18 Oct
2011
18 Oct
'11
11:51 a.m.
On 18 Oct 2011 at 18:16, vintagecoder at aol.com wrote:
If you want the link to the doc on IBM's
instructions (Principles Of
Operation) I'll send it to you. I took a quick look now but the
floating point is over my head. Somewhere along the line they added
the IEEE binary floating point support, but the original hex
implementation had normalized and unnormalized instructions too. I
just can't remember if their FORTRAN had protection against a useless
mantissa or not, but I seem to remember it did. I wouldn't swear to it
though, I'm pretty rusty on that stuff.
The "red book"? I've still got the one I bought in the 1960s.
I didn't say that S/360 didn't have unnormalized FP operations; it
certainly did, at least according to my "green card" that I still
have.
I can't find anything about FP rounding normalization bits. I
thought your reference might be to a bit in the PSW that I hadn't
noticed before, but a quick check shows otherwise.
No, the evil in the S/360 FP operations was that normalization was
carried out to, not bit boundaries, but "nibble" (half-byte)
boundaries. So the casual user, thinking that he's got 24 bits of
precision in the mantissa would find out that he was very wrong.
AFAIK IEEE floating point didn't enter into the picture until S/390.
--Chuck