Two operations are supported by the Floating Point Divide/Square Root Unit:
- Floating Point Divide
- Floating Point Square Root
The divider/root core uses a modified SRT algorithm to provide two quotient or root bits per clock cycle. A ‘go’ signal causes the unit to reset, capture the input operands, and to begin the division/root process. A ‘done’ signal is provided to signal that the result is ready. The ‘done’ signal can be configured to appear immediately when a Zero, QNaN, or Infinity result can be determined directly from the input operands.
The operation of this unit follows the IEEE 754 Standard for Floating Point Arithmetic for Single Precision (32-bit) arithmetic for divide and square root with the exception that Denormals are treated as like-signed Zeros. An additional exception flag is provided to signal when a Denormal input or what would have been a Denormal output is flushed to Zero. Results for exceptions are the IEEE Standard default results as defined for when no trap occurs.
NaN formats may be specified with a configurable QNaN bit value and bit position, and a default QNaN may be specified for Invalid results.
The signaling of Underflow is configurable for the detection of tininess either before or after rounding.
Highlighted Features
- IEEE-like Arithmetic (Denormal treated as Zeros)
- IEEE Single Precision Format
- Fully Synchronous Design
- All outputs registered
- Minimal combinational logic on inputs
- 15-clock pipeline rate
- 17-clock latency
Applications
- Microcontrollers
- Microprocessors
More Information
SLE also offers a complete suite of design services to speed the integration of SIP into your system. Contact sales for more information.
