Re: Vector Rotation

Clay S. Turner wrote:

Hello Carlos,

One of the standard tricks in hardware is to have a fixed number of iterations. This is so the algo can be pipelined. So using the constraint of a fixed number of iterations is how I arrived at a region of convergence that gets smaller for x small. The region is determined as the domain of the seed so that after the fixed number of iterations the result is within a prespecified epsilon of the true answer. The algo that pairs bits and uses trial subtractions, will produce a known precision given a priori known number of iterations. In this algo, the square root is the same order of complexity as a division.

Gotcha!

Thanks for the clarification!

Carlos
--
.

Relevant Pages

  • Re: Loop Efficiency
    ... although the array is untouched. ... Algo 1 took 24828ms for 10000 iterations ... private void makeData{ ...
    (comp.lang.java.programmer)
  • Re: Vector Rotation
    ... The region of convergence is very small for x ... is no way that the algorithm can diverge. ... About 13 iterations ... This is so the algo can be pipelined. ...
    (comp.dsp)
  • Re: Infinite Loops and Explicit Exits
    ... > GO TO -EXIT stops one cycle, not iterations. ... COBOL program is that the program branched out of a perform range and ... Unrepentant Poopie-Face isn't relevant to much of anything. ... brought about by PERFORM back in COBOL-60 or in any standard COBOL since ...
    (comp.lang.cobol)
  • Re: fast and accurate in mixed mode operations
    ... < and floating-point pipelines is slow, but often used to be done by ... Other than processors with hardware transcendental ... For DSQRT no fixed point divide, but four iterations. ...
    (comp.lang.fortran)
  • Re: fixed point math algorithms
    ... above into fixed point hardware? ... The wonderful thing about fixed point math is that a Taylor series will ... function in Matlab that allows for fixed point data types. ... of iterations, I was still not getting the accuracy that I ...
    (comp.lang.vhdl)