Review form for CSEM-reviews 2003
=================================
Paper 9, "Predicate-Aware Scheduling: A Technique for Reducing
Resource Constraints" by
M. Smelyanskiy, S.A. Mahlke E.S. Davidson, H.-H.S. Lee

Reviewer: Peter Aronsson

Summary
=======
The paper presents an improvement for list scheduling -and- modulo
scheduling that will reduce the resource constraint for VLIW
architectures with predicated execution (supported by hardware).

Their improvement is to allow sharing of resources with predicates
that will have disjunct values, e.g. at most one of these predicates
will evaluate to True. The other predicated operations will be
nullified and will therefor not affect the operation predicated with
True.

The paper also gives measures on benchmarks that indicate the
existence of suitable candidates for such predicates, basically an
if-then-else statement which have disjunct predicates for the then and
else part.

The design of adding this improvement to list and modulo
scheduling in general is given, followed by benchmarks to show that
the improvement pays off. An average speedup of about 2-7 % depending on
architecture was gained.

Main contributions
==================
The main contribution is introducing the resource sharing idea for
predicated disjunct operations and giving a design on how to implement
this for both list scheduling and iterative modulo scheduling.

Merits and Weaknesses
=====================
Well founded work with good experiments, both evaluation of existence
of code suited for the improvement and benchmark measures of speedup
in actual compiled code.

A missing part is to give some explanation on the cost (in terms of
computational and space complexity) of implementing this improvement.

Numerical ratings
=================
Significance 8
Originality 8
Interest to Journal on programming lanugage and compiler technology 8
Quality of experimental evaluation 9
Overall organization 8
Presentation 7
Length 6
References appropriate 7

Overall rating 8

Recommendation : accept

Reviewers confidence: 7


Suggestions for improvement:

Table 3 is hard to grasp. Perhaps divide it, or make diagrams
instead. It is good that you already have simpler diagrams, maybe they
should appear before. Perhaps by putting Table 3 in an appendix.