min and max of gpuArray

Hi,

why does the following fail? I cannot find any caveat about GPU support of min/max on the documentation. Plus, the error message is really not helping. Maybe it has been solved in R2020?

Is there a way to obtain the linear index of the min/max without fecthing the array from the GPU ?

>> A = rand(3, 3, 3, 'gpuArray');
>> min(A, [], 'all')

ans =

    0.0342

>> min(A, [], 'linear')
Error using gpuArray/min
Option must be 'all', 'linear', 'omitnan', or 'includenan'.
 
>> [minimum, index] = min(A, [], 'linear');
Error using gpuArray/min
Option must be 'all', 'linear', 'omitnan', or 'includenan'.

Thanks!

Answer 1

[Matt J]

I cannot find any caveat about GPU support of min/max on the documentation.

Check the "Extended Capabilities" section of doc min.

Maybe it has been solved in R2020?

Yes, it has.

Is there a way to obtain the linear index of the min/max without fecthing the array from the GPU ?

[minimum, index] = min(A(:));

or to operate along a specific dimension, use the attached function minlidx,

[minimum, index] = minlidx(A,dim);

[Massimiliano Zanoli]

awesome, thanks