unexpected order of evaluation

[dodicat]

srvaldez
Please remember that when you are trying to get ArcCot working in bigfloat-05 that the last arccot should be - Arccot (-20, 239).
You have - Arccot (20, 239).
The value of x in that function immediately is
-.10000000000000000000000000000000000000000000000000000000000000e-00017
after
x=1 : x=x/d
which throws everything off.
I messed about a bit, but couldn't get x to be .0041841004184100418410041841004184100418410041841004 or thereabouts which it should be.

[srvaldez]

dodicat
I am confused, I just downloaded my program from the link that I posted, set the precision to 1000 and checked the result against that given by the pari/gp calculator http://pari.math.u-bordeaux.fr/, the relative error was -5.003865326277570842 E-1007
would you please re-download the program and try again ?
let me know what happens

<edit> the reason that I do x=1 : x=x/d is so that x will be set to bigfloat 1, a bigfloat divided by an integer will give full bigfloat precision, but x=1/d will set x to the double precision of 1/d
you could do x=bigfloat(1)/d and it would work
you can also use quoted literals to assign bigfloat constants to a bigfloat variable: x="1" for example

[srvaldez]

also, in the arccot example function I used the non-overloaded multiply and divide so that I could limit the precision and speed things up
I implemented that trick in the fpatn function and for precision 2000 or greater it's significantly faster than without the trick

[dodicat]

I downloaded again.
result:

Code: Select all

NUMBER_OF_DIGITS =  50
d =  239
lim= 12
pr= 0.4326756663205261
d =  57
lim= 15
pr= 0.3461405330564209
d =  18
lim= 21
pr= 0.2472432378974435
 -.10000000000000000000000000000000000000000000000000000000000000e-00019
 0.003606299869716167
 

This is with 64 bit windows compiler.
The 32 bit compiler:below

Code: Select all

NUMBER_OF_DIGITS =  50
d =  239
lim= 12
pr= 0.4326756663205261
d =  57
lim= 15
pr= 0.3461405330564209
d =  18
lim= 21
pr= 0.2472432378974435
 -.10000000000000000000000000000000000000000000000000000000000000e-42949672
 0.003194000224539195
 

[srvaldez]

dodicat
since the compile command is at the top of file via #cmdline I can discount that as the problem, what version of FB are you using?

[dodicat]

I as using 1.10.1

[srvaldez]

I just downloaded FB-1.10.1-mingw-w64 gcc 11.2.0 and had no problem, I am stumped, try the following and tell me what you get
? pie-"3.14159265358979323846264338327950288419716939937510582096978610"
instead of ? pie

my suspicion is that the conversion to string is not working right, my guess is that it could be because of different text encoding
try compiling from the CLI

[dodicat]

It just freezes.
I tried with 1.2 (version shown at the end), not subtracting the string

Code: Select all

NUMBER_OF_DIGITS =  50
d =  239
lim= 12
pr= 0.4326756663205261
d =  57
lim= 15
pr= 0.3461405330564209
d =  18
lim= 21
pr= 0.2472432378974435
 -.10000000000000000000000000000000000000000000000000000000000000e-42949672
 0.00383829977363348
1.20.0
 

I have a command line editor, still the same direct from command line.

[srvaldez]

compile the original file from the download using the CLI
I think that your editor is changing the text encoding and causing problems

[dodicat]

I tried notepad, wordpad, and direct command line.

[srvaldez]

Ok, last resort
here's the compiled executable https://u.pcloud.link/publink/show?code ... 2sAhGYURL7
be sure to check for viruses
I forgot to put a sleep at the end so run from the console

[dodicat]

I used the built in
i=__builtin_clz(fac3(0))
instead of the naked asm.
OK now.

Code: Select all

 NUMBER_OF_DIGITS =  50
d =  239
lim= 12
pr= 0.4326756663205261
d =  57
lim= 15
pr= 0.3461405330564209
d =  18
lim= 21
pr= 0.2472432378974435
 3.14159265358979323846264338327950288419716939937510582096978610e+00000
 0.003082999959588051

[srvaldez]

Ok, will change that
thanks dodicat for the debug session
now the question is why the naked function didn't work for you

[dodicat]

probably because I am sitting here with my clothes on.

[srvaldez]