Re: [xsl] How to improve the performance of my matrix addition function?

["Martin Honnen martin.honnen@xxxxxx" <xsl-list-service@xxxxxxxxxxxxxxxxxxxxxx>]

Am 12.07.2020 um 23:53 schrieb Martin Honnen martin.honnen@xxxxxx:
> On 12.07.2020 22:47, Dr. Roger L Costello costello@xxxxxxxxx wrote:
> > Hi Folks,
> >
> > My application uses lots of matrix operations. I used the SAXON Tool
> > Profile (-TP) option to generate a web page that shows the performance
> > of each of my matrix operations. I found that my matrix addition
> > function is taking an appallingly large amount of time. Below is my
> > matrix addition function. Do you have suggestions on ways to improve
> > its performance?B  /Roger

> And of course in XPath 3/XSLT 3 there is the "for-each-pair" function
> made for tasks like this but I have no idea whether that would perform
> better. And to write it solely as a function expression you need XQuery
> or the XPath 4 extension function introduced in Saxon 10 to create
> elements.


Here is an XSLT function uses for-each-pair twice together with
saxon:new-element to construct the row and col elements:


    <xsl:function name="matrix:addition" as="element(Matrix)"
visibility="public">
        <xsl:param name="M" as="element(Matrix)"/>
        <xsl:param name="N" as="element(Matrix)"/>
        <Matrix>
            <xsl:sequence
                select="
                    for-each-pair(
                      $M/row,
                      $N/row,
                      function ($row1, $row2) {
                        saxon:new-element(
                          'row',
                          for-each-pair(
                            $row1/col,
                            $row2/col,
                            function ($col1, $col2) {
                              saxon:new-element('col', $col1 + $col2)
                            }
                          )
                        )
                      }
                    )"
            />
        </Matrix>
    </xsl:function>



Needs Saxon 10 PE or EE and I haven't tested whether it performs any
better than the previous suggestions.