> Weren't there some VAX models which had hardware support for
> 64- or even 128-bit data types?
Floating point, yes. (128-bit only on a few models.) Integer, not
really. The VAX architecture did have a few instructions that made it
easier to do multi-longword arithmetic, but 32-bits was the largest
native type with full instruction support.
I also find I need to qualify an earlier comment I made. Nowadays,
indeed, compiler vendors do not change the default integer and real
kinds between 32 and 64 bit architectures, but 25+ years ago, when the
industry was transitioning from 16 to 32 bits (took much longer in the
PC world), Fortran compilers did change the integer default from 16 bits
to 32 bits. There were many reasons for this, including improved
performance and that 16-bit integers were too limiting for many
applications. The landscape is a bit different nowadays and there is no
benefit to changing the default integer kind. If we did, the default
real kind would also have to change, due to language rules, and that
would impact performance for many implementations.
Steve Lionel
Software Products Development
Intel Corporation
Nashua, NH
|