We are *comparing* implementations on the same machine, with the same array, with the same simple access patterns and averaging
over 1000 runs. I even toggled which test runs first or second. IF the L1 cache misses are an issue, that would slow *both* down, so multi_array must be EVEN WORSE than the test times showed right? I wans't speaking of L1 cache miss on this
the running times particuliar test sorry for the inconvenience
I think the conclusion is that the pointer based approach has a significant (up to20 times on my system) performance improvement over
base+index approach (at least as implemented by multi_array). Yes I came to the same conclusion. I was just pointing the fatc that the base+pointer is not the only way to do something (see my attachment in my previous post) Again sorry for our misandurstanding