I have added a wrapper for dotu so in the latests julias you can write
BLAS.dotu(a,b).


2014/1/2 Sheehan Olver <[email protected]>

> My main problem is that
>
> (a.'*b)[1,1]
>
> looks a bit ugly.  I'll just implement my own dotu
>
> Sent from my iPhone
>
> On Jan 3, 2014, at 12:21 AM, Stefan Karpinski <[email protected]>
> wrote:
>
> Unless the vectors in question are small, the allocation of a one-element
> array is unlikely to be significant. If the vectors are small, writing out
> the dot product yourself is likely to be faster; In fact, it may be faster
> anyway.
>
> On Jan 2, 2014, at 7:38 AM, Ivar Nesje <[email protected]> wrote:
>
> Thanks for correcting me. I should have checked better before answering,
> especially because it is so simple in Julia to follow the functions and see
> what actually gets calculated.
>
> kl. 13:29:49 UTC+1 torsdag 2. januar 2014 skrev Andreas Noack Jensen
> følgende:
>>
>> The problem here is that the method in operators.jl is
>> At_mul_B(a,b)=transport(a)*b and therefore there is a transposed copy in
>> the calculation.
>>
>>
>> 2014/1/2 Ivar Nesje <[email protected]>
>>
>>> Julia does part of transformation automatically for you.
>>>
>>> If you look at
>>>
>>> julia> @which a.'*b
>>> At_mul_B(a,b) at operators.jl:122
>>>
>>> You can see that the call is automatically rewritten to At_mul_B(a,b)
>>> without making a transposed copy.
>>>
>>> I am not sure what you can do about the result being a 1 element
>>> Array{Complex{Float64},1} instead of just a Complex{Float64}.
>>>
>>> Ivar
>>>
>>> kl. 13:02:39 UTC+1 torsdag 2. januar 2014 skrev Sheehan Olver følgende:
>>>
>>>>
>>>> I want to do
>>>>
>>>>   a.'*b
>>>>
>>>> where a and b are Vector{Complex}, but this returns an Array, not a
>>>> constant, and probably does unneccesary memory allocation for when
>>>> constructing  a.'
>>>>
>>>> If it was
>>>>
>>>>   a'*b
>>>>
>>>> I can just replace it with dot(a,b).  Is there an equivalent that
>>>> doesn't conjugate the first argument?
>>>>
>>>
>>
>>
>> --
>> Med venlig hilsen
>>
>> Andreas Noack Jensen
>>
>


-- 
Med venlig hilsen

Andreas Noack Jensen

Reply via email to