Slashdot is powered by your submissions, so send in your scoop

 


Forgot your password?
Close
typodupeerror
×
Intel

Linux Running On Intel XScale CPU 73

Posted by timothy from the small-packages dept.
Erik Mouw writes: "Just want to let you know that Nicolas Pitre (nico@cam.org) and I got Linux running on the Intel 80200 XScale CPU. Nico did the largest part of the work during the past couple of weeks, and we did the final bug fixes in a hotel room in New York. The official announce of the patch is available at the linux-arm-announce mailing list archive." The board was on display at the MontaVista booth at Linux World Expo, one of the several tiny Linux set-ups on tantalizing display around the show floor for everything from vending machines to cheap PDAs. (No Yopy in sight, though, despite the fact that development models are available for sale.) Congratulations to Erik and Nicolas.
This discussion has been archived. No new comments can be posted.

Linux Running On Intel XScale CPU More

Linux Running On Intel XScale CPU

Comments Filter:
  • It is kind of fun to push underpowered hardware to the breaking point. Maybe some sort of high end floating point emulation can be added to ease developer's pains?
    --
    Peace,
    Lord Omlette
    ICQ# 77863057
  • I have a developer model of the Yopy and let me just say it rocks even for a dev model. And the company seems like its got a real cool philosphy of open source behind it and so far has been real supportive to the yopy developers. But seriosuly i think when this thing actually comes out if they can keep the prices down they will be nice little machines for a variety of purposes. just my 2 cents though

  • Support for XScale's Thumb instructions? (Score:3)

    by Morgaine ( 4316 ) on Saturday February 03, 2001 @07:00PM (#458409)
    Are the Thumb and ARM instruction sets on the XScale mutually exclusive, or is there some way for Linux to use Thumb instructions --- for example in user-mode processes only?

    And an even more fundamental question: does gcc support the Thumb instruction set at all?

    Finally, what is the relationship between the XScale and Atmel's 91AT series, which also features a combination of ARM and Thumb instruction sets?

  • Re:leaving out floating point is stupid (Score:3)

    by q000921 ( 235076 ) on Saturday February 03, 2001 @09:04PM (#458410)
    It's funny how when DEC had it, everybody thought the StrongARM was the best thing since sliced bread.

    Until recently, handhelds like the Palm and its WinCE clones didn't need much smarts. That's rapidly changing.

  • Just a question, if you really do have internal access to the X-scale stuff. This MHz based on voltage sounds suspiciously like asynchronous processing, in which changing the input voltage changes the rate at which signals pull up or down.

    Is it some clock like signal that is asynchronous? Or are portions of the control pipeline asynchronous? Or is this just terribly nifty synchronous logic?

    Geek dating! [bunnyhop.com]
  • No. Intel just started shipping the 80200, while the Jornada 820 already exists for quite some time. If a certain device isn't supported by Linux it is usually because the vendor refuses to give away programming information.

    One of the main reasons the Compaq iPaq has been successful in the Linux market is that Compaq opened up all programming information and actively participates on the iPaq mailing lists.


    Erik
  • I/O processors isn't all Intel is positioning this for. They are also talking about using it as a low-power chip for handhelds.

    And if you read Intel's sales pitch, they are talking about applications far beyond traditional I/O processing. Traditional I/O processing requires a little bit of bit pattern matching, maybe some error correction, some compression, a few traditional data structures, and copying lots of data around fast.

    For this chip, Intel is talking about "extremely complex applications", "processor intensive calculations", "rich content", and all that. And even in the domain of fairly traditional I/O, statistical and other numerical techniques become increasingly important: you need to predict traffic patterns, calculate optimal routes, predict resource utilization, model statistical distributions, etc. Coding and compression algorithms for media are also usually much more easily formulated with floating point.

  • I stand corrected (and my apologies for implying you are a moron), Intel's literature does mention handhelds. I was caught up in the sample RAID implementations to the point of overlooking the other applications suggested. I also incorrectly assumed you were tying the mention of the Yopy in with the port of Linux to this architecture, which share nothing but the ARM ISA.

    Now I'll ask you to reconsider your original post, from my perspective it echos the long running x86 vs PPC or CISC vs RISC arguements. Without your subsequent clarifying messages it read as an ill-informed attack on Intel and on the efforts to port Linux to this chip.

    Your original message comes through clearer now I think, why base a media capable PDA on this when you'll waste that extra processing (or battery) power on FP emulation.

  • .. that the work you and np were doing on the SA1100/1110 is gonna slow down significantly? I've been hoping for some time that we would see a unified UCB1200 driver come out so that we could do sound and touch screen (and modem) at the same time without IRQ conflicts/problems.
  • Don't bother with changing clock generators and the such, just bring Vcc lower. It's very cool. Even at full speed, it only draws a few watts.

