1.2. FPGA-Based Programmable Hardware Platforms
What constitutes a programmable hardware platform, and, in particular, one appropriate for high-performance computing? The term "platform" is somewhat arbitrary but generally refers to a known, previously verified hardware and/or software configuration that may be used as the basis for one or more specific applications. For our purposes, a programmable platform may be represented by a single FPGA (or other such programmable device), a complete board with multiple FPGAs, or even a development system such as a desktop PC or workstation that is used to emulate the behavior of an FPGA. In short, a programmable hardware platform is a platform that includes at least one programmable hardware element, such as an FPGA, and that can implement all or part of a software algorithm. As you will see, you can also extend the definition of a platform to include various "soft" components that are found within the FPGA itself, such as embedded processors and related peripheral devices, and even such things as embedded operating systems running within the FPGA. All of these elements, taken as a whole, can be used to describe a particular FPGA-based platform.
FPGA-based platforms range from individual FPGAs, with or without embedded processors, and single-board, single-FPGA prototyping platforms to high-performance FPGA computing platforms consisting of multiple FPGAs combined with other processing resources on one or more boards. New FPGA-based platforms are being announced with increasing frequency, so the sample platforms appearing in this book represent only a small fraction of the FPGA-based solutions that are available.
Today's FPGAs are capable of much higher levels of system integration than those of previous generations. In particular, the ability to combine embedded processor cores and related standard peripheral devices with custom hardware functions (intellectual property [IP] blocks as illustrated in Figure 1-1) has made it possible to custom-craft a programmable platform ideally suited to a particular task or particular application domain.
Figure 1-1. FPGA-based platforms may include embedded or adjacent microprocessors combined with application-specific blocks of intellectual property.
The challenges of programmable platform-based design are primarily (but not exclusively) in the domain of the system architect and the software application developer. The success or failure of a development projectas measured in terms of development time and the final performance achieveddepends to a large degree on how well a large application and its constituent algorithms are mapped onto platform resources. To a software application developer it may be less than obvious which portions of the design should go into hardware (for example, in one or more FPGA devices) and which should be implemented as software on a traditional processor, whether a discrete device or a processor core embedded within an FPGA platform. Even more fundamental is the need for software application developers to consider their applications and algorithms in entirely new ways, and to make use of parallel programming techniques that may be unfamiliar to them in order to increase algorithmic concurrency and extract the maximum possible performance.
The fundamental decisions made by an application developer in the early phases of the design, the effectiveness of the algorithm partitioning, and the mapping of the application to hardware resources through the use of automated compiler and hardware synthesis tools can impact the final system's performance by orders of magnitude.