c44fa3e8a9
This is a followup to #81014 and #84582: Before this patch, Clang would accept `__attribute__((assume))` and `[[clang::assume]]` as nonstandard spellings for the `[[omp::assume]]` attribute; this resulted in a potentially very confusing name clash with C++23’s `[[assume]]` attribute (and GCC’s `assume` attribute with the same semantics). This pr replaces every usage of `__attribute__((assume))` with `[[omp::assume]]` and makes `__attribute__((assume))` and `[[clang::assume]]` alternative spellings for C++23’s `[[assume]]`; this shouldn’t cause any problems due to differences in appertainment and because almost no-one was using this variant spelling to begin with (a use in libclc has already been changed to use a different attribute).
71 lines
1.8 KiB
ReStructuredText
71 lines
1.8 KiB
ReStructuredText
Call may contain unknown parallel regions. Use `[[omp::assume("omp_no_parallelism")]]` to override. [OMP133]
|
|
====================================================================================================================
|
|
|
|
.. _omp133:
|
|
|
|
This analysis remark identifies calls that prevented :ref:`OMP131 <omp131>` from
|
|
providing the generic-mode kernel with a fully specialized state machine. This
|
|
remark will identify each call that may contain unknown parallel regions that
|
|
caused the kernel to require a fallback.
|
|
|
|
Examples
|
|
--------
|
|
|
|
This will occur for any generic-mode kernel that may contain unknown parallel
|
|
regions. This is typically coupled with the :ref:`OMP132 <omp132>` remark.
|
|
|
|
.. code-block:: c++
|
|
|
|
extern void setup();
|
|
|
|
void foo() {
|
|
#pragma omp target
|
|
{
|
|
setup();
|
|
#pragma omp parallel
|
|
{
|
|
work();
|
|
}
|
|
}
|
|
}
|
|
|
|
.. code-block:: console
|
|
|
|
$ clang++ -fopenmp -fopenmp-targets=nvptx64 -O2 -Rpass-analysis=openmp-opt omp133.cpp
|
|
omp133.cpp:6:5: remark: Call may contain unknown parallel regions. Use
|
|
`[[omp::assume("omp_no_parallelism")]]` to override. [OMP133]
|
|
setup();
|
|
^
|
|
|
|
The remark suggests marking the function with the assumption that it contains no
|
|
parallel regions. If this is done then the kernel will be rewritten with a fully
|
|
specialized state machine.
|
|
|
|
.. code-block:: c++
|
|
|
|
[[omp::assume("omp_no_parallelism")]] extern void setup();
|
|
|
|
|
|
void foo() {
|
|
#pragma omp target
|
|
{
|
|
setup();
|
|
#pragma omp parallel
|
|
{
|
|
work();
|
|
}
|
|
}
|
|
}
|
|
|
|
.. code-block:: console
|
|
|
|
$ clang++ -fopenmp -fopenmp-targets=nvptx64 -O2 -Rpass=openmp-opt omp133.cpp
|
|
omp133.cpp:4:1: remark: Rewriting generic-mode kernel with a customized state machine. [OMP131]
|
|
#pragma omp target
|
|
^
|
|
|
|
Diagnostic Scope
|
|
----------------
|
|
|
|
OpenMP target offloading analysis remark.
|