A little first-hand about the Orbiter development

Jay Windley said:
The issue of runtime optimization is
strictly one of profiling -- looking at how code behaves when it runs and
optimizing it accordingly based on this information.

Not really. A dynamic optimizer has an advantage over a static
optimizer. Static optimizers are often thwarted by not knowing all the
code that is running. Things like the possibility of aliasing (non-
obvious ways of having multiple ways of referring to the same variable)
force static optimizers to be conservative. There is at least one JVM
(currently, iirc, owned by BEA) that can over-optimize the code because
it knows what other code is also loaded into the JVM. Should the
program load some new code, the JVM can deoptimize the code as needed
to prevent problems.

A large dose of engineering good taste is needed to be sure you don't
loose more than you gain in dynamic optimization.


This is not generally
available to static optimizers, but natively-compiled languages such as
C/C++ can use standalone optimizers to identify problematic code stretches.

This really isn't a language thing. Traditionally, C/C++ has been
statically optimized, but there is nothing that would prohibit a
HotSpot-like system that detected performance-sensitive code and
optimized it. With most C/C++ systems, this would be a machine-code to
machine-code optimizer, with the advantage of being usable with any
language that compiles to machine code.


The simplistic example I posted was, of course, an unambiguously invariant
loop body. Both Java and C will, if set up to optimize to the same degree,
remove that code and eliminate the loop entirely. It was based on a
comparison I read a while back which used a somewhat less straightforward
loop which the HotSpot JVM had optimized (correctly) into oblivion while the
C compiler (gcc) had not. As I said, it doesn't appear that this happens
very often and the benchmarking people don't seem interested in it.

Actually, the benchmarking people were interested in it because the
compiler writers optimized their products for best benchmark numbers,
resulting, as you say, in optimizing away entire loops. (Tuning to the
benchmark was not just a software thing. I know of a machine that
included an arc-hyperbolic-cosecant instruction, but didn't include an
arc-hyperbolic-secant function. Guess which trig functions appeared in
the popular benchmarks.) The benchmark owners eventually mandated that
the benchmarks be run with all optimizers turned off.
--
Kevin Willoughby lid

We'd spend the remaining time trying to fix the engine.
-- Neil Armstrong

I could go on and on over this topic. Given the forum, let me just say
we essentially agree on all of this stuff.

Jay Windley said:
Right now Java costs me time. My Java programmers have to wait until a JVM
becomes operational on platforms I build. That has to wait until the C/C++
environment is ready, at which point my C/C++ guys are off and running.

I'm not surprised. One of the dirty secrets of Java is that most of the
JVMs are coded in C. Can't start a new JVM until the C compiler is
working well. So the JVM guys can't start working until the time when
your C/C++ guys can also start working.


ObSpace: I am, however, interested in Java being used on the new Mars
probes. Is it written up anywhere in more detail?

Not as much detail as I suspect you'd like. If you are a member of the
Java Developers Community (or willing to register for the JDC), go to
http://servlet.java.sun.com/javaone/...home/index.jsp
and enter MARS into the search box.


| Myself, I like blackboard and chalk.

Just TRY to find an honest-to-goodness chalkboard anymore. I hate
whiteboards; not enough resistance to the stroke.

Also, getting chalk dust off my pants is trivial, getting those damn
marker inks off my pants is tough.
--
Kevin Willoughby lid

We'd spend the remaining time trying to fix the engine.
-- Neil Armstrong

(Replies consolidated to reduce off-topicness)

"Kevin Willoughby" wrote in message
...
|
| A large dose of engineering good taste is needed to be sure
| you don't loose more than you gain in dynamic optimization.

Yes, thank you. I was going to allude to this eventually. Static or
dynamic, an optimizer will always have the disadvantage of not knowing the
problem you're trying to solve. It can do lint-picking things like strength
reduction or factoring out invariants, or consolidating similar code as you
brought up. But the greatest increases in efficiency that I've seen -- and
I mean they're *big* wins -- have always been from skilled programmers
figuring out better ways to do things at the bird's-eye level with knowledge
of what each modular chunk of code is supposed to accomplish.

Funny how it always comes back to having skilled people whose understanding
is focused on the problem rather than on the tool. I run into this all the
time. People think in terms of what's available to them and what they can
do with it, not in terms of what you need to solve the problem.

| ... With most C/C++ systems, this would be a machine-code
| to machine-code optimizer, with the advantage of being usable
| with any language that compiles to machine code.

