This fixes remaining issues in my previous PR #90959.
Changes:
- Removed dependency on LLVM header in `xray_interface.cpp`
- Fixed XRay patching for some targets due to missing changes in
architecture-specific patching functions
- Addressed some remaining compiler warnings that I missed in the
previous patch
- Formatting
I have tested these changes on `x86_64` (natively), as well as
`ppc64le`, `aarch64` and `arm32` (cross-compiled and emulated using
qemu).
**Original description:**
This PR introduces shared library (DSO) support for XRay based on a
revised version of the implementation outlined in [this
RFC](https://discourse.llvm.org/t/rfc-upstreaming-dso-instrumentation-support-for-xray/73000).
The feature enables the patching and handling of events from DSOs,
supporting both libraries linked at startup or explicitly loaded, e.g.
via `dlopen`.
This patch adds the following:
- The `-fxray-shared` flag to enable the feature (turned off by default)
- A small runtime library that is linked into every instrumented DSO,
providing position-independent trampolines and code to register with the
main XRay runtime
- Changes to the XRay runtime to support management and patching of
multiple objects
These changes are fully backward compatible, i.e. running without
instrumented DSOs will produce identical traces (in terms of recorded
function IDs) to the previous implementation.
Due to my limited ability to test on other architectures, this feature
is only implemented and tested with x86_64. Extending support to other
architectures is fairly straightforward, requiring only a
position-independent implementation of the architecture-specific
trampoline implementation (see
`compiler-rt/lib/xray/xray_trampoline_x86_64.S` for reference).
This patch does not include any functionality to resolve function IDs
from DSOs for the provided logging/tracing modes. These modes still work
and will record calls from DSOs, but symbol resolution for these
functions in not available. Getting this to work properly requires
recording information about the loaded DSOs and should IMO be discussed
in a separate RFC, as there are mulitple feasible approaches.
---------
Co-authored-by: Sebastian Kreutzer <sebastian.kreutzer@tu-darmstadt.de>
This commit has completed the Extension for Resumable Non Maskable
Interrupts, adding four CRSs and one Trap-Return instruction.
Specification link:["Smrnmi"
Extension](https://github.com/riscv/riscv-isa-manual/blob/main/src/rnmi.adoc)
---------
Co-authored-by: Sam Elliott <sam@lenary.co.uk>
nsw is now added to do-variable increment when -fno-wrapv is enabled as
GFortran seems to do.
That means the option introduced by #91579 isn't necessary any more.
Note that the feature of -flang-experimental-integer-overflow is enabled
by default.
While building llvm (clang, lld) for wasm using emscripten (recipe
hosted on emscripten-forge
https://github.com/emscripten-forge/recipes/tree/main/recipes/recipes_emscripten/llvm)
I ended up with this error
```
│ │ wasm-ld: error: ../../../../lib/libclangInterpreter.a(Wasm.cpp.o): undefined symbol: lld::wasm::link(llvm::ArrayRef<char const*>, llvm::raw_ostream&, llvm:
│ │ :raw_ostream&, bool, bool)
```
This is due to the link function here
a4819bd46d/clang/lib/Interpreter/Wasm.cpp (L25-L30)
This was added through this PR (https://github.com/llvm/llvm-project/pull/86402) as an attempt to support running clang-repl and executing C++ code interactively inside a Javascript engine using WebAssembly when built with Emscripten.
The definition for link is present in lldwasm and when building for the emscripten platform we should be linking against it.
This changes the entry-point Python script of newhdrgen to use
the idiomatic main() function of no arguments that does the whole
job of the script. This replaces the unusual pattern of having
an idiosyncratic main(...) signature and having the script's
direct code do argument parsing outside the main function. The
idiomatic pattern makes it possible to usefully wrap the script
in Python and still use its full command-line functionality.
CTMark #113200 size overhead was 5.3%, now it's 4.7%.
