Factoring -1.#J% complete

Peter Nelson · 2005-04-03, 08:20

Hi, I have tried running the new v24.11 on a pentium 3. Yes I know it is a little slower than 23.8.1 for LL tests.

The new Trial Factoring benchmark is cool.

I saw that it tries running TF on M35,000,011 at different bit lengths.
It appears to be doing this on 1000 iterations (although not explicit).

My current TF assignments have been lower exponents than this.

However, I thought I would try to replicate these results (after going offline from primenet).

So in worktodo.ini I created an entry for factoring 35000011,57

I usually have screen iterations every 100 when timing manually, but what I saw also occurs in the default 1000 iterations.

That is for quite a few (though not a majority) of the screen output lines, the output appears as "-1.#J% complete" rather than the numerical percentage. Most of the status percentages appear as usual.

I think this is processed in the facpct function in the source code, and may therefore require attention. Probably the calculation is overflowing or something (but only sometimes).

This bug may only appear when testing exponents of this bigger size.

Can anyone else replicate these results on their machine?

Peter Nelson · 2005-04-04, 11:14

OK, on an pentium 4 (ie with SSE2) I also tried to trial factor M35000011 (assuming already done to 2^57).
Same errors when printing out % complete to screen (intermittent).

The exponents we are trial factoring are growing, and we will soon be in this sort of range. I chose it because of the TF benchmarks being done around this number.

GEORGE, please could you investigate this problem and include a fix for it in your code sometime.
I don't think it depends on the machine you run on, just when dealing with some large exponents to Trial Factor.

Peter

Prime95 · 2005-04-04, 15:11

Weird. I couldn't reproduce it on my P4 yesterday. Do you have a non-default number of decimal digits displayed?

Peter Nelson · 2005-04-05, 08:10

I have run this again and it still gets the same.

It appears less frequently (but have seen it) in 1000 iterations.

It is more obvious if you set screen output to 100 iterations. Try this.

Here is some of my output to give you an idea of the frequency and apparent randomness....

Starting factoring M35000011 to 2^68
[Apr 05 08:56] Factoring M35000011 to 2^58 is -1.#J% complete. Time 2.925 sec

Note it begins testing to 2^58 because I had placed "57" as done up to already in the factor line of worktodo.ini

Subsequent lines contain percentages....
4.77%
-1.#J%
9.54%
11.93%
14.31%
16.70%
19.08%
21.47%
23.85%
-1.#J%
28.62%
31.01%
33.39%
-1.#J%
-1.#J%
40.55%
42.94%
45.32%
47.71%
50.09%
52.48%
54.87%
57.25%
59.64%
62.02%
-1.#J%
66.79%
69.17%
71.56%
-1.#J%
76.33%
78.72%
etc

Prime95 · 2005-04-06, 00:17

Found it. It only happened in the release version. The debug version was bug-free. Thanks.

2005-04-03, 08:20	#1
Peter Nelson Oct 2004 529₁₀ Posts	Factoring -1.#J% complete Hi, I have tried running the new v24.11 on a pentium 3. Yes I know it is a little slower than 23.8.1 for LL tests. The new Trial Factoring benchmark is cool. I saw that it tries running TF on M35,000,011 at different bit lengths. It appears to be doing this on 1000 iterations (although not explicit). My current TF assignments have been lower exponents than this. However, I thought I would try to replicate these results (after going offline from primenet). So in worktodo.ini I created an entry for factoring 35000011,57 I usually have screen iterations every 100 when timing manually, but what I saw also occurs in the default 1000 iterations. That is for quite a few (though not a majority) of the screen output lines, the output appears as "-1.#J% complete" rather than the numerical percentage. Most of the status percentages appear as usual. I think this is processed in the facpct function in the source code, and may therefore require attention. Probably the calculation is overflowing or something (but only sometimes). This bug may only appear when testing exponents of this bigger size. Can anyone else replicate these results on their machine?

2005-04-04, 11:14	#2
Peter Nelson Oct 2004 23² Posts	Same message on another machine - bugfix request OK, on an pentium 4 (ie with SSE2) I also tried to trial factor M35000011 (assuming already done to 2^57). Same errors when printing out % complete to screen (intermittent). The exponents we are trial factoring are growing, and we will soon be in this sort of range. I chose it because of the TF benchmarks being done around this number. GEORGE, please could you investigate this problem and include a fix for it in your code sometime. I don't think it depends on the machine you run on, just when dealing with some large exponents to Trial Factor. Peter

2005-04-05, 08:10	#4
Peter Nelson Oct 2004 23² Posts	Example of -1.#J% complete I have run this again and it still gets the same. It appears less frequently (but have seen it) in 1000 iterations. It is more obvious if you set screen output to 100 iterations. Try this. Here is some of my output to give you an idea of the frequency and apparent randomness.... Starting factoring M35000011 to 2^68 [Apr 05 08:56] Factoring M35000011 to 2^58 is -1.#J% complete. Time 2.925 sec Note it begins testing to 2^58 because I had placed "57" as done up to already in the factor line of worktodo.ini Subsequent lines contain percentages.... 4.77% -1.#J% 9.54% 11.93% 14.31% 16.70% 19.08% 21.47% 23.85% -1.#J% 28.62% 31.01% 33.39% -1.#J% -1.#J% 40.55% 42.94% 45.32% 47.71% 50.09% 52.48% 54.87% 57.25% 59.64% 62.02% -1.#J% 66.79% 69.17% 71.56% -1.#J% 76.33% 78.72% etc

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
6^383+1 by GNFS is complete. Thanks!	fivemack	Factoring	50	2008-03-24 19:57
f14 complete	masser	Sierpinski/Riesel Base 5	2	2006-04-23 16:05
Complete Factorization???	Khemikal796	Factoring	13	2005-04-15 15:21
61.5 thru 62m complete to 2^60	nitro	Lone Mersenne Hunters	0	2003-12-07 13:50
sieving is complete	wfgarnett3	PSearch	1	2003-07-28 12:06

2005-04-04, 15:11	#3
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2³×1,019 Posts	Weird. I couldn't reproduce it on my P4 yesterday. Do you have a non-default number of decimal digits displayed?

2005-04-06, 00:17	#5
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2³·1,019 Posts	Found it. It only happened in the release version. The debug version was bug-free. Thanks.