Most compilers you buy will optimize both at the abstract code level and at
the machine code level. These are the compilers sold by the CPU
manufacturers. In the cases I've tested they produce much better code than
freebies like GNU's compiler. Remember, my customers write code that has to
run for several days on terascale systems. A 10% speedup is not
insignificant.

| Actually, the benchmarking people were interested in it because
| the compiler writers optimized their products for best benchmark
| numbers, resulting, as you say, in optimizing away entire loops.

Well, I don't put a lot of stock in benchmarks, especially those done by
manufacturers of either hardware or compilers/runtimes. I always find it
interesting in the Java versus C debate that the C people's benchmarks come
out faster for C while the Java people's benchmarks come out faster for
Java. Again, the spectre of the skilled programmer raises its head.

The only benchmark I pay attention to is LINPACK because it's a standard in
the supercomputing industry. And pretty much anything goes. I've got a guy
whose only job is to hand-optimize math libraries for these beasts. He can
sometimes squeeze 20% speedups out of that code, which pushes our benchmarks
along nicely. And we use compilers provided by the CPU manufacturers with
all the recommended optimizations enabled.

But what I meant was that the loop in question wasn't interesting because it
was, in fact, so trivial. The failure of one compiler to optimize it to the
bit bucket isn't that big a deal in the grand scheme of things. I certainly
don't mean to say that factorization of loop invariants is not interesting,
nor that elimination of dead loops is not interesting.

"Kevin Willoughby" wrote in message
...
| I could go on and on over this topic. Given the forum, let me
| just say we essentially agree on all of this stuff.

Yeah, I think we're wearing out our welcome. I don't mind Java being
pitched for things that it's never been used for before. I do mind Java
being pitched for everything.

| I'm not surprised. One of the dirty secrets of Java is that
| most of the JVMs are coded in C.

I never considered it that much of a secret. I worked with Sun's original
JVM that was *very poorly* coded in C. I *really* hated that code and so
did everyone else that touched it. I think that was one of the big reasons
Java didn't have immediate appeal.

| Not as much detail as I suspect you'd like. If you are a member of the
| Java Developers Community (or willing to register for the JDC), go to
| http://servlet.java.sun.com/javaone/...home/index.jsp
| and enter MARS into the search box.

LOL, I'll get one of our Java guys to do it. Thanks. I've worked alongside
several projects to embed Java, none of which came to fruition. I'd like to
see a success story.

| Also, getting chalk dust off my pants is trivial, getting those damn
| marker inks off my pants is tough.

Here's a tip (pun intended). Make this a tight "basic block" in your
whiteboard presentations:

1. Remove cap
2. Draw on whiteboard
3. Replace cap

Not only will it keep your clothes cleaner, the pens will last longer.
Nothing irritates me more than an enfeebled whiteboard marker, except for
people who stand there for five minutes talking while the cap is off the
pen. They *dry out*, people!

--
|
The universe is not required to conform | Jay Windley
to the expectations of the ignorant. | webmaster @ clavius.org

On or about Thu, 17 Jul 2003 22:44:39 -0400, Kevin Willoughby
made the sensational claim that:
On of my pet peeves: people who mash the marker's point into the
whiteboard so that is has the consistency of overcooked spaghetti. Ya
just can't do a drawing with a blunt, mushy marker.

*sniff sniff* Wazzat you say?
--
This is a siggy | To E-mail, do note | This space is for rent
It's properly formatted | who you mean to reply-to | Inquire within if you
No person, none, care | and it will reach me | Would like your ad here

"rk" wrote in message
...
|
| What did he do on the AP-101?

Writing assemblers and compilers for it. I don't know if he did any direct
support work for the shuttle. He was working for IBM at the time, but I
don't know who else might have been using the -101 while he was there.

--
|
The universe is not required to conform | Jay Windley
to the expectations of the ignorant. | webmaster @ clavius.org

"Kevin Willoughby" wrote in message
...
|
| Yah, but the hand that holds the cap will get brightly colored
| splotches.

Better the hand than the pants. Or the white shirt.

| On of my pet peeves: people who mash the marker's point into the
| whiteboard so that is has the consistency of overcooked spaghetti.

Agreed. But around here they tend to dry out before succumbing to
tenderization. The other nice thing about chalk is that you can use the
same rotation technique as for pencil to assure consistent line weight.

I like CAD. I'm commited to CAD. I've evangelized CAD. I've built CAD.
But I still like the feel of pencil on vellum.