The patch affects only signed integers.
https://alive2.llvm.org/ce/z/Lv5hyi
* The patch replaces code which extracted sign bit,
maximized/minimized it, then packed it back, with
simple sign bit flip. The another way to think about
transformation is as a subtraction of MIN_SINT from
A/B. Then we map MIN_SINT to 0, 0 to -MIN_SINT, and
MAX_SINT to MAX_UINT.
* Then to maximize/minimize A/B we don't need
to extract sign bit, we can apply shadow the
same way as to other bits.
* After sign bit flip, we had to switch to unsigned
version of the predicates.
* After change above getHighestPossibleValue/getLowestPossibleValue
became very similar, so we can combine into a single function.
* Because the function does sign bit flip and
requires unsigned predicates used for returned values,
there is no point in keeping it as a member of class,
to hide, we switch to function local lambda.
ValueObject is part of lldbCore for historical reasons, but conceptually
it deserves to be its own library. This does introduce a (link-time) circular
dependency between lldbCore and lldbValueObject, which is unfortunate
but probably unavoidable because so many things in LLDB rely on
ValueObject. We already have cycles and these libraries are never built
as dylibs so while this doesn't improve the situation, it also doesn't
make things worse.
The header includes were updated with the following command:
```
find . -type f -exec sed -i.bak "s%include \"lldb/Core/ValueObject%include \"lldb/ValueObject/ValueObject%" '{}' \;
```
This fixes a bug in the tiling implementation of tensor.unpack that was
causing an infinite loop when certain unpack ops get tiled and fused as
a producer. The tiled implementation of tensor.unpack sometimes needs to
create an additional tensor.extract_slice on the result of the tiled
unpack op, but this slice was getting added to the `generatedSlices` of
the tiling result. The `generatedSlices` are used to find the next
producers to fuse, so it caused an infinite loop of fusing the same
unpack op after it was already in the loop. This fixes the bug by adding
the slice of the source instead of the result.
Signed-off-by: Max Dawkins <max.dawkins@gmail.com>
Reuse the definition of profile density from llvm-profgen (#92144):
- the density is computed in perf2bolt using raw samples (perf.data or
pre-aggregated data),
- function density is the ratio of dynamically executed function bytes
to the static function size in bytes,
- profile density:
- functions are sorted by density in decreasing order, accumulating
their respective sample counts,
- profile density is the smallest density covering 99% of total sample
count.
In other words, BOLT binary profile density is the minimum amount of
profile information per function (excluding functions in tail 1% sample
count) which is sufficient to optimize the binary well.
The density threshold of 60 was determined through experiments with
large binaries by reducing the sample count and checking resulting
profile density and performance. The threshold is conservative.
perf2bolt would print the warning if the density is below the threshold
and suggest to increase the sampling duration and/or frequency to reach
a given density, e.g.:
```
BOLT-WARNING: BOLT is estimated to optimize better with 2.8x more samples.
```
Test Plan: updated pre-aggregated-perf.test
Reviewers: maksfb, wlei-llvm, rafaelauler, ayermolo, dcci, WenleiHe
Reviewed By: WenleiHe, wlei-llvm
Pull Request: https://github.com/llvm/llvm-project/pull/101094
Align DataAggregator (Linux perf and pre-aggregated profile reader) to
DataReader (fdata profile reader) behavior: set BF->RawBranchCount which
is used in profile density computation (#101094).
Reviewers: ayermolo, maksfb, dcci, rafaelauler, WenleiHe
Reviewed By: WenleiHe
Pull Request: https://github.com/llvm/llvm-project/pull/101093
These mismatched data layouts are exposed by the refactoring in
https://github.com/llvm/llvm-project/pull/105734 that now also
compares the non-integral pointers list of the data layout.
I dropped the explicit definition from all tests that use opt/llc since
those tools will insert the correct value, but for the llvm-as based
tests and explicit layout is required.