    This is incorrect say's the manual [intel.com].

    Getting a lower power consumption from frequency and voltage scaling is not easy. I'm the author of the ARM-Linux frequency and voltage scaling driver and do a Ph.D. on it. The problem is that a processor does not know what speed to run in. The processor has to be explicitly set in a certain frequency mode. In Crusoe the code morphing software tries to guess a good speed. Voltage and frequency scaling technique has been show to work in 1994 by Mark Weiser, it is not new. The problem has always been for the software to know what speed to set the processor in.

    Read more details ... [tudelft.nl].

    Johan.

  • http://www.ti.com/calc/docs/faq/83faq067.htm

    without the extra %0D at the end...
  • people are always porting linux to more and more systems, and claiming it's a good platform for embedded systems, when it's just too damn big and bloated for it. so fanaticism gets in the way of good technical decisions and being able as a customer to get quality products.

    Please explain why you think Linux is a bad choice for say, the Tivo video recorder, or might not be a better choice than Windows for car automation.

    People port Linux to different platforms for the same reason that people are interested in using it for embedded applications: because the source code is open, can be customized as needed, and because it's extremely configurable.

    Point me to the wonderful open source embedded OS that you have in mind, and perhaps we'll all start using it.

  • What boards did you use? What's available? (Score:3)

    by billstewart ( 78916 ) on Sunday February 04, 2001 @12:14AM (#458419) Journal
    Sounds cool - for the vast majority of us who do very little high-speed floating point, and would much rather have DSP features like a fast MAC (:-), this would be a really good machine. What kind of hardware did you use - the Intel 80310 evaluation board [intel.com]? How much hardware did you have to add around it? Looks like there are a few PCI slots, ether and RAM, so it shouldn't be too hard to add a video an d a disk controller of some sort.

  • Yes, there *IS* a lot of "overlap" between the applications of the two. Regardless, my original statements hold true on how and who for the Pentium microprocessor is designed versus how and who for the StrongARM microcontroller is designed.

    Microprocessors range from simplistic, single-issue CPUs to the nine-issue, massively buffered K7. Microcontrollers range from basic 8-bit cores to the 32-bit superpipelined StrongARM with external, co-processor options.

    Your application will be the best judge of what to use.

    -- Bryan "TheBS" Smith

  • Ick, forget about the TI's. What about my HP 48GX? ;)
  • This is trollbait, but i'll bite regardless.
    On your point with respect to detecting hardware, "Any OS should be able to automatically detect hardware", have you ever considered even for a second that not all hardware cannot be reliably detected? Yes, that's right, that isa ne2k card that you plugged in your computer 5 minutes ago might just hang your system. Go ahead windows find my hardware right now for me.

    My mother recently got a computer, it has NT 5 on it. She comes across things she HATES about windows on a regular basis, she likes some of the things my unix system at home has. She has never owned a computer until recently. Your argument about unix (you mentioned linux specifically) as being 2 parts gui and 500 parts command line, this is rather a moot point considering windows is very similar if you want the same functionality as a unix system. You also
    seemed to have entirely ignored OS-X and NeXTstep (on which I am typing this now).

    In short, you should probably go back. Erase all your preconceptions and bias and then look again. Short term and then again long term.
    In the short term perhaps windows will win. But long term windows has no fucking chance.
  • 6809, not 68k
  • Started shipping development kits 2 weeks ago. Got mine already. Cool.

  • 80200 is a nice CPU.. (Score:5)

    by toastyman ( 23954 ) <toasty@dragondata.com> on Saturday February 03, 2001 @09:21PM (#458425) Homepage
    I've got one of the Intel 80200 evaluation boards on my desk at work. They're reallllly nice CPU's..

    They run existing ARM code. Intel has a "porting guide" that's only a few pages long, mentioning hardly used features of the original chips. The only big difference is that the don't support thumb mode anymore, which isn't a huge loss.

    They change their clock speed depending on the input voltage. Yes, you heard me right. Wanna slow the system down? Don't bother with changing clock generators and the such, just bring Vcc lower. It's very cool. Even at full speed, it only draws a few watts. And of course, no heatsink needed.

    It's faaaaaaaaaaast. As long as you don't need floating point, it's actually a very competitive chip. You can check intel's site for actual benchmarks, but I was very surprised.

    It's small. It's in a BGA package less than an inch square. The PCI-PCI bridge they used on the sample board is several times larger than the CPU itself, I had to look several times to find it.

    It's cheap. I don't think my NDA with Intel allows me to discuss the pricing we have, but.... They're cheap enough to put in a PDA for sure.

