RE: DEC rant [was Re: SIGFPE resolution?]
On Wed, Nov 05, 2003 at 11:23:09AM, Phil Mendelsohn wrote:
>I don't have a beef. But, having seen DEC introduce a lot of things
>that the world hasn't really grokked yet makes me mad at the parallel
>between your concern with IEEE compliance and Kurt Vonnegut's story
>"Harrison Bergeron." (Briefly, to ensure the equality of all, those
>with exceptional abilities were fitted with handicapping devices.)
Geez, I always remembered that story as "Flowers for Algeron" or
something like that from grade school reading. We probably
read the condensed version or something.
Although an intersting analogy, I'm not sure it translates
well here in that IEEE compliance in pre-EV6 systems
is generally managed by exception (not totally, I know).
>There will never be any good solution to floating point on the Alpha,
>because IEEE spec is equivalent to, but not better than, using
>H-floats from a numerical analysis standpoint. Alpha is faster
>(normalized) faster, because it's a real architecture. It rankles to
>throw out a lovely hardware solution because masses of (software)
>people who didn't really need it, bought hardware that was cheaper.
>Now those that really want fast FP are stuck with things that aren't
>as good as we could *possibly* do.
Perhaps you are unaware, but Alphas in the EV6+ generation are
IEEE compliant. DEC chose to make previous generations faster
by removing certain limitations.
When they moved to an out-of-order design, they (and this is my
recollection) found it easier to do the OO FPU design
if they removed many of the (had something to do with needing
to permit imprecise traps) restrictions they created to make
the earlier designs faster.
In the process they are now IEEE compliant without a
performance penalty.
So to say that there will never be a good solution is wrong.
DEC removed the restrictions in newer generations and now a
small number of us with older machines are stuck fighting
what will be a never ending battle with bugs that the mass
of developers out there don't even consider.
We are asking that people take special care to sympathize
with Alpha's handicaps (from their point of view, their platforms
just work). Fighting for minority rights in the real, politcal
world has obvious value to me - doing it in the economic harsh-reality
computer hardware world seems more difficult to show the value
especially in the case of Alpha with a known EOL coming
along. It's not extinction, but it makes it more
difficult to move forward on par with everyone else.
Alpha is more expensive and difficult for distributions to
support - why do you think RH and all don't support us
anymore? I love that Debian has such a grand vision - I just
don't want Alpha to be the first dropped platform.
>Also, Debian is not about absolute conformity. I don't care if FP
>takes a performance hit for Joe User, but like the other gent, I don't
>want to have to rewrite a compiler to take advantage of a piece of
>hardware I may have that does things right, when the rest of the world
>operates on a kludge that's standardized. (Oops; opinions showing.)
I know Debian is not about conformity, but it does seem to
have a goal of supporting all platforms uniformly.
See above - the "kludge" is now managed effectively by the
architecture you respect. Were the folks at DEC delusional
when they made that change?
I can assure you they didn't make the change to specifically
be IEEE compliant - they did it because it was the right thing
to do to meet their performance specifications.
They were a particularly rebellious bunch and only did things
to either improve performance or simplify design.
It was only after significant analysis that the BWX stuff was
added to EV56 - they resisted that forever until they found out
that code size would be reduced and fewer cache misses would
occur and the memory load latencies would not be affected at
all that they finally did what most critics considered to
be obvious.
>I am *not* going to give in to the temptation of making a smart remark
>about Alpha's success being a reflection on anyone's career at DEC and
>anyone's personal success of making things palatable to the masses,
>because it would be unfair, and most likely totally untrue. DEC's
>stealth marketing is legendary. I saw the same thing happen with VAX,
>and pdp-11, and DECsystem. The thing is that low-volume, high-priced
>niche products used to be a viable business model -- so DEC succeeded
>because it made the best elitist, nerdy, schmucky stuff -- but no
>longer it seems. It wasn't DEC that changed, it was the rest of the
>world that went wrong. ;)
Make a joke - I don't care - I'm proud of my work there.
Increasing Alpha volume was the only way it would have ever
been viable in the changed world you mention - look at how IA64
is flailing, even with the full financial and marketing
support of Intel/HP. It's radically different and so far
doesn't offer much value to users and it's not being purchased.
Alpha's time has come - perhaps I'll jusy try to find
an EV6 out there with an owner that doesn't realize what
they have and let the rest of you work this out (I wish
I had that much disposable income - why do think I'm
trying to get mileage outta this box?).
I've been using my 164LX at home since 1998 I believe - I'd
hate to see it go - amazing how much life I've gotten out
of this machine - I'm just trying to migrate full time to
Debian as to get more life out of it - this is complicated
by SIGFPE's everytime I try to do something that I
can do on NT4 without a problem).
Although, after the 1GB RAM upgrade, NT4 on this machine
absolutely screams - I just cannot use my nifty USB/Firewire
devices (yes, I could use Win2k, but my heart just cannot imagine
using the so-called FX!32 that's in Win2k - I can't yet
let go of that child).
....tom
---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.532 / Virus Database: 326 - Release Date: 27-Oct-2003
Reply to: