Storage Allocation for Streaming-Based Register File 181
the individual OAT domains along the variant dimension in case of non-
overlapping lifetimes. This transformation can be done in the SRAM OAT
domain itself as it reduces the memory footprint of each of the arrays. These
steps make the VWR allocation process more efficient. However they are not
required.
For each iteration domain in order of its priority, all the arrays in one
color can be considered for folding with the insertion of load stores. Based
on the number of arrays in one color, the number of locations can be dis-
tributed to each of the individual arrays. Note that in the proposed algorithm
this distribution is done equally among the arrays in one color. However this
may be suboptimal, ...