[geeks] Best Vista story I've seen

[Charles Shannon Hendrix]

On Wed, 21 Feb 2007 10:34:35 -0600
Doug McLaren <dougmc at frenzied.us> wrote:

> On Tue, Feb 20, 2007 at 07:18:07PM -0500, Charles Shannon Hendrix wrote:
> 
> | > > Can you imagine the look someone would have given you if you asked
> them, | > > in 1986, if you could use their Cray to play Doom?
> 
> Not everything done on the big iron was serious :)

I know... but a lot of big iron shops were full of people who were far too
serious.  A lot of real work was prevented by people who thought it wasn't
"real" enough.

It's ironic that machines with a fraction of the power my desktop has now,
were once guarded so closely.

My iPod has more power and storage than the first mainframe I used.

> But the company I worked for also ported software to Crays, so I had
> access to some machines actually at Cray for porting and testing.  My
> code compiled and ran on the Crays with no problems, but was only
> about 6x faster than on the SGI I had locally.  Which was nice, but it
> wasn't *that* big.  After talking about it with my co-workers (I was
> the sysadmin, not a developer of stuff like this) it should have been
> much faster than that, but I just didn't know how all the tricks
> needed to make stuff run fast on the Crays, and it really wasn't time
> effective for me to learn.

I found the same thing.  I was mostly using C, so I didn't have the benefit
of FORTRAN compilers that tended to do a lot of the work for you, especially
on parallel machines.

Single CPU Crays were easier, you just needed to learn to write code that
kept the vector unit working efficiently.

Parallel machines were really hard, especially esoteric stuff like Maspar.
It was difficult to translate algorithms into their limited and very odd
computation systems and do so efficiently.

I wasn't a FORTRAN programmer, so I couldn't benefit from the automation
those compilers offered.

I would like to learn how to write code that is efficient on SIMD now that
AMD/Intel are moving to that for all math, and ditching the FPU.

I just hate x86 assembly, and wonder if there are techniques you can use in
vanilla C that help SIMD compilers.


-- 
shannon
     \                         | And in billows of might swell the Saxons 
<===>|==Downfall of the Gael=> | before her,-- Unite, oh unite!
     /                         | Or the billows burst o'er her!