LL tests: Integer or floating point?

E_tron · 2004-01-12, 06:15

Can the LL test be performed with an integer calculation? Would it be as good?

flava · 2004-01-12, 06:48

Yes it can, there are integer based FFT's out there, like the one used by GMP. I implemented a LL test using it and it is slow, like 10X slower than Prime95 (on a Athlon). The only advantage I can think off is it might be less error prone than float FFT's.

ColdFury · 2004-01-12, 09:05

I could swear that one of the non-windows mersenne programs uses an integer-based FFT? (glucas?)

Anyways, there's no inherent reason why an integer FFT should be slower than a floating point one, provided it's been optimized enough. The decision to use integer versus floating point is really dependant on the strength of the platform you're targetting.

aaronl · 2004-01-12, 09:52

Well integer is the only choice if you don't have a floating point unit, but other than that, you're better off using floating point. Not to mention that floating point variables tend to have a higher precision on most inexpensive computers.

ARM Prime uses integer math because the ARM CPU doesn't have a FPU.

E_tron · 2004-01-13, 19:44

Most games use floating point, so that would explain why the P4 and Athlon do so well with floating point math.

Now the VIA processors... they are geared for the integer.

2004-01-12, 06:15	#1
E_tron Sep 2002 Austin, TX 231₁₆ Posts	LL tests: Integer or floating point? Can the LL test be performed with an integer calculation? Would it be as good?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
floating point operations	ATH	Lounge	3	2006-01-01 20:29
Floating point options for Windows XP 64	dsouza123	Hardware	2	2005-03-12 17:45
Floating point precision	lunna	Hardware	11	2003-12-29 16:46
Integer and Floating point Trial Factoring in parallel ?	dsouza123	Software	3	2003-09-21 12:46
floating point exception in Version 23.4.2	mda2376	Software	2	2003-06-12 04:45

2004-01-12, 06:48	#2
flava Feb 2003 2×59 Posts	Yes it can, there are integer based FFT's out there, like the one used by GMP. I implemented a LL test using it and it is slow, like 10X slower than Prime95 (on a Athlon). The only advantage I can think off is it might be less error prone than float FFT's.

2004-01-12, 09:05	#3
ColdFury Aug 2002 2⁶·5 Posts	I could swear that one of the non-windows mersenne programs uses an integer-based FFT? (glucas?) Anyways, there's no inherent reason why an integer FFT should be slower than a floating point one, provided it's been optimized enough. The decision to use integer versus floating point is really dependant on the strength of the platform you're targetting.

2004-01-12, 09:52	#4
aaronl Aug 2003 60₈ Posts	Well integer is the only choice if you don't have a floating point unit, but other than that, you're better off using floating point. Not to mention that floating point variables tend to have a higher precision on most inexpensive computers. ARM Prime uses integer math because the ARM CPU doesn't have a FPU.

2004-01-13, 19:44	#5
E_tron Sep 2002 Austin, TX 561₁₀ Posts	Most games use floating point, so that would explain why the P4 and Athlon do so well with floating point math. Now the VIA processors... they are geared for the integer.