View Single Post
Old 2021-01-18, 20:49   #73
ewmayer's Avatar
Sep 2002
Rep├║blica de California

2·5,813 Posts

@tdulcet - How about I add support in v19.1 for the -radset flag to take either an index into the big table, or an actual set of comma-separated FFT radices? Shouldn't be difficult - if the expected -radset[whitespace]numeric arg pair is immediately followed by a comma, the code assumes it's a set of radices, reads those in, checks that said set is supported and if so runs with it.

I expect - fingers crossed, still plenty of work to do - to be able to release v19.1 around EOM, so you'd have to wait a little while, but it sounds like this is the way to go.

Edit: Why make people wait - here is a modified version of Mlucas.c which supports the above-described -radset argument. You should be able to drop into your current v19 source archive, but suggest you save the old Mlucas.c under a different name - maybe add a '.bak' - so you can diff the 2 versions to see the changes, the first and most obvious of which is the version number, now bumped up to 19.1.

Note user-supplied radix set is considered "advanced usage" in the sense that I assume users of it know what they are doing, though I have included a set of sanity checks on inputs. Most important is to understand the difference between the FFT length conventions between the -fftlen and -radset args: -fftlen supplies a real-vector FFT length in Kdoubles; -radset [comma-separated list of radices] specifies a corresponding set of complex-FFT radices. If the user has supplied a real-FFT length (in Kdoubles) via -fftlen, the product of the complex-FFT radices (call it 'rad_prod') must correspond to half that value, accounting for the Kdoubles scaling of the former. In C-code terms, we require that (rad_prod>>9) == fftlen .

Note that event though this is strictly-speaking redundant, the -fftlen arg is required even if the user supplies an actual radix set; this is for purposes of sanity checking the latter, because the above-described differing conventions make it easy to get confused. Using any of the radix sets listed in the mlucas.cfg file along with the corresponding FFT length is of course guaranteed to be OK.

Examples: After building the attached Mlucas.c file and relinking, try running the resulting binary with the following sets of command-line arguments to see what happens:

-iters 100 -fftlen 1664 -radset 0
-iters 100 -fftlen 1664 -radset 208,16,16,16
-iters 100 -fftlen 1668 -radset 208,16,16,16
-iters 100 -fftlen 1664 -radset 207,16,16,16
-iters 100 -fftlen 1664 -radset 208,8,8,8,8

Last fiddled with by ewmayer on 2021-02-13 at 20:06 Reason: Deleted attachment; code now part of v19.1
ewmayer is offline   Reply With Quote