ECM users - version 30.9-30.12 (see post#235) - Page 19

Prime95 · 2023-03-08, 16:02

Quote:

Originally Posted by kruoli

Resuming P-1 stage 1 files results in an access violation.

Try re-downloading. I updated 30.11b1 without creating b2 (I know I'm supposed to create a b2 but I'm feeling a little lazy this morning).

kruoli · 2023-03-08, 16:11

Magnificent, thank you, it now works!

cmpute · 2023-03-09, 02:16

As I mentioned above, sometimes the stage2 of P+1 will crash. The previous post is an example that always lead to core dump on my computer, here's another example, where a crash happens every several hours (not every time, about every 3-4 exponent with 2 workers).

The assignments I am doing are P+1 on exponents around 10M, Start=2 / 7, B1=2200000, B2=100000000. The stack trace this time is

Code:

Thread 36 "mprime" received signal SIGSEGV, Segmentation fault.
#0  0x0000000000480cb4 in fill_pairmap(int, void**, int, int, int, int, int, int, short*, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, unsigned char**, unsigned long*) ()
#1  0x0000000000461125 in pp1_stage2_impl(pp1handle*) ()
#2  0x00000000004766a2 in pplus1 ()
#3  0x0000000000441e10 in primeContinue ()
#4  0x0000000000442039 in LauncherDispatch ()
#5  0x00000000004b710a in ThreadStarter ()
#6  0x00007ffff781d6db in start_thread (arg=0x7ffff376c700) at pthread_create.c:463
#7  0x00007ffff689461f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

This time the crash happened in `fill_pairmap` rather than `pp1_stage2_impl`. Note that I'm still using 30.10b4, will try 30.11 soon.

Thanks for your work in advance!

Prime95 · 2023-03-09, 02:52

Quote:

Originally Posted by cmpute

Note that I'm still using 30.10b4, will try 30.11 soon.

30.11 should not behave any differently. To debug (11 days from now) I'll need a save file near the end of stage 1, the worktodo.txt extry, the CPU characteristics (AVX512?, FMA?, AVX?), and the memory settings. Thanks!

Mark Rose · 2023-03-09, 03:06

I hope save-everything-before-high-memory-usage-starts is on the list of features for the new version

Code:

[Work thread Mar 8 19:53] M83360873 stage 1 is 99.75% complete. Time: 63.229 sec.
[Work thread Mar 8 19:53] M83360873 stage 1 complete. 1062604 transforms. Total time: 4729.586 sec.
[Work thread Mar 8 19:54] Inversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 30.352 sec.
[Work thread Mar 8 19:54] Available memory is 23000MB.
[Work thread Mar 8 19:54] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Mar 8 19:54] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Mar 8 19:54] Switching to FMA3 FFT length 5M, Pass1=1280, Pass2=4K, clm=1, 4 threads
[Work thread Mar 8 19:54] Estimated stage 2 vs. stage 1 runtime ratio: 1.011
[Work thread Mar 8 19:54] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Mar 8 19:54] Using 22998MB of memory.  D: 1050, 120x460 polynomial multiplication.
[Work thread Mar 8 19:54] Setting affinity to run polymult helper thread on CPU core #2
[Work thread Mar 8 19:54] Setting affinity to run polymult helper thread on CPU core #3
[Work thread Mar 8 19:54] Setting affinity to run polymult helper thread on CPU core #4
[Work thread Mar 8 19:54] Stage 2 init complete. 2961 transforms. Time: 46.138 sec.
Killed
mark@foo:~/mprime$ ./mprime -d
[Main thread Mar 8 19:59] Mersenne number primality test program version 30.8
[Main thread Mar 8 19:59] Optimizing for CPU architecture: Core i3/i5/i7, L2 cache size: 8x256 KB, L3 cache size: 16 MB
[Main thread Mar 8 19:59] Starting worker.
[Work thread Mar 8 19:59] Worker starting
[Work thread Mar 8 19:59] Setting affinity to run worker on CPU core #1
[Work thread Mar 8 19:59] Optimal P-1 factoring of M83360873 using up to 23000MB of memory.
[Work thread Mar 8 19:59] Assuming no factors below 2^77 and 1 primality test saved if a factor is found.
[Work thread Mar 8 19:59] Optimal bounds are B1=368000, B2=58626000
[Work thread Mar 8 19:59] Chance of finding a factor is an estimated 3.67%
[Work thread Mar 8 19:59] 
[Work thread Mar 8 19:59] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Mar 8 19:59] Using FMA3 FFT length 4480K, Pass1=448, Pass2=10K, clm=2, 4 threads
[Work thread Mar 8 19:59] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Mar 8 19:59] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Mar 8 19:59] M83360873 stage 1 is 73.28% complete.
[Work thread Mar 8 20:00] M83360873 stage 1 is 75.16% complete. Time: 70.219 sec.

I didn't have the available memory for P2, the system killed mprime, and I'm repeating work.

Most of the time I'm fine, but sometimes I'll have 20 or 30 web apps open by accident haha

Prime95 · 2023-03-09, 03:33

Quote:

Originally Posted by Mark Rose

I hope save-everything-before-high-memory-usage-starts is on the list of features for the new version

I rushed that in at the last minute -- I hope it works for you.

