While reading this article the first thing that came to my mind was
one of the possible project goals.  This being the improvement of
block formation.  What I was wondering was how the optimizations that
occur prior to our path profiling vary the effectiveness of the path
profiling.  I was interested in doing some testing relating to
questions of relative performance with and without the other methods
of optimizations.  Maybe this is already a given for how to do
experimental work with the compiler, maybe I skimmed over that line,
but it seems unclear whether the other papers written utilizing
IMPACT address how this is approach.

Another thing that I noticed was that Figure 6 seems like it could be
misleading. This is because it is telling us the cost of performing
new operations relative to the IMPACT compiler.  Unfortunately
profiling may presumably not be able to be sped up at the same ratio
that commercial compilers compare with the IMPACT instructions.  So
for example say it takes IMPACT normally an hour to compile a
benchmark where gcc takes 5 minutes, was it realistic to believe that
a new profiling stage would also take 1/20th of the time on gcc and
so on for optimization.