XSL unique value key

Goal

(XSLT 1.0). My goal is to take a set of elements, S, and produce another set, T, where T contains the unique elements in S. And to do so as efficiently as possible. (Note: I don't have to create a variable containing the set, or anything like that. I just need to loop over the elements that are unique).

Example Input and Key

<!-- My actual input consists of a bunch of <Result> elements -开发者_Go百科->
<AllMyResults>
<Result>
<someElement>value</state>
<otherElement>value 2</state>
<subject>Get unique subjects!</state>
</Result>
</AllMyResults>

<xsl:key name="SubjectKey" match="AllMyResults/Result" use="subject"/>

I think the above works, but when I go to use my key, it is incredibly slow. Below is the code for how I use my key.

<xsl:for-each select="Result[count(. | key('SubjectKey', subject)[1]) = 1]">
    <xsl:sort select="subject" />
            <!-- Do something with the unique subject value -->
    <xsl:value-of select="subject" />
</xsl:for-each>

Additional Info

I believe I am doing this wrong because it slowed down my XSL considerably. As some additional info, the code shown above is in a separate XSL file from my main XSL file. From the main XSL, I am calling a template that contains the xsl:key and the for-each shown above. The input to this template is an xsl:param containing my node-set (similar to the example input shown above).

I can't see any reason from the information given why the code should be slow. It might be worth seeing if the slowness is something that happens on all XSLT processors, or if it's peculiar to one.

Try substituting

count(. | key('SubjectKey', subject)[1]) = 1

with:

generate-id() = generate-id(key('SubjectKey', subject)[1])

In some XSLT processors the latter is much faster.