common. It doesn't have
an ABI name so it crashes the register printing in the debug output.

17 months ago[RISCV] Convert segment registers to VR registers in RISCVMCInstLower.
Craig Topper [Wed, 22 Mar 2023 17:01:17 +0000 (10:01 -0700)]
[RISCV] Convert segment registers to VR registers in RISCVMCInstLower.

Similar to what we do for the LMUL>1 register classes. The printing
is only working today because the segment registers have "ABI" names
set to their base register name.

17 months ago[mlir][tosa] Canonicalize concatenate->slice sequence
Luke Hutton [Wed, 22 Mar 2023 00:54:15 +0000 (00:54 +0000)]
[mlir][tosa] Canonicalize concatenate->slice sequence

Adds a canonicalizer for the concatenate->slice sequence where
an output of slice can be replaced with an input of concatenate.

This is useful in the context of operations with complex inputs
and outputs that are legalized from a framework such as TFL.
For example, a TFL graph (FFT->FFT) will be legalized to the
following TOSA graph:

     <complex input>
         /     \
     slice    slice
         \     /
           FFT
          /   \     -+
       concatenate   |
         /     \     |  Redundant
     slice    slice  |
         \     /    -+
           FFT
         /     \
       concatenate
            |
     <complex output>

Concatenate and slice operations at the boundaries of the graph are
useful as they maintain the correct correspondance of input/output
tensors to the original TFL graph. However, consecutive
complex operations will result in redundant concatenate->slice
sequences which should be removed from the final TOSA graph.

The canonicalization does not currently handle dynamic types.

Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Reviewed By: rsuderman

Differential Revision: https://reviews.llvm.org/D144545

17 months ago[Constant] Inline ConstantInt::getSigned
Craig Topper [Wed, 22 Mar 2023 16:31:51 +0000 (09:31 -0700)]
[Constant] Inline ConstantInt::getSigned

ConstantInt::getSigned calls ConstantInt::get with the IsSigned flag set to true.
That flag normally defaults to false.

For always signed constants the code base is not consistent about whether
it uses ConstantInt::getSigned or ConstantInt::get with IsSigned set to true.
And it's not clear how to decide which way to use.

By making getSigned inline, both ways should generate the same code in
the end.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D146598

17 months ago[DebugInfo] Allow parsing line tables aligned to 4 or 8-byte boundaries
Benjamin Maxwell [Wed, 22 Mar 2023 14:48:28 +0000 (14:48 +0000)]
[DebugInfo] Allow parsing line tables aligned to 4 or 8-byte boundaries

This allows the DWARFDebugLine::SectionParser to try parsing line tables
at 4 or 8-byte boundaries if the unaligned offset appears invalid. If
aligning the offset does not reduce errors the offset is used unchanged.

This is needed for llvm-dwarfdump to be able to extract the line tables
(with --debug-lines) from binaries produced by certain compilers that
like to align each line table in the .debug_line section. Note that this
alignment does not seem to be invalid since the units do point to the
correct line table offsets via the DW_AT_stmt_list attribute.

Differential Revision: https://reviews.llvm.org/D143513

17 months ago[WebAssembly] Fix epilogue insertion for indirect tail calls
Thomas Lively [Wed, 22 Mar 2023 16:28:47 +0000 (09:28 -0700)]
[WebAssembly] Fix epilogue insertion for indirect tail calls

Previously epilogues were incorrectly inserted after indirect tail calls because
they did not have the `isTerminator` property. Add that property and test that
they get correct epilogues. To be safe, also add other properties that were
defined for direct tail calls.

Differential Revision: https://reviews.llvm.org/D146569

17 months ago[libc++] Qualifies ptrdiff_t and max_align_t.
Mark de Wever [Tue, 21 Mar 2023 18:01:12 +0000 (19:01 +0100)]
[libc++] Qualifies ptrdiff_t and max_align_t.

This has been done using the following commands
  find libcxx/test -type f -exec perl -pi -e 's|^([^/]+?)((?<!::)ptrdiff_t)|\1std::\2|' \{} \;
  find libcxx/test -type f -exec perl -pi -e 's|^([^/]+?)((?<!::)max_align_t)|\1std::\2|' \{} \;

The std module doesn't export declarations in the global namespaace.,
This is a preparation for that module.

Reviewed By: #libc, ldionne

Differential Revision: https://reviews.llvm.org/D146550

17 months ago[InstSimplify] Actually use NewOps for calls in simplifyInstructionWithOperands
Arthur Eubanks [Wed, 22 Mar 2023 01:00:08 +0000 (18:00 -0700)]
[InstSimplify] Actually use NewOps for calls in simplifyInstructionWithOperands

Resolves a TODO.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D146599

17 months ago[NFC][WebAssembly] Autogenerate test expectations for tailcall.ll
Thomas Lively [Wed, 22 Mar 2023 16:21:12 +0000 (09:21 -0700)]
[NFC][WebAssembly] Autogenerate test expectations for tailcall.ll

A follow-on commit will add tests to this file and using the
update_llc_test_checks script will make that easier.

