Re: Atkinson dither algorithm & 'for each' loop

[Lagi Pittas via use-livecode]

HI All

Hoping I'm not teaching my granny to suck eggs ... Here is a great article
I remembered reading  from a few years ago. It's easy to find on google so
you probably know of it ...

http://www.tannerhelland.com/4660/dithering-eleven-algorithms-source-code/

In any case his projects and how he writes about them are interesting

 http://www.tannerhelland.com/programming-directory/

Lagi

On 12 October 2017 at 12:05, Peter Reid via use-livecode <
use-livecode@lists.runrev.com> wrote:

> One note of caution regarding the use of the "repeat for each" loop,
> whilst you will get a loop iteration for every value in the collection
> (fldhexa3 in your example), you are not guaranteed the order in which they
> will occur.  This doesn't matter in a lot of cases but does matter when the
> sequence is significant.  In the case of your example I believe sequence is
> critical, otherwise the pixels might appear to be scrambled!
>
> The following adjusted loop guarantees the sequence at the expense of
> speed:
>
>   put 1 into i
>   repeat for each word theWord in fldhexa3
>  put word i of fldhexa3 into theWord
>  put 00 & theword & theword & theword after tVar2
>  add 1 to i
>   end repeat
>
> The original "improved" loop reduces the run-time to 25%.  However, the
> "modified improved" loop only manages to reduce the original run-time to
> 50%.
>
> The suggested loop above takes advantage of the "for each" mechanism to
> produce a set of iterations very rapidly but gets slowed by the need to
> guarantee sequence. I wonder whether the LC engine could impose strict
> sequence more effectively with a variant of the "for each" loop such as
>
>   repeat for each sequenced word x in theCollection
>  ...
>   end repeat
>
> My own tests, comparing the speed of the 4 common repeat loops, imply that
> the current "for each" form is hugely faster than the others.  I tested
> "repeat for each...", "repeat while...", "repeat until...", "repeat
> with..." and a simulated "repeat for each sequenced..." forms using a
> simple loop body that added lines of text one after another, e.g.
>
>   put empty into tData
>   repeat with i = 1 to tMaxI
> put line i of tList & return after tData
>   end repeat
>
> I ran this test for 250,000 iterations for each type of loop, which
> produced the following timings:
>
>   Starting test for 250,000 iterations...
>   repeat for each... 0 mins 0 secs 111 millisecs
>   repeat while... 0 mins 30 secs 569 millisecs
>   repeat until... 0 mins 30 secs 379 millisecs
>   repeat with... 0 mins 30 secs 341 millisecs
>   repeat for each seq... 0 mins 30 secs 524 millisecs
>
> As you can see, in this test the "repeat for each..." form was approx. 275
> times faster than the other forms.  Also the simulated "repeat for each
> sequenced..." form was no faster than the other forms.  This shows how
> variable the speed will be with the simulated "repeat for each
> sequenced...", depending on the details of the loop body.
>
> If there was a "repeat for each sequenced..." form of loop in LC, any
> speed-up could be very beneficial even if the amount of speed-up was only
> 10 times faster!
>
> Cheers
>
> Peter
> --
> Peter Reid
> Loughborough, UK
>
> > On 9 Oct 2017, at 10:18am, use-livecode-requ...@lists.runrev.com wrote:
> >
> > Message: 12
> > Date: Sat, 7 Oct 2017 15:53:44 +0200
> > From: Malte Pfaff-Brill 
> > To: use-livecode@lists.runrev.com
> > Subject: Re: Atkinson dither algorithm
> > Message-ID: <42023b36-0a4e-4251-bb0c-9cd46de55...@derbrill.de>
> > Content-Type: text/plain; charset=us-ascii
> >
> > Hi Al,
> >
> > I already posted on the forums, but for completeness also here:
> >
> > a lot can be done by replacing repeat with with repeat for each where
> you can.
> >
> > --repeat with i = 1 to the number of words of fldhexa3
> >   --   put 00 & word i of fldhexa3 & word i of fldhexa3 & word i of
> fldhexa3 after tVar2
> >   --end repeat
> >
> >   repeat for each word theWord in fldhexa3
> >  put 00 & theword & theword & theword after tVar2
> >   end repeat
> >
> >
> > A sidenode:
> >
> > I always use strict compile mode, therefore I added the needed variable
> declarations and noticed you use startTime as a variablename, which is a
> reserved keyword. That is not a good idea.  (I noticed, because I managed
> to freeze liveCode where I fixed only half of the use of startTime. Booom.)
> >
> > Cheers,
> >
> > malte
>
>
> ___
> use-livecode mailing list
> use-livecode@lists.runrev.com
> Please visit this url to subscribe, unsubscribe and manage your
> subscription preferences:
> http://lists.runrev.com/mailman/listinfo/use-livecode
>
___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Atkinson dither algorithm & 'for each' loop

[Alejandro Tejada via use-livecode]

After reading Mark Waddingham explanation, now i am using a sequential
ordered array for this handler. Check this much shorter and faster version:

on mouseUp

   put the millisecs into startTime
   set the cursor to busy

   put the alphadata of img "Image" into tAlphaData
   put the imagedata of img "Image" into tVar
   put the width of img "Image" into tImageWidth
   put the height of img "Image" into tImageHeight
   put the number of chars of tVar into tImgPixels

   if existence(sb the "ThresholdDither") then
  put thumbPos of sb the "ThresholdDither" into tThreshold
   else
  put 127 into tThreshold
   end if

   put numtochar(0) & numtochar(255) & numtochar(255) & numtochar(255) into
tWP
   put numtochar(0) & numtochar(0) & numtochar(0) & numtochar(0) into tBP
   put ((tImageWidth * tImageHeight) - tImageWidth) into tHW

   put ImgToArrayNum(ImgToCh(tVar,1)) into tArray2

   put 0 into tPixelCounter
   put tImgPixels/4 into tPixels
   repeat tPixels

  add 1 to tPixelCounter
  put tPixelCounter into tPixelPosition

  put tArray2[tPixelCounter] into tOldPixelValue

  if round(tOldPixelValue) <= tThreshold then
 put 0 into tNewPixelValue
 put tBP after fldhexa8
 else
 put 255 into tNewPixelValue
 put tWP after fldhexa8
 end if

  put (tOldPixelValue - tNewPixelValue)/8 into tDifusionError

  -- Atkinson dither add the diffusion error
  -- to 6 adjacent pixels
  -- x o o
  --  o o o
  -- o

  if tPixelPosition mod tImageWidth <> 0 then
 -- pixel position is less than image width
 put tPixelCounter + 1 into tNewKey
 put (tDifusionError) + tArray2[tNewKey] into tArray2[tNewKey]

 if tPixelPosition mod (tImageWidth - 1) <> 0 then
-- pixel position is less than image width - 1
put tPixelCounter + 2 into tNewKey
put tDifusionError + tArray2[tNewKey] into tArray2[tNewKey]
 end if
  end if

  if tPixelPosition <= tHW then -- ((tImageWidth * tImageHeight) -
tImageWidth)
 -- pixel position is not in the last line of the image

 if tPixelPosition mod tImageWidth <> 1 then
-- pixel position is not the first pixel in any line of the
image
put (tPixelPosition + tImageWidth - 1) into tNewKey
put (tDifusionError) + tArray2[tNewKey] into tArray2[tNewKey]
 end if

 put (tPixelPosition + tImageWidth) into tNewKey
 put (tDifusionError) + tArray2[tNewKey] into tArray2[tNewKey]

 if tPixelPosition mod tImageWidth <> 0 then
-- pixel position is less than image width
put (tPixelPosition + tImageWidth + 1) into tNewKey
put (tDifusionError) + tArray2[tNewKey] into tArray2[tNewKey]
 end if

 if tPixelPosition <= (tImageWidth * tImageHeight) - (tImageWidth *
2) then
-- pixel position is less than image width - 1
put (tPixelPosition + tImageWidth * 2) into tNewKey
put (tDifusionError) + tArray2[tNewKey] into tArray2[tNewKey]
 end if
  end if

   end repeat

   create img
   set the height of it to the height of img "Image"
   set the width of it to the width of img "Image"
   set the imagedata of it to fldhexa8
   set the alphaData of it to tAlphaData

   put the millisecs - startTime && "milliseconds to create Atkinson Dither
from image's single channel"

end mouseUp

Function ImgToArrayNum vImageData
   -- This function converts binary imagedata to
   -- integer numbers from 0 to 255 into
   -- an array.
   -- vImageData is a single color channel
   -- stored as binary imagedata.

   put empty into tResult
   put 0 into tCounter
   repeat for each char K in vImageData
  add 1 to tCounter
  put chartonum(K) into tResult[tCounter]
   end repeat
   return tResult
end ImgToArrayNum

Function ImgToCh tImageData tChannel
   -- This function returns binary data.
   --
   -- tImageData is unmodified original imagedata of image
   -- with 4 chars for each pixel: 1 alphadata and 3 color channels.
   --
   -- tChannel is a number from 1 to 3:
   -- 1 is red channel, 2 is green channel and 3 is blue channel

   put tImageData into tempVar

   delete char 1 to tChannel of tempVar
   -- the first char of the imagedata is part
   -- of the alphadata or maskdata, so when
   -- we delete char 1, the next char is part
   -- from red channel... if we delete 2 first
   -- chars, then next char is green channel
   -- if we delete 3 first chars, then we get
   -- the blue channel

  repeat until tempVar is empty
  put char 1 of tempVar after tResult
  delete char 1 to 4 of tempVar
   end repeat

   return tResult

end ImgToCh

Have a nice weekend!

