• Re: Combining Practicality with Perfection

    From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Sat Oct 24 00:22:30 2026
    From Newsgroup: comp.arch

    On Wed, 11 Feb 2026 23:55:29 GMT, MitchAlsup
    <user5857@newsgrouper.org.invalid> wrote:

    In this case YOU have to ask YOURSELF why are you providing any of those >strange data sizes AT ALL ??? That is who is Concertina for ???

    I think that ARM-based smartphones already existed when I started work
    on Concertina II.
    But it may be that it was before Apple switched the Macintosh over to
    Apple Silicon, so ARM was still seen as a niche product.

    So, psychologically, I felt that we were living in a world where,
    essentially, x86 - at least in its x86-64 incarnation - had *killed
    off all other computer architectures*.
    From this, I drew the conclusion that, when it comes to computer
    architectures, "There can be only one" - like Highlander.

    And from that, I came to the further conclusion... that if the world
    can have only one computer architecture, then, in order for the world
    not to be impoverished thereby, that one surviving computer
    architecture must be able to "do everything", where "everything"
    includes, say, being able to run PDP-10 code at nearly native speeds.
    Hence, strange data sizes.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From MitchAlsup@user5857@newsgrouper.org.invalid to comp.arch on Sat Oct 24 19:32:36 2026
    From Newsgroup: comp.arch

    quadibloc@invalid.com (John Savard) posted:

    On Wed, 11 Feb 2026 23:55:29 GMT, MitchAlsup <user5857@newsgrouper.org.invalid> wrote:

    In this case YOU have to ask YOURSELF why are you providing any of those >strange data sizes AT ALL ??? That is who is Concertina for ???

    I think that ARM-based smartphones already existed when I started work
    on Concertina II.
    But it may be that it was before Apple switched the Macintosh over to
    Apple Silicon, so ARM was still seen as a niche product.

    So, psychologically, I felt that we were living in a world where, essentially, x86 - at least in its x86-64 incarnation - had *killed
    off all other computer architectures*.

    Microprocessors killed off minicomputers--no doubt.
    Microprocessors did not kill off mainframes--and still cannot (RAS)

    From this, I drew the conclusion that, when it comes to computer architectures, "There can be only one" - like Highlander.

    {Duncan or Connor} ??

    One will never happen, one may become and remain dominant for long
    periods of time, but not forever.

    And from that, I came to the further conclusion... that if the world
    can have only one computer architecture, then, in order for the world
    not to be impoverished thereby, that one surviving computer
    architecture must be able to "do everything", where "everything"
    includes, say, being able to run PDP-10 code at nearly native speeds.
    Hence, strange data sizes.

    Everything I have seen over 50 years tells me "mediocracy first wins".

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From jgd@jgd@cix.co.uk (John Dallman) to comp.arch on Sun Oct 25 06:00:00 2026
    From Newsgroup: comp.arch

    In article <6adbf8ee.10622140@news.eternal-september.org>, quadibloc@invalid.com (John Savard) wrote:

    if the world can have only one computer architecture, then, in
    order for the world not to be impoverished thereby, that one
    surviving computer architecture must be able to "do everything",
    where "everything" includes, say, being able to run PDP-10 code
    at nearly native speeds.

    However, an architecture of extreme complexity, burdened with lots of
    rarely used features, is most unlikely to become the world's sole
    architecture.

    Have you tried running PDP-10 code in an emulator on a modern 64-bit
    machine? I'd bet that it significantly exceeds the speed of the native hardware.

    John
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Sun Oct 25 17:38:15 2026
    From Newsgroup: comp.arch

    On Sun, 25 Oct 2026 06:00 +0000 (GMT Standard Time), jgd@cix.co.uk
    (John Dallman) wrote:
    In article <6adbf8ee.10622140@news.eternal-september.org>, >quadibloc@invalid.com (John Savard) wrote:

    if the world can have only one computer architecture, then, in
    order for the world not to be impoverished thereby, that one
    surviving computer architecture must be able to "do everything",
    where "everything" includes, say, being able to run PDP-10 code
    at nearly native speeds.

    However, an architecture of extreme complexity, burdened with lots of
    rarely used features, is most unlikely to become the world's sole >architecture.

    True, but I haven't been able to figure out an alternative.

    Also, fortunately, my sole competition happens to be struggling under
    a burden of its own which seems to be just as great or greater. (That
    is, if you don't count ARM and RISC-V...)

