results are not apparently deterministic?

[shabbychef]

I am hunting down a bug that we have traced back to sedumi not being deterministic. I have a MWE:

sed_A = sparse([1e-06 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.000374569348444187 0 -0.000143510377232224 0 -0.000107308290246781 0 -0.00023475302385101 0 -0.000222588804655496 0 -0.00039621568504792 0 -0.000175372571085547 0 0 0 0 0 0 0 0;0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00182361766281337 0 0.000300035708922513 0 5.25148072155411e-05 0 0.000172937546376846 0 0.000368772505991557 0 0.000258802261860665 0 9.13124417751099e-05 0 -1 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00166011586827782 0 5.08618097827969e-05 0 0.000550274603301129 0 -1.24577387791141e-05 0 2.80912803114481e-05 0 0.000787997869564677 0 0 -1 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00131496023553193 0 -2.85427725918689e-05 0 -6.40917512673142e-05 0 -6.48347565119609e-05 0 -9.88411732133345e-05 0 0 0 -1 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00116855302122233 0 -1.40653733430151e-05 0 1.38084332511408e-05 0 0.00052463330628182 0 0 0 0 -1 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00177126319840008 0 0.000136498835240328 0 -0.000139615027512656 0 0 0 0 0 -1 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00218208935736517 0 -0.000117739846758266 0 0 0 0 0 0 -1 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00170708755797057 0 0 0 0 0 0 0 -1;-1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0]);
sed_b = [-5e-08;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0];
sed_c = [0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;1;0;0;0;0;0;0;0];
sed_K = struct(sprintf('f'), {0}, sprintf('l'), {17}, sprintf('q'), {[2 2 2 2 2 2 2 8]}, sprintf('r'), {0});
pars = struct(sprintf('fid'), {0}, sprintf('maxiter'), {100});

% run sedumi 200 times, and hash the output:
for i=1:200
    [x,y,info]  = sedumi(sed_A,sed_b,sed_c,sed_K,pars);
    % get the md5 hash of x:
    myhash = java.security.MessageDigest.getInstance('MD5');
    myhash.update(typecast(x,'uint8'));
    hstr = lower(sprintf('%02X',typecast(myhash.digest,'uint8')));
    disp(hstr);
end

In any one Matlab instance, I tend to get the same results. In different Matlab instances, I can get different output. I have gotten all the following output hashes:

1347dafee66c8ff3680ff91ec51cc8bc
01b524f80080f1de14e910d6878ea380
1a8829b52cdf4e970f48ce4c2c0d5003
79f20e7f39ea5e1fba8c06f45d401960

and a few more. I have confirmed that the different hashes correspond to different output. Typically the variation is on the order of 1e-7, indicating some kind of single to double conversion might be at fault.

This is in Matlab r2007b on a linux machine. I have confirmed the results on a few diffferent linux machines all running r2007b. I have code which sets the random seed used by Matlab, and it does not appear to fix the issue. I have also tried setting maxNumCompThreads(1) to see if multithreading causes the problem. It does not appear to.

edit: I am using SeDuMi 1.21

[mcg1969]

Interesting! I'm not entirely sure where randomness my be introduced.

[mcg1969]

(Sorry, didn't mean to close it. Re-opened.)

[robertgj]

deczky3_socp_test.gdb.session.txt
I have been trying to track down a
similar problem. I don't have MATLAB but I have made minor edits to SeDuMi_1_3 to build it under octave. I built octave-4.0.0 with address-sanitizer added in the environment (CFLAGS="-g -fsanitize=address -fno-omit-frame-pointer", etc) and ran a script that eventually calls SeDuMi. address sanitizer is triggered at blkchol2.c:79. This appears to be an off-by-one error. "x[idamax_(&n,x,&one)]" is trying to read past the end of x. I have attached a debugging session. I will spend some time trying to work out what is happening in the dense cholesky factorisation in cholonBlk().