--
|
The universe is not required to conform | Jay Windley
to the expectations of the ignorant. | webmaster @ clavius.org

"Jay Windley" writes:
And that was my understanding of what this thread is about (or rather, what
it has become). It's less about Java versus C or FORTRAN and more about
correct understanding of underlying principles and basic concepts. Today's
grads are being insulated from concepts that are still important.

Agreed. I'm sure we all have a story we can tell about paying for
something we bought with cash, something unexpected happens, and all
of a sudden the cashier simply can't make change for you because of "a
problem with the register". The real problem is that the person
running the cash register has no clue how to make change on their own.
We're turning out grads (at all levels) that are entirely dependent on
computers to do this sort of work for them.

Jeff
--
Remove "no" and "spam" from email address to reply.
If it says "This is not spam!", it's surely a lie.

On Fri, 18 Jul 2003 10:46:55 -0600, "Jay Windley"
wrote:

I like CAD. I'm commited to CAD. I've evangelized CAD. I've built CAD.
But I still like the feel of pencil on vellum.

Or the look of India ink in a ruling pen going onto creamy drafting
paper.

Mary

--
Mary Shafer Retired aerospace research engineer

"A MiG at your six is better than no MiG at all."
Anonymous US fighter pilot

In article ,
Mary Shafer wrote:

On Fri, 18 Jul 2003 10:46:55 -0600, "Jay Windley"
wrote:

I like CAD. I'm commited to CAD. I've evangelized CAD. I've built CAD.
But I still like the feel of pencil on vellum.

Or the look of India ink in a ruling pen going onto creamy drafting
paper.

Mary

When I was a freshman engineering undergrad (1986 - 1987) we were
required to take two quarters of Engineering Graphics, which at that
time still meant basic drafting. The classes were a bit of a rarity in
that they were four hour classes and actually met four mornings a week.
Day after day we'd all sit around drafting tables and learn to letter
properly; how to properly dimension drawings; how to derive perspectives
and elevations, etc.

By the time I'd graduated and gone to work on Space Station Freedom,
Boeing had almost entirely eliminated manual drafting. Due to an MSFC
requirement (to help a local business, of course!) we were using
Intergraph Unix-based workstations running their I/EMS software (on
machines with a then-generous 16 megabytes of RAM - about a year later
Intergraph was in the midst of upgrading the machines to an astounding
48 megs each! One of the most interesting things about the Intergraph
system was that even then they were capable of being clustered to
perform detailed 3D renderings using spare machine cycles - very common
in redering farms now but amazing new technology to us junior engineers.
Some of the models were detailed enough that one master machine and the
spare cycles of two or three dozen others took two weeks or more to
render. This was Boeing's first "paper-free" design - everything from
the pressure vessel to the fluid lines were modeled in 3D and the
drawings generated from the models.

Interestingly, a couple years after the start of the SSF contract when
the company started the Triple-7 program, they chose CATIA instead of
I/EMS . . . ;-)

--
Herb Schaltegger, Esq.
Chief Counsel, Human O-Ring Society
"I was promised flying cars! Where are the flying cars?!"
~ Avery Brooks

Bruce Palmer wrote in message t...
Mike Speegle wrote:
Even with calculators kids struggle. Last week at the movies I
bought something at the counter for $3.25. Gave the kid 3 ones and
while I'm fishing in my pocket for change he enters $3.00 tendered and
the display reads $.25 still due. I give him 3 dimes and he keys in
$3.30 tendered with $3.05 in change. ??? He actually counted out 3
ones but I saved his butt and just took the nickel shaking my head.
Sheesh. So even with electonic help some of these kids are seriously
polluting the gene pool. :-(

And to think baseball pitcher Mike Hampton left New York for the Rockies
because he liked the Colorado school system better

But seriously, here on Long Island if the cash registers didn't figure
out the change, 99% of the kids you run into couldn't do it themselves.

I can recall, not terribly long ago although it seems like it, a
summer student we had (4th yr Mech Eng, OK, I _was_ having an affair
with her but I wasn't her supervisor) who came to me with a problem
and I hand-estimated a square root (casting out 9's method, usually
gives you a decent value) and I had to explain to her what it was:
the implication being that without calculators (didn't have mine at
work that day) she was lost. So, then, what ARE they teaching
engineers, or high school students for that matter, in maths class
these days ?

BTW, it all ended amicably and I attended her wedding in France a few
years later.

Cheers all,