On 5 Jan 2011, at 11:44am, TR Shaw wrote:

> Thats why years ago IBM 1400 series did all their arith with long strings of 
> bcd numbers for finance.

BCD was awesome for fixed-point finance.  But pretty-much everything back then 
was done with overnight runs, so execution time didn't matter.  A six-hour 
overnight session is not qualitatively worse than a five-hour one.  Richard's 
comment about using fixed-point maths is probably the best way to solve it with 
modern CPUs.

>  Additionally IBM7000 series and PDP9's could add .33333333 to .666666 and 
> get 1.0. Maybe PDP10's and IBM 360s but I can't remember.

The code to do maths like that has a ton of tuning variables set up to get the 
right answers.  For instance back then the currency with the most places after 
the decimal was the Lira (four digits).  So repetitions of five digits were 
considered to be 'recurring'.  It was just as easy to find sums BCD got wrong 
as it is with binary.

Simon.
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users

Reply via email to