[9fans] 64 bit arithmetic

[9fans] 64 bit arithmetic

[Leo Caves]

 From http://plan9.bell-labs.com/magic/man2html/1/2c

The compilers accept long long variables as a 64-bit type. The standard header
typedefs this to vlong. Arithmetic on vlong values is usually emulated by a
run-time library.


What's the overhead on this?

plan 9 will be adopted for Beowulf-type applications, as this model
is naturally incorporated within the system's design.
However, most of these applications are numeric intensive and often
require 64-bit arithmetic.

is there any work internally in Bell Labs on native 64 bit libraries?

Re: [9fans] 64 bit arithmetic

[Leo Caves]

my misunderstanding of an entirely unambiguous man page.
the emulated library is for 64 bit fixed point arithmetic as you say,
not floating point.
(I must remember to engage brain before posting)

Re: [9fans] 64 bit arithmetic

[Eric Grosse]

> From: Leo Caves <caves@yorvic.york.ac.uk>
> plan 9 will be adopted for Beowulf-type applications, as this model
> is naturally incorporated within the system's design.
> However, most of these applications are numeric intensive and often
> require 64-bit arithmetic.
> is there any work internally in Bell Labs on native 64 bit libraries?

Not that I'm aware of, unless you mean something like
/sys/src/libmp/386/mpvecdigmuladd.s.

In my own "Beowulf-type applications" (domain decomposition for solving
partial differential equations) the 64-bit arithmetic was floating point.
64-bit fixed point seems more relevant to large SMP systems than to clusters.

Eric