Differential Revision: https://reviews.llvm.org/D146568

17 months ago[AArch64] Add asm aliases for MOV, LDR, STR with predicate-as-counter
Caroline Concatto [Wed, 22 Mar 2023 14:57:18 +0000 (14:57 +0000)]
[AArch64] Add asm aliases for MOV, LDR, STR with predicate-as-counter

In the 2022-12 release of the A64 ISA it was updated that the assembler must
also accept predicate-as-counter register names for the source predicate
register and the destination predicate register for:
 * *MOV: Move predicate (unpredicated)*
 * *LDR (predicate): Load predicate register*
 * *STR (predicate): Store predicate register*

Reviewed By: sdesmalen

Differential Revision: https://reviews.llvm.org/D146311

17 months ago[RISCV][NFC] Make interleaved access test more vectorizable
Luke Lau [Mon, 20 Mar 2023 17:31:40 +0000 (17:31 +0000)]
[RISCV][NFC] Make interleaved access test more vectorizable

The previous test case stored the result of a deinterleaved load and add
into the same source address, which resulted in some scatters which we
weren't testing for and made the tests harder to understand.
Store it at a separate address, which will make the tests easier to read
when the cost model is changed after D145085 is landed

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D146442

17 months ago[ARM] Improve min/max vector reductions on Arm
Caleb Zulawski [Wed, 22 Mar 2023 16:00:19 +0000 (16:00 +0000)]
[ARM] Improve min/max vector reductions on Arm

This patch adds some more efficient lowering for vecreduce.min/max under NEON,
using sequences of pairwise vpmin/vpmax to reduce to a single value.

This nearly resolves issues such as #50466, #40981, #38190.

Differential Revision: https://reviews.llvm.org/D146404

17 months ago[X86] Extend all_of(icmp_eq()) / any_of(icmp_ne()) -> scalar integer fold to AVX512...
Simon Pilgrim [Wed, 22 Mar 2023 15:56:23 +0000 (15:56 +0000)]
[X86] Extend all_of(icmp_eq()) / any_of(icmp_ne()) -> scalar integer fold to AVX512 targets

Extends 1bb95a3a99cb44f2b8b801e5137d3ac529253f3b to combine on AVX512 targets where the vXi1 type is legal

Continues work on addressing Issue #53419

17 months agoRegenerate checks for bswap.ll, NFC
Jun Zhang [Wed, 22 Mar 2023 15:41:53 +0000 (23:41 +0800)]
Regenerate checks for bswap.ll, NFC

Signed-off-by: Jun Zhang <jun@junz.org>
17 months agoFix accessing of aligned arrays in offloaded target regions
Doru Bercea [Wed, 22 Feb 2023 16:58:48 +0000 (11:58 -0500)]
Fix accessing of aligned arrays in offloaded target regions

17 months ago[gn build] Port 883dbb9c86be
LLVM GN Syncbot [Wed, 22 Mar 2023 14:45:00 +0000 (14:45 +0000)]
[gn build] Port 883dbb9c86be

17 months agoRevert "[MemProf] Context disambiguation cloning pass [patch 1a/3]"
Nikita Popov [Wed, 22 Mar 2023 14:43:46 +0000 (15:43 +0100)]
Revert "[MemProf] Context disambiguation cloning pass [patch 1a/3]"

This reverts commit d6ad4f01c3dafcab335bca66dac6e36d9eac8421.

Fails to build on at least gcc 12.2:

/home/npopov/repos/llvm-project/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp:482:1: error: no declaration matches ‘ContextNode<DerivedCCG, FuncTy, CallTy>* CallsiteContextGraph<DerivedCCG, FuncTy, CallTy>::getNodeForInst(const CallInfo&)’
  482 | CallsiteContextGraph<DerivedCCG, FuncTy, CallTy>::getNodeForInst(
      | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/npopov/repos/llvm-project/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp:393:16: note: candidate is: ‘CallsiteContextGraph<DerivedCCG, FuncTy, CallTy>::ContextNode* CallsiteContextGraph<DerivedCCG, FuncTy, CallTy>::getNodeForInst(const CallInfo&)’
  393 |   ContextNode *getNodeForInst(const CallInfo &C);
      |                ^~~~~~~~~~~~~~
/home/npopov/repos/llvm-project/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp:99:7: note: ‘class CallsiteContextGraph<DerivedCCG, FuncTy, CallTy>’ defined here
   99 | class CallsiteContextGraph {
      |       ^~~~~~~~~~~~~~~~~~~~

17 months ago[MLIR][LLVM] Copy byval attributes during inlining.
Johannes de Fine Licht [Wed, 22 Mar 2023 14:22:11 +0000 (15:22 +0100)]
[MLIR][LLVM] Copy byval attributes during inlining.

