Im trying to implement Neeedleman-Wunch with affine penalty model.

According to Algorithm in Bioinformatic: Practical Introduction (Slides) I have to initialize four tables and back-trace them,

to do back-tracing, another table is initialized and filled up with the help of a maximum function, which returns the maximum value to be added to table V(i,j) plus the associated figure ( \, | or -) to be added to back-tracing table

this approach works with simple queries, but once the the strings become complicated, things go out of control. like the following example:

Target: TGCTAGTATAAACCTTATGGTATCTGCAGCAGAGGTTTCTTTAATCTCTCAATAGTAGATGCTTTGAAAC

Read: TTATCTATAATTTGGTATTGTAATGACAGTTTGTGTTTGGTTTTTTCTTCAGTAT

Alignment:

TGC-TAG-TATAAACCT-TATGGTATCTG-CAGCA-GAGGTTTCTTTAATCTCTCAATAGTAGATGCTTTGAAAC ---TTATCTATAATTTGGTATTGTAA-TGACAGTTTGTGTTTGGTTTTTTCT-TCAGTAT---------------

score = 46

while using EMBOSS-needle

TGCTA-GTATAAACCTTATGGTATCTGCA--GCAGAG-----GTTT-CTTTAATCTCTCAATAGTAGATGCTTTGAAAC --TTATCTATAA----TTTGGTAT-TGTAATG-ACAGTTTGTGTTTGGTTTTTTCT-TCAGTAT---------------

score = 52

My question, is this the right approach to implement NW with affine gap penalty model ?
here is my implementation in C++ Github.

int max(int diagonal,  int vertical, int horizontal, char *figure) {
int  max = 0 ;
if( diagonal > vertical && diagonal > horizontal )
{
    max = diagonal ;
    *figure = '\\' ;
}

else if (vertical > horizontal)
{
    max = vertical ;
    *figure = '|' ;


}

else     
{
    max = horizontal ;
    *figure = '-' ;
}
return  max ;
}