[philus]

Hi all,

I have transformed my serial UDF with a parallel version looking the UDF manual (the pressure average example). The UDF is well compiled but the solver don't even interate one time.

The goal of this UDF is to compute an h value on a boundary from a mean Tw (wall temperature). I paste a simplified version:

DEFINE_PROFILE(name,t,i){
    real Tw, T_film;
    real h;
    real areatot=0.0;
    
    #if !RP_HOST
        face_t f;
        real A2[2];

        begin_f_loop(f,t){
            if(PRINCIPAL_FACE_P(f,t)){
                F_AREA(A2,f,t);
                areatot+=NV_MAG(A2);
                Tw+=F_T(f,t)*NV_MAG(A2);
            }
        } end_f_loop(f,t)
        
        # if RP_NODE
            areatot=PRF_GRSUM1(areatot);
            Tw=PRF_GRSUM1(Tw);
        #endif
    #endif
    
    node_to_host_real_2(areatot,Tw);
    
    #if !RP_NODE
        Tw/=areatot;
        h=Tw*2;
    #endif
    
    host_to_node_real_1(h);
    
    #if !RP_HOST
        begin_f_loop(f,t){
            if(PRINCIPAL_FACE_P(f,t)){
                F_PROFILE(f,t,i)= h;
            }
        } end_f_loop(f,t)
    #endif 
}

Can someone help me showing the errors?

Thanks in advance,
philus

[pakk]

Originally Posted by philus

The UDF is well compiled but the solver don't even interate one time.

But what does it do instead? Give an error? Or does it keep running without ever giving results, as if it were in an endless loop? Or something else?

[philus]

The window "Running calculation" appears, but nothing happens and Fluent do not responds anymore.

[pakk]

Strange. I don't know what can cause that.

You might try to simplify your UDF further, and see if that works.

For example: only calculate the average temperature, but not assign it to anything. Or: don't calculate the average temperature, and just assign the value 1 to the profile. That might give you information about which part fails.

(And minor detail: you did not initialize Tw, but that should not give the problem you see now.)

[Kokemoor]

Two quick clarifying questions: this worked in serial before you added compiler directives, global reductions, and node/host transfers? Also, this is 2D?

My first try would be to remove the '#if !RP_HOST' around the final loop with 'F_PROFILE'. It seems like that shouldn't cause any problems, but if it isn't necessary, better to leave it out.

I'm not certain, but I think global reductions don't need to be protected by compiler directives. You could try removing the '#if RP_NODE' around your 'PRF_GRSUM1's.

Definitely not the cause of this issue, but I'd recommend where you define your area vector 'A2' either defining it as a vector

real NV_VEC(A2);

or with the 'ND_ND' macro

real A2[ND_ND];

Either of those will give you a vector of length 2 in 2D and 3 in 3D. Even if you know you'll only be doing 2D, it's good practice.

[philus]

Originally Posted by Kokemoor

Two quick clarifying questions: this worked in serial before you added compiler directives, global reductions, and node/host transfers? Also, this is 2D?

Exactly, I modified the serial version by adding the "#if"s. The simulation is 2D.

Originally Posted by Kokemoor

My first try would be to remove the '#if !RP_HOST' around the final loop with 'F_PROFILE'. It seems like that shouldn't cause any problems, but if it isn't necessary, better to leave it out.

This gives me a problem (cannot compile) because the host doesn't know "f".

Thanks for the hints!

[`e`]

There are a number of lines in this code, which is a "simplified version" of your actual UDF. You need to determine where the error (or otherwise) is occurring. Start your debugging process by inserting commands to print information to the screen at key steps with the Message() function.

For example:

# if RP_NODE
    Message("I am node number %d and my total area = %e\n",myid,areatot);
    areatot=PRF_GRSUM1(areatot);
    Message("I am node number %d and now my shared total area = %e\n",myid,areatot);
    Tw=PRF_GRSUM1(Tw);
#endif