strange behaviour of mpz_jacobi

[Paul Zimmermann]

       Torbjörn,

thank you for confirming the strange behaviour of mpz_jacobi on Core 2.
I also noticed a similar "spike" around 1022 limbs:

tarte% ./speed -s 1018-1023 mpz_jacobi
overhead 0.000000002 secs, precision 10000 units of 3.53e-10 secs, CPU freq 2833.00 MHz
           mpz_jacobi
1018      0.028995000
1019      0.032995000
1020      0.035994000
1021      0.043993000
1022      0.066989000
1023      0.028995000

Do you have any explanation for this? A cache or TLB miss?
(I don't think this has something to do with GMP's tuning.)

Paul