ALGEBRAIC PROCESSOR

• US 20130185345A1
• Filed: 01/16/2012
• Published: 07/18/2013
• Est. Priority Date: 01/16/2012
• Status: Abandoned Application

• Current Assignee: Qualcomm, Inc.
• Original Assignee: DesignArt Networks Limited (Qualcomm, Inc.)
• Inventors: TSADIK, MEIR, TOUBOUL, ASSAF
• Application Number:
  US13/350,850
• Publication Number:
  US 20130185345A1
• Time in Patent Office: Days
• US Class Current: 708/235
• CPC Class Codes: G06F 7/544 for evaluating functions by calculation G06F7/4824 takes precedence

Abstract

An algebraic processor as part of a wireless telecommunication system, including pre-computed Look Up Tables (LUT), used for computing a number of different functions using linear interpolation. Preferably, the step of computing is implemented in a multiplier-accumulator having a SIMD structure.

Claims

1. An algebraic processor comprising:

a programmable hardware unit including;

at least one lookup table storing values of at least one function, said lookup table including two values for each entry;

an arithmetic engine performing a same mathematical operation on two operands from said at least one lookup table and an input operand, thereby estimating a value of said at least one function by means of linear interpolation; and

an output register.

2. The algebraic processor according to claim 1, wherein said arithmetic engine includes a multiplier-accumulator and a scaling module.

3. The algebraic processor according to claim 1, wherein said arithmetic engine performs said same mathematical operation on a plurality of pairs of operands from said at least one lookup table and corresponding input operands, in a single cycle.

4. The algebraic processor according to claim 1, wherein said programmable hardware unit is a vector device.

5. The algebraic processor according to claim 4, wherein said vector device is a multiplier-accumulator having a Single Instruction Multiple Data (SIMD) structure.

6. The algebraic processor according to claim 1, wherein said programmable hardware unit is a scalar device.

7. The algebraic processor according to claim 1, wherein said at least one look up table includes pre-calculated function values and the pre-calculated derivatives of those values and the arithmetic engine performs interpolation from one of these pre-calculated numbers to the required input value, using Taylor polynomials.

8. The algebraic processor according to claim 7, further comprising additional values for each entry, each being a pre-calculated further derivative of said function value.

9. A method for calculating a selected function for an input operand, the method comprising:

calculating, in a programmable hardware unit, an approximate value of a selected function for an input word using linear interpolation from pre-calculated values of said selected function; and

outputting said approximate value of said function of said input word to an output register.

10. A method for calculating a selected function for an input word, the method comprising:

receiving an instruction, according to a selected resolution, for dividing the input word;

dividing said input word, according to said received instruction, into an index for a lookup table and an input operand;

using said index, reading pre-calculated values from said lookup table as operands for at least one function to be calculated; and

performing a same mathematical operation on said operands from said at least one lookup table and an input operand, thereby calculate an approximation of said function of said input word by means of interpolation

11. The method according to claim 10, wherein said mathematical operation is performed on multiple groups of operands in a single cycle.

12. The method according to claim 10, wherein said step of performing is implemented in a multiplier-accumulator SIMD.

13. The method according to claim 10, wherein said mathematical operation is linear interpolation.

14. The method according to claim 10, wherein said mathematical operation includes performing interpolation according to the formula:

f(x)=f(a₀+dx)=f(a₀)+f′

(a₀)*dx

using fixed point multiplication.

15. The method according to claim 10,

wherein said operation includes multiplication and summing;

and further comprising applying a scaling shift before said summing operation.