Drop changes to consecutive-ptr-uniforms.ll since that test checks boths IR output and debug messages. I'd missed this in the original commit, and Florian pointed it out in post-commit review.
Original commit message:
These are the ones my first round of scripting couldn't handle that required a bit of manual messaging. This should be the last batch in llvm-check.
This reverts commit bbba86764a.
Adds a fallback to use the debuginfod client library (386655) in `findDebugBinary`.
Fixed a cast of Erorr::success() to Expected<> in debuginfod library.
Added Debuginfod to Symbolize deps in gn.
Updates compiler-rt/lib/sanitizer_common/symbolizer/scripts/build_symbolizer.sh to include Debuginfod library to fix sanitizer-x86_64-linux breakage.
Reviewed By: jhenderson, vitalybuka
Differential Revision: https://reviews.llvm.org/D113717
When `a` was an array type, `__decay_copy(a)` was incorrectly marking itself
noexcept(false), because it is false that `int[10]` is nothrow convertible to `int[10]`
(in fact it is not convertible at all).
We have no tests explicitly for `__decay_copy`, but the new ranges::begin
and ranges::end tests fail before this patch.
Differential Revision: https://reviews.llvm.org/D115658
This is the second part of https://reviews.llvm.org/D114993 after slicing
into 2 independent commits.
This is needed at the moment to get good codegen from 2d vector.transfer
ops that aim to compile to SIMD load/store instructions but that can
only do so if the whole 2d transfer shape is handled in one piece, in
particular taking advantage of the memref being contiguous rowmajor.
For instance, if the target architecture has 128bit SIMD then we would
expect that contiguous row-major transfers of <4x4xi8> map to one SIMD
load/store instruction each.
The current generic lowering of multi-dimensional vector.transfer ops
can't achieve that because it peels dimensions one by one, so a transfer
of <4x4xi8> becomes 4 transfers of <4xi8>.
The new patterns here are only enabled for now by
-test-vector-transfer-flatten-patterns.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D114993
There are instances where clang codegen creates stores to
address space 4 in ctors, which causes a crash in llc.
This store was being optimized out at opt levels > 0.
For example:
pragma omp declare target
static const double log_smallx = log2(smallx);
pragma omp end declare target
This patch ensures that any global const that does not
have constant initialization stays in address space 1.
Note - a second patch is in the works where all global
constants are placed in address space 1 during
codegen and then the opt pass InferAdressSpaces
will promote to address space 4 where necessary.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D115661
lld only needs DIContext.h which it gets through Symbolize.h -> SymbolizableModule.h -> DIContext.h. This replaces it with a direct include of DIContext.h to avoid any confusion and pulling in unnecessary headers.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D115659
This patch adds support for prologue and epilogue generation for
the z/OS target under the XPLINK64 ABI for functions with a stack
size of less than 1048576 bytes (huge stack frames).
Reviewed by: uweigand, Kai
Differential Revision: https://reviews.llvm.org/D114457
Link BasicTests via explicit target_link_libraries() rather than
clang_target_link_libraries() in order to fix linking when building
clang against libclang-cpp. The latter requires all listed libraries
to be part of libclang-cpp and omits them if libclang-cpp is used.
However, LLVMTestingSupport is not part of libclang-cpp, so omitting it
causes undefined symbols. Link to the library explicitly to follow suit
with the 7 other unittest programs.
Differential Revision: https://reviews.llvm.org/D115580
This is the first part of https://reviews.llvm.org/D114993 which has been
split into small independent commits.
This is needed at the moment to get good codegen from 2d vector.transfer
ops that aim to compile to SIMD load/store instructions but that can
only do so if the whole 2d transfer shape is handled in one piece, in
particular taking advantage of the memref being contiguous rowmajor.
For instance, if the target architecture has 128bit SIMD then we would
expect that contiguous row-major transfers of <4x4xi8> map to one SIMD
load/store instruction each.
The current generic lowering of multi-dimensional vector.transfer ops
can't achieve that because it peels dimensions one by one, so a transfer
of <4x4xi8> becomes 4 transfers of <4xi8>.
The new patterns here are only enabled for now by
-test-vector-transfer-flatten-patterns.
Reviewed By: nicolasvasilache
* Generalizes passes linalg-detensorize, linalg-fold-unit-extent-dims, convert-elementwise-to-linalg.
* I feel that more work could be done in the future (i.e. make FunctionLike into a proper OpInterface and extend actions in dialect conversion to be trait based), and this patch would be a good record of why that is useful.
