On Mon, Oct 17, 2016 at 9:21 AM, Maxim Uvarov <maxim.uva...@linaro.org>
wrote:

> by 'some' I mean mask of numa nodes:
>
> odp_numa_mask_t  mask;
>
> odp_pool_create(.., param.mask = 0x5);
>
> where 0x5 will say that numa node 0 and numa node 2 available for
> allocation, but not numa node 1.
>

Given that an odp_dev_t is abstract, That could be encoded in the string
that is supplied to odp_dev_id(), especially since this is intentionally
platform-dependent. Currently there is no application-visible structure in
this. I'm not sure we want to introduce one, as masks would imply.


>
>
> Maxim.
>
>
> On 17 October 2016 at 17:18, Bill Fischofer <bill.fischo...@linaro.org>
> wrote:
>
>>
>>
>> On Mon, Oct 17, 2016 at 8:59 AM, Maxim Uvarov <maxim.uva...@linaro.org>
>> wrote:
>>
>>> On 10/13/16 03:44, Bill Fischofer wrote:
>>>
>>>> +
>>>> +#ifndef ODP_API_DEV_H_
>>>> +#define ODP_API_DEV_H_
>>>> +#include <odp/visibility_begin.h>
>>>> +
>>>> +#ifdef __cplusplus
>>>> +extern "C" {
>>>> +#endif
>>>> +
>>>> +#include <odp/api/std_types.h>
>>>> +
>>>> +/** @defgroup odp_dev ODP DEVICE
>>>> + *  Operations on devices
>>>> + *  @{
>>>> + */
>>>> +
>>>> +/**
>>>> + * @typedef odp_dev_t
>>>> + * ODP Device
>>>> + */
>>>> +
>>>> +/**
>>>> + * @def ODP_DEV_NAME_LEN
>>>> + * Maximum device name length in chars
>>>> + */
>>>> +
>>>> +/**
>>>> + * @def ODP_DEV_ANY
>>>> + * Any device
>>>> + */
>>>> +
>>>> +/**
>>>> + * @def ODP_DEV_INVALID
>>>> + * Invalid device
>>>> + */
>>>>
>>>
>>> By defining INVALID and ANY will we miss SOME? Maybe on memory
>>> allocation it makes sense to provide mask of possible numa nodes and
>>> implementation will choose what is the best?
>>
>>
>> These two special symbols seem needed. What is the use case for
>> ODP_DEV_SOME and how would that have a platform-independent meaning?
>>
>>
>>>
>>>
>>> Maxim.
>>>
>>
>>
>

Reply via email to