Reviewed By: MaskRay
Pull Request: https://github.com/llvm/llvm-project/pull/107276
Identifiers `clang` and `noescape` are not reserved by the C++ standard,
so perhaps we need to use the equivalent reserved forms.
Also changes the occurrences of that attribute to a macro, following the
convention for `[[_Clang::__lifetimebound__]]`.
Addresses
https://github.com/llvm/llvm-project/pull/91651#discussion_r1807852646.
This change is part of this proposal:
https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294
- `VecFuncs.def`: define intrinsic to sleef/armpl mapping
- `LegalizerHelper.cpp`: add missing fewerElementsVector handling for
the new atan2 intrinsic
- `AArch64ISelLowering.cpp`: Add arch64 specializations for lowering
like neon instructions
- `AArch64LegalizerInfo.cpp`: Legalize atan2.
Part 5 for Implement the atan2 HLSL Function #70096.
I am currently trying to test the LLVM runtimes (including compiler-rt)
against an installed LLVM tree rather than a build tree (since that is
no longer available). Currently, the runtimes build of compiler-rt assumes
that LLVM_BINARY_DIR is writable since it uses configure_file() to write
there during the CMake configure stage. Instead, generate this file inside
CMAKE_CURRENT_BINARY_DIR, which will match LLVM_BINARY_DIR when invoked
from llvm/runtimes/CMakeLists.txt.
I also needed to make a minor change to the hwasan tests: hwasan_symbolize
was previously found in the LLVM_BINARY_DIR, but since it is generated as
part of the compiler-rt build it is now inside the CMake build directory
instead. I fixed this by passing the output directory to lit as
config.compiler_rt_bindir and using llvm_config.add_tool_substitutions().
For testing that we no longer write to the LLVM install directory as
part of testing or configuration, I created a read-only bind mount and
configured the runtimes builds as follows:
```
$ sudo mount --bind --read-only ~/llvm-install /tmp/upstream-llvm-readonly
$ cmake -DCMAKE_BUILD_TYPE=Debug \
-DCMAKE_C_COMPILER=/tmp/upstream-llvm-readonly/bin/clang \
-DCMAKE_CXX_COMPILER=/tmp/upstream-llvm-readonly/bin/clang++ \
-DLLVM_INCLUDE_TESTS=TRUE -DLLVM_ENABLE_ASSERTIONS=TRUE \
-DCOMPILER_RT_INCLUDE_TESTS=TRUE -DCOMPILER_RT_DEBUG=OFF \
-DLLVM_ENABLE_RUNTIMES=compiler-rt \
-DLLVM_BINARY_DIR=/tmp/upstream-llvm-readonly \
-G Ninja -S ~/upstream-llvm-project/runtimes \
-B ~/upstream-llvm-project/runtimes/cmake-build-debug-llvm-git
```
Reviewed By: ldionne
Pull Request: https://github.com/llvm/llvm-project/pull/86209
We have got customer reporting "v &obj" and "p &obj" reporting different
results.
Turns out it only happens for obj that is itself a reference type which
"v &obj" reports the address of the reference itself instead of the
target object the reference points to. This diverged from C++ semantics.
This PR fixes this issue by returning the address of the dereferenced
object if it is reference type.
A new test is added which fails before.
Co-authored-by: jeffreytan81 <jeffreytan@fb.com>
Add pattern that converts a `tensor.expand_shape` op to a more static
form.
This matches the pattern: `tensor.cast` -> `tensor.expand_shape` if it
has a foldable `tensor.cast` and some constant foldable `output_shape`
operands for the `tensor.expand_shape`. This makes the
`tensor.expand_shape` more static, as well as allowing the static
information to be propagated further down in the program.
For image_atomic instructions with TFE, in some cases (e.g., when
dmask=3) the disassembler produces dst register with wrong size (e.g.,
image_atomic_smin v5, v1, s[8:15] dmask:0x3 tfe, instead of v[5:7]).
This patch fixes the VDataDwords values for image atomic instructions.
