https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116717
Bug ID: 116717
Summary: bpf: fix built-in functions for memory model aware
atomic operations.
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: jemarch at gcc dot gnu.org
Target Milestone: ---
Target: bpf
At the moment the BPF target ignores the memory ordering arguments passed to
atomic builtins, generating code that always implements monotonic ordering. We
should change the target to emit instructions for the two ordering schemas
supported by BPF: total and relaxed.
__ATOMIC_RELAXED
lock* instruction if result is not used, atomic_fetch_OP if it is
used.
__ATOMIC_CONSUME
__ATOMIC_ACQUIRE
__ATOMIC_RELEASE
__ATOMIC_ACQ_REL
__ATOMIC_SEQ_CST
BPF: atomic_fetch_OP instructions.
Note this is WIP in clang as well, with corresponding LLVM PR:
https://github.com/llvm/llvm-project/pull/107343
