I was disappointed with what I felt were somewhat inconclusive results in
this paper.  When they concluded the article they basically under my
interpretation said that  both are good and path profiling has a greater
accuracy.  They gave situations for which edge and path perform well, but
were I trying to write a pratical compiler I wouldn't know which one to
choose.  To remedy this I would like to have seen more data relating to how
each affect compile times and what relative performance increases will be
seen by applying path profiling as opposed to edge on multiple execution
processors.  This leads to a question I'm wondering about the profiling
techniques as to whether they measure actual executed instructions, all
instructions including NOPs.  My last question relates to what paths do
edge techniques miss and how hard are those paths to optimize, is there
something that characterizes them that would make them hard or easy to
improve.