Below I present tcpl "two's-complement plus".  It works with the 
transpose of the arrangement below for adding two's-complement numbers:

  1 1 0 0  (value _4)
  1 1 0 1  (value _3)
  -------
  1 1 0 0  (carries shifted to the right)
  1 0 0 1  (answer, value _7)

Working above from right to left, you say 0 plus 1 is 0 1 and write 0 1 
vertically below the line. Next you say 0 plus 0 plus 0 is 0 0 (the 
first 0 is the carry from the preceding step) and write 0 0 vertically 
below the line.  Continuing, you say 0 plus 1 plus 1 is 1 0 and 1 plus 1 
plus 1 is 1 1 .  The answer 1 0 0 1 is in the bottom line.

     hc =: {.@#:@(,:  2 * |)  NB. Raul's improved #: (hash colon)

     hcinv =: ([: -/ [: #. (,: [: +: 1 {. ]))"1  NB. Henry Rich

     Table =: 2 2 2 2 $ 0 0,0 1,0 1,1 0,0 1,1 0,1 0,1 1

     (< 1 1 1) { Table  NB. sum of three bits
  1 1

     stack =: ,.&.|:  NB. stacks x over y

     hv =: (0 {:: <"1) :: ]  NB. returns head vector

     op =: ] stack~ (2 {. [) , Table {~ [: < (2 {. [) , 2 { [: hv ]

     ba =: 0 ,~ 0 ,.~ 0 ,.~ ,.  NB. build argument

     1 1 0 0 ba 1 1 0 1  NB. starting table
  1 1 0 0
  1 1 0 0
  0 0 0 0
  0 1 0 0
  0 0 0 0

     op/ 1 1 0 0 ba 1 1 0 1
  1 1 1 1
  1 1 1 0
  0 0 0 0
  0 1 0 1
  0 0 0 0

     NB. above see transpose of arrangement at top, with an added
     NB. row of 0's

     tcpl =: [: }: [: {:"1 [: op/ ba  NB. two's-complement plus

     1 1 0 0 tcpl 1 1 0 1  NB. _4 plus _3 is _7
  1 0 0 1

     hc _4 _3 _7
  1 1 0 0
  1 1 0 1
  1 0 0 1

On 12/10/2011 6:48 PM, Kip Murray wrote:
> Cool. I think it is an improvement because it neatly avoids a case
> statement. Now I wonder if we could implement two's-complement addition
> and multiplication with overflow, basing these on bitwise operations
>
> 0 + 0 is 0, 0 + 1 is 1 + 0 is 1, and 1 + 1 is 1 0
>
> 0 * 0 is 0 * 1 is 1 * 0 is 0, and 1*1 is 1
>
> That is, I do not want hc and hcinv to be used except to produce data
> and check answers, and I want n-bit two's-complement answers for n-bit
> two's-complement data so some of them will be wrong because of overflow.

ADDED LATER -- ACCEPT 2n-bit multiplication answers from n-bit data

>
> Here for Linda is Henry's hcinv expressed without conjunctions other
> than " .
>
> hcinv =: ([: -/ [: #. (,: [: +: 1 {. ]))"1
>
> On 12/10/2011 10:00 AM, Henry Rich wrote:
>> Same idea, better implementation
>>
>> hcinv =. -/@:#.@:(,: +:@(1&{.))"1
>>
>> Henry Rich
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to