Intel Corporation

Duke University

Ralph Nathan

Shih-Lien L Lu

Helia A Naeimi

Daniel J Sorin

John L Gustafson

A mechanism for recycling error bits in a floating point unit is disclosed. A system of the disclosure includes a memory and a processing device communicably coupled to the memory. In one embodiment, the processing device comprising a floating point unit (FPU) to generate a result value from applying an operation on floating point number inputs to the FPU and generate an error value using the result value. The FPU also writes the result value to a first register of the processing device dedicated to storing results from the operation of the FPU and writes the error value to a second register of the processing device dedicated to storing errors from the operation of the FPU. The second register is separate from the first register.

Recycling error bits in floating point units

A mechanism for recycling error bits in a floating point unit is disclosed. A system of the disclosure includes a memory and a processing device communicably coupled to the memory. In one embodiment, the processing device comprising a floating point unit (FPU) to generate a result value from applying an operation on floating point number inputs to the FPU and generate an error value using the result value. The FPU also writes the result value to a first register of the processing device dedicated to storing results from the operation of the FPU and writes the error value to a second register of the processing device dedicated to storing errors from the operation of the FPU.

A method, device and system for representing numbers in a computer including storing a floating-point number M in a computer memory; representing the floating-point number M as an interval with lower and upper bounds A and B when it is accessed by using at least two floating-point numbers in the memory; and then representing M as an interval with lower and upper bounds A and B when it is used in a calculation by using at least three floating-point numbers in the memory. Calculations are performed using the interval and when the data is written back to the memory it may be stored as an interval if the size of the interval is significant, i.e. larger than a first threshold value. A warning regarding the suspect accuracy of any data stored as an interval may be issued if the interval is too large, i.e. larger than a second threshold value.

Growing community of inventors

Durham, NC, United States of America

Ralph Nathan