srsieve/sr2sieve enhancements

[bur]

Originally Posted by storm5510

George Woltman, a.k.a. "Prime95", wrote something about screen savers. He indicated to never use one which would require mathematical calculations when running. The majority of them do. He recommends just a blank screen

I know this is an older post, but a screen saver isn't required anymore. In the past it really did save the screen because on the old CRTs the constant bombardment by electrons could alter the screen if it showed the same image for hours. Though I never really got it how apparently the Word icons or start menu icon didn't matter...


With LCDs this doesn't happen anymore, since the LED backlight is on all the time. No matter if blank screen or not. It's different for OLEDs, since in that case the individual pixel is really switched on and off and to prevent uneven aging a blank screen could help.


But the easiest solution is not to use a screen saver, but just switch off the display after 10 minutes or so via energy saver settings. Doesn't use any CPU power and saves electrical power and you display won't age.

[storm5510]

Originally Posted by bur

I know this is an older post, but a screen saver isn't required anymore. In the past it really did save the screen because on the old CRTs the constant bombardment by electrons could alter the screen if it showed the same image for hours. Though I never really got it how apparently the Word icons or start menu icon didn't matter...

With LCDs this doesn't happen anymore, since the LED backlight is on all the time. No matter if blank screen or not. It's different for OLEDs, since in that case the individual pixel is really switched on and off and to prevent uneven aging a blank screen could help.

But the easiest solution is not to use a screen saver, but just switch off the display after 10 minutes or so via energy saver settings. Doesn't use any CPU power and saves electrical power and you display won't age.

I gradually stopped using screen savers when I began using LCD's. I have one here I came across which was made in 2003. It has one reversed pixel. Otherwise, it function fine. Another I bought new in 2008. It still works good. I have a recent new one which is 1920 x 1080. I was amazed at the clarity. If I leave, or am sleeping, I turn them off manually.

[Gelly]

It's likely that I'm very dense (as is standard), but I'm having whack-o issues attempting to run srsieve2 on sequences like 3^n+2 or 3^n-2. This might be along the lines of trying to use a screwdriver to open a can of tomatoes, but I was assuming that having k=1 would be sufficient to get things working and would be fine to use as a tool - even trying to sidestep things with some k = b^(small power) to avoid issues if necessary.

I was greeted with "Sequence 27*3^n+2 has 1000000 terms removed has they have the factor 5" in the +2 case, which definitely seems weird, as numbers like 3^14+2 are prime and are thusly not divisible by 5, and as for -2 it comes down to "Fatal Error: Expected 18446744073709324636 terms when building sequences, but counted only 235736.", which, at the very least, strikes fear into my heart.

[rogue]

Originally Posted by Gelly

It's likely that I'm very dense (as is standard), but I'm having whack-o issues attempting to run srsieve2 on sequences like 3^n+2 or 3^n-2. This might be along the lines of trying to use a screwdriver to open a can of tomatoes, but I was assuming that having k=1 would be sufficient to get things working and would be fine to use as a tool - even trying to sidestep things with some k = b^(small power) to avoid issues if necessary.

I was greeted with "Sequence 27*3^n+2 has 1000000 terms removed has they have the factor 5" in the +2 case, which definitely seems weird, as numbers like 3^14+2 are prime and are thusly not divisible by 5, and as for -2 it comes down to "Fatal Error: Expected 18446744073709324636 terms when building sequences, but counted only 235736.", which, at the very least, strikes fear into my heart.

What did you use on the command lnie?

[Gelly]

Originally Posted by rogue

What did you use on the command lnie?

For 3^n+2, I did srsieve2 -n 1 -N 1000000 -s"27*3^n+2".
For 3^n-2, I did the same but with "27*3^n-2".

Similar results occur for -s"1*3^n+2" and fails to parse -s"3^n+2", as expected.

Edit: I suppose I should mention that the issue with the "removing 1000000 factors" bit isn't that it removes factors, but it removes the entire range, full stop.

[rogue]

Originally Posted by Gelly

For 3^n+2, I did srsieve2 -n 1 -N 1000000 -s"27*3^n+2".
For 3^n-2, I did the same but with "27*3^n-2".

Similar results occur for -s"1*3^n+2" and fails to parse -s"3^n+2", as expected.

Edit: I suppose I should mention that the issue with the "removing 1000000 factors" bit isn't that it removes factors, but it removes the entire range, full stop.

The problem is a piece of algebraic factor code that is outputting the message should only be used if c = +1 or -1. I'll fix as soon as I can.

[LaurV]

Windows? Did you double the hat?

[rogue]

Attached is sr1sieve 1.4.7. The major change is for non-x86 users. It will now work for p up tp 2^64 whereas it was limited to 2^52 previously. Based upon a few tests it is actually only about 5% slower than when running with the 64-bit ASM routines. There is a minor change in an attempt to get more accurate timestamp details so that the best baby steps method is chosen a runtime. This costs a second or two upon start up. If you use -vv you will see which baby steps method it chooses.

[rogue]

To compile for ARM or other non-x86 CPUs, use ASSEMBLER=no in the makefile and use # to comment all architectures in the makefile as well.

[yoyo]

I compiled it on ARM 64 and make check reports an error

./srtest 1000000 0
Testing calculation of a^n (mod p) for random a,n < p < 2^62, seed=1610277202.
*** 5^0 (mod 13): expected 1, got 3.
*** 2746100997775259206^3224756103013107157 (mod 3821170955791617625):
    expected 2787118976035927961, got 2537629473793175596.
Stopping after 2 errors.
make: *** [Makefile:329: check] Error 1
pi@raspberrypi2:~/sr1sieve/sr1sieve_1.4.7$

[rogue]

Originally Posted by yoyo

I compiled it on ARM 64 and make check reports an error

./srtest 1000000 0
Testing calculation of a^n (mod p) for random a,n < p < 2^62, seed=1610277202.
*** 5^0 (mod 13): expected 1, got 3.
*** 2746100997775259206^3224756103013107157 (mod 3821170955791617625):
    expected 2787118976035927961, got 2537629473793175596.
Stopping after 2 errors.
make: *** [Makefile:329: check] Error 1
pi@raspberrypi2:~/sr1sieve/sr1sieve_1.4.7$

Use the attached srtest. You can see those when you look at the code. Extra steps are needed to use use the new generic mulmod correctly.