Summary:
We have a CMake condition to not include this static library on NVPTX
because their linker doesn't support it. There was a typo that made this
trigger on all builds not just the NVPTX ones.
The RPC interface can support multiple independent clients. This support
currently only supports many single-thread warps / workgroups
coordinating over a single lock. This patch uses the support added in
the previous patch to test the RPC interface with multiple blocks.
Note that this does not work with multiple threads currently because of
the effect of warps / workgroups executing in lockstep incorrectly. This
will be added later.
Depends on D148485
Reviewed By: lntue, sivachandra
Differential Revision: https://reviews.llvm.org/D148486
Currently, the RPC interface with the loader is only tested if the other
tests fail. This test adds a direct test that runs a simple integer
increment over the RPC handshake 10000 times.
Depends on https://reviews.llvm.org/D148288
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D148342
This patch reworks the RPC interface to allow more generic memory
operations using the shared better. This patch decomposes the entire RPC
interface into opening a port and calling `send` or `recv` on it.
The `send` function sends a single packet of the length of the buffer.
The `recv` function is paired with the `send` call to then use the data.
So, any aribtrary combination of sending packets is possible. The only
restriction is that the client initiates the exchange with a `send`
while the server consumes it with a `recv`.
The operation of this is driven by two independent state machines that
tracks the buffer ownership during loads / stores. We keep track of two
so that we can transition between a send state and a recv state without
an extra wait. State transitions are observed via bit toggling, e.g.
This interface supports an efficient `send -> ack -> send -> ack -> send`
interface and allows for the last send to be ignored without checking
the ack.
A following patch will add some more comprehensive testing to this interface. I
I informally made an RPC call that simply incremented an integer and it took
roughly 10 microsends to complete an RPC call.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D148288
This was a bit confused because nofpclass expresses the opposite
from what an assume of class expresses. We need to assume
the intersection of assumed classes, which also needs to be inverted
to convert to nofpclass.
This patch should address an issue that caused the process public run
lock to not be updated during a process launch/attach when the process
stops.
That caused the public run lock to report its state as running while the
process state is stopped. This prevents the users to interact with the
process (through the command line or via the SBAPI) since it's
considered still running.
To address that, this patch refactors the name of the internal hijack
listeners to a specific pattern `lldb.internal.<action>.hijack` that
are used to ensure that we've attached to or launched a process successfully.
Then, when updating the process public state, after updating the state
value, if the process is not hijacked externally, meaning if the process
doens't have a hijack listener that matches the internal hijack
listeners pattern, we can update the public run lock accordingly.
rdar://108283017
Differential Revision: https://reviews.llvm.org/D148400
Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
Same issue as was fixed in commit 3d95323, but for hexadecimal floats.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D148545
If any LLVM subprojects are built separately, the LLVM build directory
LLVM_LIBRARY_DIR is added to both the build and install runpaths in
llvm_setup_rpath(), which is incorrect when installed. Separate the
build and install runpaths on ELF platforms and finally remove the
incorrect call to this function for compiler-rt, as previously attempted
in 21c008d5a5b. That prior attempt was reverted in 959dbd1761c, where it
was said to break the build on macOS and Windows, so I made sure to keep
those platforms the same.
Two examples of incorrect runpaths that are currently added, one from
the latest LLVM 16 toolchain for linux x86_64:
> readelf -d clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.*so | ag "File:|runpath"
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.asan.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.dyndd.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.hwasan_aliases.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.hwasan.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.memprof.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.scudo_standalone.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.tsan.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.ubsan_minimal.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
File: clang+llvm-16.0.0-x86_64-linux-gnu-ubuntu-18.04/lib/clang/16/lib/x86_64-unknown-linux-gnu/libclang_rt.ubsan_standalone.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/tmp/llvm_release/final/Phase3/Release/llvmCore-16.0.0-final.obj/./lib]
Another is in the Swift toolchain, which builds lldb separately:
> readelf -d swift-5.9-DEVELOPMENT-SNAPSHOT-2023-03-24-a-ubuntu20.04/usr/{bin/lldb*,lib/liblldb.so}|ag "File:|runpath"
File: swift-5.9-DEVELOPMENT-SNAPSHOT-2023-03-24-a-ubuntu20.04/usr/bin/lldb
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/home/build-user/build/buildbot_linux/llvm-linux-x86_64/./lib]
File: swift-5.9-DEVELOPMENT-SNAPSHOT-2023-03-24-a-ubuntu20.04/usr/bin/lldb-argdumper
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/home/build-user/build/buildbot_linux/llvm-linux-x86_64/./lib]
File: swift-5.9-DEVELOPMENT-SNAPSHOT-2023-03-24-a-ubuntu20.04/usr/bin/lldb-server
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/home/build-user/build/buildbot_linux/llvm-linux-x86_64/./lib]
File: swift-5.9-DEVELOPMENT-SNAPSHOT-2023-03-24-a-ubuntu20.04/usr/lib/liblldb.so
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/../lib:/home/build-user/build/buildbot_linux/llvm-linux-x86_64/./lib:/home/build-user/build/buildbot_linux/swift-linux-x86_64/lib/swift/linux:$ORIGIN/../lib/swift/linux]
This patch should fix this problem of absolute paths from the build host
leaking out into the toolchain's runpaths.
Differential revision: https://reviews.llvm.org/D146918
This patch handles unevaluated contexts to ensure no warnings are produced by the machinery
for buffer access made within an unevaluated contexts. However, such accesses must be
considered by a FixableGadget and produce the necessary fixits.
Reviewed by: NoQ, ziqingluo-90, jkorous
Differential revision: https://reviews.llvm.org/D144905
This is an initial patch that lowers acc.enter_data copyin/create/attach
clauses to the newly added data operand operations. Follow up patches will
add support for array section and derived type and derived type component
as well as support in other data operation.
