You can rewrite *."0 1"1 _ as #
]L3=. _3((|."1@:#:@:i.@:(2^#)) (+./"2 @: #) ])\Y
0 0 0 0 0 NB. zero
0 0 0 0 1 NB. 0 { Y
0 0 0 1 0 NB. 1 { Y
0 0 0 1 1 NB. +./ 0 1 { Y
0 0 1 0 0 NB. 2 { y
0 0 1 0 1 NB. 0 2 { Y
0 0 1 1 0 NB. +./ 0 1 { Y
0 0 1 1 1 NB. +./ 0 1 2 { Y
0 0 0 0 0 NB. zero
0 1 0 0 0 NB. 3 { Y
1 0 0 0 0 NB. 4 { Y
1 1 0 0 0 NB. +./ 3 4 { Y
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
On Wed, Oct 12, 2016 at 11:34 PM, Michal Dobrogost <
[email protected]> wrote:
> I've tried to clean up the code to be more explanatory. All we really care
> about is generating L2 given an arbitrary Y. Rewriting this has also helped
> me catch a subtle bug with arrays going left-to-right but #. and #:
> interpreting numbers right-to-left.
>
> NB. X and Y are completely arbitary and are chosen to be
> NB. easy to trace.
> ] X =. 0 1 1 0 1
> 0 1 1 0 1
> ] Y =. |.=/~i.5
> 0 0 0 0 1
> 0 0 0 1 0
> 0 0 1 0 0
> 0 1 0 0 0
> 1 0 0 0 0
>
> NB. This is the final result but we are actually trying to
> NB. generate lookup tables for highly optimized C++ code.
> X (+./@:*.) Y
> 1 0 1 1 0
>
> NB. Break up x into 2-bit chunks.
> NB. Note: reversed with |. because #. below interprets right-most
> NB. digit as most significant but J's array has first
> NB. item as left-most.
> X
> 0 1 1 0 1
> _2<@|.\X
> ┌───┬───┬─┐
> │1 0│0 1│1│
> └───┴───┴─┘
>
> NB. We convert 2-bit chunks of x into indices.
> _2#.@|.\X
> 2 1 1
>
> ]L2=. _2((|."1@:#:@:i.@:(2^#)) (+./"2 @: (*."0 1"1 _)) ])\Y
> 0 0 0 0 0 NB. zero
> 0 0 0 0 1 NB. 0 { y
> 0 0 0 1 0 NB. 1 { y
> 0 0 0 1 1 NB. +./ 0 1 { y
>
> 0 0 0 0 0 NB. zero
> 0 0 1 0 0 NB. 2 { y
> 0 1 0 0 0 NB. 3 { y
> 0 1 1 0 0 NB. +./ 2 3 { y
>
> 0 0 0 0 0 NB. zero
> 1 0 0 0 0 NB. 4 { y
> 0 0 0 0 0
> 0 0 0 0 0
>
> NB. Compute by 2-bit lookups (same result as +./@:*. above).
> NB. We don't really care, this just demonstrates that the LUT works.
> +./ (_2#.@|.\x) {"0 2 L2
> 1 0 1 1 0
>
> Cheers,
>
> Mike
>
> On Wed, Oct 12, 2016 at 7:00 PM, Michal Dobrogost <
> [email protected]> wrote:
>
>> Hi Raul,
>>
>> What is the definition of *inds*?
>>
>> I would describe the higher level operation as: select the rows of Y
>> where X is 1. Then take the column-wise OR of them.
>>
>> However the real thing I'm interested in is generating the lookup tables
>> which serve as an intermediate step in the higher level operation.
>>
>> On Wed, Oct 12, 2016 at 9:35 AM, Raul Miller <[email protected]>
>> wrote:
>>
>>> I'm not quite sure I understand what you are doing here.
>>>
>>> Here's what I understand from your description:
>>>
>>> You're looking for rows where a bit set in X has a bit set in Y, and
>>> you want to split up X and Y into smaller pieces for your intermediate
>>> result, and you want to use indices rather than bit operations for
>>> your final operation.
>>>
>>> That gives me something like this.
>>>
>>> X =: ? 5 $ 2
>>> Y =: 30 > ? 5 6 $ 100
>>> X
>>> 0 1 1 1 1
>>> Y
>>> 0 0 0 0 0 0
>>> 1 0 0 0 0 0
>>> 1 1 1 0 0 1
>>> 1 0 0 1 0 0
>>> 0 0 0 0 0 1
>>> (_2 {. X) inds _2 {. Y
>>> 0 3 5
>>> (3 {. X) inds 3 {. Y
>>> 0 1 2 5
>>> 0 3 5 ([ -. -.) 0 1 2 5
>>> 0 5
>>>
>>> So maybe you would be looking at making an adverb along the lines of:
>>>
>>> L=: adverb define
>>> :
>>> (m {. x) inds m {. y
>>> )
>>>
>>> Or, more concisely:
>>>
>>> L=: adverb define
>>> inds&(m&{.)
>>> )
>>>
>>> But when I look at your calculations, I don't see anything like this.
>>>
>>> If I am off base here, could you describe how what you are looking for
>>> differs from what I am understanding?
>>>
>>> Thanks,
>>>
>>> --
>>> Raul
>>>
>>>
>>> On Wed, Oct 12, 2016 at 12:14 PM, Michal Dobrogost
>>> <[email protected]> wrote:
>>> > Hi All,
>>> >
>>> > I've been mucking around generating look up tables in C++. Getting
>>> > frustrated, I wrote up this J single liner. Can you think of ways to
>>> > simplify the LUT computation (the expression we assign to L2 and L3
>>> below)?
>>> >
>>> > *Original Operator (no LUT)*
>>> >
>>> > ] x =. ? 5 $ 2
>>> > 1 0 1 1 0
>>> > ] y =. 30 > ? 5 6 $ 100
>>> > 1 0 0 0 0 0
>>> > 0 1 0 1 1 0
>>> > 0 1 0 0 0 1
>>> > 0 0 0 0 1 1
>>> > 0 1 1 1 0 1
>>> > x (+./@:*.) y
>>> > 1 1 0 0 1 1
>>> >
>>> > *LUT Explanation*
>>> >
>>> > The idea is to break up x into smaller chunks (2-bits, 3-bits, etc) and
>>> > precompute the operation for the corresponding chunks of y. Then we
>>> just
>>> > convert the chunks into indices and look them up in the LUT.
>>> >
>>> > _2<\x
>>> > ┌───┬───┬─┐
>>> > │1 0│1 1│0│
>>> > └───┴───┴─┘
>>> > _2#.\x
>>> > 2 3 0
>>> >
>>> > *2-bit LUT*
>>> >
>>> > ]L2=. _2((#:@:i.@:(2&^)@:#) (+./"2 @: (*."0 1"1 _)) ])\y
>>> > 0 0 0 0 0 0
>>> > 0 1 0 1 1 0
>>> > 1 0 0 0 0 0
>>> > 1 1 0 1 1 0
>>> >
>>> > 0 0 0 0 0 0
>>> > 0 0 0 0 1 1
>>> > 0 1 0 0 0 1
>>> > 0 1 0 0 1 1
>>> >
>>> > 0 0 0 0 0 0
>>> > 0 1 1 1 0 1
>>> > 0 0 0 0 0 0
>>> > 0 0 0 0 0 0
>>> >
>>> > +./ (_2#.\x) {"0 2 L2 NB. Compute by 2-bit lookups
>>> > 1 1 0 0 1 1
>>> >
>>> > *3-Bit LUT*
>>> >
>>> > ]L3=. _3((#:@:i.@:(2&^)@:#) (+./"2 @: (*."0 1"1 _)) ])\y
>>> > 0 0 0 0 0 0
>>> > 0 1 0 0 0 1
>>> > 0 1 0 1 1 0
>>> > 0 1 0 1 1 1
>>> > 1 0 0 0 0 0
>>> > 1 1 0 0 0 1
>>> > 1 1 0 1 1 0
>>> > 1 1 0 1 1 1
>>> >
>>> > 0 0 0 0 0 0
>>> > 0 1 1 1 0 1
>>> > 0 0 0 0 1 1
>>> > 0 1 1 1 1 1
>>> > 0 0 0 0 0 0
>>> > 0 0 0 0 0 0
>>> > 0 0 0 0 0 0
>>> > 0 0 0 0 0 0
>>> >
>>> > +./ (_3#.\x) {"0 2 L3 NB. Compute by 3-bit lookups
>>> > 1 1 0 0 1 1
>>> > ----------------------------------------------------------------------
>>> > For information about J forums see http://www.jsoftware.com/forums.htm
>>> ----------------------------------------------------------------------
>>> For information about J forums see http://www.jsoftware.com/forums.htm
>>
>>
>>
>
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
