From: "mihalco" <mihalco@pacbell.net>
Date: February 10, 2006 4:40:43 PM MST
To: Astro_IIDC@yahoogroups.com
Subject: Re: Intel based Mac question
Hi Milton,
Thanks for the update. I know it would be a huge undertaking, and
you definitely clarified where you are headed.
Regards,
Kur
--- In Astro_IIDC@yahoogroups.com, Milton Aupperle <milton@...> wrote:
Hi Kurt;
We are hard at work on Astro IIDC 3.0 (6+ months now in Alpha and
hope to begin Beta testing in four to six weeks). Apples' Intel
announcement didn't fit with our development schedule nor are
their
sales going to make much impact in the existing Mac user base -
(roughly 15 million are running OSX 10.3 or higher). So were not
concerned about WacIntel at this point. Also there is no reason
that
you can't use Astro IIDC within Rosetta - although I'm pretty sure
you can not use KeySpan or any serial devices for scope control
because they have KEXT dependencies for their POSIX driver.
I've scoped it out and we have roughly 600,000+ lines of scalar
code
to sort through and re-write to "switch" to Intel, which will take
3
to 6 months to write and likely another 3+ months to debug - both
our
code and Apples (lots of issues with FireWire). We optimized our
code
to run very fast on G3, G4 and G5 processors and all of the
critical
performance stuff has to be completely re-thought out. (For
example
some of our code never touches the main cache Ram except for data
input / output, we utilize all 32 CPU register and write the code
so
we don't overflow them - intel has at most 7 CPU registers and you
can't make that work). Your going to loose all our SIMD
optimizations, as SSE3 is missing several crucial instructions
that
have no equivalent to Altivec. So you'll have to suffer a
significant
performance hit and Apple's VecLib / VecDSP libraries are of no
use
to us for what we do.
Also the switch from MetroWorks CodeWarrior 9.x to Apple's XCode
2.2.x has had massive performance hit to our our PPC code base -
so
we will not be doing Universal Binaries. When I say massive, I'm
talking about compiling the same code base with all stable
optimizations on in both CodeWarrior and XCode and the XCode build
is
a minimum half as fast so that it wastes 100% more CPU cycles to
do
the same tasks. Although XCode is better than it was before (PB was
a
complete joke), Apple has really done an exceedingly poor job with
it's compiler optimizations for PowerPC with XCode. They could
have
bought or Licensed Code Warrior G4 and IBM's G5 compilers for a
song
and delivered really great performance, but they didn't and
everyone
pays for it. And this isn't just "me", a lot of developers are
saying
the same thing on the SIMD lists too and have comparable numbers.
Lastly, were having a really tough time justifying the expense for
a
MacIntel port, especially since Astro IIDC is a niche product for
science and Apple doesn't care about niche products or niche
developers. Apple's market share is 2.1% world wide and 2.5% in
the
USA (based on Apples' SEC Mac sales numbers as compared to Gartner
and IDC numbers of PC sold at
http://www.macworld.com/news/2006/01/19/
pcmarket/index.php) and frankly, their are more XBox systems out
their right now than their are Macs running OS X. Also their is
considerable doubt whether Apple will continue to support FireWire
on
new Macs or not, as iPod's don't do FireWire, the built in camera
on
all new Macs are USB 2, Apple has dumped FW 800 support on their
Laptops and their Intel partner is big on USB 2 support - not
FireWire. Since iBooks, MacMini's and iMacs' don't have expansion
slots and the bulk of our users use them, the loss of FireWire
would
kill off all FireWire products on the Mac. Their are a lot of very
unhappy FireWire hardware and software developers out their right
now
and Apple is saying nothing about what their intentions are.
So we have not made a decision on whether we will port MacinTel.
We
may very well rewrite everything for x86 including the GUI part
(which is maybe another 100,000 to 200,000 lines of code) and do
a
Windows XP/Vista port instead. What I do know is that if we do
offer
the MacIntel port, it will be a new product and you will be
buying
new - we won't offer an upgrade.
Hope that Helps..
Milton J. Aupperle
President
ASC - Aupperle Services and Contracting
Mac Software (Drivers, Components and Application) Specialist
#1005 - 815 14th Avenue. S.W.
Calgary Alberta Canada T2R0N5
1-(403)-229-9456
milton@...
www.outcastsoft.com