[bloodflow]

Hi all,

I have two UDF's that I am using to define the viscosity of a secondary phase, one UDF works perfectly and doesnt slow down the solution, the other causes the iterations to be incredibly slow (for larger meshes). I think this is due to my if statement, but is there any way to improve the efficiency? currently the slow UDf is unusable due to how slow it is.

PHP Code:

DEFINE_PROPERTY(rbc_visco,c,thread)
{
Domain *mix_domain = Get_Domain(1);
Thread *mix_thread;
Thread **pt;
real phi, rbcvisco, shear, muplasma, k0, kinf, gammac, kval;
mp_thread_loop_c(mix_thread, mix_domain, pt)
{
phi=C_VOF(c,pt[1]);
shear=C_STRAIN_RATE_MAG(c,pt[1]);
muplasma=0.00123;
k0=((0.06106410882322652)+(2/(0.047770935254055194+phi)));
kinf=exp(1.8009617109358729-(3.6660219116019688)*(phi)+(2.574119363866957)*(pow(phi,2))+(0.023458247768672405)*(pow(phi,3)));
gammac=exp(-7.013323148316977+(34.3877131444968)*(phi)-(39.80154094066538)*(pow(phi,2))+(13.991671192652591)*(pow(phi,3)));
kval=((k0)+(kinf)*(pow((shear)/(gammac),0.5)))/(1+(pow((shear)/(gammac),0.5)));
rbcvisco=((((muplasma)*(pow(1-((phi*kval)/2),-2))-((muplasma)*(1-phi)))/(phi)));
}
return rbcvisco;
} 


This is the UDF that works without a problem, and is fast regardless of mesh but this one:

PHP Code:

DEFINE_PROPERTY(rbc_visco,c,thread)
{
Domain *mix_domain = Get_Domain(1);
Thread *mix_thread;
Thread **pt;
real rbcvof, rbcvisco, shear, muplasma, rbcvof_crit, b, charlie, beta, lamb, flickv, ninf, dimshear, nst, nsum;
muplasma=0.00123;
rbcvof_crit=0.98;
b=8.780739943551884;
charlie=2.823542271766962;
beta=16.277748270259284;
lamb=1252.6440693060042;
flickv=0.14274532071176718;

mp_thread_loop_c(mix_thread, mix_domain, pt)
{

begin_c_loop(c,pt[1])
{

rbcvof=C_VOF(c,pt[1]);

if (rbcvof<0.2)
        {
        shear=C_STRAIN_RATE_MAG(c,pt[1]);
        ninf=(b)*(exp(-charlie*(rbcvof)));
        nsum=ninf;
        rbcvisco=((muplasma)*(rbcvof+(pow((1-(rbcvof/rbcvof_crit)),-nsum))-1))/(rbcvof);
        }
else
        {
        shear=C_STRAIN_RATE_MAG(c,pt[1]);
        ninf=(b)*(exp(-charlie*(rbcvof)));
        dimshear=1+pow((lamb*shear),2);
        nst=beta*(pow((dimshear),(-flickv)));
        nsum=(ninf+nst);
        rbcvisco=((muplasma)*(rbcvof+(pow((1-(rbcvof/rbcvof_crit)),-nsum))-1))/(rbcvof);
        }

}

end_c_loop(c,pt[1])
}

return rbcvisco;
} 


Works okay for small meshes, but for my detailed models it is incredibly slow. What can I do to improve this, is it the additional "beging_c_loop"? I am not 100% on which part of my second UDF is relevant.


Thanks you for any advice.

[AlexanderZ]

why do you use additional cell loop
remove

begin_c_loop(c,pt[1])
{

this is the reason, why it is slow, you have loop over all cells on each cell
c is already defined in the header of macro

best regards