We shouldn't offload the responsibility of the read_begin flush for shared data mutations to the user. read_end requires a load barrier at the least, not a store barrier.ck_pring
parent
ab6cbec48b
commit
4fc1717975
Loading…
Reference in new issue