optimize/vectorize code CUDA Matlab - matlab

Question

my question is about vectorize my for loop in Matlab using CUDA.
This arrays are gpuArrays.
The main problem is that the array isn't completed at start and it's getting bigger with every k iteration. Also it is creating by previous probes from array. Is it possbile to make this loop parallel/vectorize or any other way to optimize the code for CUDA to make it faster?
for k=3:5000
q_=gWmacierz((1:N_/2+1),k-1);
p_d1_=[p0_;gWmacierz((N_/2+2:N_+1),k-1)];
p_d2_=[gWmacierz((N_/2+2:N_+1),k-1);pN_];
F_=arrayfun(#funkcja,q_,p_d1_,p_d2_);
C1_=diag(F_+4*Gc);
C_=[C1_ C2_;C3_ C4_];
tic;
if k>3000
Wwektor = ginvA_* ((gB_*gWmacierz(:,k-2)) + (C_*gWmacierz(:,k-1)) + (gD_*gu01_) + (gE_*gu00_) - gQ*gleak);
else
Wwektor = ginvA_* ((gB_*gWmacierz(:,k-2)) + (C_*gWmacierz(:,k-1)) + (gD_*gu01_) + (gE_*gu00_));
end
toc;
gWmacierz(:,k)= Wwektor;
end
Pin=(gWmacierz(N_/2+2,:)+gWmacierz(N_/2+3,:))/2;
Pout=(gWmacierz(N_,:)+gWmacierz(N_+1,:))/2;
for k=3:5000
q=gmacierz((1:N/2+1),k-1);
p_d1=[Pin(k);gmacierz((N/2+2:N+1),k-1)];
p_d2=[gmacierz((N/2+2:N+1),k-1);Pout(k)];
F=arrayfun(#funkcja,q,p_d1,p_d2);
C1=diag(F+4*Gc);
C=[C1 C2;C3 C4];
wektor = ginvA* ((gB*gmacierz(:,k-2)) + (C*gmacierz(:,k-1)) + (gD*[Pin(k-1);Pout(k-1)]) + (gE*[Pin(k);Pout(k)]));
gmacierz=[gmacierz wektor];
end
Qin=(gWmacierz(2,:));
Qout=(gWmacierz(N/2,:));
Qin_ref=(gmacierz(1,:));
Qout_ref=(gmacierz(N/2+1,:));
deltaQin=Qin-Qin_ref;
deltaQout=Qout-Qout_ref;