Support inlining of function calls with the byval attribute on function
arguments by copying the pointee into a newly alloca'ed pointer at the
callsite before inlining.

The alignment attribute is not yet taken into account.

Reviewed By: ftynse, gysit

Differential Revision: https://reviews.llvm.org/D146616

17 months ago[gn build] Port e4ceb5a7bb9b
LLVM GN Syncbot [Wed, 22 Mar 2023 14:28:29 +0000 (14:28 +0000)]
[gn build] Port e4ceb5a7bb9b

17 months ago[GVN] Fix change reporting when removing assume (PR61574)
Nikita Popov [Wed, 22 Mar 2023 14:22:33 +0000 (15:22 +0100)]
[GVN] Fix change reporting when removing assume (PR61574)

Report a change when removing a true/false assume.

Fixes https://github.com/llvm/llvm-project/issues/61574.

17 months ago[X86] Create extra prolog/epilog for stack realignment
Luo, Yuanke [Wed, 22 Mar 2023 12:41:36 +0000 (20:41 +0800)]
[X86] Create extra prolog/epilog for stack realignment

Fix some bugs and reland e4c1dfed38370b4 and 614c63bec6d67c.
1. Run argument stack rebase pass before the reserved physical register
   is finalized.
2. Add LEA pseudo instruction to prevent the instruction being
   eliminated.
3. Don't support X32.

17 months ago[Assignment Tracking] Fix mem2reg misidentifying unlinked stores
OCHyams [Wed, 22 Mar 2023 13:29:46 +0000 (13:29 +0000)]
[Assignment Tracking] Fix mem2reg misidentifying unlinked stores

updateForDeletedStore updates the assignment tracking debug info for a store
that is about to be deleted by mem2reg. For each variable backed by the target
alloca, if a dbg.assign exists it is kept (well - it's downgraded to a
dbg.value). A dbg.value is inserted if there's not a linked dbg.assign for a
variable which is backed by the target alloca. This patch fixes a bug whereby a
store with a linked dbg.assign that describes a fragment different to the one
linked to the alloca was not counted for the variable, leading to both keeping
the dbg.assign (downgrading it) and inserting a new dbg.value.

Reviewed By: StephenTozer

Differential Revision: https://reviews.llvm.org/D146299

17 months ago[NFC] Add DebugVariableAggregate class
OCHyams [Wed, 22 Mar 2023 13:27:35 +0000 (13:27 +0000)]
[NFC] Add DebugVariableAggregate class

A DebugVariableAggregate is a DebugVariable that discards FragmentInfo; it
represents a whole variable instance.

Reviewed By: StephenTozer

Differential Revision: https://reviews.llvm.org/D146298

17 months ago[gn build] Port d6ad4f01c3da
LLVM GN Syncbot [Wed, 22 Mar 2023 14:06:06 +0000 (14:06 +0000)]
[gn build] Port d6ad4f01c3da

17 months ago[MemProf] Context disambiguation cloning pass [patch 1a/3]
Teresa Johnson [Thu, 29 Dec 2022 20:11:38 +0000 (12:11 -0800)]
[MemProf] Context disambiguation cloning pass [patch 1a/3]

Support for building, printing, and displaying CallsiteContextGraph
which represents the MemProf metadata contexts. Uses CRTP to enable
support for both IR (regular LTO) and summary (ThinLTO). This patch
includes the support for building it in regular LTO mode (from
memprof and callsite metadata), and the next patch will add the
handling for building it from ThinLTO summaries.

Also includes support for dumping the graph to text and to dot files.

Follow-on patches will contain the support for cloning on the graph and
in the IR.

The graph represents the call contexts in all memprof metadata on
allocation calls, with nodes for the allocations themselves, as well as
for the calls in each context. The graph is initially built from the
allocation memprof metadata (or summary) MIBs. It is then updated to
match calls with callsite metadata onto the nodes, updating it to
reflect any inlining performed on those calls.

Each MIB (representing an allocation's call context with allocation
behavior) is assigned a unique context id during the graph build. The
edges and nodes in the graph are decorated with the context ids they
carry. This is used to correctly update the graph when cloning is
performed so that we can uniquify the context for a single (possibly
cloned) allocation.

Depends on D140786.

