Bucle anidado en GPU con OpenACC devuelve Fallo de segmentación

there are weeks that I lose my mind with OpenACC! For example I have to parallelize the code

float temp;
for( i_v = 0; i_v < ntv; i_v++ )
{
    temp = 0;

    for( i_el = 0; i_el < len_tv; i_el++ )
        temp += pow( tva[i_v*len_tv + i_el], (float)2.0 );

    tv_sq[i_v]=temp;
}    

to do that, I have tried several ways, the last after the reading of Nvidia Technology Group that explains very well the function of gang and vector. So I try:

float temp;
#pragma acc data copyin(tva[:nfa]) copyout(tv_sq[:ntv]) create(temp)
{
    #pragma acc kernels loop independent 
        for( i_v = 0; i_v < ntv; i_v++ )
        {
            temp = 0;

            #pragma acc loop independent gang vector reduction(+:temp)
                for( i_el = 0; i_el < len_tv; i_el++ )
                    temp += pow( tva[i_v*len_tv + i_el], (float)2.0 );

            tv_sq[i_v]=temp;
        }
}

the compiler said

 710, Generating create(temp)
     Generating copyout(tv_sq[0:ntv])
     Generating copyin(tva[0:nfa])
712, Loop is parallelizable
     Accelerator kernel generated
    712, #pragma acc loop gang /* blockIdx.x */
    716, #pragma acc loop vector(128) /* threadIdx.x */
712, Generating present_or_copyout(tv_sq[0:ntv])
     Generating present_or_copyin(tva[0:nfa])
     Generating NVIDIA code
     Generating compute capability 3.5 binary
716, Loop is parallelizable

but when I try to execute:

Segmentation fault (core dumped)

the same with cuda-memcheck, it answers only segmentation fault. Previously I insert the pragmas in "parallel" ways and not "kernels", but I had problems of "Invalid global write of size 8 (...) Out of bounds".

Clearly there is something that I haven't understand, so if somebody can tell me the way to parallelize these code I'll try to understand my errors

Muchas muchas gracias