Messages in this thread | | | Date | Tue, 17 Mar 2009 00:10:16 -0400 | From | Mathieu Desnoyers <> | Subject | Re: cli/sti vs local_cmpxchg and local_add_return |
| |
* David Miller (davem@davemloft.net) wrote: > From: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> > Date: Mon, 16 Mar 2009 21:32:20 -0400 > > > If some of you would be kind enough to run my test module provided below > > and provide the results of these tests on a recent kernel (2.6.26~2.6.29 > > should be good) along with their cpuinfo, I would greatly appreciate. > > Here's sparc64, but cycles is always computed as zero. > > Probably that's because get_cycles() on sparc64 counts system > bus clock cycles, not CPU cycles, and you the loop iteration > isn't expensive enough to get a system clock tick in. > > This is a dual UltraSPARC-IIIi at 1.2 GHz >
Hi David,
Thanks for running those tests. Actually, I did not expect good results for sparc64 because the local_t primitives map to atomic_t. Looking at sparc atomic_64.h, I notice that all atomic operations except cmpxchg are done through function calls even when those functions only contain few instructions. Is there any particular reason for that ? These function calls can be quite costly. We could easily inline those.
And to "unleash" the full power of local_t, we should see if there are variants of the atomic operations which are safe only on UP and if there are some memory barriers currently embedded in the atomic_t ops we could remove in a local_t version. Actually, all the BACKOFF_SETUP/BACKOFF_SPIN is specific to SMP, and therefore the local_t version probably does not need that because it touches specifically per-cpu data. That could give very interesting results.
The reason why the results shows 0 cycles per loop is just because there is less that a bus clock cycle per loop. But the total time (in bus cycles) for the whole 20000 cycles gives us equivalent information.
Mathieu
> [1052598.484452] test init > [1052598.486230] test results: time for baseline > [1052598.487878] number of loops: 20000 > [1052598.489485] total time: 752 > [1052598.491061] -> baseline takes 0 cycles > [1052598.492649] test end > [1052598.494874] test results: time for locked cmpxchg > [1052598.496460] number of loops: 20000 > [1052598.498005] total time: 7879 > [1052598.499521] -> locked cmpxchg takes 0 cycles > [1052598.501060] test end > [1052598.503194] test results: time for non locked cmpxchg > [1052598.504733] number of loops: 20000 > [1052598.506213] total time: 7879 > [1052598.507722] -> non locked cmpxchg takes 0 cycles > [1052598.509229] test end > [1052598.511347] test results: time for locked add return > [1052598.512821] number of loops: 20000 > [1052598.514265] total time: 8254 > [1052598.515682] -> locked add return takes 0 cycles > [1052598.517130] test end > [1052598.519427] test results: time for non locked add return > [1052598.520850] number of loops: 20000 > [1052598.522230] total time: 11259 > [1052598.523561] -> non locked add return takes 0 cycles > [1052598.524939] test end > [1052598.526393] test results: time for enabling interrupts (STI) > [1052598.527767] number of loops: 20000 > [1052598.529085] total time: 1877 > [1052598.530373] -> enabling interrupts (STI) takes 0 cycles > [1052598.531713] test end > [1052598.533240] test results: time for disabling interrupts (CLI) > [1052598.534594] number of loops: 20000 > [1052598.535892] total time: 3189 > [1052598.537189] -> disabling interrupts (CLI) takes 0 cycles > [1052598.538551] test end > [1052598.540176] test results: time for disabling/enabling interrupts (STI/CLI) > [1052598.541579] number of loops: 20000 > [1052598.542900] total time: 3940 > [1052598.544207] -> enabling/disabling interrupts (STI/CLI) takes 0 cycles > [1052598.545595] test end >
-- Mathieu Desnoyers OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68
| |