    I hope this chip really takes off, because there are so many cool things I'd love to do with it. :)

    -- Kevin
  • Again, this is a microcontroller, and NOT a general purpose "microprocessor."

    It's an ARM chip that Intel is positioning for "compute intensive" applications, "low power consumption", "rich media", and "handhelds".

    More importantly, the context of this posting is that someone booted Linux on it and that people built PDAs based on this processor, so it's not just being used as an I/O controller for RAIDs.

    Besides, I was under the impression most handwriting recognition was image processing anyway, which is usually integer based.

    Handwriting and speech recognition involve lots of statistical modeling, which is floating point based. You can convert those algorithms to fixed point, but that's a lot of work, since you need to make sure that you scale everything correctly.

    32-bit interpolated interger math is much faster than single-precision floating-point -- important for things like real-time speech recognition.

    I suggest you do some actual benchmarks on real processors. On a few processors, what you say is true. On most modern processors, you actually often get faster performance if you implement using FP because integer arithmetic (for loops and indexes) and FP arithmetic can happen in parallel.

  • if any one wants better information on youp go to http://www.gmate.co.kr which is the most up to date site for the yopy pretty sweet, Perl for all os's and micro patforms :)
  • Er, no. That's for PA-RISC processors. The HP48G (and 49G, for that matter) series runs using an NEC Saturn CPU at 4MHz.
  • The main issues were the architecture specific start code, interrupts, memory management. Oh, and some small changes in the MTD drivers to support the XScale.


    Erik
  • Ack, you're right... It doesn't support 26-bit mode.. Sorry, it's late here. :)

    -- Kevin
  • I'd assume that would be a better bet
  • Is this the XScale that can run at up to a 1000mhz, yet consumes less power than a 206mhz ARM (same CPU used in the Compaq IPAQ PocketPC). If such a CPU was made this year... and was very inexpensive, well, ok, its too good to be true. Lets stop dreaming.
  • Intel did not buy ARM. Intel bought DEC's StrongARM division (and their ARM licenses). ARM is still independent from intel, see www.arm.com

  • It installed just fine on mine ... maybe you mean it doesn't support 486SX / 386 (no FPU emulation) by default?
  • My company is also looking at the XScale...

    The logic is sychronous. All modern, high-speed processors are fed with a clock signal that is a fraction of the actual internal clock rate. Inside the part, a phase-locked loop (PLL) multiplies the clock speed up.

    I believe what the XScale does is dynamically change the clock multplier based on input voltage. So, it's a neat trick, but it isn't async logic or anything bleeding edge like that.

    This is not a bad thing. Consider what "bleeding edge" implies....
  • "Only, if one wishes to use Linux, then one needs the patience of a stone to find all the functionality of it."

    With how-many-years since win95 has been around I can find it very easy to understand that people forget what it was like for "ordinary people" to learn the windows GUI. I can also appreciate the fact that people still dont know how to use it and if they do, they dont know what good they are doing. I'm not talking about people that can open up Word or play a cd after they accidently closed CDPlayer. I'm talkin' about the people that say "Man, my computer sucks rotten worm turds, but it used to be so nice." You know, the people that havent defragged their win95 box since they upgraded from 3.1.

    "There seems to be no integration of design in the system, only a bunch of utilities that do not match, or work well together."

    Hmmm, maybe you should take a look at the inner-workings (or non-workings) of win2k (wow, is the entire OS a bug?). The entire active directory setup sucks major crud outta a garden hose their is no discerning between it and other major players in DNS (sounds like an analogy between the win kernel and the GUI). Now granted, im talking about a Server supposedly made for geeks, but, in responce to "However, the usefulness of Linux must derive from total introverts all coding in separate parts of the world, and trying to put a OS together with bubble gum, and tooth picks." it seems as though a bunch of introverts residing in a plant that houses billions of dollars of hard and software cant put an OS together that doesn't need to be patched with the aformentioned bubble gum and toothpicks.

  • You say a lot about why ARM/XScale processors are not used for general-purpose - and particularly mobile - computing, but not much about why they can not or should not be used in such applications. Seems to me that a processor that's a little slower than a Pentium/Athlon but that dissipates a ton less power/heat would be a big win in many general-purpose environments. Crusoe, which has similar characteristics, is already being touted for both mobile and high-density server applications - by people who know a lot more about pipelines and functional units than you do.

    So, the question is: what's wrong with considering an XScale processor for general-purpose use? Is there a real technical stumbling block involved, or is it just a matter of "you should follow the pack"?

  • there are so many cool things I'd love to do with it.