[robertgj]

This hack seems to fix my problem:

--- SeDuMi_1_3.orig/blkchol2.c  2009-10-19 22:09:54.000000000 +1100
+++ SeDuMi_1_3/blkchol2.c   2015-11-19 11:20:07.062226915 +1100
@@ -124,7 +124,7 @@ void cholonBlk(double *x, double *d, mwI
    ------------------------------------------------------- */
     xkk = x[inz];
     if(xkk > lb[k]){ /* now xkk > 0 */
-      if(xkk < ub){
+      if((m>1) && (xkk < ub)){
         ubk = maxabs(x+inz+1,m-1) / maxu;
         if(xkk < ubk){
 /* ------------------------------------------------------------

[mcg1969]

Thanks for finding this!

[shabbychef]

I left my former job and no longer have access to Matlab, so I cannot test this issue. I have notified a former colleague (@cerebellum-github ?) regarding this fix, hoping he can confirm the issue is resolved.

[robertgj]

The following, copied from the original post, works for me in Octave (built on my machine under Fedora 23 from the octave-4.0.0.tar.xz source without native graphics using "./configure --without-fltk --disable-gui --disable-java"):

more off
addpath('SeDuMi_1_3');
sed_A = sparse([1e-06 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.000374569348444187 0 -0.000143510377232224 0 -0.000107308290246781 0 -0.00023475302385101 0 -0.000222588804655496 0 -0.00039621568504792 0 -0.000175372571085547 0 0 0 0 0 0 0 0;0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00182361766281337 0 0.000300035708922513 0 5.25148072155411e-05 0 0.000172937546376846 0 0.000368772505991557 0 0.000258802261860665 0 9.13124417751099e-05 0 -1 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00166011586827782 0 5.08618097827969e-05 0 0.000550274603301129 0 -1.24577387791141e-05 0 2.80912803114481e-05 0 0.000787997869564677 0 0 -1 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00131496023553193 0 -2.85427725918689e-05 0 -6.40917512673142e-05 0 -6.48347565119609e-05 0 -9.88411732133345e-05 0 0 0 -1 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00116855302122233 0 -1.40653733430151e-05 0 1.38084332511408e-05 0 0.00052463330628182 0 0 0 0 -1 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00177126319840008 0 0.000136498835240328 0 -0.000139615027512656 0 0 0 0 0 -1 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00218208935736517 0 -0.000117739846758266 0 0 0 0 0 0 -1 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00170708755797057 0 0 0 0 0 0 0 -1;-1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0]);
sed_b = [-5e-08;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0];
sed_c = [0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;1;0;0;0;0;0;0;0];
sed_K = struct(sprintf('f'), {0}, sprintf('l'), {17}, sprintf('q'), {[2 2 2 2 2 2 2 8]}, sprintf('r'), {0});
pars = struct(sprintf('fid'), {0}, sprintf('maxiter'), {100});

% run sedumi 200 times, and hash the output:
for i=1:200
  [x,y,info]  = sedumi(sed_A,sed_b,sed_c,sed_K,pars);
  myhash=md5sum(x,true);
  disp(myhash);
  if strcmp(myhash,'93b885adfe0da089cdf634904fd59f71')==false
    error('Expected myhash to be 93b885adfe0da089cdf634904fd59f71');
  endif
end

[robertgj]

Sorry, I didn't read the documentation of md5sum carefully enough. Try again. This works for me:

more off
format long e
addpath('SeDuMi_1_3');
sed_A = sparse([1e-06 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.000374569348444187 0 -0.000143510377232224 0 -0.000107308290246781 0 -0.00023475302385101 0 -0.000222588804655496 0 -0.00039621568504792 0 -0.000175372571085547 0 0 0 0 0 0 0 0;0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00182361766281337 0 0.000300035708922513 0 5.25148072155411e-05 0 0.000172937546376846 0 0.000368772505991557 0 0.000258802261860665 0 9.13124417751099e-05 0 -1 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00166011586827782 0 5.08618097827969e-05 0 0.000550274603301129 0 -1.24577387791141e-05 0 2.80912803114481e-05 0 0.000787997869564677 0 0 -1 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00131496023553193 0 -2.85427725918689e-05 0 -6.40917512673142e-05 0 -6.48347565119609e-05 0 -9.88411732133345e-05 0 0 0 -1 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00116855302122233 0 -1.40653733430151e-05 0 1.38084332511408e-05 0 0.00052463330628182 0 0 0 0 -1 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00177126319840008 0 0.000136498835240328 0 -0.000139615027512656 0 0 0 0 0 -1 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00218208935736517 0 -0.000117739846758266 0 0 0 0 0 0 -1 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00170708755797057 0 0 0 0 0 0 0 -1;-1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0]);
sed_b = [-5e-08;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0];
sed_c = [0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;1;0;0;0;0;0;0;0];
sed_K = struct(sprintf('f'), {0}, sprintf('l'), {17}, sprintf('q'), {[2 2 2 2 2 2 2 8]}, sprintf('r'), {0});
pars = struct(sprintf('fid'), {0}, sprintf('maxiter'), {100});

