This is a small first step towards reorganization of the ORC libraries:
Declarations for types and function names (as strings) to be found in the
"ORC runtime bootstrap" set are moved into OrcRTBridge.h / OrcRTBridge.cpp.
The current implementation of the "ORC runtime bootstrap" functions is moved
into OrcRTBootstrap.h and OrcRTBootstrap.cpp. It is likely that this code will
eventually be moved into ORT-RT proper (in compiler RT).
The immediate goal of this change is to make these bootstrap functions usable
for clients other than SimpleRemoteEPC/SimpleRemoteEPCServer. The first planned
client is a new RuntimeDyld::MemoryManager that will run over EPC, which will
allow us to remove the old OrcRemoteTarget code.
The code was using getTypeStoreSize to calculate the difference
between consecutive objects. The calculation was incorrect due
to padding that is added between consecutive objects. The
getTypeAllocSize includes the padding amount. For example,
if the type is [19 x i8], the difference between consecutive
objects is 32 bytes, not 19 bytes.
A second case for getTypeAllocSize is needed when computing
the pointer values for the vector accesses. The calculation needs
to account for the padding as well.
Differential Revision: https://reviews.llvm.org/D109403
This is for Swift VFE support. In some vtable forms that Swift emits, the "base" of a relative pointer is not the global symbol itself directly, but a GEP into it -- so the pointer is relative to a particular field in the global. So getPointerAtOffset() needs to be able to see through the GEP and allow it in a SUB expression, to correctly recognize the offset as a vtable slot.
Differential Revision: https://reviews.llvm.org/D109169
For multithreaded modules (i.e. modules with a shared memory), lld injects a
synthetic Wasm start function that is automatically called during instantiation
to initialize memory from passive data segments. Even though the module will be
instantiated separately on each thread, memory initialization should happen only
once. Furthermore, memory initialization should be finished by the time each
thread finishes instantiation. Since multiple threads may be instantiating their
modules at the same time, the synthetic function must synchronize them.
The current synchronization tries to atomically increment a flag from 0 to 1 in
memory then enters one of two cases. First, if the increment was successful, the
current thread is responsible for initializing memory. It does so, increments
the flag to 2 to signify that memory has been initialized, then notifies all
threads waiting on the flag. Otherwise, the thread atomically waits on the flag
with an expected value of 1 until memory has been initialized. Either the
initializer thread finishes initializing memory (i.e. sets the flag to 2) first
and the waiter threads do not end up blocking, or the waiter threads succesfully
start waiting before memory is initialized so they will be woken by the
initializer thread once it has finished.
One complication with this scheme is that there are various contexts on the Web,
most notably on the main browser thread, that cannot successfully execute a
wait. Executing a wait in these contexts causes a trap, and in this case would
cause instantiation to fail. The embedder must therefore ensure that these
contexts win the race and become responsible for initializing memory, since that
is the only code path that does not execute a wait.
Unfortunately, since only one thread can win the race and initialize memory,
this scheme makes it impossible to have multiple threads in contexts that cannot
wait. For example, it is not currently possible to instantiate the module on
both the main browser thread as well as in an AudioWorklet. To loosen this
restriction, this commit inserts an extra check so that the wait will not be
executed at all when memory has already been initialized, i.e. when the flag
value is 2. After this change, the module can be instantiated on threads in
non-waiting contexts as long as the embedder can guarantee either that the
thread will win the race and initialize memory (as before) or that memory has
already been initialized when instantiation begins. Threads in contexts that can
wait can continue racing to initialize memory.
Fixes (or at least improves) PR51702.
Reviewed By: dschuff
Differential Revision: https://reviews.llvm.org/D109722
This is a fix on top of D106525's Case 2. In D106525, in
`runEHOnFunction` which handles Emscripten EH, We rethrow `longjmp` only
when the module has any usage of `setjmp` or `longjmp`. But now Wasm
object files are linked using wasm-ld, the module this pass sees is not
the whole program, and even if this module does not contain any
`longjmp`, another file can contain it and can be linked with the
current module. This enables the rethrowing of longjmp whenever
Emscripten SjLj is enabled, regardless of whether it is used in this
module or not.
Reviewed By: dschuff
Differential Revision: https://reviews.llvm.org/D109670
This reduces the maintenance burden by using globs, which is the
tradeoff we make in the other LLVM Bazel build files as well.
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D109720
This patch adds parsing support for the nontemporal clause. Also adds a couple of test cases.
Reviewed By: clementval
Differential Revision: https://reviews.llvm.org/D106896
std::is_convertible has no defined behavior when its arguments
are incomplete, even if they are equal. In practice, it returns false.
Adding std::is_same allows us to use the constructor using a callable,
even if the return value is incomplete. We also check the case where
we convert a T into a const T.
Reviewed By: DaniilSuchkov
Differential Revision: https://reviews.llvm.org/D104703
Committer: Daniil Suchkov <dsuchkov@azul.com>
* Revert https://reviews.llvm.org/D107307 so that both LHS and RHS have
the same layout with K0 as the innermost dimension.
* Continuing from https://reviews.llvm.org/D107003, move also 'K'
to the outer side, so that now the inter-tile dimensions as all outer,
and the intra-tile dimensions are all inner.
Reviewed By: asaadaldien
Differential Revision: https://reviews.llvm.org/D109692
Fixes the following warning:
$llvm_project/lldb/source/Plugins/ObjectFile/Minidump/MinidumpFileBuilder.cpp:744:11: warning:
format specifies type 'long' but the argument has type 'lldb::offset_t' (aka 'unsigned long long') [-Wformat]
m_data.GetByteSize());
^~~~~~~~~~~~~~~~~~~~
This revision allows hoisting static alloc/dealloc pairs as high as possible during ComprehensiveBufferization.
