Variable Precision DSP Blocks User Guide: Agilex™ 5 FPGAs and SoCs

ID 813968
Date 9/20/2024
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Visible to Intel only — GUID: lss1690831835653

Ixiasoft

Document Table of Contents

2.3.4. Fixed-point to Floating-point Converter for Tensor Mode

The fixed-point to floating-point converter consists of:
  • A 2's complement converter—Converts the negative DOT product result to a positive number.
  • A leading-zero-detector—Identifies the leading zero position of the positive number.
  • A barrel shifter—Shifts the positive number to the left according to the number of leading zeroes to obtain the upper 19 bits of the 32-bit floating-point number. The most significant bit of the barrel shifter output represents the leading one of the 32-bit floating-point format and is discarded by the barrel shifter.

The converter also calculates the exception values and applies the values to the output data. These exception values include overflow, underflow, and Not-a-Number (NaN).