Please correct me if I'm totally off the right track.
I'm interested in finding out which task contributes to higher log_y
and because an individual might do more than one task during each
measuring Y round, so I thought I could use the weight statement. So
if an individual worked longer in task Z than task Y, then task Z
would give more weight to contributing Y. But I'm not sure if this is
how the code should be like.
Your help is really appreciated.

proc mixed data=tmp2 covtest method=reml cl;
        class round Subject task;
        model log_y=task/solution;
        random int / subject=Subject;
        lsmeans task/ pdiff cl;
        weight duration;
run;