[Octave-bug-tracker] [bug #63184] [MXE Octave] AMD ZEN AOCL library support for blas, lapack and fftw

[Dmitri A. Sergatskov]

Follow-up Comment #19, bug #63184 (project octave):

Test failures with blis:


***** warning <rcond = > assert (inv (Inf (2,2)), NaN (2,2))
!!!!! warning failed.
Expected warning <rcond = >, but got error <ASSERT errors for:  assert (inv
(Inf (2, 2)),NaN (2, 2))

  Location  |  Observed  |  Expected  |  Reason
   (1,1)          0           NaN        'NaN' mismatch
   (2,1)          0           NaN        'NaN' mismatch>

***** test
 [xinv, rcond] = inv (Inf (2,2));
 assert (xinv, NaN (2,2));
 assert (rcond, NaN);
!!!!! test failed
ASSERT errors for:  assert (xinv,NaN (2, 2))

  Location  |  Observed  |  Expected  |  Reason
   (1,1)          0           NaN        'NaN' mismatch
   (2,1)          0           NaN        'NaN' mismatch



***** assert <*60797> (sqrtm (ones (4))^2, ones (4), 5*eps)
!!!!! regression: https://octave.org/testfailure/?60797
ASSERT errors for:  assert (sqrtm (ones (4)) ^ 2,ones (4),5 * eps)
 
  Location  |  Observed  |  Expected  |  Reason
   (1,1)       NaN+NaNi        1         'NaN' mismatch
   (2,1)       NaN+NaNi        1         'NaN' mismatch
   (3,1)       NaN+NaNi        1         'NaN' mismatch
   (4,1)       NaN+NaNi        1         'NaN' mismatch
   (1,2)       NaN+NaNi        1         'NaN' mismatch
   (2,2)       NaN+NaNi        1         'NaN' mismatch
   (3,2)       NaN+NaNi        1         'NaN' mismatch
   (4,2)       NaN+NaNi        1         'NaN' mismatch
   (1,3)       NaN+NaNi        1         'NaN' mismatch
   (2,3)       NaN+NaNi        1         'NaN' mismatch
   (3,3)       NaN+NaNi        1         'NaN' mismatch
   (4,3)       NaN+NaNi        1         'NaN' mismatch
   (1,4)       NaN+NaNi        1         'NaN' mismatch
   (2,4)       NaN+NaNi        1         'NaN' mismatch
   (3,4)       NaN+NaNi        1         'NaN' mismatch
   (4,4)       NaN+NaNi        1         'NaN' mismatch



***** test
 warning ("off", "Octave:nearly-singular-matrix", "local");
 x = 1000 + (-5:5);
 xn = (x - mean (x)) / std (x);
 pn = ones (1,5);
 y = polyval (pn, xn);
 n = numel (pn) - 1;
 [p, s, mu] = polyfit (x, y, n);
 [p2, s2] = polyfit (x, y, n);
 assert (p, pn, s.normr);
 assert (s.yf, y, s.normr);
 assert (mu, [mean(x), std(x)]);
 assert (s.normr/s2.normr < sqrt (eps));
!!!!! test failed
ASSERT errors for:  assert (p,pn,s.normr)

  Location  |  Observed  |  Expected  |  Reason
    (3)           1            1         Abs err 1.7764e-15 exceeds tol
1.6241e-15 by 2e-16

(that looks fairly benign)

Performance-wise it seems to outperform openblas at least 
in some benchmarks. E.g.:


$ BLIS_NUM_THREADS=16
LD_PRELOAD=/opt/AMD/aocl/aocl-linux-gcc-3.2.0/lib_LP64/libblis-mt.so octave
-q
octave:1> a=randn(5000);
octave:2> tic; inv(a)*a; toc
Elapsed time is 1.43057 seconds.

$ OMP_NUM_THREADS=16 octave -q
octave:1> a=randn(5000);
octave:2> tic; inv(a)*a; toc
Elapsed time is 1.8783 seconds.



Dmitri.
-- 




    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?63184>

_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/