    Do you mean "cool" figuratively, or literally? Or both? ;-)

  • Since XScale uses the ARM ISA and Linux already supports ARM, I would assume only minor changes (drivers and model-specific stuff) are needed. But I can't resist asking anyway.

  • Linux on a calculator (Score:5)

    by whydna ( 9312 ) <.whydna. .at. .hotmail.com.> on Saturday February 03, 2001 @06:19PM (#458440)
    How much longer until I can get linux on my TI-92?? I can imagine that compile times might be a little high, but hey... at least it'd look like you're doing work in class... =)

    -Andy
  • Can imagine how much time you could waste with an xmame port to a TI calculator?

    -antipop

  • Running low on content lately, /.? Keep those babbling linux idiots all fueled up with stupid stories like these. Maybe they won't notice what a complete bunch of shit their OS is and how quickly it's fading away.

    Scott


    Your last name wouldn't happen to be Nudds, would it? :)

  • Stan (Score:4)

    by deran9ed ( 300694 ) on Saturday February 03, 2001 @07:35PM (#458443) Homepage

    Slashdot's gone cold I'm wondering why I got out of bed at all
    The morning rain clouds up my window and I can't see at all
    And even if I could it'll all be gray but your picture on my wall
    It reminds me, that it's not so bad -- it's not so bad

    Dear Rob, I wrote but you still ain't callin
    I left my email, my ICQ, and my yahoo chat at the bottom
    I sent two emails back in autumn, you must not-a got 'em
    There probably was a problem with your sendmail or somethin
    Sometimes I scribble email addees too sloppy when I jot 'em
    but anyways; fsck it, what's been up? Man how's your boxes?
    My boxes is linux too, I'm bout to be a compiler
    once I learn gcc,
    I'ma compile for miles ah
    I read about your Palm Pilot too I'm sorry
    I had a friend lose his Palm over at the airport in Maradonna
    I know you probably hear this everyday, but I'm your biggest fan
    I even read all your bullshit Linux news and BSD scams
    I got a room full of your posters and your pictures man
    I like the way you sold your ass too that shit was fat
    Anyways, I hope you get this man, hit me back,
    just to chat, truly yours, your biggest fan
    This is Stan

    Dear Rob, you still ain't called or wrote, I hope you have a chance
    I ain't mad - I just think it's FUCKED UP you don't answer fans
    If you didn't wanna talk to me outside your Linux World
    you didn't have to, but you coulda signed an autograph for Matthew
    That's my Senior sys admin he's only 26 years old
    We waited on a 9600 baud for you,
    four hours and you just said, "No."
    That's pretty shitty man - you're like his fsckin idol
    He wants to be just like you man, he likes you more than I do
    I ain't that mad though, I just don't like bein lied to
    Remember when we met in Boston - you said if I'd write you
    you would write back - see I'm just like you in a way
    I never had a clue about shit either
    I gcc'd shit with my wife then beat her
    I can relate to what you're saying in your page
    so when I feel like rmusering I read Slashdot to being the rage
    cause I don't really got shit else so that shit helps when I'm depressed
    I even got a tattoo of slashdot across the chest
    Sometimes I even packet myself to see how much it floods
    It's like adrenaline, the DDoS is such a sudden rush of blood
    See everything you say is real, and I respect you cause you tell it
    My girlfriend's jealous cause I talk about you 24/7
    But she don't know you like I know you Rob, no one does
    She don't know what it was like for people like us growin up
    You gotta call me man, I'll be the biggest fan you'll ever lose
    Sincerely yours, Stan -- P.S.
    We should be together too
    Dear Mister-I'm-Too-Good-To-Call-Or-Write-My-Fans,
    this'll be the last packet I ever send your ass
    It's been six months and still no word - I don't deserve it?
    I know you got my last two emails
    I wrote the @ signs on 'em perfect
    So this is my payload I'm sending you, I hope you hear it
    I'm on my modem now, I'm doing 9600 on the infohiway
    Hey Rob, I drank a fifth of vodka, you dare me to code?
    You know the song by Deep Purple by Depache Mode
    its irrelevant by playing on my linux player
    while I write some php scripts and play some Dragonslayer
    That's kinda how shit is, you coulda rescued me from drowning
    Now it's too late - I'm on a 1000 downloads now, I'm drowsy
    and all I wanted was a lousy letter or a call
    I hope you know I ripped +ALL+ of your pictures off the wall
    I love you Rob, we coulda been together, think about it
    You ruined it now, I hope you can't sleep and you dream about it
    And when you dream I hope you can't sleep and you SCREAM about it
    I hope your conscience EATS AT YOU and you can't BREATHE without me
    See Rob {*screaming*} Shut up bitch! I'm tryin to page
    Hey Rob, that's my senior admin screamin in the cage
    but I didn't cut the power off, I just rebooted, see I ain't like you
    cause if he works some harder he'll suffer more, and then the boxes die too
    Well, gotta go, I'm almost BGP bridged
    Oh shit, I forgot, how'm I supposed to send this packet out?

    Dear Stan, I meant to write you sooner but I just been busy
    You said your box is running now, how'd you like your gcc?
    Look, I'm really flattered you would install 7.0 Redhat
    and here's an autograph for your senior admin
    I wrote it on the Starter cap
    I'm sorry I didn't see you at the show, I musta missed you
    Don't think I did that shit intentionally just to diss you
    But what's this shit you said about you like to DDoS lamers too?
    I say that shit just clownin dogg,
    c'mon - how fucked up is you?
    You got some issues Stan, I think you need some counseling
    so heres some more Linux stories to keep your as busy when you get down some
    And what's this shit about us meant to be together?
    I already have a boyfriend Timothy he gets me wetter
    I really think you and your boxes need each other
    or maybe you just need to treat them better
    I hope you get to read this letter, I just hope it reaches you in time
    before you hurt yourself, I think that you'll be doin just fine
    if you relax a little, I'm glad I inspire you but Stan
    why are you so mad? Try to understand, that Linux and MS is just grand
    I just don't want you to do some crazy shit
    I seen this one shit on the news a couple weeks ago that made me sick
    Some dude was drunk and switched his router for a bridge
    and his packets were blackholed, and his DNS couldn't get digged
    and in the colo they found a tape, but they didn't say who it was to
    Come to think about, his name was.. it was you
    Damn!

  • Hey now, I wouldn't say that just yet [kernelnotes.org]. It's probably harder than installing RedHat 7 on a 486, (boot disk doesn't support it by default) but if this guy thinks it's doable, then I might have to get a TI-92+. :)
    ---
    pb Reply or e-mail; don't vaguely moderate [ncsu.edu].
  • If you have a look at Intel's website, i.e. read before you post, you can see what kind of applications Intel has in mind. It's routers, controllers, etc. This kind of hardware hardly needs floating point performance at all, but lots of integer throughput. and for the few things that require a fast FPU you can always code it up in software. Just think of integer and floating point DSPs.

    Besides, the Motorola Dragonball of the Palm isn't that fast in FP performance either. And you can always buy the coprocessor if need be. I think it's a great processor.

    Just my 5c.

  • First off, please take note of the ISA (instruction set architecture) compatibility: it's 32-bit StrongARM! Of which, like ARM, is strictly a microcontroller without floating-point! So your "analysis" of Intel's design is from a standpoint of complete ignorance.

    Again, this is a microcontroller, and NOT a general purpose "microprocessor." It's usually made for throwing data around in non-user devices like RAID controllers, backbone network switches (ones with lots of ports that require more than a basic ASIC, application specific integrated circuit), and the like. End-user devices make up only a fraction of the microcontroller market, and their integer-only functionality is quite sufficient.

    And since it is only 32-bit, 32-bit interpolated integer math is just as good (or better) than 32-bit, single-precision floating-point for the end-user applications you mentioned. Besides, I was under the impression most handwriting recognition was image processing anyway, which is usually integer based. It has SIMD (single instruction, multiple data) that you'll find in general purpose microprocessor extensions like MMX and SSE. I cannot think of any application where you'd need the definition of 64-bit, or higher, double-precision outside of the range of non-science/engineering, end-user applications. 32-bit interpolated interger math is much faster than single-precision floating-point -- important for things like real-time speech recognition.

    -- Bryan "TheBS" Smith

  • OMFG, i havent laughed this much in such a long time...
    thank you
    that was just perfect
    :')
    ROTFLMAO
    please mod this up, it deserves at least a 5 funny...

  • Re:80200 is a nice CPU.. (Score:1)

    by Anonymous Coward
    It does support Thumb.

  • Don't be a moron (Score:3)

    by crisco ( 4669 ) on Saturday February 03, 2001 @08:01PM (#458449) Homepage
    This isn't a processor for those kind of applications.

    This processor is for RAID controllers and similar I/O processing situations. Its not 'crippled', it is designed for a certain task, just as the x86 processors are designed for backwards compatibility.

    The real moron is the moderators who marked this insightful, it is neither insightful nor particularly relevant.

  • Pentium and StrongARM are NOT comparable! (Score:5)

    by BitMan ( 15055 ) on Saturday February 03, 2001 @08:10PM (#458450)

    Please, please, stop any comparison, debate or otherwise general ignorant commentary on Intel's StrongARM microcontrollers versus their Pentium microprocessors. They are two very different breeds of products!

