https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121022
Bug ID: 121022
Summary: Suboptimal code generation for atomic_thread_fence
Product: gcc
Version: 15.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: mmyangfl at gmail dot com
Target Milestone: ---
On x86, atomic_thread_fence usually results in nop (except for
memory_order_seq_cst). However,
#include <stdatomic.h>
#define test(mode) \
int mode(int res) { \
if (res) atomic_thread_fence(memory_order_ ## mode); \
return res; \
}
test(relaxed)
test(consume)
test(acquire)
test(release)
test(acq_rel)
test(seq_cst)
Compiled with -O3:
consume:
mov eax, edi
test edi, edi
jne .L9 <- Superfluous, identical branches
ret
.L9:
ret
