Incorrect roots found about rpolyplusplus HOT 3 OPEN

I cannot reproduce the issue. This is the result I get:

     (0.673943,0.3241)
    (0.673943,-0.3241)
   (-0.85084,0.525425)
  (-0.85084,-0.525425)
  (-0.979328,0.202277)
 (-0.979328,-0.202277)
   (0.533286,0.845935)
  (0.533286,-0.845935)
 (-0.0357928,0.999359)
(-0.0357928,-0.999359)
  (-0.546374,0.837541)
 (-0.546374,-0.837541)
    (1.20511,0.579537)
   (1.20511,-0.579537)

from rpolyplusplus.

I did few other tests, accuracy varies greatly depending on parameters (tolerances, number of shifts, etc.). What parameters do you use? What are residuals for each root? I use defaults (Intel C++ on Windows).

I work with reciprocal polynomials with integer coefficients, most of them have ill-conditioned roots (especially cyclotomic polys).

So far I was not able to get consistent accuracy with RpolyPlusPlus (one set of parameters work for one polynomial but not for the other). I compare it with DHSEQR from MKL/LAPACK, which is very stable (but slow).

One more thing, it is more numerically robust to use std::hypot to compute magnitude of a complex number. It takes care of the overflow/underflow and other accuracy losses.

from rpolyplusplus.

I used the default parameters. Did you enable any particular optimization flags when compiling using Intel C++?

from rpolyplusplus.