    Here's some major design differences:

    • Speed v. Power: Pentium is sold at its top speed, irrespective of power usage (which varies little between speeds) -- plus Intel sells a separate, redesigned "mobile" line for when power becomes an issue. In contrast, the single StrongARM product series's speed is "scaled" up and down and power usage depends on its speed -- plus a 733MHz part uses 1.3W (an order of magnitude lower than even the mobile Pentium), and the 333MHz model only disapates around 0.5W (and there is an older StrongARM design that vary from 40-600MHz, which makes power usage fluctuate from as low as 0.04W to 450mW).
    • Superscalar v. Superpipelined: The Pentium is superscalar, with multiple pipelines (5-6 in most Pentium products, 8-20 stages each, depending on model) and uses advanced concepts like out-of-order execution and branch prediction to keeps its pipes full and executing. The StrongARM has but a single, 7-stage pipe that is kept full by a optimized compiler tuned for applications with fewer branchs and less interactive code.
      [ Side note: the AMD Athlon uses 9 pipes. Which goes a long way to describing why a 9 pipe x 20 stage Athlon kicks the living crap out of a 6 pipe x 20 stage Pentium IV, MHz for MHz -- especially when combined with the fact that the Athlon only loses an average of 10 stages on a branch mis-predict, whereas all Pentiums have to flush all pipes -- a loss of all 20 stages on a Pentium IV. Even the 6 pipe x 10 stage Pentium III can handle itself against a Pentium IV -- more stages is usually less efficient and more troublesome (especially on branch mis-predicts), although required for OO, timing and other scalability issues. ]
    • Applications: Again, the applications the Pentium is used for is radically different than the StrongARM. In the case of the Pentium, most apps will require floating-point, whereas applications on the later usually do not. And the former can do 64/80-bit math, whereas the later can only do 32-bit. As far as multitasking, the later is great at multiple tasks and instruction rescheduling (again, micro"processor"), whereas the later is more akin to less random operations and control of data flow (again, micro"controller"). The StrongARM does NOT make an ideal, general-purpose CPU for heavy, multi-tasking workstations, but it does make a great, low-power CPU for support as well as standalone devices that only do a couple of things (or run only a couple of user apps simultaneously).

    -- Bryan "TheBS" Smith

  • Remember that the XScale is an embedded CPU, not a general purpose CPU. Embedded devices should be low power, having a full blown FPU makes no sense on such devices.

    On the other hand, the XScale also implements the ARM v5 DSP extensions, so it can be used for signal processing.


    Erik

  • More importantly, the context of this posting is that someone booted Linux on it and that people built PDAs based on this processor, so it's not just being used as an I/O controller for RAIDs.

    Yes. But, and I hope you agree with:

    1. There are a crapload of non-consumer, ARM-based systems running Linux today. I know they are getting very popular for military and scientific applications. That is a significant share of even the Linux market -- so were still talking about more than just PDAs, even when talking only running Linux.
    2. PDAs run far fewer, simultaneous processes than a desktop (let alone a server). Even if you figure in the networking stack and a few running apps, it's still far less intensive.

    On most modern processors, you actually often get faster performance if you implement using FP because integer arithmetic (for loops and indexes) and FP arithmetic can happen in parallel.

    A well informed post, greatly appreciate the insight. So now I ask:

    1. Would it still be as fast on a microcontroller, like the ARM products, with a single pipe? That would prevent multiple, parallel operations, right? So you back to talking about multi-issue microprocessors, when the post was about a single-issue microcontroller.
    2. From my understanding, much of the use of the FPU on Intel microprocessors stem from their poor ALU implementations. Sometimes its just faster to run 4 operations on a Pentium FPU than deal with the single, proper instruction through its ALU. I've seen and compared how fast interpolated, 32-bit integer math runs on a K6 (let alone an Athlon) compared to a Pentium III -- and the old K6 smokes it.

    -- Bryan "TheBS" Smith

  • what are you talking about, doom didn't use floating point anyway. it ran just fine on 386's with no FPU. doom was all int based.
  • Answer #1: No, ARM and Thumb are not mutually exclusive, they can be mixed in a single program. As far as I know Thumb can also be used in kernel mode.

    Answer #2: Yes, GCC supports thumb, though the upcoming GCC-3.0 supports it much better than GCC-2.95.x.

    Answer #3: I don't know the Atmel 91AT, but AFAIK the XScale is the first CPU that implements the ARM v5 instruction set. So the answer is: the XScale is an ARM v5 CPU, while the 91AT is (probably) ARM v4.


    Erik
  • How does the Thumb restriction on the kernel work? On a typical ARM7 or ARM9 based core, there aren't any such restrictions. What does XScale do that is different?

