Prime95, v25.5, ECM, error on non base2
 

Version 25.5

[SIZE=2]ContinueECM=1[/SIZE]

[SIZE=2][SIZE=2]ECM2=1,3,4096,1,2000,0,9[/SIZE]



[SIZE=2][INDENT][Main thread Dec 27 19:03] Mersenne number primality test program version 25.5
[/SIZE][SIZE=2][Main thread Dec 27 19:03] Starting worker thread.[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] Work thread starting[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] Using generic reduction FFT length 640[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] ECM on 3^4096+1: curve #1 with s=7359417470999745, B1=2000, B2=200000[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] 3^4096+1 has a factor: 2[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] ECM on 3^4096+1: curve #1 with s=5740293094245417, B1=2000, B2=200000[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] Stage 1 complete. 151721 transforms, 1 modular inverses. Time: 0.769 sec.[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] Using 3MB of memory in stage 2.[/SIZE]
[SIZE=2][Work thread Dec 27 19:03] Stage 2 init complete. 10959 transforms, 1 modular inverses. Time: 0.069 sec.[/SIZE]
[/INDENT][SIZE=2]After finishing stage-1 it begins stage-2. After 16 minutes, it is still working on stage-2. Memory usage has very slowly crept up and is at 9,724K after 12 minutes.[/SIZE]

Note that the older v24.14 that I've been using has had major problems with non base2 stage-2 ECM work. Usually it gets into stage-2 and then terminates. Termination is either with invalid memory access, or with a SUMOUT error.



[/SIZE]

I just tried v25.6 build-3 with the same results. Starts stage-2, but never finishes.

I'm looking into it

Fixed in 25.6 build 5. Download the "c" version of 25.6 to test it.

[quote=Prime95;121771]Fixed in 25.6 build 5. Download the "c" version of 25.6 to test it.[/quote]

Brief examination shows the original problem as being fixed. Thank you for the quick turn-around.

I'll run through some additional ECM testing for things that had issues in the past. But I'm hoping that your fix will encompass them all.

Will you be releasing source again? I will need to retool it again so as to add a fork to GmpEcm stage-2 work.