Differential Revision: https://reviews.llvm.org/D140908

17 months ago[Dexter] Add timeout options
Stephen Tozer [Wed, 1 Mar 2023 13:05:11 +0000 (13:05 +0000)]
[Dexter] Add timeout options

Adds a pair of options for Dexter that allow the user to specify a
timeout duration. These options are:

* --timeout-total: Times out if the total run-time of the debugger session
  exceeds <timeout-total> seconds.
* --timeout-breakpoint: Times out if the time without hitting a
  breakpoint exceeds <timeout-breakpoint> seconds.

Reviewed By: Orlando

Differential Revision: https://reviews.llvm.org/D145063

17 months agodocs: add some documentation on Windows SDK search
Saleem Abdulrasool [Wed, 15 Mar 2023 20:06:56 +0000 (13:06 -0700)]
docs: add some documentation on Windows SDK search

Add some documentation on the flags and the process by which clang
identifies the headers and libraries for the Windows environment.  It
should identify the flags and their interactions as well as the order in
which the various sources of information are consulted.

Differential Revision: https://reviews.llvm.org/D146165
Reviewed By: hans, mstorjo

17 months ago[flang] add -flang-experimental-hlfir flag to flang-new
Tom Eccles [Tue, 14 Mar 2023 13:27:30 +0000 (13:27 +0000)]
[flang] add -flang-experimental-hlfir flag to flang-new

This flag instructs flang-new to use the new HLFIR lowering. It is
marked as experimental and not included in --help.

This was added to make it more convenient to test the performance of
code generated by the HLFIR lowering.

Extra diffs are from running clang-format on CLOptions.inc (which was
being forced by CI).

Differential Revision: https://reviews.llvm.org/D146278

17 months ago[SimplifyCFG][LICM] Add metadata speculation tests (NFC)
Nikita Popov [Wed, 22 Mar 2023 13:21:51 +0000 (14:21 +0100)]
[SimplifyCFG][LICM] Add metadata speculation tests (NFC)

17 months ago[ConstraintElim] Add tests with GEP constant expressions.
Florian Hahn [Wed, 22 Mar 2023 13:12:49 +0000 (13:12 +0000)]
[ConstraintElim] Add tests with GEP constant expressions.

17 months ago[libc][Bazel] Bazel fix for 494734b06d62b
Dmitry Chernenkov [Wed, 22 Mar 2023 12:50:22 +0000 (12:50 +0000)]
[libc][Bazel] Bazel fix for 494734b06d62b

17 months ago[mlir][AffineMap] NFC - Refactor getProjectedMap and split into projectDims and proje...
Nicolas Vasilache [Wed, 22 Mar 2023 09:19:20 +0000 (02:19 -0700)]
[mlir][AffineMap] NFC - Refactor getProjectedMap and split into projectDims and projectSymbols

The default behavior of getProjectedMap may be surprising as it implicitly compresses the dims and
the unused symbols.

Make these explicit in the API and refactor to more idiomatic implementations with better reuse.

Differential Revision: https://reviews.llvm.org/D146611

17 months ago[mlir] Add a pattern to fold tensor.cast into scf.forall.
Alexander Belyaev [Wed, 22 Mar 2023 11:50:24 +0000 (12:50 +0100)]
[mlir] Add a pattern to fold tensor.cast into scf.forall.

Differential revision: https://reviews.llvm.org/D146558

17 months ago[InstCombine] enhance icmp with sub folds
Zhongyunde [Wed, 22 Mar 2023 11:52:41 +0000 (19:52 +0800)]
[InstCombine] enhance icmp with sub folds

The new IR with And removes a use of the input variable, which is better for analysis.
Fix https://github.com/llvm/llvm-project/issues/60818

Reviewed By: nikic, spatel

Differential Revision: https://reviews.llvm.org/D145846

17 months ago[tests] precommit tests for D145846
Zhongyunde [Wed, 22 Mar 2023 11:47:35 +0000 (19:47 +0800)]
[tests] precommit tests for D145846

17 months ago[LV] Set inbounds flag using CreateGEP in vectorizeInterleaveGroup(NFC).
Florian Hahn [Wed, 22 Mar 2023 11:28:49 +0000 (11:28 +0000)]
[LV] Set inbounds flag using CreateGEP in vectorizeInterleaveGroup(NFC).

This avoids having to cast the result of the builder to
GetElementPtrInst.

17 months ago[LoopPredication] Add a test demonstrating bug.
Serguei Katkov [Wed, 22 Mar 2023 10:19:48 +0000 (17:19 +0700)]
[LoopPredication] Add a test demonstrating bug.

LoopPredication may introduce undefined behavior.

17 months ago[MLIR][Affine] Fix assumption on int type in memref elt size method
Uday Bondhugula [Wed, 22 Mar 2023 01:46:35 +0000 (07:16 +0530)]
[MLIR][Affine] Fix assumption on int type in memref elt size method

Fix assumption on memref element type being int/float in memref elt size
related method and affine data copy generate.

