3-D Phase-shift Migration by Matrix Processing
, by Shuki Ronen
Utilizing the full power of vector processors required that many algorithms be adapted for vector processing. Future introduction of systolic-matrix-processors will have similar requirements. The motivation is that matrix-matrix multiplications have a high ratio of arithmetic operations to I/O operations, that effectively reduces internal I/O limitations and allows utilization of the full power of systolic machines.
3-D migration is a computation-intense process that might require a systolic computer if is it to be done routinely. In a previous SEP report, Stew Levin showed how finite-differencing migration would fit into a parallel processor (1984). This paper shows how 3-D phase-shift migration would be done by matrix processing.
STANFORD