On May 16, 2007, at 8:40 AM, Bert Wesarg wrote:
>> - Perhaps we should not use the type name cpuset_t since it is used
>> for multiple different kinds of items (sockets, nodes, etc.). How
>> about bitmap_t? E.g. (regardless of whatever the function names end
>> up being):
>>
>> map_from_node_id(int node_id, plpa_bitmap_t *cpuset);
>> get_node_set(plpa_bitmap_t *nodeset);
>> ...etc.
> Yes, this cpu-centric name bothered me too. But maybe bitmap_t
> sounds to
> generic, how is idset_t?
Works for me. Wiki updated.
> The problem with a new naming is, finally we need
> a cpu_set_t for the affinity function! But that shouldn't be a
> problem, or?
Yes, there will need to be a conversion.
BTW, I'm assuming that there will be a prefix on all of these
functions and type names (e.g., topo_<foo> or whatever).
>> - Can we combine map_to_tuple() and map_to_node_id()? It seems silly
>> to have 2 functions for that -- perhaps:
>>
>> map_to_tuple(int processor_id, int *socket, int *core, int *thread,
>> int *node);
> Yes, but than overload the map_to_processor_id function too:
>
> map_to_processor_id(int socket, int core, int thread, int
> *processor_id,
> int *node_id);
Good call.
>> - I wasn't fond of the use of "get_<foo>map" function names, mainly
>> because it seems like we're overloading the use of the word
>> "map" (map_from_<foo> and get_<bar>map). I suggest using some plain
>> words, for example:
>>
>> orig suggestion: map_from_node_id(int node_id, bitmap_t *cpuset);
>> 2nd suggestion: node_get_cpumap(int node_id, bitmap_t *cpumap);
>> JMS suggestion: node_get_processor_ids(int node_id, bitmap_t
>> *processor_ids);
> Definitive more readable, except your last name:
>
> tuple_get_threads(int socket, int core, bitmap_t *threads);
>
> Here 'tuple' is a 2-tuple, but all other mentions of a 'tuple' is a
> 3-tuple'. On the other side socket_core_get_threads() is very long.
Hmm; true -- we're not using "tuple" consistently. Also this one:
map_to_tuple(int processor_id, int *socket, int *core, int
*thread, int *node);
How about for this one (^^):
map_from_processor_id(int processor_id, int *socket, int *core,
int *thread, int *node);
That makes it symmetric with map_to_processor_id().
And for tuple_get_threads(), how about:
...Hmm. I'm drawing a blank. Anyone have any good suggestions?
--
Jeff Squyres
Cisco Systems
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}