* Note for downstreams:
* Since these passes are now generic, they do not automatically nest with pass managers set up for implicit nesting.
* The Detensorize pass must run on a FunctionLike, and this requires explicit nesting.
* Addressed missed comments from the original and per-suggestion removed the assert on FunctionLike in ElementwiseToLinalg and DropUnitDims.cpp, which also is what was causing the integration test to fail.
This reverts commit aa8815e42e.
Differential Revision: https://reviews.llvm.org/D115671
See the discussion in D30906 for how it must be robust to downstream
projects that might not use `GNUInstallDirs`.
I think an easier approach is just to include `GNUInstallDirs` so that
we can always rely on it. Looking at its definition in
https://gitlab.kitware.com/cmake/cmake/-/blob/master/Modules/GNUInstallDirs.cmake
it appears safe to include multiple times.
I am trying this out because we will face the same problem many times
over if we use `GNUInstallDirs` more widely, as proposed in D99484 and
its parent revisions, since we do not know what downstream projects do
and rather not break them.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D115601
MachineOutliner may outline a "patchable-function-entry" function whose body has
a TargetOpcode::PATCHABLE_FUNCTION_ENTER MachineInstr. This is incorrect because
the special code sequence must stay unchanged to be used at run-time.
Avoid outlining PATCHABLE_FUNCTION_ENTER. While here, avoid outlining FENTRY_CALL too
(which doesn't reproduce currently) to allow phase ordering flexibility.
Fixes#52635
Reviewed By: paquette
Differential Revision: https://reviews.llvm.org/D115614
Add lldb support for a Mach-O "load binary" LC_NOTE which provides
a UUID, load address/slide, and possibly a name of a binary that
should be loaded when examining the core.
struct load_binary
{
uint32_t version; // currently 1
uuid_t uuid; // all zeroes if uuid not specified
uint64_t load_address; // virtual address where the macho is loaded, UINT64_MAX if unavail
uint64_t slide; // slide to be applied to file address to get load address, 0 if unavail
char name_cstring[]; // must be nul-byte terminated c-string, '\0' alone if name unavail
} __attribute__((packed));
Differential Revision: https://reviews.llvm.org/D115494
rdar://85069250
This patch makes the process of clauses of atomic directive more clear
and preparation for the support for `atomic compare capture`.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D115586
One of the uses of `LTOCodeGenerator` is to take it as a middle+back end. Sometimes
it is very helpful to access, especially get information from the optimized module.
If the information can be changed in optimization, it cannot be get before the
module is added to `LTOCodeGenerator`. This patch adds a function
`LTOCodeGenerator::getMergedModule` to access the `MergedModule`.
Reviewed By: steven_wu
Differential Revision: https://reviews.llvm.org/D114201
explores various sparsity combinations of
the SDMM kernel and verifies that the computed
result is the same for all cases
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D115476
This reverts commit bbfaf0b170.
Post commit review noted a case where my manual update lost intentional check lines. Given I've abandoned the motivating patch, I'm just reverting the autogen prep.
The gn script for lld's COFF lib adds an unnecessary dependency on llvm/lib/DebugInfo/Symbolize. There is no such dependency in lld/COFF/CMakeLists.txt. This can be safely removed.
Reviewed By: phosek, thakis
Differential Revision: https://reviews.llvm.org/D115554
Add convertFromMLIRSparseTensor to the supporting C shared library to convert
SparseTensorStorage to COO-flavor format.
Add Python routine sparse_tensor_to_coo_tensor to convert sparse tensor storage
pointer to numpy values for COO-flavor format tensor.
Add a Python test for sparse tensor output.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D115557
StructuredDataImpl ownership semantics is unclear at best. Various
structures were holding a non-owning pointer to it, with a comment that
the object is owned somewhere else. From what I was able to gather that
"somewhere else" was the SBStructuredData object, but I am not sure that
all created object eventually made its way there. (It wouldn't matter
even if they did, as we are leaking most of our SBStructuredData
objects.)
Since StructuredDataImpl is just a collection of two (shared) pointers,
there's really no point in elaborate lifetime management, so this patch
replaces all StructuredDataImpl pointers with actual objects or
unique_ptrs to it. This makes it much easier to resolve SBStructuredData
leaks in a follow-up patch.
Differential Revision: https://reviews.llvm.org/D114791
* Generalizes passes linalg-detensorize, linalg-fold-unit-extent-dims, convert-elementwise-to-linalg.
* I feel that more work could be done in the future (i.e. make FunctionLike into a proper OpInterface and extend actions in dialect conversion to be trait based), and this patch would be a good record of why that is useful.
