This causes modern compilers (i.e., not MSVC) to emit a single bit-test-and-set instruction instead of a cmpxchg. It's still an atomic operation, so it's not that much faster (if at all), but it has simpler encoding. Previous: 000000000026bca0 <QEventDispatcherUNIX::wakeUp()>: 26bca0: mov 0x8(%rdi),%rdx 26bca4: xor %eax,%eax 26bca6: mov $0x1,%ecx 26bcab: lock cmpxchg %ecx,0x98(%rdx) 26bcb3: jne 26bcc5 <QEventDispatcherUNIX::wakeUp()+0x25> 26bcb5: mov 0x90(%rdx),%edi 26bcbb: mov $0x1,%esi 26bcc0: jmp c01d0 <eventfd_write@plt> 26bcc5: ret Now: 26b3a0: mov 0x8(%rdi),%rax 26b3a4: lock btsl $0x0,0x98(%rax) 26b3ad: jae 26b3b0 <QEventDispatcherUNIX::wakeUp()+0x10> 26b3af: ret 26b3b0: mov 0x90(%rax),%edi 26b3b6: mov $0x1,%esi 26b3bb: jmp c11d0 <eventfd_write@plt> Change-Id: I53335f845a1345299031fffd176fa8ac1de3ad13 Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>