Messages in this thread | | | Date | Wed, 31 Mar 1999 11:23:22 +0100 (BST) | From | Tigran Aivazian <> | Subject | Re: [patch] __volatile__ needed in get_cycles()? |
| |
On Wed, 31 Mar 1999, Andrea Arcangeli wrote: > I am not talking about bugs, I am simply talking about precise profiling > of the asm between two get_cycles_serialized(). Note, I always thought > from time zero that get_cycles_serialized() is not needed at all for > production code, but Tigran asked for an additional get_cycles() that > would take care of reordering issues and since it looked fun I implemented > it :). Well, originally - I merely pointed out that __volatile__ is needed in get_cycles() of 2.2.4 and it was accepted in 2.2.5. It was also said that with __volatile__ get_cycles() can be used for profiling but the word I chose was "pseudo-profiling" because of the speculative execution nature of P6 family. To make "pseudo-profiling" a real "profiling" a separate version of get_cycles() would be needed (i.e. not to cause unneeded overhead in the current use in schedule()/SMP) and you implemented various forms of it (since it looked fun :).
Regards, Tigran.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |