<strong>It is normally called 'asynchronous' chip. No master clock synchronization needed on the whole chip. One big advantage of it is less power consumption.</strong><hr></blockquote>
Since this may go over your head (or at least the head of most other readers), I'll explain it a bit better:
A regular chip uses a clock for every instruction. One part of the chip can only hand off something to another part of the chip when the clock ticks. That means that the hardware can't just be optimized to take as little time as possible. Hardware designers instead have to divide instructions into steps and sync these to the clock. This is very hard to do and not very efficient. Furthermore, once you get into higher clockspeeds, a good part of the chip (and current) is used by the circuitry for the clock.
Asynchronous chips just deliver the result when it is done. Of course, this sounds much easier than it actually is. The receiver must be able to proces the signal when it is presented and it must be able to determine when the end result is available. The latter is difficult because circuitry often takes on various values during computation. There are various solutions for that, but that would go too far for this brief explanation.
<strong> [quote]And yeah, intel is investing a lot on this technology. Not sure whether they used it or not.</strong><hr></blockquote>
Most research is going into smartcards now. Those are the chips that need to be most efficient with power. Fully asynchronous CPU's are quite a way off AFAIK (unfortunately). It has a lot to do with the retraining of engineers. It's quite different, so you cannot expect a fast switch. A lot of experience has to be gained first.
Moki's "options" concept seems a lot more feasible than what I/others thought Steve meant before then.
Changing platforms isn't increasing option, it's simply a change.
Changing PPC suppliers can still get you into a situation Apple is in now, when the 60x line was replaced with an embedded-orientated 7xx/7xxx line.
Bundles are where it's at. Screw Mac OS 9 style FAT binaries, bring on bundles with multiple compiles in the /content/MacOS folder (which I assume is how it would be done).
There is the AltiVec question though (Paging Moki, Programmer). Would developers bother to optimise for AltiVec anymore if only, say 1/2 Apple's pro machines used it? Because if Apple went multiplatform, the PowerPC only really competes when VMX/AltiVec/VE is used.
Barto
EDIT: There is a hack to get Mac OS X to run on the AmigaOne G3, and there will possibly be hacks for other PPCs in future. Like Moki said, make it hard for the 90%, because you'll never get the 10%.
There is the AltiVec question though (Paging Moki, Programmer). Would developers bother to optimise for AltiVec anymore if only, say 1/2 Apple's pro machines used it? Because if Apple went multiplatform, the PowerPC only really competes when VMX/AltiVec/VE is used.</strong><hr></blockquote>
The unfortunate fact is that very few programs/programmers optimize for AltiVec now anyway. Apple certainly does, because they need the underlying OS to be as fast as possible, but the few programs that do optimize for AltiVec are the ones that need every ounce of performance, such as PhotoShop.
The long and the short of it is that applications that really need the extra performance will still optimize for it. On the x86 side, they'll use the SIMD mnemonics there, on the PPC side, they'll use AltiVec.
AltiVec really does kick some serious ass -- I think Apple hasn't really evangelized using it as much as they should. A number of cool chunks of sample code would go a long way towards having more developers adopt it.
<strong>AltiVec really does kick some serious ass -- I think Apple hasn't really evangelized using it as much as they should. A number of cool chunks of sample code would go a long way towards having more developers adopt it.</strong><hr></blockquote>
I've seen some very nice examples. They were interesting to me and easy to understand. I'm not even into low-level coding, so that means they were pretty good. Go take a look around at altivec.org and use google. If you are really interested into low-level optimizations, I think you can find enough examples. Of course, you shouldn't assume that Altivec programming is easy. Like all performance optimizations, they take a lot of knowledge and experience to do right.
What I would really want is for Apple to buy VAST, an autovectorizing code optimizer. This will let every app use Altivec without any effort for the programmer (although a programmer can always do better of course).
I've seen some very nice examples. They were interesting to me and easy to understand. I'm not even into low-level coding, so that means they were pretty good. Go take a look around at altivec.org and use google. If you are really interested into low-level optimizations, I think you can find enough examples. Of course, you shouldn't assume that Altivec programming is easy. Like all performance optimizations, they take a lot of knowledge and experience to do right.
What I would really want is for Apple to buy VAST, an autovectorizing code optimizer. This will let every app use Altivec without any effort for the programmer (although a programmer can always do better of course).
vast is a source to source optimizer. it would not automatically let every app use altivec (with no effort for the programmer). if apple were to buy it, and if apple were to distribute it free to all developers, and if all developers were to use it, most apps would not see any significant performance increase.
p.s. altivec code really isn't that difficult. it does require a little learning and good design, but mostly it requires the willingness to write special case code.
<strong>VAST is a source to source optimizer. it would not automatically let every app use altivec (with no effort for the programmer).</strong><hr></blockquote>
Yes, it would. You either add it to the compilation pipeline or include it in gcc and CodeWarrior. The developer wouldn't need to do everything different, just press compile in ProjectBuilder or CW and the optimizer is automatically called.
[quote]<strong>most apps would not see any significant performance increase.</strong><hr></blockquote>
The creators of VAST disagree. They claim a 25% to 30% boost for large regular apps. Unfortunately I cannot afford a copy, nor do I have the source code of Office, so I can't check the figures. Even if this were exaggerated, it might already be worth it for the apps that do get optimized. I do know that some people on Apple's scitech mailing list are very interested. Those are the people that use hundreds of machines and influence the buying decisions for universities and R&D labs.
[quote]<strong>P.S. Altivec code really isn't that difficult. It does require a little learning and good design, but mostly it requires the willingness to write special case code.</strong><hr></blockquote>
Well, Graig Hunter seems to agree with you, so maybe I was wrong:
[quote]As someone who was in the same boat a few years ago (and who currently qualifies as an AltiVec dummy) let me point out a few things that might help. First is Apple's AltiVec page, at:
The goal of the article was to make it look simple, and it really is simple when you come right down to it. You can get better performance with additional tweaking, but the basic invocation of AltiVec is very simple.
Third, post to this list with specific questions once you get the basics figured out. We have a good assortment of AltiVec expertise here, from basic to advanced.
Dr. Craig A. Hunter
NASA Langley Research Center<hr></blockquote>
On the other hand, we have this Caltech guy:
[quote]
Part of the difficulty in writing a beginner level website for AltiVec is that there is so little to actually learn to use AltiVec in C. There isn't much to say. To a first order approximation, it should be no harder than learning to use any other C API like the ANSI math libraries. There are functions that operate on data. You call them. Work gets done. Simple documentation about what each "function" does should be sufficient. Most beginners expect AltiVec to be more difficult than it is. The C API is documented here:
That said, there are always gotchas, and getting high performance out of AltiVec is not this simple. First of all, thinking in vectors as opposed to scalars is a mind change akin to the difference between procedural and object oriented programming. You will need to master this if you want your code to truly scale with the increased parallelism in the vector unit. This difference is covered here:
There are a number of things about AltiVec which are different from standard types you are used to in C. Type coercion operates a little bit differently. (Bits never change.) AltiVec makes use of function overloading, even in C, where this is normally not a feature. These things should not be a problem. The biggest difference is that alignment is not handled automatically for you. This means that if you try to load a vector from an address that is not 16 byte aligned, you will get the vector from the nearest address that is 16 byte aligned and less than the unaligned address. (Addr & ~15)
In addition, writing high performance code -- any high performance code -- has always involved a whole set of techniques that many programmers don't learn very well, at least not right away. Many choose to rely instead on hearsay and speculation about what is and what is not likely to make things fast. Most such speculation and hearsay is wrong, relying on 20 year old assumptions about basic computer design that are no longer true on modern processors. The fundamental thing to emphasize in this area is that you must take a scientific approach to optimizing code. Run experiments to determine what is slow, fix those things, repeat. Never guess what to optimize. Never guess why a particular function is slow. You will just waste time. Apple has a suite of advanced performance tools called CHUD that will tell you exactly what is slow and why from the system wide level down to specific instructions. You should learn to use these.
The biggest thing to realize is that it doesn't matter how fast your code is if you can't get data into the processor fast enough. This is a substantial problem on its own, and one you need to solve first before spending time rewriting the code itself to execute more quickly is going to be effective.
[quote]Yes, it would. You either add it to the compilation pipeline or include it in gcc and CodeWarrior. The developer wouldn't need to do everything different, just press compile in ProjectBuilder or CW and the optimizer is automatically called.<hr></blockquote>
this approach is fraught with problems, including but not limited to longer compile times and confusion in debugging. i can't see anyone believing this would be a good thing for day to day production. this would also require apple purchase/licence the technology only to give it away. since this would cripple veridian's ability to sell it, it wouldn't be cheap. would most developers like a free copy to use? absolutely! would it be a practical move for apple? not that i can see.
[quote]The creators of VAST disagree. They claim a 25% to 30% boost for large regular apps<hr></blockquote>
it is dependent on the application. i feel comfortable saying 'most apps', since they (like office) share 2 characteristics. they spend most of their time waiting for user interaction and they don't work on large chunks of data at any given time. since they don't work on large chunks of data, they aren't really good candidates for vectorization. and since he are constantly waiting for user interaction, what little benefit is gained in vectorization is lost. vast is a fine product but it can't do magic.
[quote]I do know that some people on the scitech list are very interested<hr></blockquote>
their applications could quite possibly benefit from vectorization. quite a bit of my own work is in scientific research and my altivec routines definitely enhance performance. unfortunately, in the greater scheme of things, the market for these apps is only a tiny fraction of the macintosh software universe.
btw, 'this Caltech guy' ian is now employed by apple.
this getting pretty far afield. i suggest if you want to continue, we move it to pm.
I just think this Mac OS X on Intel is a bunch of horse sh*t. Steve Jobs didn't like the clones why would he want the Mac OS on Intel? It just doesn't wash.
If they moved to Intel the same reasons not to would apply to that as the clone business. It would effectively kill their hardware business. And one more : No one would develop for the Mac anymore.
If by some wild chance this is for real it's probably just what Steve Jobs said........an option. I think it more likely that they will find another chip maker or Motorola has something up their sleeve that no one knows about. Yes, this can happen. Especially since they clamped down on the rumor sites. Look around nobody knows anything.
I think this is one of those wild fire rumors that has no basis. Hell, Apple's probably encouraging it.
[QB]I just think this Mac OS X on Intel is a bunch of horse sh*t. Steve Jobs didn't like the clones why would he want the Mac OS on Intel? It just doesn't wash.
If they moved to Intel the same reasons not to would apply to that as the clone business. It would effectively kill their hardware business. And one more : No one would develop for the Mac anymore.
If by some wild chance this is for real it's probably just what Steve Jobs said........an option. I think it more likely that they will find another chip maker or Motorola has something up their sleeve that no one knows about. Yes, this can happen. Especially since they clamped down on the rumor sites. Look around nobody knows anything.
I think this is one of those wild fire rumors that has no basis. Hell, Apple's probably encouraging it.
[QB]<hr></blockquote>
Did you even read the half dozen messages above that explain why you have made several invalid assumptions? Moki's proposed direction has nothing to do with clones or Apple licensing the OS. It only has to do with Apple introducing an additional line of machines which happen to use an x86 processor.
Over and over and over people assume that "OSX on x86" must mean "OSX will run on any Wintel clone," which I agree would be a bad idea. It would knock Apple almost completely out of the hardware business. But Apple can make Powermacs with Intel or AMD x86 chips, capable of running some version or other of OSX, without making a version of OSX that can be run on a crappy Dell.
<strong>Over and over and over people assume that "OSX on x86" must mean "OSX will run on any Wintel clone," which I agree would be a bad idea. It would knock Apple almost completely out of the hardware business. But Apple can make Powermacs with Intel or AMD x86 chips, capable of running some version or other of OSX, without making a version of OSX that can be run on a crappy Dell.</strong><hr></blockquote>
All true.
But it won't be cheaper.
And it won't be faster than the Dell.
So why anyone would be 'hoping' for it is the confusing bit.
With $4B, it would seem Apple could buy whatever they might want for some time 2-3 years or so after they decide they _have_ to jump ship. If they didn't decide that after sitting at 500MHz for 18mo, Steve needs to leave the RDF for a bit. The whole 'Cell' venture is a $1B venture as I understand it. I could see Apple pitching money into that for a top end one. I could see them asking IBM very nicely for a gutted Power4-like chip (NOT a Power4, it'd have to be custom designed.)
Well, there seems to be a lot going on in this thread, but according to the MacNN main page another analyst has said that he fully expects to see "Intel Inside" (tm) Macs.
There might be something to this, it's not exactly something that a publicly traded company could flatly deny (if they were indeed planning it) and Steve's recent comments have NOT been unequivocal denials.
OS X will never go to say X86 yes its dead we all know that...
What I would want is that it work work on any computer you want. I deal with a lot of telecomm's and ISP's for their advertising and what I mean say if Cisco made the best routers or some new fangled equipment for super cheap, and they wanted to sign a deal to sell them to the whole company exclusively, they couldn't because a lot of companies won't let their business hinge on a sole supplier. We compete for a lot of business and this is always a reason why we get some stupid competitor that thinks they can do what we do.
Its just sad cuz OS X rocks so much, i know they don't make much money on software, but i wish they could switch their business model around. I have been asked so much by people wanting to run it on their PC at home, and I say buy a mac, so they rant to me about how having to buy a computer to run cool software and it sux, I think that you should have a choice.
If they did release OS X to the open market directly against windows, and it would work on any computer just think of the turmoil.
If they did this though they would have to put product keys in their stuff just to try and keep a cap on it all.
THey could still make hardware and you bet it would be better than most PC suppliers, but I mean if Dell made mac compatible computers and though they sold them for cheaper they would have to pay licensing for OS X, just think of the gravy train for apple.
in all I think it would bring a lot of innovation through all facets of the industry. Not just Mac Software and Mac Hardware, but for anybody and anyone.
<strong>What I would want is that it work work on any computer you want. I deal with a lot of telecomm's and ISP's for their advertising and what I mean say if Cisco made the best routers or some new fangled equipment for super cheap, and they wanted to sign a deal to sell them to the whole company exclusively, they couldn't because a lot of companies won't let their business hinge on a sole supplier. We compete for a lot of business and this is always a reason why we get some stupid competitor that thinks they can do what we do.
Its just sad cuz OS X rocks so much, i know they don't make much money on software, but i wish they could switch their business model around. I have been asked so much by people wanting to run it on their PC at home, and I say buy a mac, so they rant to me about how having to buy a computer to run cool software and it sux, I think that you should have a choice.</strong><hr></blockquote>
I'm only going to say this once.
Apple is a hardware company.
[quote]<strong>If they did release OS X to the open market directly against windows, and it would work on any computer just think of the turmoil.
If they did this though they would have to put product keys in their stuff just to try and keep a cap on it all.</strong><hr></blockquote>
No, they'd have to close up shop and go home since they couldn't make enough money selling software to continue to exist. There's no way anyone would buy Apple's OS on a Dell when Windows was available for it.
[quote]<strong>THey could still make hardware and you bet it would be better than most PC suppliers, but I mean if Dell made mac compatible computers and though they sold them for cheaper they would have to pay licensing for OS X, just think of the gravy train for apple.</strong><hr></blockquote>
Sure. Right up until MS said drop the Mac OS or Pay triple for Windows. Guess what? Dell's not going to upset it's gravy train just for Apple.
[ 08-07-2002: Message edited by: Tomb of the Unknown ]</p>
if apple is a hardware company? then why is the majority of its R&D on software?
If they are a hardware company then why do they make the operating system? or all the iapps? they don't manufacture anything but logic boards. If they are just a hardware company they wouldn't be any better than dell. Just another slap some compenents together and lets make another nasty, excuse to check e-mail and play video games PC.
The only reason you use a mac is cuz of the great software that is on it.
OS X is on NeXt correct? and NeXt's greatest benefit was coding for one, and working for all.
Now I know now that apples business model is solely on some hardware. But I bet you that with recent developments in apples advertising, and their overall emphasis on how OS X makes things easy, that they will push software more over than hardware. Look at the recent services such as .mac and having us upgrade to 10.2, its simply because they are finding their software will be their new beer ticket.
<strong>if apple is a hardware company? then why is the majority of its R&D on software?
[...]
The only reason you use a mac is cuz of the great software that is on it.</strong><hr></blockquote>
That doesn't describe me. The Mac is not the OS, and it's not the hardware. It's the whole Mac.
Apple is a computer maker. Most of their R&D might be spent on software, but most of their revenue comes from hardware sales. More to the point, Apple software cannot be neatly separated from their hardware. I don't mean that technically, because it's of course possible to port OS X to any hardware with the oomph to run it. I mean it in terms of considering the hardware and software as a single functional unit. Apple software is engineered with Apple hardware in mind, and vice versa. iMovie's creation and release was contingent upon AltiVec and Firewire, and the imminent SuperDrive. Rendezvous ties into the Mac's pervasive Ethernet, and also into AirPort. This is what Apple refers to as the "whole widget." The kind of integration and interoperability becomes much, much more difficult if you don't know what kind of hardware your OS and your apps will run on: A lot of the frustration and complexity attributed to Windows is due to this lack of control (one of the people I work with just bought a desktop PC that supplies ethernet via a PCI card!). If Mac OS X suddenly could run on any PC, it would inherit those problems: IRQs exist in hardware, for example. And Apple would have to do a lot of extra work to gracefully handle cases where a piece of hardware that they had previously simply assumed the presence of was missing, or inadequate.
It is not an overstatement to say that Apple's ability to offer consumer DV, and seamless networking, and true plug and play, and all the other things they are known for, is due precisely to their ability to design hardware and software in tandem. Take that away, and the Mac as a platform loses a lot of its appeal.
The idea of a Mac-with-an-x86 is more reasonable (which, IMO, is not saying much), but would-be adopters would still have to purchase a machine (until someone reverse engineered the board - but that's another kettle of fish).
PS. Does this help, moki?</strong><hr></blockquote>
Well, I'm sure it'll help someone -- I've been doing AltiVec code for some time now, and I absolutely love it. The number of apps that use it are still miniscule, though, and honestly, Apple doesn't push the technology to developers all that much.
<strong>if apple is a hardware company? then why is the majority of its R&D on software?</strong><hr></blockquote>
Maybe because it's not?
[quote]<strong>If they are a hardware company then why do they make the operating system? or all the iapps? they don't manufacture anything but logic boards. If they are just a hardware company they wouldn't be any better than dell. Just another slap some compenents together and lets make another nasty, excuse to check e-mail and play video games PC.
The only reason you use a mac is cuz of the great software that is on it.</strong><hr></blockquote>
See Amorph's excellent summary of why your assumptions are off base here.
[quote]<strong>OS X is on NeXt correct?</strong><hr></blockquote>
No, not correct. there is no NeXT platform and Mac OS X has always been a Mac OS (hence the name).
[quote]<strong>Now I know now that apples business model is solely on some hardware. But I bet you that with recent developments in apples advertising, and their overall emphasis on how OS X makes things easy, that they will push software more over than hardware. Look at the recent services such as .mac and having us upgrade to 10.2, its simply because they are finding their software will be their new beer ticket.</strong><hr></blockquote>
I can see why you would think so, but your first sentence sums up everything else that follows. Apple does all of that to do what? Answer: to sell Macs.
Moki: As president of a company that sees Apple as a competitor in your market space, I think you can be forgiven for not agreeing with my position.
[ 08-08-2002: Message edited by: Tomb of the Unknown ]</p>
Comments
<strong>It is normally called 'asynchronous' chip. No master clock synchronization needed on the whole chip. One big advantage of it is less power consumption.</strong><hr></blockquote>
Since this may go over your head (or at least the head of most other readers), I'll explain it a bit better:
A regular chip uses a clock for every instruction. One part of the chip can only hand off something to another part of the chip when the clock ticks. That means that the hardware can't just be optimized to take as little time as possible. Hardware designers instead have to divide instructions into steps and sync these to the clock. This is very hard to do and not very efficient. Furthermore, once you get into higher clockspeeds, a good part of the chip (and current) is used by the circuitry for the clock.
Asynchronous chips just deliver the result when it is done. Of course, this sounds much easier than it actually is. The receiver must be able to proces the signal when it is presented and it must be able to determine when the end result is available. The latter is difficult because circuitry often takes on various values during computation. There are various solutions for that, but that would go too far for this brief explanation.
<strong> [quote]And yeah, intel is investing a lot on this technology. Not sure whether they used it or not.</strong><hr></blockquote>
Most research is going into smartcards now. Those are the chips that need to be most efficient with power. Fully asynchronous CPU's are quite a way off AFAIK (unfortunately). It has a lot to do with the retraining of engineers. It's quite different, so you cannot expect a fast switch. A lot of experience has to be gained first.
[ 08-06-2002: Message edited by: wfzelle ]</p>
Changing platforms isn't increasing option, it's simply a change.
Changing PPC suppliers can still get you into a situation Apple is in now, when the 60x line was replaced with an embedded-orientated 7xx/7xxx line.
Bundles are where it's at. Screw Mac OS 9 style FAT binaries, bring on bundles with multiple compiles in the /content/MacOS folder (which I assume is how it would be done).
There is the AltiVec question though (Paging Moki, Programmer). Would developers bother to optimise for AltiVec anymore if only, say 1/2 Apple's pro machines used it? Because if Apple went multiplatform, the PowerPC only really competes when VMX/AltiVec/VE is used.
Barto
EDIT: There is a hack to get Mac OS X to run on the AmigaOne G3, and there will possibly be hacks for other PPCs in future. Like Moki said, make it hard for the 90%, because you'll never get the 10%.
[ 08-07-2002: Message edited by: Barto ]</p>
<strong>
There is the AltiVec question though (Paging Moki, Programmer). Would developers bother to optimise for AltiVec anymore if only, say 1/2 Apple's pro machines used it? Because if Apple went multiplatform, the PowerPC only really competes when VMX/AltiVec/VE is used.</strong><hr></blockquote>
The unfortunate fact is that very few programs/programmers optimize for AltiVec now anyway. Apple certainly does, because they need the underlying OS to be as fast as possible, but the few programs that do optimize for AltiVec are the ones that need every ounce of performance, such as PhotoShop.
The long and the short of it is that applications that really need the extra performance will still optimize for it. On the x86 side, they'll use the SIMD mnemonics there, on the PPC side, they'll use AltiVec.
AltiVec really does kick some serious ass -- I think Apple hasn't really evangelized using it as much as they should. A number of cool chunks of sample code would go a long way towards having more developers adopt it.
<strong>AltiVec really does kick some serious ass -- I think Apple hasn't really evangelized using it as much as they should. A number of cool chunks of sample code would go a long way towards having more developers adopt it.</strong><hr></blockquote>
I've seen some very nice examples. They were interesting to me and easy to understand. I'm not even into low-level coding, so that means they were pretty good.
What I would really want is for Apple to buy VAST, an autovectorizing code optimizer. This will let every app use Altivec without any effort for the programmer (although a programmer can always do better of course).
[ 08-07-2002: Message edited by: wfzelle ]</p>
<strong>
I've seen some very nice examples. They were interesting to me and easy to understand. I'm not even into low-level coding, so that means they were pretty good.
What I would really want is for Apple to buy VAST, an autovectorizing code optimizer. This will let every app use Altivec without any effort for the programmer (although a programmer can always do better of course).
[ 08-07-2002: Message edited by: wfzelle ]</strong><hr></blockquote>
vast is a source to source optimizer. it would not automatically let every app use altivec (with no effort for the programmer). if apple were to buy it, and if apple were to distribute it free to all developers, and if all developers were to use it, most apps would not see any significant performance increase.
p.s. altivec code really isn't that difficult. it does require a little learning and good design, but mostly it requires the willingness to write special case code.
<strong>VAST is a source to source optimizer. it would not automatically let every app use altivec (with no effort for the programmer).</strong><hr></blockquote>
Yes, it would. You either add it to the compilation pipeline or include it in gcc and CodeWarrior. The developer wouldn't need to do everything different, just press compile in ProjectBuilder or CW and the optimizer is automatically called.
[quote]<strong>most apps would not see any significant performance increase.</strong><hr></blockquote>
The creators of VAST disagree. They claim a 25% to 30% boost for large regular apps. Unfortunately I cannot afford a copy, nor do I have the source code of Office, so I can't check the figures. Even if this were exaggerated, it might already be worth it for the apps that do get optimized. I do know that some people on Apple's scitech mailing list are very interested. Those are the people that use hundreds of machines and influence the buying decisions for universities and R&D labs.
[quote]<strong>P.S. Altivec code really isn't that difficult. It does require a little learning and good design, but mostly it requires the willingness to write special case code.</strong><hr></blockquote>
Well, Graig Hunter seems to agree with you, so maybe I was wrong:
[quote]As someone who was in the same boat a few years ago (and who currently qualifies as an AltiVec dummy) let me point out a few things that might help. First is Apple's AltiVec page, at:
<a href="http://developer.apple.com/hardware/ve/index.html"; target="_blank">http://developer.apple.com/hardware/ve/index.html</a>;
This hasn't always been useful, but Apple keeps improving it. Also, check out my recent article on AltiVec at O'Reilly:
<a href="http://www.macdevcenter.com/pub/a/mac/2002/04/05/altivec.html"; target="_blank">http://www.macdevcenter.com/pub/a/mac/2002/04/05/altivec.html</a>;
The goal of the article was to make it look simple, and it really is simple when you come right down to it. You can get better performance with additional tweaking, but the basic invocation of AltiVec is very simple.
Third, post to this list with specific questions once you get the basics figured out. We have a good assortment of AltiVec expertise here, from basic to advanced.
Dr. Craig A. Hunter
NASA Langley Research Center<hr></blockquote>
On the other hand, we have this Caltech guy:
[quote]
Part of the difficulty in writing a beginner level website for AltiVec is that there is so little to actually learn to use AltiVec in C. There isn't much to say. To a first order approximation, it should be no harder than learning to use any other C API like the ANSI math libraries. There are functions that operate on data. You call them. Work gets done. Simple documentation about what each "function" does should be sufficient. Most beginners expect AltiVec to be more difficult than it is. The C API is documented here:
\t<a href="http://e-www.motorola.com/brdata/PDFDB/docs/ALTIVECPIM.pdf"; target="_blank">http://e-www.motorola.com/brdata/PDFDB/docs/ALTIVECPIM.pdf</a>;
That said, there are always gotchas, and getting high performance out of AltiVec is not this simple. First of all, thinking in vectors as opposed to scalars is a mind change akin to the difference between procedural and object oriented programming. You will need to master this if you want your code to truly scale with the increased parallelism in the vector unit. This difference is covered here:
\t<a href="http://developer.apple.com/hardware/ve/simd.html"; target="_blank">http://developer.apple.com/hardware/ve/simd.html</a>;
\t<a href="http://developer.apple.com/hardware/ve/data_handling.html"; target="_blank">http://developer.apple.com/hardware/ve/data_handling.html</a>;
There are a number of things about AltiVec which are different from standard types you are used to in C. Type coercion operates a little bit differently. (Bits never change.) AltiVec makes use of function overloading, even in C, where this is normally not a feature. These things should not be a problem. The biggest difference is that alignment is not handled automatically for you. This means that if you try to load a vector from an address that is not 16 byte aligned, you will get the vector from the nearest address that is 16 byte aligned and less than the unaligned address. (Addr & ~15)
\t<a href="http://developer.apple.com/hardware/ve/model.html"; target="_blank">http://developer.apple.com/hardware/ve/model.html</a>;
\t<a href="http://developer.apple.com/hardware/ve/model_details.html"; target="_blank">http://developer.apple.com/hardware/ve/model_details.html</a>;
\t<a href="http://developer.apple.com/hardware/ve/alignment.html"; target="_blank">http://developer.apple.com/hardware/ve/alignment.html</a>;
In addition, writing high performance code -- any high performance code -- has always involved a whole set of techniques that many programmers don't learn very well, at least not right away. Many choose to rely instead on hearsay and speculation about what is and what is not likely to make things fast. Most such speculation and hearsay is wrong, relying on 20 year old assumptions about basic computer design that are no longer true on modern processors. The fundamental thing to emphasize in this area is that you must take a scientific approach to optimizing code. Run experiments to determine what is slow, fix those things, repeat. Never guess what to optimize. Never guess why a particular function is slow. You will just waste time. Apple has a suite of advanced performance tools called CHUD that will tell you exactly what is slow and why from the system wide level down to specific instructions. You should learn to use these.
<a href="http://developer.apple.com/tools/debuggers.html"; target="_blank">http://developer.apple.com/tools/debuggers.html</a>;
Things to think about when writing fast code on a modern processor are covered here:
\t<a href="http://developer.apple.com/hardware/ve/throughput_vs_latency.html"; target="_blank">http://developer.apple.com/hardware/ve/throughput_vs_latency.html</a>;
\t<a href="http://developer.apple.com/hardware/ve/code_optimization.html"; target="_blank">http://developer.apple.com/hardware/ve/code_optimization.html</a>;
The biggest thing to realize is that it doesn't matter how fast your code is if you can't get data into the processor fast enough. This is a substantial problem on its own, and one you need to solve first before spending time rewriting the code itself to execute more quickly is going to be effective.
\t<a href="http://developer.apple.com/hardware/ve/performance_memory.html"; target="_blank">http://developer.apple.com/hardware/ve/performance_memory.html</a>;
Ian Ollmann, Ph.D.<hr></blockquote>
PS. Does this help, moki?
[ 08-07-2002: Message edited by: wfzelle ]</p>
this approach is fraught with problems, including but not limited to longer compile times and confusion in debugging. i can't see anyone believing this would be a good thing for day to day production. this would also require apple purchase/licence the technology only to give it away. since this would cripple veridian's ability to sell it, it wouldn't be cheap. would most developers like a free copy to use? absolutely! would it be a practical move for apple? not that i can see.
[quote]The creators of VAST disagree. They claim a 25% to 30% boost for large regular apps<hr></blockquote>
it is dependent on the application. i feel comfortable saying 'most apps', since they (like office) share 2 characteristics. they spend most of their time waiting for user interaction and they don't work on large chunks of data at any given time. since they don't work on large chunks of data, they aren't really good candidates for vectorization. and since he are constantly waiting for user interaction, what little benefit is gained in vectorization is lost. vast is a fine product but it can't do magic.
[quote]I do know that some people on the scitech list are very interested<hr></blockquote>
their applications could quite possibly benefit from vectorization. quite a bit of my own work is in scientific research and my altivec routines definitely enhance performance. unfortunately, in the greater scheme of things, the market for these apps is only a tiny fraction of the macintosh software universe.
btw, 'this Caltech guy' ian is now employed by apple.
this getting pretty far afield. i suggest if you want to continue, we move it to pm.
If they moved to Intel the same reasons not to would apply to that as the clone business. It would effectively kill their hardware business. And one more : No one would develop for the Mac anymore.
If by some wild chance this is for real it's probably just what Steve Jobs said........an option. I think it more likely that they will find another chip maker or Motorola has something up their sleeve that no one knows about. Yes, this can happen. Especially since they clamped down on the rumor sites. Look around nobody knows anything.
I think this is one of those wild fire rumors that has no basis. Hell, Apple's probably encouraging it.
[ 08-07-2002: Message edited by: jimmac ]</p>
[QB]I just think this Mac OS X on Intel is a bunch of horse sh*t. Steve Jobs didn't like the clones why would he want the Mac OS on Intel? It just doesn't wash.
If they moved to Intel the same reasons not to would apply to that as the clone business. It would effectively kill their hardware business. And one more : No one would develop for the Mac anymore.
If by some wild chance this is for real it's probably just what Steve Jobs said........an option. I think it more likely that they will find another chip maker or Motorola has something up their sleeve that no one knows about. Yes, this can happen. Especially since they clamped down on the rumor sites. Look around nobody knows anything.
I think this is one of those wild fire rumors that has no basis. Hell, Apple's probably encouraging it.
[QB]<hr></blockquote>
Did you even read the half dozen messages above that explain why you have made several invalid assumptions? Moki's proposed direction has nothing to do with clones or Apple licensing the OS. It only has to do with Apple introducing an additional line of machines which happen to use an x86 processor.
<strong>Over and over and over people assume that "OSX on x86" must mean "OSX will run on any Wintel clone," which I agree would be a bad idea. It would knock Apple almost completely out of the hardware business. But Apple can make Powermacs with Intel or AMD x86 chips, capable of running some version or other of OSX, without making a version of OSX that can be run on a crappy Dell.</strong><hr></blockquote>
All true.
But it won't be cheaper.
And it won't be faster than the Dell.
So why anyone would be 'hoping' for it is the confusing bit.
With $4B, it would seem Apple could buy whatever they might want for some time 2-3 years or so after they decide they _have_ to jump ship. If they didn't decide that after sitting at 500MHz for 18mo, Steve needs to leave the RDF for a bit. The whole 'Cell' venture is a $1B venture as I understand it. I could see Apple pitching money into that for a top end one. I could see them asking IBM very nicely for a gutted Power4-like chip (NOT a Power4, it'd have to be custom designed.)
There might be something to this, it's not exactly something that a publicly traded company could flatly deny (if they were indeed planning it) and Steve's recent comments have NOT been unequivocal denials.
What I would want is that it work work on any computer you want. I deal with a lot of telecomm's and ISP's for their advertising and what I mean say if Cisco made the best routers or some new fangled equipment for super cheap, and they wanted to sign a deal to sell them to the whole company exclusively, they couldn't because a lot of companies won't let their business hinge on a sole supplier. We compete for a lot of business and this is always a reason why we get some stupid competitor that thinks they can do what we do.
Its just sad cuz OS X rocks so much, i know they don't make much money on software, but i wish they could switch their business model around. I have been asked so much by people wanting to run it on their PC at home, and I say buy a mac, so they rant to me about how having to buy a computer to run cool software and it sux, I think that you should have a choice.
If they did release OS X to the open market directly against windows, and it would work on any computer just think of the turmoil.
If they did this though they would have to put product keys in their stuff just to try and keep a cap on it all.
THey could still make hardware and you bet it would be better than most PC suppliers, but I mean if Dell made mac compatible computers and though they sold them for cheaper they would have to pay licensing for OS X, just think of the gravy train for apple.
in all I think it would bring a lot of innovation through all facets of the industry. Not just Mac Software and Mac Hardware, but for anybody and anyone.
<strong>What I would want is that it work work on any computer you want. I deal with a lot of telecomm's and ISP's for their advertising and what I mean say if Cisco made the best routers or some new fangled equipment for super cheap, and they wanted to sign a deal to sell them to the whole company exclusively, they couldn't because a lot of companies won't let their business hinge on a sole supplier. We compete for a lot of business and this is always a reason why we get some stupid competitor that thinks they can do what we do.
Its just sad cuz OS X rocks so much, i know they don't make much money on software, but i wish they could switch their business model around. I have been asked so much by people wanting to run it on their PC at home, and I say buy a mac, so they rant to me about how having to buy a computer to run cool software and it sux, I think that you should have a choice.</strong><hr></blockquote>
I'm only going to say this once.
Apple is a hardware company.