Fixes https://github.com/llvm/llvm-project/issues/61310

Differential Revision: https://reviews.llvm.org/D146495

17 months ago[GVN] Regenerate test checks (NFC)
Nikita Popov [Wed, 22 Mar 2023 10:40:07 +0000 (11:40 +0100)]
[GVN] Regenerate test checks (NFC)

17 months ago[IR] Rename dropUndefImplying to dropUBImplying (NFC)
Nikita Popov [Wed, 22 Mar 2023 09:48:26 +0000 (10:48 +0100)]
[IR] Rename dropUndefImplying to dropUBImplying (NFC)

Clarify that this is only about immediate undefined behavior,
not about undef or poison.

17 months ago[dataflow] Log flow condition to the correct stream.
Sam McCall [Tue, 21 Mar 2023 14:34:31 +0000 (15:34 +0100)]
[dataflow] Log flow condition to the correct stream.

Differential Revision: https://reviews.llvm.org/D146527

17 months ago[llvm][ARM] Fix bounds check after use in ARMAsmParser
David Spickett [Wed, 22 Mar 2023 09:49:02 +0000 (09:49 +0000)]
[llvm][ARM] Fix bounds check after use in ARMAsmParser

Fixes #61605

17 months ago[lldb] Add test for unavailable registers
David Spickett [Mon, 13 Mar 2023 14:28:39 +0000 (14:28 +0000)]
[lldb] Add test for unavailable registers

Prior to this the only check was that we did not print
this message when reading registers that should exist.

I thought there was an indentation bug here so I wrote a test
for it. There is not, but we could do with the coverage anyway.

Reviewed By: rupprecht

Differential Revision: https://reviews.llvm.org/D145940

17 months ago[mlir][llvm] Add experimental alias scope decl intrinsic.
Tobias Gysi [Wed, 22 Mar 2023 09:20:55 +0000 (10:20 +0100)]
[mlir][llvm] Add experimental alias scope decl intrinsic.

The revision adds the llvm.experimental.noalias.scope.decl intrinsic
to the LLVM dialect and updates the import and export accordingly.

Reviewed By: Dinistro

Differential Revision: https://reviews.llvm.org/D146504

17 months ago[mlir][Transforms][NFC] Improve builder/listener API of OperationFolder
Matthias Springer [Wed, 22 Mar 2023 08:04:01 +0000 (09:04 +0100)]
[mlir][Transforms][NFC] Improve builder/listener API of OperationFolder

The constructor of `OperationFolder` takes a listener. Therefore, the remaining API should not take any builder/rewriters. This could lead to double notifications in case a listener is attached to the builder/rewriter.

As an internal cleanup, `OperationFolder` now has an `IRRewriter` instead of a `RewriterBase::Listener`. In most cases, `OperationFolder` no longer has to notify/deal with listeners. This is done by the rewriter.

Differential Revision: https://reviews.llvm.org/D146134

17 months ago[mlir][Bazel] Add missing dependency.
Adrian Kuegel [Wed, 22 Mar 2023 08:20:35 +0000 (09:20 +0100)]
[mlir][Bazel] Add missing dependency.

17 months ago[GuardWidening] Improve analysis of potential widening into hotter block, try 2
Max Kazantsev [Wed, 22 Mar 2023 07:29:12 +0000 (14:29 +0700)]
[GuardWidening] Improve analysis of potential widening into hotter block, try 2

The initial version was reverted because it looped infinitely if the likely successor
isn't properly dominated by the predecessor. In practice it means that we went up the
CFG through backedge and looped infinitely.

I also added some paranoid assertion checks to make sure that every other invariant
holds. I also found a hypothetical situation when we may go past the dominated block
while following the likely successors (it means that in fact the dominated block is
dynamically not reachable from dominating block) and explicitly prohibited this, though
I don't have a motivating test showing that it's a real problem.

https://reviews.llvm.org/D146276

17 months ago[mlir] Argument and result attribute handling during inlining.
Tobias Gysi [Wed, 22 Mar 2023 07:38:55 +0000 (08:38 +0100)]
[mlir] Argument and result attribute handling during inlining.

The revision adds the handleArgument and handleResult handlers that
allow users of the inlining interface to implement argument and result
conversions that take argument and result attributes into account. The
motivating use cases for this revision are taken from the LLVM dialect
inliner, which has to copy arguments that are marked as byval and that
also has to consider zeroext / signext when converting integers.

All type conversions are currently handled by the
materializeCallConversion hook. It runs before isLegalToInline and
supports only the introduction of a single cast operation since it may
have to rollback. The new handlers run shortly before and after
inlining and cannot fail. As a result, they can introduce more complex
ir such as copying a struct argument. At the moment, the new hooks
cannot be used to perform type conversions since all type conversions
have to be done using the materializeCallConversion. A follow up
revision will either relax this constraint or drop
materializeCallConversion in favor of the new and more flexible
handlers.