This also aligns such allocated buffers to 128B by default.
This change exhibited some issues wrt insertion points and a missing copy that are also fixed in this revision; tests are updated accordingly.
Differential Revision: https://reviews.llvm.org/D109684
APInt is used to describe a bit mask in a variety of value tracking and demanded bits/elts functions.
When traversing through dst/src operands, we have a number of places where these masks need to widened/narrowed to translate through bitcasts, reductions etc. to a different type.
This patch add a APIntOps::ScaleBitMask common helper, adds unit test coverage, and updates a number of cases to use the the helper instead of their own implementation.
This came up on D109065 where we currently have to add yet another implementation of the same code.
Differential Revision: https://reviews.llvm.org/D109683
mlir-cpu-runner/math_polynomial_approx.mlir
This test case is currently failing on SystemZ, but it does not appear to
necessarily be a target specific problem. See discussion at
https://bugs.llvm.org/show_bug.cgi?id=51204.
Patch by @dpalermo
The corrupt bitcode reported in https://bugs.llvm.org/show_bug.cgi?id=51647 seems to be a result of a later pass changing the workfn variable to addrspace(5) (thread private, on the stack). That seems reasonable for an alloca without an address space so it's an open question why that can crash the bitcode reader.
This change puts it in the thread private address space to begin with which means whatever misfired further down the pipeline does not break it. That matches the codegen from clang where stack variables are always annotated (5) and then addrspace cast prior to following use.
This therefore patches around whatever unsuccessfully moved the alloca variable to addrspace(5). That solves the problem of openmp opt producing code that crashes the bitcode reader. It should be possible to create a minimal repro for the underlying bug based on some handwritten IR that uses an alloca in a generic address space.
Reviewed By: ronlieb, jdoerfert, dpalermo-phab
Differential Revision: https://reviews.llvm.org/D109500
Remove the parent output checks, as they make the test flaky while
serving no real purpose. If the parent crashed/hanged, it will never
resume the child and the test would fail anyway.
The aim is to add the missing z/OS specific implementations for mbsnrtowcs and wcsnrtombs, as part of libc++.
Differential Revision: https://reviews.llvm.org/D98207
Moved out the checks for profitability of TryToSinkInstructions
into a lambda function.
This will also allow us to easily add checks for bailing out if the
transform is not profitable.
Tests-Run: instCombine tests.
D105819 Added NoOwnershipChangeVisitor, but it is only registered when an
off-by-default, hidden checker option was enabled. The reason behind this was
that it grossly overestimated the set of functions that really needed a note:
std::string getTrainName(const Train *T) {
return T->name;
} // note: Retuning without changing the ownership of or deallocating memory
// Umm... I mean duh? Nor would I expect this function to do anything like that...
void foo() {
Train *T = new Train("Land Plane");
print(getTrainName(T)); // note: calling getTrainName / returning from getTrainName
} // warn: Memory leak
This patch adds a heuristic that guesses that any function that has an explicit
operator delete call could have be responsible for deallocating the memory that
ended up leaking. This is waaaay too conservative (see the TODOs in the new
function), but it safer to err on the side of too little than too much, and
would allow us to enable the option by default *now*, and add refinements
one-by-one.
Differential Revision: https://reviews.llvm.org/D108753
First step in reducing redundancy in `addRelocations()` implementations across ELF JITLink backends. The patch factors out common logic for ELF relocation traversal into the new helper function `forEachRelocation()` in the `ELFLinkGraphBuilder` base class. For now, this is applied to the Aarch64 implementation. Others may follow soon.
Reviewed By: lhames
Differential Revision: https://reviews.llvm.org/D109516
When nonexistent linker inputs are passed to the driver, the linker
now errors out, instead of the compiler. If the linker does not run,
clang now emits a "warning: linker input unused" instead of an error
for nonexistent files.
The motivation for this change is that I noticed that
`clang-cl /winsysroot sysroot main.cc ole32.lib` emitted a
"ole32.lib not found" error, even though the linker finds it just fine when
I run `clang-cl /winsysroot sysroot main.cc /link ole32.lib`.
The same problem occurs if running `clang-cl main.cc ole32.lib` in a
non-MSVC shell.
The problem is that DiagnoseInputExistence() only looked for libs in %LIB%,
but MSVCToolChain uses much more involved techniques.
For this particular problem, we could make DiagnoseInputExistence() ask
the toolchain to see if it can find a .lib file, but in general the
driver can't know what the linker will do to find files, so it shouldn't
try. For example, if we implement PR24616, lld-link will look in the
registry to determine a good default for %LIB% if it isn't set.
This is less or a problem for the gcc driver, since .a paths there are
either passed via -l flags (which honor -L), or via a qualified path
(that doesn't honor -L) -- but for example ld.lld's --chroot flag
can also trigger this problem. Without this patch,
`clang -fuse-ld=lld -Wl,--chroot,some/dir /file.o` will complain that
`/file.o` doesn't exist, even though
`clang -fuse-ld=lld -Wl,--chroot,some/dir -Wl,/file.o` succeeds just fine.
This implements rnk's suggestion on the old bug PR27234.
Differential Revision: https://reviews.llvm.org/D109624
When back-deploying Swift async code we can't always toggle the flag showing an
extended frame is present because it will confuse unwinders on systems released
before this feature. So in cases where the code might run there, we `or` in a
mask provided by the runtime (as an absolute symbol) telling us whether the
unwinders can cope.
When deploying only for newer OSs, we can still hard-code the bit-set for
greater efficiency.
LLVM GN Syncbot