Gedare Bloom started a new discussion on cpukit/score/cpu/no_cpu/include/rtems/score/cpu.h: https://gitlab.rtems.org/rtems/rtos/rtems/-/merge_requests/1256#note_152197 > + * -# **Reduced Thermal Load:** It prevents the core from wasting energy > and > + * generating heat on speculative execution paths that will ultimately be > flushed. > + * -# **Shared Resource Yielding:** On SMT (Simultaneous Multithreading) > architectures, > + * it temporarily frees up ALUs and execution units for the sibling > hardware thread. > + * -# **Reduced Bus Contention:** It drastically lowers the rate of memory > read > + * requests, preventing the polling core from saturating the memory > controller. > + * > + * **Deterministic Hedged Read Library (DHRL) Context:** > + * In RTEMS_SMP configurations utilizing the DHRL API via the RTEMS_DHRL > compiler flag, > + * deterministic timing and predictable memory latency across multiple > cores are paramount. > + * When a core is blocked waiting for a hedged read operation to resolve or > a synchronization > + * flag to toggle, a standard tight loop would monopolize the shared memory > bus. > + * `_CPU_Spin_wait()` is utilized within DHRL polling routines to "quiet" > the > + * waiting core. By silencing the memory bus traffic of the blocked core, > it > + * guarantees that the active cores performing the critical path operations > have > + * unimpeded, deterministic access to maximum memory bandwidth. Remove this block, it makes the commit depend on the DHRL library ,which it should not. anyway I don't think this adds much to this place. Maybe add the comment to where you use the `_CPU_Spin_wait()` in the DHRL library instead. -- View it on GitLab: https://gitlab.rtems.org/rtems/rtos/rtems/-/merge_requests/1256#note_152197 You're receiving this email because of your account on gitlab.rtems.org.
_______________________________________________ bugs mailing list [email protected] http://lists.rtems.org/mailman/listinfo/bugs