The revision also extends the CallableOpInterface to provide access
to the argument and result attributes if available.

Reviewed By: rriddle, Dinistro

Differential Revision: https://reviews.llvm.org/D145582

17 months ago[mlir][IR] Add ReverseDominanceIterator for IR walkers
Matthias Springer [Wed, 22 Mar 2023 07:53:38 +0000 (08:53 +0100)]
[mlir][IR] Add ReverseDominanceIterator for IR walkers

Blocks are enumerated depth-first, but post-order. I.e., a block is enumerated when its successors have been enumerated. This iteration style is suitable when deleting blocks in a regions: in the absence of cycles, uses are deleted before their definitions.

Differential Revision: https://reviews.llvm.org/D146125

17 months ago[analyzer] Fix crashing getSValFromInitListExpr for nested initlists
Balazs Benics [Wed, 22 Mar 2023 07:43:09 +0000 (08:43 +0100)]
[analyzer] Fix crashing getSValFromInitListExpr for nested initlists

In the following example, we will end up hitting the `llvm_unreachable()`:
https://godbolt.org/z/5sccc95Ec
```lang=C++
enum class E {};
const E glob[] = {{}};
void initlistWithinInitlist() {
  clang_analyzer_dump(glob[0]); // crashes at loading from `glob[0]`
}
```

We should just return `std::nullopt` instead for these cases.
It's better than crashing.

Reviewed By: xazax.hun

Differential Revision: https://reviews.llvm.org/D146538

17 months ago[llvm] Use llvm::isNullConstant (NFC)
Kazu Hirata [Wed, 22 Mar 2023 07:31:48 +0000 (00:31 -0700)]
[llvm] Use llvm::isNullConstant (NFC)

17 months ago[Test] Add hanging test for D146276
Max Kazantsev [Wed, 22 Mar 2023 07:25:34 +0000 (14:25 +0700)]
[Test] Add hanging test for D146276

The patch was reverted because of hang, adding the test so that this doesn't
happen again.

17 months ago[X86] Use llvm::isOneConstant (NFC)
Kazu Hirata [Wed, 22 Mar 2023 07:25:13 +0000 (00:25 -0700)]
[X86] Use llvm::isOneConstant (NFC)

17 months ago[flang] Feature list plugin
Ethan Luis McDonough [Thu, 2 Mar 2023 18:29:05 +0000 (12:29 -0600)]
[flang] Feature list plugin

Plugin that counts the number of times each tree node occurs in a given program.  Used for test coverage.

Updated to fix build issues.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D143704

17 months ago[MCContext] Use `const Twine &` in symbol creation methods. NFC
paperchalice [Wed, 22 Mar 2023 06:13:59 +0000 (23:13 -0700)]
[MCContext] Use `const Twine &` in symbol creation methods. NFC

All of these methods will invoke `getOrCreateSymbol(const Twine &Name)`, using `Twine` here makes these methods more flexible.

Differential Revision: https://reviews.llvm.org/D145923

17 months ago[mlir][memref] Use folded composed affine apply ops in FoldMemRefAliasOps
Nicolas Vasilache [Mon, 20 Mar 2023 21:22:00 +0000 (14:22 -0700)]
[mlir][memref] Use folded composed affine apply ops in FoldMemRefAliasOps

Creating maximally folded and composd affine.apply operation during
FoldMemRefAliasOps composes better with other transformations without having
to interleave canonicalization passes.

Differential Revision: https://reviews.llvm.org/D146515

17 months ago[clang][driver] Enable '-flto' on AVR
Ben Shi [Thu, 9 Mar 2023 04:04:54 +0000 (12:04 +0800)]
[clang][driver] Enable '-flto' on AVR

Reviewed By: MaskRay

Closes https://github.com/llvm/llvm-project/issues/55940

Differential Revision: https://reviews.llvm.org/D145646

17 months ago[Coroutines] Implement fix for cwg2563 issue and enable RVO under certain conditions
Bruno Cardoso Lopes [Thu, 9 Mar 2023 00:00:02 +0000 (16:00 -0800)]
[Coroutines] Implement fix for cwg2563 issue and enable RVO under certain conditions

- The cwg2563 issue is fixed by delaying GRO initialization only when the types
  mismatch between GRO and function return.
- When the types match directly initialize, which indirectly enables RVO to
  kick in, partially restores behavior introduced in
  https://reviews.llvm.org/D117087.
- Add entry to release notes.

