Transputer

http://en.wikipedia.org/wiki/Transputer

  A transputer is an device/emulator that runs Communicating Sequental Proccesses.

Also the basis for the Transterpreter system[L1 ] written in pure ANSI C.


 (dreging info from http://svn.transterpreter.org/transterpreter/trunk/interpreter/dispatch_ins.c)

 stack based system
 instruction set is bytecode based with the distinction that
 first nybble is the main opcode and the second nybble is appended to
 an (unspecified long/32 bit?) operand register and the contents of that register is then taken as immideate operand.

  opcode: memnomic: description
   0x0     j         Jump — add immediate operand to instruction pointer.
   0x1     ldlp      Load Local Pointer — load a Workspace-relative pointer onto the top of the register stack
   0x2     pfix      Prefix — general way to increase lower nibble of following primary instruction
   0x3     ldnl      Load non-local — load a value offset from address at top of stack
   0x4     ldc       Load constant — load constant operand onto the top of the register stack
   0x5     ldnlp     Load Non-local pointer — Load address, offset from top of stack
   0x6     nfix      Negative prefix — general way to negate (and possibly increase) lower nibble
   0x7     ldl       Load Local — load value offset from Workspace
   0x8     adc       Add Constant — add constant operand to top of register stack
   0x9     call      Subroutine call — push instruction pointer and jump
   0xA     cj        Conditional jump — depending on value at top of register stack
   0xB     ajw       Adjust workspace — add operand to workspace pointer
   0xC     eqc       Equals constant — test if top of register stack equals constant operand
   0xD     stl       Store local - store at constant offset from workspace
   0xE     stnl      Store non-local - store at address offset from top of stack
   0xF     opr       Operate - general way to extend instruction set

  the opr instruction extends the instruction set by switching on the value of the operand register.

  the extended instruction set is as follows:
   0x00    rev
   0x01    lb
   0x02    bsub
   0x03    endp
   0x04    diff
   0x05    add
   0x06    gcall
   0x07    in
   0x08    prod
   0x09    gt
   0x0A    wsub 
   0x0B    out
   0x0C    sub
   0x0D    startp
   0x0E    outbyte
   0x0F    outword
   0x10    seterr
   0x11    mreleasep __PI_SUPPORT__
   0x13    csub0
   0x14    extvrfy __PI_SUPPORT__
   0x15    stopp
   0x16    ladd
   0x19    norm
   0x1A    ldiv
   0x1B    ldpi
   0x1D    xdble
   0x1F    rem
   0x20    ret
   0x21    lend
   0x22    ldtimer
   0x23    boolinvert
   0x24    widenshort
   0x25    fficall
   0x26    lend3
   0x27    lendbw
   0x28    reschedule
   0x2B    tin
   0x2C    div
   0x2E    dist
   0x2F    disc
   0x30    diss
   0x31    lmul
   0x32    not
   0x33    xor
   0x35    lshr
   0x36    lshl
   0x37    lsum
   0x38    lsub
   0x39    runp
   0x3B    sb
   0x3C    gajw
   0x3E    saveh
   0x40    shr
   0x41    shl
   0x42    mint
   0x43    alt
   0x44    altwt
   0x45    altend
   0x46    and
   0x47    enbt
   0x48    enbc
   0x49    enbs
   0x4A    move
   0x4B    or
   0x4C    csngl
   0x4D    ccnt1
   0x4E    talt
   0x4F    ldiff
   0x51    taltwt
   0x52    sum
   0x53    mul
   0x55    stoperr
   0x56    cword
   0x5A    _dup __FPU_SUPPORT__
   0x60    extin __PI_SUPPORT__
   0x61    extout __PI_SUPPORT__
   0x62    minn __PI_SUPPORT__
   0x63    unpacksn __FPU_SUPPORT__
   0x64    moutn __PI_SUPPORT__
   0x65    xminn __PI_SUPPORT__
   0x6C    postnormsn __FPU_SUPPORT__
   0x6D    roundsn __FPU_SUPPORT__
   0x71    ldinf __FPU_SUPPORT__
   0x72    fmul
   0x79    pop
   0x7A    seminit __PI_SUPPORT__
   0x7B    semclaim __PI_SUPPORT__
   0x7C    semrelease __PI_SUPPORT__
 #ifdef __FPU_SUPPORT__
                case \: ins_wsubdb(); break;
 #endif
 #ifdef __FPU_SUPPORT__  //in case of fpu support include the below
   0x82    fpldnldbi
   0x83    fpchkerr
   0x84    fpstnldb
   0x86    fpldnlsni
   0x87    fpadd
   0x88    fpstnlsn
   0x89    fpsub
   0x8A    fpldnldb
   0x8B    fpmul
   0x8C    fpdiv
   0x8E    fpldnlsn
   0x91    fpnan
   0x92    fpordered
   0x93    fpnotfinite
   0x94    fpgt
   0x95    fpeq
   0x96    fpi32tor32
   0x98    fpi32tor64
   0x9A    fpb32tor64
   0x9D    fprtoi32
   0x9E    fpstnli32
   0x9F    fpldzerosn
   0xA0    fpldzerodb
   0xA1    fpint
 #endif
 #ifdef __PI_SUPPORT__
   0xA2    getpri
 #endif
 #ifdef __FPU_SUPPORT__  //in case of fpu support include the below
   0xA3    fpdup
   0xA4    fprev
   0xA6    fpldnladddb
   0xA8    fpldnlmuldb
   0xAA    fpldnladdsn
   0xAC    fpldnlmulsn
 #endif
   0xAD    savecreg
   0xAE    restorecreg
   0xB0    barinit
   0xB1    barsync
   0xB2    barresign
   0xB3    barenroll
 #ifdef __FPU_SUPPORT__  //in case of fpu support include the below
   0xCF    fprem
   0xD0    i64toreal
   0xD1    fpdivby2
   0xD2    fpmulby2
   0xD3    fpsqrt
   0xD6    fprz
   0xD7    fpr32to64
   0xD8    fpr64to32
   0xD9    fpexpdec32
   0xDB    fpabs
   0xDF    fpchki64
 #endif
 #ifdef __PI_SUPPORT__
   0xE0    mnew
   0xE1    mfree
   0xE2    malloc
   0xE3    mrelease
   0xE4    min
   0xE5    mout
   0xE6    min64
   0xE7    mout64
   0xE8    xable
   0xE9    xin
   0xEA    xmin
   0xEB    xmin64
   0xEC    xend
   0xFD    null
 #endif