[quote]<strong>If they did release OS X to the open market directly against windows, and it would work on any computer just think of the turmoil.
If they did this though they would have to put product keys in their stuff just to try and keep a cap on it all.</strong><hr></blockquote>
No, they'd have to close up shop and go home since they couldn't make enough money selling software to continue to exist. There's no way anyone would buy Apple's OS on a Dell when Windows was available for it.
[quote]<strong>THey could still make hardware and you bet it would be better than most PC suppliers, but I mean if Dell made mac compatible computers and though they sold them for cheaper they would have to pay licensing for OS X, just think of the gravy train for apple.</strong><hr></blockquote>
Sure. Right up until MS said drop the Mac OS or Pay triple for Windows. Guess what? Dell's not going to upset it's gravy train just for Apple.
[ 08-07-2002: Message edited by: Tomb of the Unknown ]</p>
If they are a hardware company then why do they make the operating system? or all the iapps? they don't manufacture anything but logic boards. If they are just a hardware company they wouldn't be any better than dell. Just another slap some compenents together and lets make another nasty, excuse to check e-mail and play video games PC.
The only reason you use a mac is cuz of the great software that is on it.
OS X is on NeXt correct? and NeXt's greatest benefit was coding for one, and working for all.
Now I know now that apples business model is solely on some hardware. But I bet you that with recent developments in apples advertising, and their overall emphasis on how OS X makes things easy, that they will push software more over than hardware. Look at the recent services such as .mac and having us upgrade to 10.2, its simply because they are finding their software will be their new beer ticket.
<strong>if apple is a hardware company? then why is the majority of its R&D on software?
[...]
The only reason you use a mac is cuz of the great software that is on it.</strong><hr></blockquote>
That doesn't describe me. The Mac is not the OS, and it's not the hardware. It's the whole Mac.
Apple is a computer maker. Most of their R&D might be spent on software, but most of their revenue comes from hardware sales. More to the point, Apple software cannot be neatly separated from their hardware. I don't mean that technically, because it's of course possible to port OS X to any hardware with the oomph to run it. I mean it in terms of considering the hardware and software as a single functional unit. Apple software is engineered with Apple hardware in mind, and vice versa. iMovie's creation and release was contingent upon AltiVec and Firewire, and the imminent SuperDrive. Rendezvous ties into the Mac's pervasive Ethernet, and also into AirPort. This is what Apple refers to as the "whole widget." The kind of integration and interoperability becomes much, much more difficult if you don't know what kind of hardware your OS and your apps will run on: A lot of the frustration and complexity attributed to Windows is due to this lack of control (one of the people I work with just bought a desktop PC that supplies ethernet via a PCI card!). If Mac OS X suddenly could run on any PC, it would inherit those problems: IRQs exist in hardware, for example. And Apple would have to do a lot of extra work to gracefully handle cases where a piece of hardware that they had previously simply assumed the presence of was missing, or inadequate.
It is not an overstatement to say that Apple's ability to offer consumer DV, and seamless networking, and true plug and play, and all the other things they are known for, is due precisely to their ability to design hardware and software in tandem. Take that away, and the Mac as a platform loses a lot of its appeal.
The idea of a Mac-with-an-x86 is more reasonable (which, IMO, is not saying much), but would-be adopters would still have to purchase a machine (until someone reverse engineered the board - but that's another kettle of fish).
<strong>
PS. Does this help, moki?</strong><hr></blockquote>
Well, I'm sure it'll help someone -- I've been doing AltiVec code for some time now, and I absolutely love it. The number of apps that use it are still miniscule, though, and honestly, Apple doesn't push the technology to developers all that much.
<strong>I'm only going to say this once.
Apple is a hardware company.
</strong><hr></blockquote>
(nah, better not)
[ 08-08-2002: Message edited by: moki ]</p>
<strong>if apple is a hardware company? then why is the majority of its R&D on software?</strong><hr></blockquote>
Maybe because it's not?
[quote]<strong>If they are a hardware company then why do they make the operating system? or all the iapps? they don't manufacture anything but logic boards. If they are just a hardware company they wouldn't be any better than dell. Just another slap some compenents together and lets make another nasty, excuse to check e-mail and play video games PC.
The only reason you use a mac is cuz of the great software that is on it.</strong><hr></blockquote>
See Amorph's excellent summary of why your assumptions are off base here.
[quote]<strong>OS X is on NeXt correct?</strong><hr></blockquote>
No, not correct. there is no NeXT platform and Mac OS X has always been a Mac OS (hence the name).
[quote]<strong>Now I know now that apples business model is solely on some hardware. But I bet you that with recent developments in apples advertising, and their overall emphasis on how OS X makes things easy, that they will push software more over than hardware. Look at the recent services such as .mac and having us upgrade to 10.2, its simply because they are finding their software will be their new beer ticket.</strong><hr></blockquote>
I can see why you would think so, but your first sentence sums up everything else that follows. Apple does all of that to do what? Answer: to sell Macs.
Moki: As president of a company that sees Apple as a competitor in your market space, I think you can be forgiven for not agreeing with my position.
[ 08-08-2002: Message edited by: Tomb of the Unknown ]</p>