Background:
https://github.com/llvm/llvm-project/issues/56532
https://cplusplus.github.io/CWG/issues/2563.html
https://github.com/cplusplus/papers/issues/1414

Differential Revision: https://reviews.llvm.org/D145641

17 months ago[Coroutines] Fix premature conversion of return object
Bruno Cardoso Lopes [Tue, 7 Mar 2023 20:51:34 +0000 (12:51 -0800)]
[Coroutines] Fix premature conversion of return object

Fix https://github.com/llvm/llvm-project/issues/56532

Effectively, this reverts behavior introduced in https://reviews.llvm.org/D117087,
which did two things:

1. Change delayed to early conversion of return object.
2. Introduced RVO possibilities because of early conversion.

This patches fixes (1) and removes (2). I already worked on a follow up for (2)
in a separated patch. I believe it's important to split these two because if the RVO
causes any problems we can explore reverting (2) while maintaining (1).

Notes on some testcase changes:
- `pr59221.cpp` changed to `-O1` so we can check that the front-end honors
  the value checked for. Sounds like `-O3` without RVO is more likely
  to work with LLVM optimizations...
- Comment out delete members `coroutine-no-move-ctor.cpp` since behavior
  now requires copies again.

Differential Revision: https://reviews.llvm.org/D145639

17 months ago[PowerPC][NFC] add const-nonsplat-array-init.ll
Ting Wang [Wed, 22 Mar 2023 04:32:18 +0000 (00:32 -0400)]
[PowerPC][NFC] add const-nonsplat-array-init.ll

When doing store constant vector/scalar, some duplicated values can be reused.
Add test case and will show combiner can improve these.

Reviewed By: shchenz

Differential Revision: https://reviews.llvm.org/D146500

17 months agoupdate dependency for TransformOpsPyTdFiles
yijia1212 [Wed, 22 Mar 2023 04:26:09 +0000 (21:26 -0700)]
update dependency for TransformOpsPyTdFiles

update dependency for TransformOpsPyTdFiles

Differential Revision: https://reviews.llvm.org/D146605

17 months ago[RISCV] Clear mayRaiseFPException for fclass.d instruction.
Craig Topper [Wed, 22 Mar 2023 02:48:20 +0000 (19:48 -0700)]
[RISCV] Clear mayRaiseFPException for fclass.d instruction.

We got it right for fclass.s and fclass.h.

17 months ago[RISCV] Clear mayRaiseFPException for Zfa fmvh.x.d and fmvp.d.x instructions.
Craig Topper [Wed, 22 Mar 2023 02:42:27 +0000 (19:42 -0700)]
[RISCV] Clear mayRaiseFPException for Zfa fmvh.x.d and fmvp.d.x instructions.

17 months ago[RISCV] Move fli selection in RISCVISelDAGToDAG.cpp. NFC
Craig Topper [Wed, 22 Mar 2023 02:15:30 +0000 (19:15 -0700)]
[RISCV] Move fli selection in RISCVISelDAGToDAG.cpp. NFC

We custom isel for ConstantFP that has higher priority than isel
patterns. We were previously detecting valid FP constants for fli
to early exit from the custom code. This detection called
getLoadFPImm. Then we would run the isel patterns which would call
getLoadFPImm a second time.

With a little bit more code we can directly select the fli instruction
in the custom handler and avoid a second call.

Remove the incorrect mayRaiseFPException flag from the FLI instructions.

Reviewed By: joshua-arch1

Differential Revision: https://reviews.llvm.org/D146093

17 months ago[AIX][r] Do not call AddFilePathLibArgs with -r
Michael Francis [Tue, 21 Mar 2023 22:06:47 +0000 (22:06 +0000)]
[AIX][r] Do not call AddFilePathLibArgs with -r

We do not want to add file path lib args when -r is specified.

Differential Revision: https://reviews.llvm.org/D146578

17 months ago[LFTR] Minor style cleanup [nfc]
Philip Reames [Wed, 22 Mar 2023 02:12:35 +0000 (19:12 -0700)]
[LFTR] Minor style cleanup [nfc]

17 months ago[LFTR] Use evaluateAtIteration in genLoopLimit [nfc]
Philip Reames [Wed, 22 Mar 2023 02:08:34 +0000 (19:08 -0700)]
[LFTR] Use evaluateAtIteration in genLoopLimit [nfc]

Note that the comments being removed appear to be very out of sync with the actual code in question.

Differential Revision: https://reviews.llvm.org/D146468

17 months ago[LFTR] Simplify another case under assumption exit counts are integers [nfc]
Philip Reames [Wed, 22 Mar 2023 02:04:14 +0000 (19:04 -0700)]
[LFTR] Simplify another case under assumption exit counts are integers [nfc]

This invariant was introduced in 8f3d16905d75b07a933d01dc29677fe5867c1b3e.

