MEGAPOST apps to compute mersenne primes, factorization and computing decimals of pi on android

[ET_]

Originally Posted by thorken

ok i see but now i dont to publish the code.

...then your claims won't be trusted anymore...

[Uncwilly]

Originally Posted by ET_

...then your claims won't be trusted anymore...

It may be that Thorken meant "Ok, I see that I didn't publish the code as yet", not "I see, but I won't publish the code.
Let's wait and see.

[PhilF]

Originally Posted by Uncwilly

It may be that Thorken meant "Ok, I see that I didn't publish the code as yet", not "I see, but I won't publish the code.
Let's wait and see.

I was going to grab his exponent for double-checking, but it appears that if I did so it would be poaching.

I didn't know there were double checks assigned up that high.

[Batalov]

Originally Posted by PhilF

I was going to grab his exponent for double-checking, but it appears that if I did so it would be poaching.

I didn't know there were double checks assigned up that high.

Which exponent? Maybe you refer to "the other thread"? That exponent is just the first prime > 90000000, nothing to see there.

[kriesel]

Originally Posted by thorken

M216091 ScreenShot - 22m 5s

Cool. Please post separate (individual) timings for:
44497;
132049;
756839 (if you can spare your phone for a few days).

[thorken]

756839 5h 17m 19sec

[Uncwilly]

Originally Posted by thorken

756839 5h 17m 19sec

Why doesn't your program output to a text file? That would make the uploading of the data easier. When you post images of text output and videos of text, that makes me suspect that you don't understand the culture of the group. Screenshots are good for error message documentation and to show the state of something, but for source code and program output, text is best.

[ewmayer]

Originally Posted by thorken

756839 5h 17m 19sec

What FFT lengths were used for this one and for 216091? A decently-optimized DP FFT implementation would use lengths ~12K and 40K for these two exponents, implying a theoretical runtime increase (based on the O(n^2 log n) scaling for an FFT-based LL test) of ~14, which is pretty close to what you get. So at least your code (or whosever code you are using - too lazy to dig back through the thread to find the Spanish-language output someone posted that listed the author) appears to be FFT-based, the main question which remains is how close to optimal is it for the hardware in question?

[kriesel]

Originally Posted by ewmayer

What FFT lengths were used for this one and for 216091? A decently-optimized DP FFT implementation would use lengths ~12K and 40K for these two exponents, implying a theoretical runtime increase (based on the O(n^2 log n) scaling for an FFT-based LL test) of ~14, which is pretty close to what you get. So at least your code (or whosever code you are using - too lazy to dig back through the thread to find the Spanish-language output someone posted that listed the author) appears to be FFT-based, the main question which remains is how close to optimal is it for the hardware in question?

gpuOwL at its low end produces _lower than the expected order_ scaling, as does CUDALucas, and also code I wrote myself decades ago. I think it has to do with irreducible setup and other overhead affecting the low-p timings more as a percentage. The O(p^2 log p log log p) is asymptotic behavior, at high p and long run time that washes out constant or low order setup overhead. Some observed scalings (which are sensitive to sampling ranges, so power variations of +-0.05 are unremarkable):

gpuowl ranges from linear in p, to p^1.822
CUDALucas p^2.047
hypothetical p^2 log p log log p; p^2.117 (over p< 3.33E9)
prime95 p^2.172
perl ntheory p^2.386
inefficient grammar school with high overhead: p^2.425 (but p^3 expected)
thorken phone app v1 p^2.452
thorken phone app v2 p^2.262

Plots of actual timings tend to bend upward to higher power at high p, such as for gpuowl https://www.mersenneforum.org/showpo...6&postcount=10 as the asymptotic behavior is approached / overhead is diluted.

Re v2, "I programmed it" https://www.mersenneforum.org/showpo...99&postcount=1
Re v1, "I found on a mobile that I found this application in the folder / system / xbin of the phone" https://www.mersenneforum.org/showpo...58&postcount=1

[kriesel]

Originally Posted by Uncwilly

Why doesn't your program output to a text file? That would make the uploading of the data easier. When you post images of text output and videos of text, that makes me suspect that you don't understand the culture of the group. Screenshots are good for error message documentation and to show the state of something, but for source code and program output, text is best.

Amen!
A good writer begins with his audience in mind; their perspective, their needs, and makes an effort, once, to be clear and convenient for them, so that they don't all have to make an effort to access and understand the message.
What that would look like here is:
English language program output, screen shot ok as demonstration of having run it, and captured text, so that it can be read more easily, can be quoted as needed, etc.

There's nothing inherently wrong with screen shots, in some instances. I use them frequently to communicate about application issues or requests, such as with prime95. But that has an intrinsically graphical interface on Windows.

The phone app relating to this thread is the equivalent of a command prompt console text based application on Windows. The phone app is tall and narrow while my laptop is short and wide. The text in the screen shot is double spaced, making it either too tiny to read or too tall to display all at once. It's also unnecessarily wordy and verbosely repetitiously redundant. It's an image file, so the text can not be copied and pasted into an online translator for reading by non-Spanish speakers / readers (posted on an almost entirely English forum), unless tediously hand transcribed by an intended recipient. It's an image file, so it takes many more bytes (and more seconds) to transmit, every time, than a simple text file. That's an issue for those stuck with slow internet. If someone were intending to make postings INconvenient and INaccessible to the typical reader of this forum, it might resemble what is already being done.

[thorken]

Then my app its bad or good?
Keep in mind that the mobile reduces speed as soon as it reaches 70 degrees, low ups, low ups, with which the scaling index would improve a lot if the mobile will not resort to protect the CPU, in the 5 hours of calculation it reached 79 several times I looked at it and suddenly banana 69 and after a few seconds I went back up, so if you do not lower the speed by temperature you would have easily lowered the test one hour, that is instead of 5h then 4, it is over at 5min to start calculating the estimated time my app said was 4h 3m and it stayed until the temperatures went up and with them the calculation time, I think my program is quite good, I've done it with cxxdroid and I'm trying to port it to pc but I get an error with the gmp. I can not find it in dev-cpp and there is no way to do it, I have not had much time either.