    Have you tried running PDP-10 code in an emulator on a modern 64-bit
    machine? I'd bet that it significantly exceeds the speed of the native >hardware.

    But it doesn't exceed the speed of the modern 64-bit machine.

    And? Why does that matter?

    It's not enough for people to be able to emulate a PDP-10 for merely _sentimental_ reasons. I want to breathe life into those dusty decks!

    I want there to be other ways of computing!

    Even though I remain sedulously unimpressed by the notion that
    psilocybin beats empiricsm as a way of knowing.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From John Levine@johnl@taugh.com to comp.arch on Sun Oct 25 18:47:58 2026
    From Newsgroup: comp.arch

    According to John Dallman <jgd@cix.co.uk>:
    However, an architecture of extreme complexity, burdened with lots of
    rarely used features, is most unlikely to become the world's sole >architecture.

    Have you tried running PDP-10 code in an emulator on a modern 64-bit
    machine? I'd bet that it significantly exceeds the speed of the native >hardware.

    The KL-10 was about 1.8 MIPS. On modern hardware you'd have to write a pretty bad
    simulator to run that slow.

    I had a lot of fun programming a PDP-10 fifty years ago, but it ran out of address bits
    (even the expanded address hack only gave 30 bits, or about 4GB) and the handwriting was
    on the wall that only 8 bit byte addressed machines would survive.
    --
    Regards,
    John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
    Please consider the environment before reading this e-mail. https://jl.ly
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From Stephen Fuld@sfuld@alumni.cmu.edu.invalid to comp.arch on Sun Oct 25 22:43:01 2026
    From Newsgroup: comp.arch

    On 10/25/2026 10:38 AM, John Savard wrote:
    On Sun, 25 Oct 2026 06:00 +0000 (GMT Standard Time), jgd@cix.co.uk
    (John Dallman) wrote:
    In article <6adbf8ee.10622140@news.eternal-september.org>,
    quadibloc@invalid.com (John Savard) wrote:

    if the world can have only one computer architecture, then, in
    order for the world not to be impoverished thereby, that one
    surviving computer architecture must be able to "do everything",
    where "everything" includes, say, being able to run PDP-10 code
    at nearly native speeds.

    However, an architecture of extreme complexity, burdened with lots of
    rarely used features, is most unlikely to become the world's sole
    architecture.

    True, but I haven't been able to figure out an alternative.

    Also, fortunately, my sole competition happens to be struggling under
    a burden of its own which seems to be just as great or greater. (That
    is, if you don't count ARM and RISC-V...)

    Have you tried running PDP-10 code in an emulator on a modern 64-bit
    machine? I'd bet that it significantly exceeds the speed of the native
    hardware.

    But it doesn't exceed the speed of the modern 64-bit machine.

    And? Why does that matter?

    It's not enough for people to be able to emulate a PDP-10 for merely _sentimental_ reasons. I want to breathe life into those dusty decks!

    What exactly does that mean? You aren't providing an exact match for
    the PDP-10 ISA, so you need some kind of emulator. Surely no one is
    writing new PDP-10 code, so there isn't a market for that.

    I want there to be other ways of computing!

    Again, what does that mean? If you mean different ISA, there are two currently popular ones, RISC 5 seems to be gaining, and I expect that
    Mitch hopes to add to the list. I suppose you could mean some sort of AI/neural net thing, but your architecture shows no signs of that.
    --
    - Stephen Fuld
    (e-mail address disguised to prevent spam)
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From jgd@jgd@cix.co.uk (John Dallman) to comp.arch on Mon Oct 26 09:35:00 2026
    From Newsgroup: comp.arch

    In article <6ade3c57.66203562@news.eternal-september.org>, quadibloc@invalid.com (John Savard) wrote:

    It's not enough for people to be able to emulate a PDP-10 for merely _sentimental_ reasons. I want to breathe life into those dusty
    decks!

    Software that ran on PDP-10s, and other obsolete architectures, which
    people or organisations _needed_ has either been re-written for other architectures by now, or is run under emulation.

    Given that there isn't a commercial emulator for PDP-10, it seems
    unlikely to me that there's any significant business advantage to be
    gained from a new architecture that emulates it faster. Business
    advantage drives processor chip manufacture.

    John
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From MitchAlsup@user5857@newsgrouper.org.invalid to comp.arch on Mon Oct 26 13:40:11 2026
    From Newsgroup: comp.arch

    quadibloc@invalid.com (John Savard) posted:

    On Sun, 25 Oct 2026 06:00 +0000 (GMT Standard Time), jgd@cix.co.uk
    (John Dallman) wrote:
    In article <6adbf8ee.10622140@news.eternal-september.org>, >quadibloc@invalid.com (John Savard) wrote:

    if the world can have only one computer architecture, then, in
    order for the world not to be impoverished thereby, that one
    surviving computer architecture must be able to "do everything",
    where "everything" includes, say, being able to run PDP-10 code
    at nearly native speeds.

    However, an architecture of extreme complexity, burdened with lots of >rarely used features, is most unlikely to become the world's sole >architecture.

    True, but I haven't been able to figure out an alternative.

    Also, fortunately, my sole competition happens to be struggling under
    a burden of its own which seems to be just as great or greater. (That
    is, if you don't count ARM and RISC-V...)

    But it has multiple companies with hundreds of architects and engineers
    (each) working to move them forward. You only have you.

    Have you tried running PDP-10 code in an emulator on a modern 64-bit >machine? I'd bet that it significantly exceeds the speed of the native >hardware.

    But it doesn't exceed the speed of the modern 64-bit machine.

    And? Why does that matter?

    It's not enough for people to be able to emulate a PDP-10 for merely _sentimental_ reasons. I want to breathe life into those dusty decks!

    I want there to be other ways of computing!

    Even though I remain sedulously unimpressed by the notion that
    psilocybin beats empiricsm as a way of knowing.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From John Levine@johnl@taugh.com to comp.arch on Mon Oct 26 19:53:10 2026
    From Newsgroup: comp.arch

    According to John Dallman <jgd@cix.co.uk>:
    In article <6ade3c57.66203562@news.eternal-september.org>, >quadibloc@invalid.com (John Savard) wrote:

    It's not enough for people to be able to emulate a PDP-10 for merely
    _sentimental_ reasons. I want to breathe life into those dusty
    decks!

    Software that ran on PDP-10s, and other obsolete architectures, which
    people or organisations _needed_ has either been re-written for other >architectures by now, or is run under emulation.

    Given that there isn't a commercial emulator for PDP-10, it seems
    unlikely to me that there's any significant business advantage to be
    gained from a new architecture that emulates it faster.

    Foonly, XKL, and Systems Concepts built PDP-10 architecture machines
    after DEC gave up. As far as I can tell, technically they were fine,
    ran all the software faster than the KL, lower power, and used more
    modern peripherals, but most PDP-10 users had already moved on.
    --
    Regards,
    John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
    Please consider the environment before reading this e-mail. https://jl.ly
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Mon Oct 26 22:05:30 2026
    From Newsgroup: comp.arch

    On Mon, 26 Oct 2026 13:40:11 GMT, MitchAlsup
    <user5857@newsgrouper.org.invalid> wrote:
    quadibloc@invalid.com (John Savard) posted:

    Also, fortunately, my sole competition happens to be struggling under
    a burden of its own which seems to be just as great or greater. (That
    is, if you don't count ARM and RISC-V...)

    But it has multiple companies with hundreds of architects and engineers >(each) working to move them forward. You only have you.

    Oh, I am well aware of that. I'm not totally deluded!

    I've tried to do without block structure, which I admit is off-putting
    to many. The current Concertina II includes a means of embedding
    48-bit instructions in code without block headers, thus getting most
    of the instruction set in without block headers.

    What I _can't_ do, though, without block headers without crippling the instruction set, is to include 16-bit short instructions. The short instructions wind up nearly useless, and the opcode space left leaves
    the memory-reference instructions severely limited.

    Obviously, of course, I can avoid that by not having 32 registers.
    With 8 registers, the 68020 is an existence proof, and the System/360
    proves that even 16 registers will work. But RISC with 32 registers
    exists, is popular, and having 32 registers adds performance.

    And the more exotic things for which I wish to at least specify
    opcodes - like System/360 "Hexadecimal Floating-Point" - are things
    that are not part of the basic subset of the architecture.

    So to have 32 registers, immediate values in instructions, short
    instructions nearly 16 bits in length (17 bits plus 32/15 bits header overhead), and dumb fast decoding... I need the block structure.

    And if I have it, of course, it lets me cheat the Pigeonhole Principle shamelessly - since, as pointed out, it's mode bits, but mode bits
    hackers can't exploit!

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Mon Oct 26 22:09:12 2026
    From Newsgroup: comp.arch

    On Mon, 26 Oct 2026 13:40:11 GMT, MitchAlsup
    <user5857@newsgrouper.org.invalid> wrote:

    But it has multiple companies with hundreds of architects and engineers >(each) working to move them forward. You only have you.

    Also...

    The way things are going these days, someone is going to leave lying
    around under a BSD license the plans for an OoO RISC-V core which, no
    doubt, I can modify to execute code for a different instruction set,
    even my own.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Mon Oct 26 22:12:40 2026
    From Newsgroup: comp.arch

    On Mon, 26 Oct 2026 22:05:30 GMT, quadibloc@invalid.com (John Savard)
    wrote:

    Obviously, of course, I can avoid that by not having 32 registers.
    With 8 registers, the 68020 is an existence proof, and the System/360
    proves that even 16 registers will work. But RISC with 32 registers
    exists, is popular, and having 32 registers adds performance.

    I may not have exhausted all possible alternatives, though. Concertina
    II has 32 registers for regular use - but it has another set of 128
    registers (each for integer and floating-point) which require longer instructions to use them. So I could convince myself that a machine
    which uses banks of 8 registers most of the time, but can switch to
    128 registers, at the cost of using 48-bit instructions to do that,
    could still provide adequate performance.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From EricP@ThatWouldBeTelling@thevillage.com to comp.arch on Mon Oct 26 18:20:51 2026
    From Newsgroup: comp.arch

    On 2026-Oct-26 18:09, John Savard wrote:
    On Mon, 26 Oct 2026 13:40:11 GMT, MitchAlsup <user5857@newsgrouper.org.invalid> wrote:

    But it has multiple companies with hundreds of architects and engineers
    (each) working to move them forward. You only have you.

    Also...

    The way things are going these days, someone is going to leave lying
    around under a BSD license the plans for an OoO RISC-V core which, no
    doubt, I can modify to execute code for a different instruction set,
    even my own.

    John Savard

    Do you mean like this:

    https://opencores.org/projects?expanded=Processor


    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From antispam@antispam@fricas.org (Waldek Hebisch) to comp.arch on Mon Oct 26 22:51:34 2026
    From Newsgroup: comp.arch

    John Savard <quadibloc@invalid.com> wrote:
    On Sun, 25 Oct 2026 06:00 +0000 (GMT Standard Time), jgd@cix.co.uk
    (John Dallman) wrote:
    In article <6adbf8ee.10622140@news.eternal-september.org>, >>quadibloc@invalid.com (John Savard) wrote:

    if the world can have only one computer architecture, then, in
    order for the world not to be impoverished thereby, that one
    surviving computer architecture must be able to "do everything",
    where "everything" includes, say, being able to run PDP-10 code
    at nearly native speeds.

    However, an architecture of extreme complexity, burdened with lots of >>rarely used features, is most unlikely to become the world's sole >>architecture.

    True, but I haven't been able to figure out an alternative.

    Also, fortunately, my sole competition happens to be struggling under
    a burden of its own which seems to be just as great or greater. (That
    is, if you don't count ARM and RISC-V...)

    Have you tried running PDP-10 code in an emulator on a modern 64-bit >>machine? I'd bet that it significantly exceeds the speed of the native >>hardware.

    But it doesn't exceed the speed of the modern 64-bit machine.

    And? Why does that matter?

    It's not enough for people to be able to emulate a PDP-10 for merely _sentimental_ reasons. I want to breathe life into those dusty decks!

    I want there to be other ways of computing!

    Nobody sane will spend serious effort writing PDP-10 assembly.
    Note that classic PDP-10 has address space slightly larger
    than megabyte, that is seriously inadeqate now. Extended
    addresing allows bigger address space, but why bother when
    32 bit and 64 bit machines are widely available.

    Look at sorry state of x86_64-32. This uses 32-bit pointers on
    common 64 bit PC-s. In a sense it is widely available:
    there is a lot of hardware, there is Linux kernel and C
    compiler. And theoretically it gives significant advantages,
    as long as program fits into 32 bit address space it uses less
    storage for pointers and has less data to copy. IIUC realistic
    claim is that programs run 10-20% faster than programs using
    64 bit pointers. Yet a lot of software does not work.
    10% faster is too weak motivation to make sure that
    programs work.

