ISBN-13: 9783844323498 / Angielski / Miękka / 2011 / 148 str.
Reconfigurable computing, in which a general purpose processor (GPP) is augmented with one or more FPGAs, is increasingly used for high-performance computing where massive fine-grain parallelism and pipelining can be exploited. This book proposes a novel transformation called Recursive Variable Expansion (RVE) for reconfigurable computing, which can be applied to an important category of loops. RVE removes all data dependences in an expression. Hence, it can provide more parallelism as compared to the existing loop transformations. Each chapter of this book contains a simple motivational example to understand the basic problem. Later, the problem and its solution is discussed in more technical depth. Finally, the idea is supported by results on the relevant kernels. The contents of the book is useful for researchers and practitioners involved in getting high performance from Reconfigurable Systems.
Reconfigurable computing, in which a general purpose processor (GPP) is augmented with one or more FPGAs, is increasingly used for high-performance computing where massive fine-grain parallelism and pipelining can be exploited. This book proposes a novel transformation called Recursive Variable Expansion (RVE) for reconfigurable computing, which can be applied to an important category of loops. RVE removes all data dependences in an expression. Hence, it can provide more parallelism as compared to the existing loop transformations. Each chapter of this book contains a simple motivational example to understand the basic problem. Later, the problem and its solution is discussed in more technical depth. Finally, the idea is supported by results on the relevant kernels. The contents of the book is useful for researchers and practitioners involved in getting high performance from Reconfigurable Systems.