Testing ICC Auto-vectoriser with part struct
I have created a quick test case that compares how the Intel compiler vectorises loops containing updates to particles that have different struct formats.
Either:
struct array {
float x[3];
}
or
struct flat {
float x, y, z;
}
The results show that the loops containing the flat structures are faster than the array structures. See the attached file, which runs each version and prints the timings. @matthieu