Sure does. However, it has to be handled at some point. Especially with C++.


On Mon, Apr 28, 2014 at 3:45 AM, vlab <[email protected]> wrote:

>  Thanks Daniel for the clarification and link
>
> Garbage collection is handy, but does slow things up - just sayin.
>
>
> On 4/28/2014 3:27 AM, Daniel Bell wrote:
>
> OpenCV is pretty fast and they use smart pointers to reduce overhead.
> Normally you only have a header to a matrix not the matrix itself with a
> fair bit of garbage collection going on automatically.  Then, each of these
> matrices are thread safe. Therefore, all of the classes that are made up
> from these matrices are in effect thread safe and offer some degree of
> garbage collection.
>
>  TBB is Threading Building Blocks. OpenCV is from intel and so is TBB.
>
>  https://www.threadingbuildingblocks.org/
>
>
> On Mon, Apr 28, 2014 at 1:23 AM, vlab <[email protected]> wrote:
>
>>  I have worked a fair amount with parallelism of code.  But not familar
>> with 'TBB' - anyone care to clarify?
>> I have recently come to know of NuPIC and just recently downloaded but
>> have yet to look, build nor install.  It will be a week or so till my
>> schedule opens up enough.
>>
>> I'm all for speed and speed means minimal layers of code, working more
>> directly with the HW, such at the GPU using the libraries directly.  But, I
>> don't know the efficacy of OpenCV (another thing for me to look into).
>>
>>
>>
>> On 4/27/2014 9:18 PM, Daniel Bell wrote:
>>
>> Well, i will keep you updated regarding any response from OpenCV. I
>> believe that they use TBB but unsure what role that plays in thread safety
>> as opposed to creating threads.
>>
>>
>> On Mon, Apr 28, 2014 at 4:11 PM, Chetan Surpur <[email protected]>wrote:
>>
>>>  I would think so, especially since it would allow more people to
>>> contribute.
>>>
>>> On April 27, 2014 at 8:58:22 PM, Daniel Bell (
>>> [email protected]) wrote:
>>>
>>>   Would OpenCL be a better direction than pure GPU?
>>>
>>>  I've asked on the OpenCV forum how they have approached this.  There
>>> are a number of options.
>>>
>>>
>>> On Mon, Apr 28, 2014 at 3:46 PM, Chetan Surpur <[email protected]>wrote:
>>>
>>>> I would love to discuss how we would go about making NuPIC parallel. We
>>>> could really use the boost in bandwidth, especially as we approach the
>>>> future. If we could easily refactor the C++ implementation to use the
>>>> graphics card, that could enable applying the CLA to problems that require
>>>> low latency computation.
>>>>  On Apr 27, 2014 8:35 PM, "Daniel Bell" <
>>>> [email protected]> wrote:
>>>>
>>>>> HI Chetan,
>>>>>
>>>>>  I can certainly do that in the next couple of months. If we could
>>>>> find an answer to this, then i can make some progress towards this in a
>>>>> massively parallel and async way. ;)
>>>>>
>>>>>  As far as i am aware, OpenCV is completely thread safe. Maybe we
>>>>> could take some inspiration from there.
>>>>>
>>>>>  Regards,
>>>>>
>>>>>  Daniel
>>>>>
>>>>>
>>>>> On Mon, Apr 28, 2014 at 3:28 PM, Chetan Surpur <[email protected]>wrote:
>>>>>
>>>>>>  Daniel,
>>>>>>
>>>>>>  Great question! I'm not entirely sure, but I just wanted to say
>>>>>> that it would be great if someone experimented with parallelizing NuPIC 
>>>>>> :)
>>>>>>
>>>>>>  - Chetan
>>>>>>
>>>>>>  On April 27, 2014 at 8:17:41 PM, Daniel Bell (
>>>>>> [email protected]) wrote:
>>>>>>
>>>>>>   Hello,
>>>>>>
>>>>>>  I am interested in the progress of this project. From the activity
>>>>>> on the mailing lists, i can see that the next few months will be very
>>>>>> interesting as more material becomes available. I am yet to download and
>>>>>> install due to time restrictions on my current workload.
>>>>>>
>>>>>>  My intention is to use this project within my application once it
>>>>>> is in a more stable state.  When designing my application i am attempting
>>>>>> to use a large amount multi-threading and so i want to know what
>>>>>> considerations i will need to think about.
>>>>>>
>>>>>>  How thread safe are the C++ bindings? What aspects would require
>>>>>> locking for async tasks.
>>>>>>
>>>>>>  My application will lean toward detection and classification
>>>>>> problems for 2D matrices.
>>>>>>
>>>>>>  Regards,
>>>>>>
>>>>>>  Daniel
>>>>>>   _______________________________________________
>>>>>> nupic mailing list
>>>>>> [email protected]
>>>>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> nupic mailing list
>>>>>> [email protected]
>>>>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>>>>
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> nupic mailing list
>>>>> [email protected]
>>>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>>>
>>>>>
>>>> _______________________________________________
>>>> nupic mailing list
>>>> [email protected]
>>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>>
>>>>
>>>  _______________________________________________
>>> nupic mailing list
>>> [email protected]
>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>
>>>
>>> _______________________________________________
>>> nupic mailing list
>>> [email protected]
>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>
>>>
>>
>>
>> _______________________________________________
>> nupic mailing 
>> [email protected]http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>
>>
>>
>> _______________________________________________
>> nupic mailing list
>> [email protected]
>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>
>>
>
>
> _______________________________________________
> nupic mailing 
> [email protected]http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>
>
>
> _______________________________________________
> nupic mailing list
> [email protected]
> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>
>
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to