20040911, 15:25  #1 
Aug 2004
way out west
1A_{16} Posts 
fftlen change...
Could someone explain the significance of the fftlen portion of LLR? My computer had always used the Mersenne fftlen 65536. When I downloaded the 1300 range yesterday, it started using fftlen 81920 for no apparent reason. If this is appropriate, OK, but it is slower.
It still seems to be chugging along merrily, but the time/iteration is up 50%. 
20040911, 16:48  #2 
Sep 2002
Database er0rr
2×7×281 Posts 
You can see the various 321 FFT lengths here
As our numbers get bigger and bigger we expect them to take longer and longer. A number that is twice as long as another will take over four times as long to test. Thomas Ritschel and Jean Penne liased to form the FFT breaks. With the older LLR it would get say 90% through a number realise that the FFT size was too small and switch to a higher one  of course the number had to be restarted with the new size. This was wasting a lot of time. HTH. I hope someone else might give a better explaination... 
20040913, 12:06  #3  
Feb 2003
5×383 Posts 
Quote:
It seems that the new LLR/P4 uses (almost) the same algorithm for the FFT breaks as the new PRP3, for which George Woltman has given an explanation here. If we use Georges algorithm, we get the following: log2(3)/2 = 0.8, which leaves 19.2 bits per FFT word. Therefore the 64K FFT can handle n up to 64K*19.2 = 1258291. In practice the switch 64K > 80K occured at n=1273063, so it seems that the algorithm used in LLR/P4 is slightly different. 

20040913, 13:45  #4 
P90 years forever!
Aug 2002
Yeehaw, FL
16775_{8} Posts 
Each FFT length supports a different number of bits per word. Smaller FFTs support more bits per word than larger FFTs. A length 32 FFT supports about 23 bits per word, while a length 4M FFT supports about 19 bits per word.
In the current sources, if you look in mult.asm the jmptable lists the maximum Mersenne exponent each FFT length can handle. The bits per word is simply this exponent divided by the FFT length. 
20040915, 13:13  #5 
3222_{8} Posts 
?
Why is the table offset from the mersenne exponents?
Is this editable? 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Climate Change  David John Hill Jr  Science & Technology  1589  20210331 13:28 
Name Change?  Fred  Lounge  8  20160131 17:42 
Requests for change...  WraithX  FactorDB  32  20141218 03:40 
Change the world!  Xyzzy  Lounge  5  20090831 12:41 
How can I change worktype?  Andriy  Information & Answers  1  20090620 12:39 