New Z196 (IBM mainframe) instructions

[hanc...@bbs.cpcn.com]

An article describing the new instructions added for Z196.

I'm not sure if any of the instructions are priviledged, I think some
are.

Some are for floating point operations. Given the many enhancements
to floating point architecture I wonder how a conventional modern IBM
mainframe would perform in a super computing ("number crunching")
application.

http://share.confex.com/share/116/webprogram/Handout/Session8546/New%20CPU%20Facilities%20in%20the%20z196.pdf

[John W Kennedy]

IBM hasn't even tried to make the 360 ancestral line competitive in
supercomputing since the 90s, and the Fortran compiler is ludicrously
out of date. Supercomputing these days is done on systems with
thousands of what used to be called "processors" but are now called
"cores", and the z196 maxes out at a mere eighty. Hexadecimal floating
point is used for legacy, binary floating point is used for data
exchange, and decimal floating point is mainly aimed at replacing
packed decimal, in the hopes that it will finally render moot the
wretched intermediate-precision problem.

--
John W Kennedy
"You can, if you wish, class all science-fiction together; but it is
about as perceptive as classing the works of Ballantyne, Conrad and W.
W. Jacobs together as the 'sea-story' and then criticizing _that_."
-- C. S. Lewis. "An Experiment in Criticism"

[Nomen Nescio]

> IBM hasn't even tried to make the 360 ancestral line competitive in
> supercomputing since the 90s,

They are very different, as we know, they were never designed to be
supercomputers.

> and the Fortran compiler is ludicrously out of date.

Is that true? I understood XL/Fortran is being updated to the very latest
standards.

[John W Kennedy]

On 2012-02-02 14:49:25 +0000, Nomen Nescio said:

>> IBM hasn't even tried to make the 360 ancestral line competitive in
>> supercomputing since the 90s,
>
> They are very different, as we know, they were never designed to be
> supercomputers.

Not "never". The 91, 95, and 195 were, and the 3090 had a vector-math
feature. (In between, IBM had a series of channel-attached vector
processors.)

>> and the Fortran compiler is ludicrously out of date.
>
> Is that true? I understood XL/Fortran is being updated to the very latest
> standards.

As far as I can tell from IBM's website, XL/Fortran is no longer
available on 360-line systems. VS Fortran, for z/OS and z/VM, is still
limited to Fortran 77, and doesn't even support IEEE binary
floating-point.

Apple is paying more attention to supercomputing (Grand Central
Dispatch, OpenCL) than IBM's mainframe division is nowadays.