Percolation scheduling

A. Nicolau, Percolation Scheduling A Parallel Compilation Technique, Ithaca, NY Dept, of Computer Science, Cornell University, TR 85-678, May 1985. 

First, uses percolation scheduling to allow data-independent operations to percolate toward earlier control steps, not considering resource constraints. To meet resource constraints, the mobility of each operation in the control step is computed, and operations with higher mobility are delayed until later control steps. 

Different solutions to this problem have been proposed. In [7], Fisher described a series of transformations used in trace scheduling for microcode compaction. This method compacts the critical path independently of the control block boundaries. A refinement of this tqjproach was proposed recently by Potasman et al and is called percolation scheduling [8]. Camposano and Ber unaschi use an alternative approach called path-based scheduling [9], wh e all possible paths are compacted independently. A post-processing step merges separate paths to Iowa the controller costs. 

Its main features are given by the use of a stream of inert carrier gas which percolates through a bed of an adsorbent covered with adsorbate and heated in a defined way. The desorbed gas is carried off to a detector under conditions of no appreciable back-diffusion. This means that the actual concentration of the desorbed species in the bed is reproduced in the detector after a time lag which depends on the flow velocity and the distance. The theory of this method has been developed for a linear heating schedule, first-order desorption kinetics, no adsorbable component in the entering carrier gas (Pa = 0), and the Langmuir concept, and has already been reviewed (48, 49) so that it will not be dealt with here. An analysis of how closely the actual experimental conditions meet the idealized model is not available. 

For any given cooling schedule, when an amorphous polymer undergoes a glass transition while the free volume no longer percolates through the structure, the available fraction is still systematically reduced by structural aging if the polymer... 

The controlflow/dataflow graphs are rearranged and optimized. Using a percolation algorithm, the vertices are scheduled into specific states, subject to data dependencies, resource limits, and other constraints. 

---