* Note for downstreams:
* Since these passes are now generic, they do not automatically nest with pass managers set up for that.
* If running them over nested functions, you must nest explicitly. Upstream has adopted this style but *-opt still has some uses of implicit pipelines via args. See tests for argument changes needed.
Differential Revision: https://reviews.llvm.org/D115645
Adapt the LinalgStrategyVectorizationPattern pass to apply the vectorization patterns in two stages. The change ensures the generic pad tensor op vectorization pattern does not run too early. Additionally, the revision adds the transfer op canonicalization patterns to the set of applied patterns, since they are needed to enable efficient vectorization for rank-reduced convolutions.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D115627
We had previously been using a different workaround for pretending that
we were inside a system header, however it had some undesirable effects
on dependency parsing for build systems, as explained in [1].
This patch changes the workaround to use `#pragma GCC system_header`,
which shouldn't suffer from the same issue. Unfortunately, it is a lot
more verbose. The issue is that `#pragma GCC system_header` is ignored
when we are inside a source file, so we have to create a header just for
the sake of using it. IMO this seems like an artificial restriction
without much benefit, but investigating that is a different story.
For now, this should at least solve build system problems at the
cost of some readability.
[1]: https://reviews.llvm.org/D95972#3178968
Differential Revision: https://reviews.llvm.org/D115334
A small collection of utils called osutils has been added. These utils
are now used in loader tests instead of the equivalent entrypoints.
The loader can now be built and tested without depending on any entrypoint.
In a follow up change, the relevent entrypoints will be updated to use the
newly added osutils.
Reviewed By: lntue, michaelrj
Differential Revision: https://reviews.llvm.org/D115655
I noticed we weren't propagating tail flags on calls when
FortifiedLibCallSimplifier.optimizeCall() was replacing calls to runtime
checked calls to the non-checked routines (when safe to do so). Make
sure to check this before replacing the original calls!
Also, avoid any libcall transforms when notail/musttail is present.
PR46734
Fixes: https://github.com/llvm/llvm-project/issues/46079
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D107872
This patch provides a draft overlay to support compilation of llvm libc with Bazel.
Tested on linux x86-64 with
```
cd git/llvm-project/utils/bazel
bazelisk-linux-amd64 build --sandbox_base=/dev/shm --config=generic_clang @llvm-project//libc:all
```
Differential Revision: https://reviews.llvm.org/D114712
LiveRangeEdit::allUsesAvailableAt checks that VNI at use is the same
as at the original use slot. However, the VNI can be the same while
a specific subrange needed for use can be dead at the new index.
This patch adds subrange liveness check if there is a subreg use.
Fixes: SWDEV-312810
Differential Revision: https://reviews.llvm.org/D115278
This commit should fix a heap-use-after-free bug that was caught by the
sanitizer bot.
The issue is that we were reading memory from a second target into a
`SBData` object in Python, that was passed to lldb's internal
`ScriptedProcess::DoReadMemory` C++ method.
The ScriptedPythonInterface then extracts the underlying `DataExtractor`
from the `SBData` object, and is used to read the memory with the
appropriate address size and byte order.
Unfortunately, it seems that even though the DataExtractor object was
still valid, it pointed to invalid, possibly garbage-collected memory
from Python.
To mitigate this, the patch uses `SBData::SetDataWithOwnership` to copy
the pointed buffer to lldb's heap memory which prevents the
use-after-free error.
rdar://84511405
Differential Revision: https://reviews.llvm.org/D115654
Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
This patch introduces a new method to SBData: SetDataWithOwnership.
Instead of referencing the pointer to the data, this method copies the
data buffer into lldb's heap memory.
This can prevent having the underlying DataExtractor object point to
freed/garbage-collected memory.
Differential Revision: https://reviews.llvm.org/D115652
Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
If the gather node is a mix of undefvalues and exractelement
instructions, need to take the ordering for such nodes into account too.
It allows to reorder some (sub)trees and remove some extra shuffles,
improving overall vectorization.
Also, outlined common functionality into a separate function.
Differential Revision: https://reviews.llvm.org/D115358
Due to the differences between the types of long double, this function
is effectively three functions in one. This patch adds basic support for
the types of long double, although it's just using the fast path and the
fallback for the moment. I still need to implement a version of
Eisel-Lemire for performance, but the existing algorithms should be
correct.
Reviewed By: sivachandra, lntue
Differential Revision: https://reviews.llvm.org/D113710
Nikolas Klauser