Floating-Point Processor Cores
GB3 offers an array of float-point processors for embedded
multi-media applications. All of the processors are designed to
the IEEE-754 standard, IEEE Standard for Binary Floating-Point
Arithmetic, ANSI/IEEE Std 754-1985. The cores implement the
traditional single (32-bit) and double (64-bit) precision
formats, as well as the new half precision format (16-bit).
The FPSDZ
design implements a streamlined version of IEEE-754 standard that
supports both single and double precision operands. For maximum
performance with minimum size, the FPSDZ design omits two of the
high overhead, low utility features of the standard: gradual
underflow and denormal numbers. The core offers a high performance
3-stage pipeline, and an easy-to-use interface. The core is
designed for the user that needs double and single precision
operations, high throughput performance, and nothing else.
The FPSX
is a collection of four single precision floating-point
modules designed for high performance applications. Each module
implements a pipeline with only 10-12 gate delays in each clock
cycle for greater than 1 GHz performance. All modules can
sustain a 1-cycle throughput. Four modules are provided: add,
multiply, conversions, and compare. The multiply module
implements a fused floating-point multiply-add, and also supports
several 32-bit integer multiply functions. FPSZ offers a clean
interface and the same IEEE-754 compliance level as the FPSDZ, and
is an ideal design collection for compute intensive applications.
The FPSZ is a single
precision floating-point unit designed for low-power, minimal area implementations.
Since low-power applications usually necessitate low clock frequencies,
the FPSZ is designed with a 1-stage pipeline, where most instructions
execute in one clock cycle. This feature ensures maximum performance
at low clock frequencies, and as a bonus, eliminates the extra logic
needed to support a multi-stage pipeline. The FPSZ offers the same clean
interface and IEEE-754 compliance level as the FPSDZ, and is an ideal
design for a wide range of low-power applications.
The A7FPSZ combines
the FPSZ core with an ARM7® co-processor interface, and a user-configurable
register file. The result is a small but powerful floating-point co-processor
to one of the world's most popular embedded processor architectures.
The newest addition to the family is the FPHZ
core, a small but very powerful half precision floating-point
unit. Like it's single precision cousin, the FPHZ implements a
1-stage pipeline that maximizes performance in low-power
designs. The FPHZ is truly tiny - less than 1/3 the size of the
single precision unit! The FPHZ is an ideal choice for a pixel
processor in image processing applications.
|