الفهرس 
IntrodductionOnly 14 pages are availabe for public view
 |  | from | 105 |  |  |
| | | from | 105 | | |
Abstract
High performance computing is needed for evaluating computationally in¬tensive problems. For years, computers have progressed in architecture and hardware aspects to exploit parallelism, but the algorithmic or software as¬pects have not been developed fully in many fields.
High performance multidimensional volume transpose or permutation lies at the heart of data intensive applications generally in most of the academic, scientific and important industrial fields. Transpose is used specially in mul¬timedia production like 4D cinema, nuclear medicine like both 3D and 4D CT (Computed Tomography) and PET (Positron Emission Tomography) imag¬ing, seismic, oil industry, business intelligence like OLAP (On Line Analytical Processing applications) systems and in other fields.
Thus, there is a great need for developing higWy efficient parallel permuta¬tion algorithm.<; for multi-dimensional volumes resulting in shorter execution time and thus maximizing the overall performance.
In this thesis, a contribution to An Efficient in-Place 3D Tronspose for Multicore Processors with Software Managed memory Hier¬archy” algorithm is presented by extending its functionalities to deal with higher dimensions, and not only 3 dimensions (according to the original al¬gorithm’s future work in [1]) while maximizing the performance and exploiting the advantages of Graphics Processing Units.
The complexity analysis of the proposed algorithm L” studied, performance L” me&¬sured and a new parallel hardware (Graphics Processing Card) L” introduced. Re¬sults of experiments demonstrate that the proposed algorithm fulfills its objectives.
**This 3D transpose algorithm will be aliased as either IBM’s algorithm or original algorithm throughout the rest of thesis