What functions does pagefun support?

In Matlab help, you said there are many functions that pagefun supports.

But I check it and found not so many functions are supported by pagefun.

What was worse, sum is not supported by pagefun!!!

PAGEFUN does not support the specified function 'sum'.

When do you make pagefun support the elementary functions, like sum, size.... I am so looking forward to...

[Stephen23]

The pagefun documentation has a list of supported functions.

[Jan]

 When do you make pagefun support the elementary functions, like
 sum, size...

This is the public forum, not the official support of MathWorks. So "we" cannot modify the pagefun code.

[sungho Kang]

Thank you for your answers... I forgot this site is not official... :(

[Stephen23]

@sungho Kang: what MATLAB version are you using?

[sungho Kang]

I am using Mat 16a.

Answer 1

[Matt J]

If you want the page-wise sum of a gpuArray, you don't need pagefun. Examples:

>> X=gpuArray.rand(100,100,200);

>> Y1=sum(X,1); 
>> Y2=sum(X,2);
>> Yall=sum(reshape(X,1,[],200),2);

>> whos X Y1 Y2 Yall

     Name        Size                 Bytes  Class       Attributes

      X         100x100x200                4  gpuArray              
      Y1          1x100x200                4  gpuArray              
      Y2        100x1x200                  4  gpuArray              
      Yall        1x1x200                  4  gpuArray

That said, I agree that it would be nice if TMW advanced the functionality of pagefun(), e.g., to include things like interp2.

[Joss Knight]

Batch interp2 is just interp3. You just need to provide an appropriate query.

The main missing pagefun functions are other batch BLAS and LAPACK operations like non-square mldivide, and svd. Also, some people ask us to relax the restrictions that all the matrices be the same size.

[Matt J]

The problem with posing batch interp2 as an interp3 operation

Vq = interp3(X,Y,Z,V,Xq,Yq,Zq)

is that the query data take up a lot of space. In scenarios where Xq, Yq are the same for every page, you have to do a lot of replication. Even when Xq and Yq are not the same for every page, the Zq data will comprise a whole volume of replicated data just to indicate the page number.

Answer 2

[Joss Knight]

sum already supports batch operations, and with accumarray you can sum arbitrary data partitions.

pagefun supports all the matrix functions listed by the help, and every element-wise gpuArray function. Given how many element-wise operations there are ( plus, times, sqrt, log, sin, acos etc etc etc), isn't it fair to say that's 'many'? Or is it a little disingenuous?!

[Matt J]

But why would one wish to use pagefun for element-wise operations? Isn't

A=pagefun(@times,B,C);

just the same as doing

 A=B.*C;

directly?

[Joss Knight]

That's why I'm suggesting it's a little disingenuous!

It wasn't the same, until they added automatic dimension expansion to the basic binary operators.

[sungho Kang]

Ok.

The reason that I asked the question is that I wanted to use ismember with pagefun, like

pagefun(@ismember, A, B)

because I have very many small matrices(or vectors) to check that each of them is subset of each other matrices(or vectors).

But I got error message with the command, so I looked into how I can do that with pagefun and elementary functions like sum, size, transpose etc.

And I got that the elementary functions of sum, size is not supported by pagefun.

The original problem is that I want to check each row of A is subset of each row of B.

ismember(A(i, :), B(j, :))

To do this very fast, I need to pagefun with ismember. But it's not possible, so I asked the question.