high-performance finite differ ence partial differential equation solvers. The

motivating applicati on is exploration seismology where methods such as

Full-Waveform Inv ersion (FWI) and Reverse-Time Migration (RTM) are used to invert

ter abytes of seismic data to create images of the earth's subsurface. Even us ing modern

supercomputers, it can take weeks to process a single sei smic survey and create

a useful subsurface image. The computational cost is dominated by the numerical

solution of wave equations and th eir corresponding adjoints. Additionally,

the actual set of par tial differential equations being solved and their

numerical discret ization is under constant innovation as increasingly realistic

repre sentations of the physics are developed, further ratcheting up the cost of

practical solvers. By embedding a domain-specific language &nd ash; based on the symbolic

mathematics library SymPy – wi thin Python and exploiting sophisticated compiler

technology, D evito makes it possible to quickly develop highly-optimized fini te

difference solvers. In this talk, we will present the l anguage, compiler architecture,

applications and performance of Devito, with emphasis on automatically generated

MPI-Open MP production-grade wave propagators for high-frequency RTM. Fun damental

features such as boundary conditions and interpolation will also be discussed. END:VEVENT END:VCALENDAR