This field corresponds to the Syntax enumeration, and that gained
another entry in 1fdf952dee. However,
the bit-field for storing the syntax used was not adjusted to handle
the extra field.
This turns out to be unobservable for HLSL attributes at the moment, so
there is no test coverage. But it's also not really an NFC change
either.
This adds tests checking the behavior of const variables declared with
weak attribute.
Both checking that they can not be used in places where a constant
expression is required and that a dynamic initializer is emitted when
used as an initializer expression.
Differential Revision: https://reviews.llvm.org/D126578
Improved/fixed cost modeling for shuffles by providing masks, improved
cost model for non-identity insertelements.
Differential Revision: https://reviews.llvm.org/D115462
The state is now stored on the thread's stack memory. This enables
implementing pthread API like pthread_detach which takes the pthread_t
structure argument by value.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D126716
This patchs adds the arguments necessary to allocate the size of the
dynamic shared memory via the `LIBOMPTARGET_SHARED_MEMORY_SIZE`
environment variable. This patch only allocates the memory, AMDGPU has a
limitation that shared memory can only be accessed from the kernel
directly. So this will currently only work with optimizations to inline
the accessor function.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D125252
The script uses llvm-link to link LLVM bitcode files.
5426da8ffa used -DLLVM_DISABLE_ASSEMBLY_FILES=ON
to ignore object files compiled from lib/Support/BLAKE3/*.S.
A better approach (which fits Bazel better) is to ignore non-bitcode files.
Reviewed By: akyrtzi
Differential Revision: https://reviews.llvm.org/D126728
sanitizer_intercept_overriders.h might override SANITIZER_INTERCEPT_CRYPT_R to
be undefined. There's no need to require crypt.h in that case.
(The motivation is that crypt() moved from glibc into its own package at some
point, which makes intercepting it and building with a single sysroot that
supports both pre-bullseye and post-bullseye a bit hairy.)
Differential Revision: https://reviews.llvm.org/D126696
There is no direct ctypes for MLIR's complex (and thus np.complex128
and np.complex64) yet, causing the mlir python binding methods for
memrefs to crash. This revision fixes this by passing complex arrays
as tuples of floats, correcting at the boundaries for the proper view.
NOTE: some of these changes (4 -> 2) were forced by the new "linting"
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D126422
Don't build atomic fetch nand libcall functions when the required
compiler builtin isn't available. Without this compiler-rt can't be
built with LLVM 13 or earlier.
Not building the libcall functions isn't optimal, but aligns with the
usecase in FreeBSD where compiler-rt from LLVM 14 is built with an LLVM
13 clang and no LLVM 14 clang is built.
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D126710
Extractelement instructions may come from different basic blocks, need
to take it into account when looking for a last instruction in the
bundle to prevent compiler crash.
Differential Revision: https://reviews.llvm.org/D126777
`isExternCContext()` is returning false for functions in C files
Reviewed By: rnk, aaron.ballman
Differential Revision: https://reviews.llvm.org/D126559
Rename CalleeSavedRegs defs to avoid being overly specific:
* CSR_AMDGPU_AGPRs_32_255 => CSR_AMDGPU_AGPRs
* CSR_AMDGPU_SGPRs_30_31 + CSR_AMDGPU_SGPRs_32_105 => CSR_AMDGPU_SGPRs
* CSR_AMDGPU_SI_Gfx_SGPRs_4_29 + CSR_AMDGPU_SI_Gfx_SGPRs_64_105 =>
CSR_AMDGPU_SI_Gfx_SGPRs
* CSR_AMDGPU_HighRegs => CSR_AMDGPU
* CSR_AMDGPU_HighRegs_With_AGPRs => CSR_AMDGPU_GFX90AInsts
* CSR_AMDGPU_SI_Gfx_With_AGPRs => CSR_AMDGPU_SI_Gfx_GFX90AInsts
Introduce a class RegMask to mark the cases where we use the
CalleeSavedRegs class purely as an expedient way to produce a mask.
Update the names of these masks to not mention "CSR". Other targets also
seem to do this, so a reasonable alternative is to actually update
table-gen to include a new class to do this explicitly, but the current
approach seems harmless so I opted to just make it more explicit.
Reviewed By: arsenm, sebastian-ne
Differential Revision: https://reviews.llvm.org/D109008
As per issue #1196, the loop induction variable, which is an argument
in the omp.wsloop operation, does not have a memory location, so when
passed to a function or subroutine, the reference to the value is not
a memory location, but the value of the induction variable. The callee
function/subroutine is then trying to dereference memory at address 1
or some other "not a good memory location".