Mark Rose · 2023-03-09, 03:43

Quote:

Originally Posted by Prime95

I rushed that in at the last minute -- I hope it works for you.

Wonderful!

I await the next Linux binary to be posted

Prime95 · 2023-03-09, 04:49

Quote:

Originally Posted by Mark Rose

I await the next Linux binary to be posted

It is in 30.11

kruoli · 2023-03-09, 08:38

It looks like 30.11 is auto-selecting higher B2 even for P-1 in comparison to 30.8.
30.8:

Code:

[Work thread Mar 8 11:55] P-1 on M111149 with B1=10000000000, B2=TBD
[Work thread Mar 8 11:55] Available memory is 61440MB. 
[Work thread Mar 8 11:55] Switching to FMA3 FFT length 8K, Pass1=128, Pass2=64, clm=2, 8 threads
[Work thread Mar 8 11:55] With trial factoring done to 2^119, optimal B2 is 67065*B1 = 670650000000000.
[Work thread Mar 8 11:55] If no prior P-1, chance of a new factor is 2.74%
[Work thread Mar 8 11:55] Estimated stage 2 vs. stage 1 runtime ratio: 0.653
[Work thread Mar 8 11:55] Using 61440MB of memory.  D: 1771770, 167040x810354 polynomial multiplication

30.11:

Code:

[Worker Mar 9 08:25] P-1 on M111091 with B1=10000000000, B2=TBD
[Worker Mar 9 08:25] Using FMA3 FFT length 6K, Pass1=128, Pass2=48, clm=2, 6 threads
[Worker Mar 9 08:25] Available memory is 61440MB.
[Worker Mar 9 08:25] Switching to FMA3 FFT length 8K, Pass1=128, Pass2=64, clm=2, 8 threads
[Worker Mar 9 08:25] With trial factoring done to 2^119, optimal B2 is 108799*B1 = 1087990000000000.
[Worker Mar 9 08:25] If no prior P-1, chance of a new factor is 2.86%
[Worker Mar 9 08:25] Estimated stage 2 vs. stage 1 runtime ratio: 0.664
[Worker Mar 9 08:25] Using 61440MB of memory.  D: 2552550, 230400x780395 polynomial multiplication.

Also, it seems to use more memory than 30.8, it now thrashes for me when it never did before (1 screen output every 30 s vs 120 s).

Prime95 · 2023-03-09, 14:29

That is a bit strange. I haven't made many changes to the P-1 code. For now, try reducing your memory settings. I'll look into the issue later.

Edit: Thinking more, there was a change regarding the actual and estimated memory use with high D values and multithreading. What does "top" say for prime95's memory usage?

henryzz · 2023-03-09, 14:45

Has the optimal B2 calculation been done with an FFT of 6K in 30.11 and 8K in 30.8?

2023-03-09, 02:16	#201
cmpute "JZ" Mar 2023 31 Posts	Another crash example for P+1 tasks As I mentioned above, sometimes the stage2 of P+1 will crash. The previous post is an example that always lead to core dump on my computer, here's another example, where a crash happens every several hours (not every time, about every 3-4 exponent with 2 workers). The assignments I am doing are P+1 on exponents around 10M, Start=2 / 7, B1=2200000, B2=100000000. The stack trace this time is Code: Thread 36 "mprime" received signal SIGSEGV, Segmentation fault. #0 0x0000000000480cb4 in fill_pairmap(int, void*, int, int, int, int, int, int, short, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, unsigned char*, unsigned long) () #1 0x0000000000461125 in pp1_stage2_impl(pp1handle) () #2 0x00000000004766a2 in pplus1 () #3 0x0000000000441e10 in primeContinue () #4 0x0000000000442039 in LauncherDispatch () #5 0x00000000004b710a in ThreadStarter () #6 0x00007ffff781d6db in start_thread (arg=0x7ffff376c700) at pthread_create.c:463 #7 0x00007ffff689461f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95 This time the crash happened in `fill_pairmap` rather than `pp1_stage2_impl`. Note that I'm still using 30.10b4, will try 30.11 soon. Thanks for your work in advance! Last fiddled with by cmpute on 2023-03-09 at 02:43*

2023-03-09, 14:29	#208
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2·4,127 Posts	That is a bit strange. I haven't made many changes to the P-1 code. For now, try reducing your memory settings. I'll look into the issue later. Edit: Thinking more, there was a change regarding the actual and estimated memory use with high D values and multithreading. What does "top" say for prime95's memory usage? Last fiddled with by Prime95 on 2023-03-09 at 14:36

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
That's a Lot of Users!!!	jinydu	Lounge	9	2006-11-10 00:14
Beta version 24.6 - Athlon users wanted	Prime95	Software	139	2005-03-30 12:13
For Old Users	Citrix	Prime Sierpinski Project	15	2004-08-22 16:43
Opportunity! Retaining new users post-M40	GP2	Lounge	55	2003-11-21 21:08
AMD USERS	ET_	Lounge	3	2003-10-11 16:52

2023-03-08, 16:11	#200
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 1,559 Posts	Magnificent, thank you, it now works!

2023-03-09, 14:45	#209
henryzz Just call me Henry "David" Sep 2007 Liverpool (GMT/BST) 3³·227 Posts	Has the optimal B2 calculation been done with an FFT of 6K in 30.11 and 8K in 30.8?