[petsc-users] Vec operations load/stores

[Justin Chang]

Okay, thanks

On Tue, May 19, 2015 at 4:43 PM, Matthew Knepley <knepley at gmail.com> wrote:

> On Tue, May 19, 2015 at 4:30 PM, Justin Chang <jychang48 at gmail.com> wrote:
>
>> Hi everyone,
>>
>> To somewhat follow up on a previous mail, I have a few questions
>> regarding the total bytes transferred for the following Vec operations
>> (assuming Vec size N and sequential mode):
>>
>> VecTDot()
>> VecDot()
>> VecNorm()
>> VecScale()
>> VecSet()
>> VecAXPY()
>> VecAYPX()
>> VecWAXPY
>> VecPointwiseMult()
>>
>> 1) for the first three operations, I am loading two vectors so that's
>> 2*N*8 bytes transferred. For storing, am I simply storing one scalar? Or am
>> I individually storing all N components as they are being summed up?
>>
>
> First 2 operations. No need to count the store of 1 scalar.
>
>
>> 2) for the next five operations where I now have a scalar, am I loading
>> it only once or is it loaded N times?
>>
>
> Once, so don't bother counting.
>
>
>> 3) Do any of the above operations "overlap" or depend on one another? For
>> instance if my solver invokes VecTDot X times does it also
>> invoke, say, VecPointwiseMult X times?
>>
>
> No.
>
>   Thanks,
>
>     Matt
>
>
>> This is all theoretically speaking (i.e., assuming I am bypassing the
>> cache and write-allocate policy).
>>
>> Thanks,
>> Justin
>>
>>
>>
>>
>
>
> --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which their
> experiments lead.
> -- Norbert Wiener
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150519/603886f7/attachment.html>