Do you use the breadth-first mode of GMP-ECM?

[akruppa]

Do you use the breadth-first mode of GMP-ECM?

[ATH]

You forgot the poll option "breadth-first mode ???"

I had to read up on README to see what it was.

[akruppa]

If you consider yourself a regular GMP-ECM user and you've never heard of breadth-first mode, please vote! I'm toying with the thought of removing it, mostly because I don't understand what our main() function does, and wonder how many users would miss it.

[R.D. Silverman]

Originally Posted by akruppa

If you consider yourself a regular GMP-ECM user and you've never heard of breadth-first mode, please vote! I'm toying with the thought of removing it, mostly because I don't understand what our main() function does, and wonder how many users would miss it.

I've never heard of it, but I can guess what it is. When doing
convolutions for products of (p_i - p_j) as p_i and p_j vary over
(different) arithmetic sequences, one can fix p_i and let p_j vary or
vice-versa.

I am guessing that one is referred to as depth-first, and the other
as breadth-first.

Is my guess close?

[Karl M Johnson]

From the tarball's README:

NOTE that in -b "breadth-first" mode, GMP-ECM reads all candidate numbers in 
the input stream and keeps them in memory, so if there are many large numbers
to be tested, the memory requirement will increase noticeably.

[akruppa]

Not remotely. Breadth-first mode reads all input numbers into memory, then does one curve on each number in turn, up to the desired number of curves. May have a use if you want any one factor from a set of input numbers, but I don't seen much use beyond that. Most of its functionality could be replicated by a shell script that executes GMP-ECM repeatedly, doing one curve per input number per execution.

[R.D. Silverman]

Originally Posted by akruppa

Not remotely. Breadth-first mode reads all input numbers into memory, then does one curve on each number in turn, up to the desired number of curves. May have a use if you want any one factor from a set of input numbers, but I don't seen much use beyond that. Most of its functionality could be replicated by a shell script that executes GMP-ECM repeatedly, doing one curve per input number per execution.

Ah. Thanks for the clarification.

I agree that this feature does not seem all that useful.

[axn]

I've used it for brilliant number search.

[bsquared]

I've never used it, but why remove functionality if it works?

[akruppa]

The main function is a mess of loops and gotos, I can't get it through my head. Every time I tried I was overcome by the thought "why can't I just bin this?"

[chris2be8]

I've used it to avoid having both tasks on a dual core machine running stage 1 and stage 2 at the same time. By running 1 task depth first and 1 breadth first I can get them to process the numbers in a different order.

But I'm not sure if it gives any real speedup, even when the two tasks are on 1 CPU with hyperthreading.

Chris K