[with SPKG] Wrap Bernstein 's primegen

[robertwb]

See http://cr.yp.to/primegen.html

Some code at http://thread.gmane.org/gmane.comp.python.cython.devel/2579/focus=2581

CC: @sagetrac-mvngu @sagetrac-kevin-stueve

Component: number theory

Issue created by migration from https://trac.sagemath.org/ticket/3925

[wjp]

comment:1

I added some checks and doctests, and made Primes() use this.

This is still a work-in-progress. TODO:

• Turn primegen-0.97 into a .spkg. It needs an added -fPIC option since it will be linked into a cython extension module.

• Make sage.rings.arith.primes(start,stop) also use this for n not too small and not too large.

• Determine if there are predefined portable uint32/uint64 types available for use in cython.

[wjp]

Attachment: primegen-0.97.spkg.gz

[wjp]

comment:4

Attachment: trac-3925-spkg_deps.patch.gz

I've added an attempt at an .spkg for primegen-0.97 as an attachment, and also took the liberty of patching (untested...) spkg/install and spkg/standard/deps to build it automatically when installing sage. (In the 'spkg_deps' patch.)

I'm not entirely confident the build system of the library will work everywhere, since it is rather non-standard, but hopefully it is portable enough.

The library is tiny, with the .spkg only 32KB, and the compiled (Linux x86_64) library only 17KB.

Timing:

def f():    
    P = Primes()
    for p in P:
        if p > 10^8:
            break
time f()

goes from 84.17s (without this spkg+patch) to 20.77s (with spkg+patch) on a 2GHz Opteron.

[JohnCremona]

comment:5

Attachment: trac-3925-primegen.patch.gz

Hello Willem. I successfully installed the spkg and the second patch but I don't know how to install the first patch as it changes files not in the usual code tree. If you tell me how, I would like to test this. -- John

[wjp]

comment:6

Hi John. The first patch isn't necessary to use the spkg. It's only for making a fresh 'make' of sage automatically build the spkg. I'm not too sure sure if that patch is right, actually; that part should probably be left to a release manager.

[JohnCremona]

comment:7

Replying to @wjp:

Hi John. The first patch isn't necessary to use the spkg. It's only for making a fresh 'make' of sage automatically build the spkg. I'm not too sure sure if that patch is right, actually; that part should probably be left to a release manager.

OK, I'll have another go sometime this weekend. I'm glad about the first patch, since I'm not really competent to say if it's right (though it looks ok).

[JohnCremona]

comment:8

To adopt this spkg as part of Sage
proper would need a vote on sage-devel. I suggest that wjp helps that process
by collecting some data (before and after). For example:

sage: time P = prime_range(10^8)
CPU times: user 1.83 s, sys: 0.50 s, total: 2.32 s
Wall time: 2.33 s
sage: len(P)
5761455

but this does not use the new PrimeGen class. I tried this (with the
new spkg + patch):

sage: pg=Primes().pg
sage: pg.reset()
sage: N=pg.count(10^8)
sage: pg.reset()
sage: time P=[pg.next() for _ in range(N)]
CPU times: user 4.98 s, sys: 0.03 s, total: 5.01 s
Wall time: 5.02 s

which is slower but it's using a more stupid method to collect the
primes that in prime_range().

[aghitza]

comment:9

Changing this to "needs work" given John's latest comments. Note that "work" here would mean making the case to sage-devel that the spkg should be adopted, and asking for a vote.