% run sedumi 200 times, and hash the output:
for i=1:200
  [x,y,info]  = sedumi(sed_A,sed_b,sed_c,sed_K,pars);
  xstr=sprintf('%16.10g ',x)
  myhash=md5sum(xstr,true);
  disp(myhash);
  if strcmp(myhash,'d50f2fb51d0c438fe218383a145f2534')==false
    error('Expected myhash to be d50f2fb51d0c438fe218383a145f2534');
  endif
end

[siko1056]

Finally closed by #58 thanks to @robertgj 😊 👍

This program runs in Octave 5.2.0 and Matlab R2020b (with the old and new mexa64 binaries 😕 maybe only an Octave issue).

install_sedumi
format long e
sed_A = sparse([1e-06 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.000374569348444187 0 -0.000143510377232224 0 -0.000107308290246781 0 -0.00023475302385101 0 -0.000222588804655496 0 -0.00039621568504792 0 -0.000175372571085547 0 0 0 0 0 0 0 0;0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0;-2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00182361766281337 0 0.000300035708922513 0 5.25148072155411e-05 0 0.000172937546376846 0 0.000368772505991557 0 0.000258802261860665 0 9.13124417751099e-05 0 -1 0 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00166011586827782 0 5.08618097827969e-05 0 0.000550274603301129 0 -1.24577387791141e-05 0 2.80912803114481e-05 0 0.000787997869564677 0 0 -1 0 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00131496023553193 0 -2.85427725918689e-05 0 -6.40917512673142e-05 0 -6.48347565119609e-05 0 -9.88411732133345e-05 0 0 0 -1 0 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00116855302122233 0 -1.40653733430151e-05 0 1.38084332511408e-05 0 0.00052463330628182 0 0 0 0 -1 0 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00177126319840008 0 0.000136498835240328 0 -0.000139615027512656 0 0 0 0 0 -1 0 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00218208935736517 0 -0.000117739846758266 0 0 0 0 0 0 -1 0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00170708755797057 0 0 0 0 0 0 0 -1;-1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0]);
sed_b = [-5e-08;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0];
sed_c = [0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;0;1;0;0;0;0;0;0;0];
sed_K = struct(sprintf('f'), {0}, sprintf('l'), {17}, sprintf('q'), {[2 2 2 2 2 2 2 8]}, sprintf('r'), {0});
pars = struct(sprintf('fid'), {0}, sprintf('maxiter'), {100});

if (exist ("OCTAVE_VERSION", "builtin") == 5)
  expected = '287ab8afc72fef3240296fe3d62c81be'; % Octave 5.2.0
else
  expected = '71ed00585640dbc3f14990e38a2db1aa'; % Matlab R2020b
end

% run sedumi 200 times, and hash the output:
for i=1:200
  [x,y,info]  = sedumi(sed_A,sed_b,sed_c,sed_K,pars);
  xstr=sprintf('%16.10g ',x);
  if (exist ("OCTAVE_VERSION", "builtin") == 5)
    myhash = hash("md5", xstr);
  else
    mhash = java.security.MessageDigest.getInstance('MD5');
    mhash.update(typecast(x,'uint8'));
    myhash = lower(sprintf('%02X',typecast(mhash.digest,'uint8')));
  end
  if (~strcmp (myhash, expected))
    error('Expected myhash to be %s, but got %s', expected, myhash);
  end
end

disp ('done')