Leopold Toetsch <[EMAIL PROTECTED]> wrote: > Here are snippets from the profile:
> with SharedRef > CODE OP FULL NAME CALLS TOTAL TIME AVG T. ms > ---- ----------------- ------- ---------- ---------- > 753 new_p_ic_p 100000 0.157785 0.0016 > 905 set_p_i 100000 0.049269 0.0005 > with Ref > CODE OP FULL NAME CALLS TOTAL TIME AVG T. ms > ---- ----------------- ------- ---------- ---------- > 753 new_p_ic_p 100000 0.051330 0.0005 > 905 set_p_i 100000 0.011356 0.0001 A have redefined the locking macros to use a rwlock in SharedRef: CODE OP FULL NAME CALLS TOTAL TIME AVG T. ms ---- ----------------- ------- ---------- ---------- 753 new_p_ic_p 100000 0.168588 0.0017 -3 DOD 212 0.106973 0.5046 905 set_p_i 100000 0.087742 0.0009 So pthread_rwlock_rdlock() is about 100% slower then pthread_mutex_lock. Acquiring a rwlock seems only be reasonable for lengthy operations like sorting an array or such. leo