  • Yes and no.

    Yes, this is indeed the same XScale core that can run up to 1GHz (I don't have exact power figures for that speed, though).

    No, this particular CPU (the 80200) is not the comparable to the SA-1110 as used in the Compaq iPaq. The SA-1110 has a lot of IO on board, like two serial ports, IrDA, USB client, LCD controller, PCMCIA controller, etc. The 80200 is more comparable to the SA-110 as used in the Netwinder.


    Erik
  • title says it all

    (-;

    john jones

  • The cost performance does not work out in the case where power is cheap. The ARM family processors are cheaper to fabricate, but their performance tops out far below the ia32 top end. You can pick up lowend ia32 processors for a few dollars more than the ARM family and have greater performance plus family unity across your product line.

    Example, for compiling code my 200MHz StrongARM benchmarks about like a Pentium 90. The more recent ia32s do even more work per cycle. For a rough rule of thumb divide the Xscale clock rate by 3 to compare to modern ia32 machines. (The missing part of that factor comes from intel's xscale v. strongarm slides.)

    That said, for two months of the year I live in an area with limited electricty. I use an ARM based Linux desktop with an LCD display and get my job done just fine. Runs for ages on my 12v battery bank. Sort like a laptop with a 15.1" screen, a detached keyboard and mouse, and a 180lb battery. It works fine, but I can do much better cost and performance wise if you give me an electricty outlet thats on all the time and costs nearly nothing.

    Now, if you offer me a Xscale based belt system with a battery, small LCD, svga connector, PS2, USB, enet, and ieee1394 I would snap it up in a heartbeat. I can always bum a keyboard and monitor. I can use USB for a modem if needed and ieee1394 for disks, tapes, whatever. Use it like a fat PDA, but its also my desktop.
  • The 68000 has no MMU. You might be able to use some code from Linux and hack something together, but you'll never get Linux proper running on there.
  • Here's [kernelnotes.org] a link to a few kernel hackers talking about it.
    Doesn't look good for Linux on a TI-89/92(+).

    --

  • Re:what's the use? (Score:3)

    by Wesley Felter ( 138342 ) <wesley@felter.org> on Saturday February 03, 2001 @06:38PM (#458461) Homepage
    XScale isn't just an embedded CPU; its predecessor StrongARM was used in the NetWinder desktop machine a few years back. Now that XScale is up to ~700 MHz we might see an ARM comeback in non-embedded systems.
  • Intel delivered this kind of crippled hardware before and got a big market share compared to slightly more expensive and considerably more usable processors.

    XScale is nothing but the next generation StrongARM processor, which was developed by DEC at least five years ago.

    It's funny how when DEC had it, everybody thought the StrongARM was the best thing since sliced bread. That Netwinder machine shipped with Linux and was at the top of ever geek's wantlist. ALl without FP. Suddenly Intel bought it, and now it's poison. It is obvious that your agenda is not based on technical merit; you are merely out to debase Intel's brand name, and don't care about the atcual quality of the products.

    FYI, XScale is 17 times faster, and uses 1/3 the power of, the Dragonball processor which is shipped in Palm's (which also doesn't do floating point). The Dragonball has much higher marketshare, at least in the PDA market. So why don't you go around spending your time debasing the Dragonball?

    Let's not let them get away with it again: if it doesn't do floating point, don't waste your time porting stuff that is more easily expressed using floating point to it.

    Better yet: Don't waste your time being an Intel hater.
  • I know its slightly off topic, but I feel that it needs to be said (and no flames fromt eh HP calc people)

    How about some form of unix (maybe NetBSD) for the TI-89? Its got a MOT 68k running at 8MHz/10MHz (rev a/b), 2MB EEPROM, and 1MB RAM. How about it?

    Mark Duell
  • The TI-89 has a Motorola 68000 processor at 10 or 12 MHz. See the TI calculator processor chip and speed FAQ. [ti.com]

    And next time, don't correct someone when you don't know what the hell you're talking about.

  • Re:hah (Score:1)

    by tcc ( 140386 )
    The point is alternatives are always a good thing, and If you have to choice to developpe let's say on WinCE and you don't know it, and you know very well linux, you'll have fun doing your projects on an environment on which you are already familiar. I can't beleive people are bitching when there are alternatives, and these same people are bitching on monopolies... oh well.

  • Oh that's why all those doom clone on my palmpilot sucks, dang...
  • This is some of the funniest stuff I have seen on /. for a long time
  • Dynamically change the clock multiplier?

    I was actually thinking that if you have some simple asynch logic to generate a clock signal, you could probably pretty trivially do the clock scaling.

