Option PRE_FULL_PSI is slower than PRE_PSI

[caliarim]

Dear maintainers,

if I run the simple_test.m, two dimensional example, logN=10:11 (just to fix the ideas) and compare the options PRE_PSI and PRE_FULL_PSI, I consistently get the elapsed times t1 in the first case smaller than in the second. I understood that PRE_FULL_PSI should be the faster method.

[tvolkmer]

I think the documentation about the flag PRE_FULL_PSI is a little bit out-dated. It used to be the faster flag in many cases. However, on modern computers with fast double precision arithmetic (and relatively slow memory access), PRE_FULL_PSI may be distinctly slower in various cases.

In general, I recommend using PRE_PSI (or possibly PRE_LIN_PSI if you use a small NFFT window length m).

[tvolkmer]

Just for comparison. My runtimes of simple_test.m single thread Intel(R) Core(TM) i5-2400 CPU @ 3.10GHz on openSUSE Linux

average runtime of 3 runs d=2 logN=10: PRE_PSI = 0.34 seconds, PRE_FULL_PSI = 0.30 seconds d=2 logN=11: PRE_PSI = 1.41 seconds, PRE_FULL_PSI = 1.50 seconds

So on my computer, the non-threaded PRE_FULL_PSI is a little bit faster for logN=10 but slower for logN=11.