Wrong behaviour in BackDifEq of VMatrix

[Víctor de Buen Remiro]

Function BackDifEq of VMatrix doesn't works fine.
Running this random check for Matrix and VMatrix BackDifEq instances returns different results being the second ones missing sometimes or very large numbers in other cases.

Real m = 300;
Real  s1 = 5;
Real  s2 = 26;
Real  p1 = IntRand(1,2);
Real  q1 = IntRand(0,2);
Real  p2 = IntRand(1,1);
Real  q2 = IntRand(0,1);
Polyn ar1 = RandStationary(p1,s1);
Polyn ar2 = RandStationary(p2,s2);
Polyn ma1 = RandStationary(q1,s1);
Polyn ma2 = RandStationary(q2,s2);
Polyn ar = ar1*ar2;
Polyn ma = ma1*ma2;

Real p = Degree(ar);
Real q = Degree(ma);
Real n = Max(p,q);


NameBlock CF1 = ARMAProcess::FastCholeskiCovFactor(ar1, ma1, m);
NameBlock CF2 = ARMAProcess::FastCholeskiCovFactor(ar2, ma2, m);
NameBlock CF  = ARMAProcess::FastCholeskiCovFactor(ar,  ma,  m);

VMatrix A = Gaussian(m,1,0,1);

VMatrix Li_ar.f = CholeskiFactor(CF::_.Li_ar,"XXt",True);

VMatrix L_ma.s  = Convert(CF::_.L_ma,"Cholmod.R.Sparse");

VMatrix X = CholeskiSolve(Li_ar.f,L_ma.s*A,"PtL");

VMatrix CF.X = CF::filter(X);

VMatrix cmp.CF   = A | CF.X;

VMatrix CF12.X = CF2::filter(CF1::filter(X));

VMatrix A.n   = Sub(CF12.X, m-n+1, 1, n,  1);
VMatrix A.q   = Sub(CF12.X, m-n+1, 1, q,  1);
VMatrix X.p   = Sub(X,      m-n+1, 1, p,  1);
VMatrix X.m_n = Sub(X,          1, 1, m-n,1);
VMatrix A.m_n = BackDifEq(ar/ma,X.m_n,X.p,A.q);

VMatrix A.m_n_ = Mat2VMat(BackDifEq(ar/ma,VMat2Mat(X.m_n),VMat2Mat(X.p),VMat2Mat(A.q)));

Real error = VMatMax(Abs(A.m_n-A.m_n_));
Real quality = Max(0,1-error);