Black-box approximation of bivariate functions

Add to your list(s) Download to your calendar using vCal

• David B Thomas (Imperial College)
• Monday 19 February 2018, 16:00-17:00
• Level 9 Seminar Room, EE Dept..

If you have a question about this talk, please contact George A Constantinides.

There is often a need to evaluate bivariate functions in hardware, using as little area as possible while maintaining full throughput. Custom solutions are available for certain functions such as atan2(x,y), but for an arbitrary f(x,y) there is no general solution for automatically creating a hardware operator. This talk outlines a black-box method for bivariate function approximation, which is intended to work with bivariate input ranges up to _{24-bits, and to provide faithfully rounded solutions at up to} 24 bits of accuracy. In order to minimise RAM and DSP usage, this requires high-order polynomial patches, as well as non-linear range-reduction which can adapt to local curvature. Initial results show the method is quite general, and can handle functions including atan2, incomplete gamma, bessel functions, (and inverses of those functions) in a practical number of resources (i.e. it doesn’t swallow the whole FPGA :)

This talk is part of the CAS Talks series.

This talk is included in these lists:

• Andrea Picciau's list
• CAS Talks
• Circuits and Systems Group: Internal Seminars
• Level 9 Seminar Room, EE Dept.

Note that ex-directory lists are not shown.