The measure [Rnk] computes the ranking across all products in the table, regardless of any applied filters. (it behaves much like a calculated column, but this can be modified if you need flexibility/filters)
Rnk = RANKX(ALL('Table'[PRODUCT]),calculate(sum('Table'[Val])))
The cumulative value needs to summarize the value for the "current" product and all products that have a "higher" (ie smaller) rank. However this doesn't work for the Row Subtotal as that would then overcount the sum (it would add 1+1+2+1+2+3+1+2+3+4 etc).
This issue can be turned sideways to make it an advantage.
var ta = SUMMARIZE(ALLSELECTED('Table'),'Table'[PRODUCT],"SM",sum('Table'[Val]))
ta is a table variable that lists the individual products and their sums. If we are in a table cell we need to grab the subset of the results as per [Rnk] but in the Row SubTotal we can use the entire table.
var t = topn([Rnk],ta,[SM],DESC)
t is another table variable, but this time it only fetches the rows from 1 to [Rnk] from our summary table based on the sum being sorted descending. This is a little cheeky as we are abusing the [Rnk] measure for row selection. Technically you could have done the same with a proper RANKX, but where's the fun in that?
The last thing we have to do is figure out where in the table/matrix we are. if HASONEVALUE is true then we are in the table cell, otherwise we are in the row subtotal.
return if(hasonevalue('Table'[PRODUCT]),sumx(t,[SM]),sumx(ta,[SM]))
for the table cell we provide the sum of sums of the sub-table t, for the row subtotal it is the sum of sums of the full table ta.
