An astronomical number

[raymond]

In order to get 20 digit accuracy for 2^64, I may need to retain the 40 most significant digits of each computation. The timing may then increase by a factor of 4.

By retaining the 36 most significant digits of each computation, I get 19 digit accuracy for 2^64. The timing was 0.5 ms.

[raymond]

One minor modification, rounding the 37th digit before cutting it to 36 digits, resulted in at least the 20-digit accuracy for 2^64. Still takes 0.5 ms.

Attached is the latest (and probably last) source file for the record on this project.

[jack]

 

[jack]

Raymond, how do you run your program?
I assembled it ok but when I run it all I get to see is this message box
(I included umdtoa.asm from your fpu library and it assembles without error)
---------------------------
Error
---------------------------
284
---------------------------
OK   
---------------------------
the number varies between runs

[raymond]

Hi jack

You must have a more modern/faster computer than mine to show a computation speed of 0.28 ms. The reason for the "error" title, is that I don't usually bother to give a title to such message boxes; so it uses the 'default' title!!!!

[daydreamer]

One minor modification, rounding the 37th digit before cutting it to 36 digits, resulted in at least the 20-digit accuracy for 2^64. Still takes 0.5 ms.

Attached is the latest (and probably last) source file for the record on this project.

great
I tried make SSE2 (real8s) version earlier,wonder if the rounding up/down or to integer works best with the formula?
what about windows asm version for this:
https://rosettacode.org/wiki/Barnsley_fern