|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The Vector Calculator is a fully functional example
application of the RARE
|
|
project. Its hardware modules are designed for
floating point operations
|
|
|
and optimized for speed. Each hardware modules runs
at 50 MHz
|
|
|
|
|
|
|
|
|
|
Device utilization for hardware modules are
Adder/Subtractor 451
|
|
(28%) and Multiplier/Divider� 969 (60%). Each module is highly
|
|
|
pipelined and can handle one or more vector sets at a
time.
|
|
|
|
|
|
|
|
|
For comparison purposes, the calculator is
implemented in C++, Java and
|
|
FPGA hardware and can be executed on a local machine
or at a remote machine.
|
|
|
|
|
|
|
|
|
|
|
|
The PNN Image Classifier is another example
application developed under
|
|
the RARE project. The Classifier is implemented using
fixed point modules.
|
|
|
|
|
|
|
The hardware design is divided into two sections and
each section is mapped
|
|
onto an FPGA chip. Device utilizations are 659 (41%)
and 928 (58%).
|
|
|
|
|
|
|
|
|
|
|
In this application, the host first initializes the
board by storing weights values
|
|
|
into the PE1 memory.
Then, the host sends input data to the hardware one pixel
|
|
at a time through FIFO1. The
hardware then calculates the class values for each
|
|
class. After that the host receives these values and
determines the winning class.
|
|
|
|