A Complete History Of Mainframe Computing

[Warren Wolfe]

Chuck Guzis wrote:
> He describes the CDC6600 as "was very limited in the instructions it 
> could perform".  Huh??  Compared to what?  One little-known aspect 
> was that the 6600 smoked any of the IBM iron of that time when 
> running COBOL.  

The same is true of ForTran.  The key was the bizarre way that floating 
point numbers were stored. in a 60 bit word.  The first 18 bits (high 
order) were the exponent, and the remainder were the mantissa.  To 
assemble a FP value, place the mantissa in the lower bits.  1's 
complement it.  Then put in the exponent, and 1's complement it again.  
Voila.

When I first saw that, i just couldn't figure it out.  I finally asked 
one of the gurus about it, and he told me that by doing that, if you 
compared FP numbers as if they were integers, the larger FP number would 
be the larger integer.  Tests, loops, and compares all went at integer 
speed, avoiding a great deal of FP math.  Weird, but FAST.  Still makes 
me shudder a bit to think about it, though...

Warren