    However, I don't know how linear the ramp would be, nor do I know if the XScale has a linear ramp. Does XScale have a small set of pre-defined clock rates, or is it literally some continuous well defined function determinate upon the input voltage?

    Oh well, neat trick, regardless.

    Geek dating! [bunnyhop.com]
  • No, that's not what it means. Nico has been working on this for quite some time (sponsored by MontaVista [mvista.com]), but he still maintains the SA1100 Linux tree. I am still working on the LART [tudelft.nl].

    Nico and I have been sharing a hotel room during the Linux World Expo [linuxworldexpo.com] in New York, and we wanted to have the XScale shown running on the expo. So we hacked on thursday night and showed the result on friday.

    About the UCB1200 driver: there is a unified UCB1200 driver in Nico's latest SA1100 patch. I promised to review it, but still got no time to do it. I'll do it Real Soon Now [tm].


    Erik
  • Yup, we used the Intel IQ80310 [intel.com].

    The IQ80310 can be used as a PCI host with a passive PCI backplane, but it can also be used as a normal PCI card in a PC (or whatever else with a PCI bus).

    Because we didn't have a passive PCI backplane in New York, MontaVista [mvista.com] bought the cheapest PC available at a local computer store and we used that as a backplane. I have no idea what kind of PC it was, we just used it as backplane and power supply to get the XScale running. The PC was definitively worth its price :-)


    Erik
  • The XScale was certainly not running on its breaking point, we had it only running at a mere 400 MHz because getting it to run at all was much more important that running at full speed.

    Oh, floating point emulation has been a part of ARM Linux for ages.


    Erik
  • Intel's StrongARM microcontrollers versus their Pentium microprocessors

    What??? Take a look at the original link XScale Processor [intel.com] and you'll see the're not talking about a microcontroller but a microprocessor. Semantics, maybe, but the facts are that it is a capable microprocessor used in desktop hardware [microdigital.co.uk], not just a microcontroller destined for life controlling a washing machine.

    Of course, the nice thing is that the StrongARM core also gets embedded into chips that look like microcontrollers, but are really embedded processors. We are talking a seriously different league to the good ol' 68HC11!

    Other than that I entirely agree with your comments :-)

  • Until recently, handhelds like the Palm and its WinCE clones didn't need much smarts. That's rapidly changing.

    • The Palm STILL doesn't need much "smarts" in the way of hardware performance; It gets the job done with "primitive" hardware. (Basically a 68000 with some added IO, really.)
    • WinCE definitely needs to be on some very fast hardware (compared to a 16MHz dragonball anyway) to keep up with palm, let alone to surpass it. The UI is slow, because they spent the time to make it pretty. It IS nice to have that added power, so you can play games on it, but other than that it's fairly pointless. The only thing I'd really want to use that kind of horsepower for (IE, a PDA with >100MHz power) would be perhaps to play mp3s. Any game requiring a 100MHz chip probably isn't that well suited to a PDA.

    --

  • The Thumb instructions are encoded into 16 bits instead of the ARM's 32. This achieves higher code compression, but costs a few instructions. Among those instructions you can't use in Thumb mode are the ones to control the processor configuration.. system registers. You have to be in ARM mode to do things like mask interrupts, so I'd imagine a kernel would have to have at least some ARM-mode code in it. However, at least on the ARM7, it's pretty easy to switch back and forth. (It cues off the low bit of the instruction address, so switching "modes" is little more than branching to a new odd/even address.)
  • You're missing one vital point - with a StrongARM processor, assembly language is extremely easy, and because of this, more applications can be written in assembly, and thus faster. Compare the responsiveness of a 5-year-old StrongARM RiscPC to a Windoze, or even Linux box and you'll see what I mean.
  • ...is the Intel 80200 Processor based on Intel XScale microarchitecture?

    :)

    1st Law Of Networking: Loose ends are bad, termination is good.

  • as if Intel will kill its cash cow pentium line.

    anyway, you miss my whole point. people are always porting linux to more and more systems, and claiming it's a good platform for embedded systems, when it's just too damn big and bloated for it. so fanaticism gets in the way of good technical decisions and being able as a customer to get quality products.

  • Quite a bit, considering how long it would take to play a game at 0.23 frames a second. ;)

    1st Law Of Networking: Loose ends are bad, termination is good.

  • "as if Intel will kill its cash cow pentium line."

    No, instead, they bought ARM, and continued making and promoting their products to kill that line?

    Also, linux *is* good for embedded systems - juts not super-tiny embedded systems. There's a continuum from microwave to supercomputer. Linux fills some range of that, and that range intersects with the range of systems that we cll "embedded".

Slashdot Top Deals

One man's constant is another man's variable. -- A.J. Perlis

Close