Clients can now pass multiple parent classes to RTTIExtends. Each parent
class will be inherited from publicly (and non-virtually). The isa,
cast, and dyn_cast methods will now work as expected for types with
multiple inheritance.
Use getVectorLoopRegion to retrieve the vector loop header instead of
relying on getting the successor of the entry block. This makes sure the
code still works correctly as more parts of the skeleton are modeled in
VPlan.
WebAssembly's `memory.fill` and `memory.copy` instructions trap if the
pointers are out of bounds, even if the length is zero. This is
different from LLVM, which expects that it can call `memcpy` on
arbitrary invalid pointers if the length is zero. To avoid spurious
traps, branch around `memory.fill` and `memory.copy` when the length is
zero.
---------
Co-authored-by: Heejin Ahn <aheejin@gmail.com>
Some platforms (e.g. 64-bit CHERI) have stronger alignment requirements
on values returned from allocators. For all other platforms this does
not result in any functional change.
Reviewed By: cjappl, vitalybuka
Pull Request: https://github.com/llvm/llvm-project/pull/84440
While these flags semantically are relevant only for C++, we do add them
to CMAKE_REQUIRED_FLAGS if they are detected. All flags in that variable
are used both when testing compilation of C and C++ (and for detecting
libraries, which uses the C compiler driver).
Therefore, to be sure we safely can add the flags to
CMAKE_REQUIRED_FLAGS, test for the option with the C language.
This should fix compilation with GCC; newer versions of GCC do support
the -nostdlib++ option, but it's only supported by the C++ compiler
driver, not the C driver. (However, many builds of GCC also do accept
the option with the C driver, if GCC was compiled with Ada support
enabled, see [1]. That's why this issue isn't noticed in all
configurations with GCC.)
Clang does support these options in both C and C++ driver modes.
This should fix#90332.
[1]
https://github.com/llvm/llvm-project/issues/90332#issuecomment-2325099254
The use of CLANG_NO_DEFAULT_CONFIG in the tests was added because some
Linux distributions had a global default config file, that added flags
relating to hardening, which interfere with the sanitizer tests. By
setting CLANG_NO_DEFAULT_CONFIG, the global default config files that
are found are ignored, and the sanitizers get the expected default
compiler behaviour.
(This was https://github.com/llvm/llvm-project/issues/60394, which was
fixed in 8ab762557fb057af1a3015211ee116a975027e78.)
However, some toolchains may rely on default config files for mandatory
parts required for functioning at all - setting things like sysroots,
-rtlib, -unwindlib, -stdlib, -fuse-ld etc. In such a case we can't
forcibly disable any default config, because it will break the otherwise
working toolchain.
Add a test for whether the compiler works while passing
--no-default-config to it. If the option is accepted and the toolchain
still works while that is set, set CLANG_NO_DEFAULT_CONFIG while running
tests.
(This adds a little bit of inconsistency, as we're testing for the
command line option, while using the environment variable. However doing
compile testing with an environment variable isn't quite as easily
doable, and passing an extra command line flag to all compile commands
while testing, is a bit clumsy - therefore this inconsistency.)
byval arguments in NVPTX are special. We're only allowed to read from
them using a special instruction, and if we ever need to write to them
or take an address, we must make a local copy and use it, instead.
The problem is that local copies are very expensive, and we create them
very late in the compilation pipeline, so LLVM does not have much of a
chance to eliminate them, if they turn out to be unnecessary.
One way around that is to create such copies early on, and let them
percolate through the optimizations. The copying itself will never
trigger creation of another copy later on, as the reads are allowed. If
LLVM can eliminate it, it's a win. It the full optimization pipeline
can't remove the copy, that's as good as it gets in terms of the effort
we could've done, and it's certainly a much better effort than what we
do now.
This early injection of the copies has potential to create undesireable
side-effects, so it's disabled by default, for now, until it sees more
testing.