This is fixed by creating a temporary memory location and storing the
value of the induction variable in that.
Test fixes as a consequence of the changed code generated.
Add checking for some of the omp-unstructured.f90 to check for alloca,
store and load operations, to ensure the correct flow. Add a test
for CYCLE inside a omp-do loop.
Also convert to use -emit-fir in the omp-unstructrued, and make
the symbol matching consistent in the omp-wsloop-variable test.
Reviewed By: peixin
Differential Revision: https://reviews.llvm.org/D126711
This fillRow(..., 0) is redundant because when the size of the tableau is
consistent, the resize always creates a new row, which is zero-initialized.
Also added asserts throughout to ensure the dimensions of the tableau remain
consistent.
Reviewed By: Groverkss
Differential Revision: https://reviews.llvm.org/D126709
Formatting a string-literal had an off-by-one issue where the NUL
terminator became part of the formatted output.
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D126665
Fix the warning: comparison of unsigned expression in ‘>= 0’ is always
true.
Reviewed By: kiranchandramohan, shraiysh
Differential Revision: https://reviews.llvm.org/D126784
"std::has_unique_object_representations<_BitInt(N)>" was always true,
even if the type has padding bits (since the trait assumes all integer
types have no padding bits). The standard has an explicit note that
this should not hold for types with padding bits.
Differential Revision: https://reviews.llvm.org/D125802
We enable a custom handler to optimize conversions between scalars
and fixed vectors. Unfortunately, the custom handler picks up scalar
to scalar conversions as well. If the scalar types are both legal,
we wouldn't match any of the fixed vector cases and would return SDValue()
causing the LegalizeDAG to expand the bitcast through memory.
This patch fixes this by checking if it's a scalar to scalar conversion
and returns `Op` if both types are legal.
Differential Revision: https://reviews.llvm.org/D126739
This patch supports the following checks:
```
[5.1] 2.21.2 THREADPRIVATE Directive
The threadprivate directive must appear in the declaration section of
a scoping unit in which the common block or variable is declared.
[5.1] 2.14.7 Declare Target Directive
The directive must appear in the declaration section of a scoping unit
in which the common block or variable is declared.
```
Reviewed By: kiranchandramohan, shraiysh, NimishMishra
Differential Revision: https://reviews.llvm.org/D125767
D124631 added special processing for STATEPOINT instructions.
It appears that assertion added there is too strong. We can get two
tied operands with the same register tied to different defs. If we
hit such case, do not process it in statepoint-specific code and
delegate it to common case.
As discussed on D77804, instcombine will have already performed a similar SimplifyMultipleUseDemandedBits call which will break the UXTB16 pattern that was being match in these DAG tests
I've updated the existing tests so that it match the instcombine IR (with a suitable FIXME) and added an equivalent test pattern suggested by @dmgreen
We may need to peek through a bitcast when identifying an fneg idiom
via its pool constant, but we can't allow a different-sized constant
in that match.
This is noted in issue #55758 with an example that needs fast-math,
but as the test here shows, this has potential to miscompile more
generally (no fast-math required).
Differential Revision: https://reviews.llvm.org/D126775
Avoid the dependency on TargetInstrInfo, which depends on the subtarget
and therefore the individual function.
Currently AMDGPU is constructing PseudoSourceValue instances in MachineFunctionInfo.
In order to facilitate copying MachineFunctionInfo, we need to stop allocating these
there. Alternatively we could allow targets to subclass PseudoSourceValueManager,
and allocate them similarly to MachineFunctionInfo.
Note, this is a re-submission of D125894 with `features = ["-header_modules"]`
added to the main BUILD.bazel file.
Some functions like `stpncpy` are implemented in terms of `memset` but are not
currently using `-fno-builtin-memset`. This is somewhat hidden by the fact that
we use `-ffreestanding` globally and that `-ffreestanding` implies
`-fno-builtin` for Clang.
This patch also removes `-mllvm -combiner-global-alias-analysis` that is Clang
specific and that does not bring substantial gains on modern processors.
Also we keep `-mllvm --tail-merge-threshold=0` for aarch64 in CMakeLists.txt
but we omit it in the Bazel config. This is because Bazel consumes the source
files directly and so it can use PGO to take optimal decisions locally.
Differential Revision: https://reviews.llvm.org/D126773
I'm a bit confused by what's actually stored for the allocation
hints. The MIR parser only handles the "simple" case where there's a
single hint. I don't really understand the assertion in
clearSimpleHint, or under what circumstances there are multiple hint
registers.
Stella Laurenzo