Prev: fortran sensibility at specification order
Next: FFTW and Subroutines, copying arrays
From: glen herrmannsfeldt on 21 Nov 2009 18:56
robin <robin_v(a)bigpond.com> wrote:

> OS/360 didn't, in general, have 16-byte reals.
> real F.P. occupied 32 bits and 64 bits.

It was supported in hardware on the 360/85, and software
on other processors. My first hand dissassembly was from
a hex dump of the OS/360 DXR emulation routine.

> How function results were returned depended not on the
> OS but on the hardware and the particular compilers.

Mixed language programming is much easier if they use the
same convention. Even more between different compilers
for the same language.

> It wasn't until the S/370 that quad precision was available
> (and then only for +. -. * but not division),

Yes only +, -, and * (until about half way through ESA/390
when DXR was finally supported by hardware). IBM did a
study on the cost vs. advantage of each and decided that
divide was relatively less common, and software, which used
the other operations in hardware, was the best way to go.
(Published in IBM Systems Journal.)

> and the OS then was OS/370. (And once again, the OS has nothing
> to do with the way that function results are returned.)

There is no OS/370. OS/360 ran on S/370 hardware, otherwise
there is OS/VS1, OS/VS2 and MVS/370.

-- glen
First  |  Prev  | 
Pages: 1 2 3 4 5 6 7 8 9 10
Prev: fortran sensibility at specification order
Next: FFTW and Subroutines, copying arrays