Reviewed By: razvanlupusoru
Differential Revision: https://reviews.llvm.org/D148721
These vectors are resized in the constructor and never change size.
We can manually allocate two arrays instead.
This reduces the size of TreePatternNode by removing the
unneeded capacity end pointer fields from the std::vector.
This does not help. The build is broken because compiler-rt-clear
deletes cmake generated files in STAMP_DIR/
This reverts commit da89ed99a16920e5986b14853d297322ccf7109e.
The data operations added in D148389 hold two data clause fields:
"dataClause" and "decomposedFrom". However, in most cases, dataClause
field holds a default value (except for acc_copyin_readonly,
acc_create_zero, and acc_copyout_zero).
The decomposedFrom field holds the original clause specified by user.
As work began on lowering to these new operations [1], it seems that
having both fields adds a bit of ambiguity. There is only one scenario
where we actually intended to use both:
acc data copyout(zero:)
The original intent was that this clause would be decomposed to
the following operations:
acc.create {dataClause = acc_create_zero, decomposedFrom =
acc_copyout_zero}
...
acc.copyout {dataClause = acc_copyout_zero}
However, we can encode the zero semantics like so without need of both:
acc.create {dataClause = acc_copyout_zero}
...
acc.copyout {dataClause = acc_copyout_zero}
Thus get rid of the decomposedFrom field and update verifier checks
to check for all data clauses that can be decomposed to the particular
operation.
So now the dataClause holds the original user's clause which simplifies
understanding of the operation.
[1] https://reviews.llvm.org/D148721
Reviewed By: clementval
Differential Revision: https://reviews.llvm.org/D148731
These probably do not need to be in the ConstString StringPool as they
don't really need any of the advantages that ConstStrings offer.
Lifetime for these things is always static and we never need to perform
comparisons for setting descriptions.
Differential Revision: https://reviews.llvm.org/D148679
Mingw toolchains implicitly add an .exe suffix if the output linked
file doesn't have a suffix. In many cases the extra suffix doesn't
cause any issues, but in some tests, this discrepancy between expected
output file name and actual output file does affect the tests.
In one test, a rm command fails to remove the executable since it
doesn't have the expected name. By failing to remove the executable, the
later llvm-profdata command tries to read the executable as if it was
a profile data file.
In another test, when the Python executor executes commands, it can
resolve executable names without the extra .exe suffix for absolute
paths (when most binaries are executed as e.g. "%t/foo"), but it fails
to resolve the executables for relative paths such as "./foo". Making
the paths absolute by using %t here shouldn't affect what the test tries
to validate.
Differential Revision: https://reviews.llvm.org/D148169
The buildvector cost for the case shown in the test should be 0 but it is -1, causing the code to get vectorized, whenit shouldn't.
Differential Revision: https://reviews.llvm.org/D148732
This patch removes the symbols defined in the library for std::allocator_arg,
std::defer_lock, std::try_to_lock, std::adopt_lock, and std::piecewise_construct.
Those were defined in the library because we provided them in C++03 as an
extension, and in C++03 it was impossible to define them as `constexpr`
variables, like in the spec.
This is technically an ABI break since we are removing symbols from the
library. However, in practice, only programs compiled in C++03 mode who
take the address of those objects (or pass them as a reference) will have
an undefined ref to those symbols. In practice, this is expected to be
rare. First, those are C++11 features that we happen to provide in C++03,
and only the C++03 definition can potentially lead to code referencing
the dylib definition. So any code that is using these objects but compiling
in C++11 mode (as they should) is not at risk. Second, all uses of these
types in the library is done by passing those types by value to a function
that can get inlined. Since they are empty types, the compiler won't
generate an undefined reference if passed by value, since there's nothing
to pass anyway.
Long story short, the risk for code actually containing an undefined
reference to one of these types is rather small (but non-zero). I also
couldn't find any app on the App Store that referenced these symbols,
which supports my impression that this won't be an issue in practice.
Differential Revision: https://reviews.llvm.org/D145587
Instead of creating a std::string from the `SmallString`,
let's just use a std::string from the start. I initially tried to make
`SmallString` work but getting it right proved complicated because
`LogHandler::Emit` will take its `StringRef` parameter and touch the raw
`const char *` from it directly, which isn't guaranteed to be
null-terminated with a `SmallString`.
I changed `WriteMessage` to take a `StringRef` instead of a
`const std::string &` for flexibility.
Differential Revision: https://reviews.llvm.org/D148676
Fixes warning treated as error from
https://lab.llvm.org/buildbot/#/builders/13/builds/34440:
'llvm::LibCallSimplifier::optimizeNew': not all control paths return a value
after a35206d78280e0ebcf48cd21bc5fff5c3b9c73fa.
Change assert in default case to a return nullptr, so we have a
reasonable fallback if invoked on a different library function, and no
longer get the above warning.
Gracefully handle non-1 stack sizes in printCompactDWARFExpr rather than
assert. Add support for DW_OP_nop and test the zero-sized stack case.
This is intended to be nearly NFC.
Differential Revision: https://reviews.llvm.org/D147269
Repurpose the DW_OP_LLVM_aspace_implicit_pointer encoding (0xe9) as the
encoding for a new operation DW_OP_LLVM_user which prefixes all other
new operations.
Differential Revision: https://reviews.llvm.org/D147265
Summary:
This variable was not being reset, which caused the options to be
compounded when building multiple architectures. This was very
problematic as the architectures are not compatible.
The new message is a bit verbose, but it makes it clearer that this isn't
a problem in libc++abi, but instead a problem in the user's code. Otherwise,
we get bugs sent down to libc++abi because people see this message in their
crash logs.
Differential Revision: https://reviews.llvm.org/D148630
