Hi,

as Michael mentioned in another thread, i should help my self to get fpsieve rocking on an above average number of cores. Currently fpsieve got a hard limit of 32 worker threads (33 in total). Is Geoff still around to elaborate on why he came up with this number? My first guess, there could be a performance decrease because fpsieve waits for all threads to finish on checkpoints, so if there is a thread which cant finish in a timely manner, it locks more or less all other threads.

Findings:
I found geoffs code is still in the state of 2009 and version 0.2.4
On my endeavour on finding other sources, i stumbled upon Roger's repo at github and his integration of compositorial sieving code.

Questions:
Is there any more current code around?
Is there a plain C implementation of geoffs asm sieving code available?
What did Roger change on his code base apart of new conditions to include his sieving code?
____________

First of all, Randy (cncr04s) has already done some work there. You might want to talk to him. Secondly, there's not much point. I'm shutting down sieving on both factorial and primorial shortly. There is no point in sieving for the maximum size candidate when we're never going to get close to it. Right now a user can test 3.27 factorial candidates in the time it takes to sieve one out on the CPU. It's much closer on primorial where one can test 1.64 candidates in the time it takes to sieve one out. In both cases, I'm not talking about factors found, I'm talking about new sieve removals. The GPU app for factorial mostly generates factors for candidates no longer in the sieve.

Factorial and primorial are never going to be doublechecked as far as I know. They can't be run in LLR, only in PFGW and there aren't any plans for a PFGW wrapper. So right now we're oversieved on both of them. Sieving can be reopened at any future point when needed, but right now it's not an efficient use of computer time.

Hi HAmsty,

Is there any more current code around?

Is there a plain C implementation of geoffs asm sieving code available?

What did Roger change on his code base apart of new conditions to include his sieving code?