The Art of High Performance Computing - Volume 2: Parallel Programming for Science and Engineering

Livro digital

Título:
The Art of High Performance Computing - Volume 2: Parallel Programming for Science and Engineering

Autor:
Victor Eijkhout

Categoria:
Tecnologia > Geral

Doador:
Raffaello D. N.

Sinopse:
Writing a program that uses thousands of cores is not just a matter of adding more processors — it requires mastering a distinct set of abstractions, and this volume treats MPI and OpenMP as complementary toolkits rather than competing philosophies. The MPI half moves from first send/receive through collectives, derived datatypes, communicators, one-sided RMA, and parallel I/O, with each concept illustrated in C and Fortran side by side. The OpenMP half covers parallel regions, worksharing, explicit tasks, SIMD directives, and GPU offloading — the full breadth of the specification — before a dedicated chapter on the PETSc library shows how higher-level abstractions can shield the scientist from low-level message passing. Chapters on hybrid MPI+OpenMP, GPU acceleration with CUDA and HIP, and mixed-language linking round out the practical toolkit. Published under CC-BY and backed by a full set of downloadable slides, this volume serves the computational scientist who has mastered sequential programming and now needs to scale — whether across a workstation's cores or a cluster's nodes.