    In slightly different spirit, Python seem now to be most
    popular programming language. Most web servers seem to
    run PHP. Both are dynamically typed and typically use
    bytecode interpreters. On low-level style code statically
    typed language compiler to machine code can get 100 times
    better speed. Yet bytecode-compiled dynamically typed
    languages are quite popular.

    How is this possible? One thing is that performance critical
    code is a small part of all code. People can and are willing
    to spend significant effort to improve speed of small
    critical pieces of code. Say, application needs a lot of 37
    bit integers and using more bits would waste space. One can
    write packing and unpacking routines so that say unpacking
    routine unpacks 37 words of 64 bits each into 64 word, puting
    37 bit integer in each 64-bit word and similar for packing
    routine. If used smartly, this can give computational speed
    the same as using 37 bit integers, but memory use of 37
    integers. For addition one can try to put three 37 bit integers
    int 2 words and use add with carry. Not as bit efficient
    as dedicated 37 bit adder, but uses 111 bit out of 128,
    so is 86% efficient. And actually, modern machines are
    generously equipped into execution units and are more data
    bound, so even using full 64 bit arithemetic may be good
    enough.

    Note that performance of modern machines is statistical. For
    important problems vendors add special instrutions. If
    support for 37 bit integers or 41 bit floats was critical,
    then vendors would add special handling. If for specific
    application some nuber size is especially important, then
    software can try to do something. For example, for computing
    information about groups one can use coset enumeration.
    This may use a lot of memory. Popular program for doing
    this offers use of 40 bit integers. That is purely software
    thing, but runs runs fast enough (probably faster than version
    using 64 bit integers) and saves nontrival amount of memory.
    Sure, having hardware 40 bit integers would make it faster
    and some more saving would be possible if 37 bits are
    enough on hardware that supports 37 bit integers. For
    somewhat different mathematical calculation one need
    computation in field having 2 to power 60 elements. So
    60 bit integers and special computational instructions
    (like carryless mutiplication) would be desirable. But
    computation like this form rather small market and
    hardware vendors will not complicate machines to improve
    performance for small niche market.

    Realistically, what one could hope for are general purpose
    extentions which also help in niche cases. Say, ability to
    relatively cheaply load any small size bit string at
    arbitrary bit address. Note that on current machines this
    may easily take 20 instructions including conditional jump.
    With better instruction set this could be reduced to few
    instructions or even one, for example instrucion having
    two register arguments and destination, so that first
    register is bit address, second is length which loads
    bitstring to the destination.
    --
    Waldek Hebisch
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Tue Oct 27 03:00:10 2026
    From Newsgroup: comp.arch

    On Mon, 26 Oct 2026 18:20:51 -0400, EricP
    <ThatWouldBeTelling@thevillage.com> wrote:

    Do you mean like this:

    https://opencores.org/projects?expanded=Processor

    Yes. I found only one out-of-order processor there, though, and it's
    in the planning stage.

    I had heard of other academic projects to create an out-of-order
    RISC-V processor, and there's something called BOOM.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From Stephen Fuld@sfuld@alumni.cmu.edu.invalid to comp.arch on Mon Oct 26 20:13:18 2026
    From Newsgroup: comp.arch

    On 10/26/2026 3:05 PM, John Savard wrote:

    And the more exotic things for which I wish to at least specify
    opcodes - like System/360 "Hexadecimal Floating-Point" - are things
    that are not part of the basic subset of the architecture.

    At the risk of feeding your obsession . . .

    I think you will admit that the use of S/360 floating point will be
    rare. And let me posit that the need to mix that with IEEE floating
    point in the same program is non-existent. If that is true, you could
    use the same op codes for both formats and have a global mode that
    specifies which one to use. This would save some opcodes. And since in practice no one will want the S/360 stuff, make that mode an optional
    feature.
    --
    - Stephen Fuld
    (e-mail address disguised to prevent spam)
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Tue Oct 27 14:30:08 2026
    From Newsgroup: comp.arch

    On Mon, 26 Oct 2026 20:13:18 -0700, Stephen Fuld
    <sfuld@alumni.cmu.edu.invalid> wrote:

    If that is true, you could
    use the same op codes for both formats and have a global mode that
    specifies which one to use.

    Some hacker might find a way to exploit that!

    I'm just using that as an example of the many data types I wish to
    offer. But indeed S/360 floating-point support will be optional even
    in Concertina II as it is.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2