Zarutian 23. maí 2007: this page is still under construction

Zarutian 30. maí 2007: and yet I continue

Zarutian 14. september 2007:

  namespace eval transputer {}
  namespace eval transputer::assam {
    variable assamble_buffer {}
    proc get_assamble_buffer {} {
      variable assamble_buffer
      return [set assamble_buffer]
    }
    proc j {relative} {
      variable assamble_buffer
      if {$relative > 15} {
        set tmp [expr $relative % 16]
        pfix [expr $relative / 16]
        set realtive $tmp
      }
      set tmp [format "%x" $relative]
      append assamble_buffer "0[set tmp]"
    }
    proc pfix {literal} {
      variable assamble_buffer
      set tmp {}
      if {$literal == 0} {
        append assamble_buffer "20"
      } else {
        while {$literal > 0} {
          set tmp [format "%x" [expr $literal % 16]][set tmp]
          set literal [expr $literal / 16]
        }
        foreach i [split $tmp ""] {
          append assamble_buffer "2[set i]"
        }
      }
    }
    proc ldc {constant} {
      variable assamble_buffer
      if {$constant > 15} {
        set tmp [expr $constant % 16]
        pfix [expr $constant / 16]
        set constant $tmp
      }
      set tmp [format "%x" $constant]
      append assamble_buffer "4[set tmp]"
    }
    proc ldnl {offset} {
      variable assamble_buffer
      if {$offset > 15} {
        set tmp [expr $offset % 16]
        pfix [expr $offset / 16]
        set offset $tmp
      }
      set tmp [format "%x" $constant]
      append assamble_buffer "3[set tmp]"
    }
  }

CMcC has worked on transputers and would like to say that the architecture is not well-suited to programming. For one thing, there are *no* global stores, and so writing malloc is quite difficult. Another oddity is that evaluating an expression like 1/0 hung the transputer by design.

FPX has also worked on transputers and disagrees with that statement. Transputers may not be well-suited for certain general purpose programming, but they shine at stream processing and other tasks. Lack of a global store is not an issue with stateless components (such as an image filter). "Hanging a transputer" with a division by zero is system-specific and not an inherent issue with transputers in general. Most transputers run single user operating systems, not having kernel/user space separation due to the lack of a MMU. These days, GPUs (graphics processing units) follow similar designs than Transputers. Yes, you wouldn't want them to run a word processor, but there are certain tasks that they do so much better than your general CPUs.


http://wotug.org/parallel/transputer/documentation/st020-450/datasheets/

Category Concurrency | Category Parallelism