17 months ago[OpenMP] Remove shadow pointer map and introduce consistent locking
Johannes Doerfert [Tue, 21 Mar 2023 20:40:36 +0000 (13:40 -0700)]
[OpenMP] Remove shadow pointer map and introduce consistent locking

The shadow pointer map was problematic as we scanned an entire list if
an entry had shadow pointers. The new scheme stores the shadow pointers
inside the entries. This allows easy access without any search. It also
helps us, but also makes it necessary, to define a consistent locking
scheme. The implicit locking of entries via TargetPointerResultTy makes
this pretty effortless, however one has to:

- Lock HDTTMap before locking an entry.
- Do not lock HDTTMap while holding an entry lock.
- Hold the entry lock to read or modify an entry.

The changes to submitData and retrieveData have been made to ensure 2
when the LIBOMPTARGET_INFO flag is used. Most everything else happens by
itself as TargetPointerResultTy acts as a lock_guard for the entry. It
is a little complicated when we deal with multiple entries, especially
as they can be equal. However, one can still follow the rules with
reasonable effort.

LookupResult are now finally also locking the entry before it is
inspected. This is good even if we haven't run into a problem yet.

Differential Revision: https://reviews.llvm.org/D123446

17 months ago[OpenMP] Remove restriction on the thread count for parallel regions
Johannes Doerfert [Wed, 22 Mar 2023 00:43:26 +0000 (17:43 -0700)]
[OpenMP] Remove restriction on the thread count for parallel regions

Differential Revision: https://reviews.llvm.org/D112194

17 months ago[OpenMP] Avoid zero size copies to the device
Johannes Doerfert [Tue, 21 Mar 2023 22:48:11 +0000 (15:48 -0700)]
[OpenMP] Avoid zero size copies to the device

This unblocks one of the XFAIL tests for AMD, though we need to work
around the missing printf still.

Differential Revision: https://reviews.llvm.org/D146592

17 months ago[libc] Don't install the GPU startup code for now
Joseph Huber [Wed, 22 Mar 2023 02:00:46 +0000 (21:00 -0500)]
[libc] Don't install the GPU startup code for now

Summary:
This startup code is only intended to be used internally, we shouldn't
export it under a conflicting name. In the future we may package this in
an exportable format.

17 months ago[RISCV] Use LBU for extloadi8.
Craig Topper [Wed, 22 Mar 2023 01:52:05 +0000 (18:52 -0700)]
[RISCV] Use LBU for extloadi8.

The Zcb extension has c.lbu, but not c.lb. This patch makes us
prefer LBU over LB if we have a choice which will enable more
compression opportunities.

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D146270

17 months ago[amdgpu][nfc] clang-format AMDGPULowerModuleLDS for easier merging
Jon Chesterfield [Wed, 22 Mar 2023 01:49:41 +0000 (01:49 +0000)]
[amdgpu][nfc] clang-format AMDGPULowerModuleLDS for easier merging

17 months ago[SCEV] Add coverage for a missing flag inference case
Philip Reames [Wed, 22 Mar 2023 01:34:40 +0000 (18:34 -0700)]
[SCEV] Add coverage for a missing flag inference case

17 months ago[amdgpu][nfc] Extract more functions in LowerModuleLDS, mark more methods static
Jon Chesterfield [Wed, 22 Mar 2023 01:23:19 +0000 (01:23 +0000)]
[amdgpu][nfc] Extract more functions in LowerModuleLDS, mark more methods static

17 months ago[llvm] Use ConstantInt::{isZero,isOne} (NFC)
Kazu Hirata [Wed, 22 Mar 2023 00:40:35 +0000 (17:40 -0700)]
[llvm] Use ConstantInt::{isZero,isOne} (NFC)

17 months ago[mlir][sparse] fix crash when using pure constant index in indexing mapping (fixes...
Peiming Liu [Tue, 21 Mar 2023 20:47:47 +0000 (20:47 +0000)]
[mlir][sparse] fix crash when using pure constant index in indexing mapping (fixes #61530)

To address https://github.com/llvm/llvm-project/issues/61530

Reviewed By: aartbik, wrengr

Differential Revision: https://reviews.llvm.org/D146563

17 months ago[amdgpu][nfc] Comment and extract two functions in LowerModuleLDS
Jon Chesterfield [Tue, 21 Mar 2023 23:15:44 +0000 (23:15 +0000)]
[amdgpu][nfc] Comment and extract two functions in LowerModuleLDS

17 months ago[lldb][CMake] Enforce not linking against plugin libs in core libs
Alex Langford [Tue, 21 Mar 2023 18:20:31 +0000 (11:20 -0700)]
[lldb][CMake] Enforce not linking against plugin libs in core libs

Non-plugin lldb libraries should generally not be linking against lldb
plugin libraries. Enforce this in CMake.

Differential Revision: https://reviews.llvm.org/D146553