Al
___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:

Re: Atkinson dither algorithm & 'for each' loop

[hh via use-livecode]

@Al.
For multiples of 1/8 (Atkinson) you need 8*256=2048 integers,
that is 11 Bit.
For multiples of 1/16 (Floyd-Steinberg) you need 16*256=4096
integers, that is 12 Bit.
In 2 chars = 16 Bit fit even multiples of 1/256.
No dither-algorithm uses such tiny diffusion-errors.

___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Atkinson dither algorithm & 'for each' loop

[Alejandro Tejada via use-livecode]

Hi Hermann,

Hermann Hoch wrote:
> So all considerations to improve the LCS script are,
> TMHO, of rather theoretical value. Here are some more:
> The Atkinson algorithm uses multiples of 1/8 = 0.125 for
> the diffusion-error, the Floyd-Steinberg algorithm uses
> multiples of 1/16 = 0.0625.
> And have this in mind:
> Whenever you use numToChar(Number) in your script you
> are wrong by Number - round(Number) of the accumulated
> pixel value ...

This is correct, for this reason, in the most recent version of
this handler, I am using 2 chars to store the new pixel value.
first char is a whole number from 0 to 255 and the second
char represent a decimal part from 0 to 99.

I will test if using 5 chars for storing the new pixel value.
In this way, the new pixel could store values of
.00 that should be enough for this task...

After reading Mark's answer, I learned that using a numerical
ordered array to store pixel values could be really fast too.
This is another path to explore.

Al
___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Re: Atkinson dither algorithm & 'for each' loop

[hh via use-livecode]

> Al wrote:
> Please, check this new handler in the forum and suggest how to
> make this faster and more precise.

I already gave you a method that is, with a 500x500 image, at
least 1000 (thousand) times faster than your script.

Of course this is not due my special skills but the ability of
javascript (used via a browser widget) to let the GPU/graphic
card do the essential job of pixel manipulation.

So all considerations to improve the LCS script are, TMHO, of
rather theoretical value. Here are some more:

> Al wrote:
> ... the result STILL is not visually identical to that handler.
> My educated guess is that Atkinson dither algorithm requires
> numbers with more than 2 numbers after the decimal point.
> That is: 243.643578 instead of just 243.64

Conclude from the following:
The Atkinson algorithm uses multiples of 1/8 = 0.125 for the
diffusion-error, the Floyd-Steinberg algorithm uses multiples
of 1/16 = 0.0625. 

And have this in mind:
Whenever you use numToChar(Number) in your script you are wrong by
Number - round(Number) of the accumulated pixel value ...

___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Re: Atkinson dither algorithm & 'for each' loop

[Mark Waddingham via use-livecode]

On 2017-10-12 19:35, Alejandro Tejada via use-livecode wrote:

Peter Read wrote:

One note of caution regarding the use of the "repeat for each" loop,
whilst you will get a loop iteration for every value in the collection
(fldhexa3 in your example), you are not guaranteed the order in
which they will occur. This doesn't matter in a lot of cases but
does matter when the sequence is significant.  In the case of your
example I believe sequence is critical, otherwise the pixels might
appear to be scrambled!


Prior to 7, you had to be careful not to modify either the container 
being iterated over *or* the var holding the chunk for the current 
iteration as it could cause issues as described (actually, it was 
fundamentally unsafe to do that, as it could result in a crash). i.e.


  repeat for each item tItem in tContainer
replace "Foo" with "B" in tContainer -- BAD - MIGHT CAUSE PROBLEMS
put "a' into char 2 of tItem -- BAD - MIGHT CAUSE PROBLEMS
  end repeat

From 7 onwards, this is no longer the case. You can do whatever you want 
with either the container. Indeed, in 7, the engine essentially operates 
as if it was iterating over a copy of the container and with a copy of 
the chunk (i.e. tItem and tContainer in the above snippet).


Upshot - you can do whatever you want with your variables - the engine 
ensures that what you do won't affect what you asked for to begin with 
(in particular, if you asked to iterate over tContainer, then it will 
iterate over tContainer as it was at the point you first asked - any 
changes to tContainer after the initial repeat for each line will have 
no effect on the repeat for each clause).


Warmest Regards,

Mark.

--
Mark Waddingham ~ m...@livecode.com ~ http://www.livecode.com/
LiveCode: Everyone can create apps

___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Atkinson dither algorithm & 'for each' loop

[Alejandro Tejada via use-livecode]

A few hours ago, I send a message to the mail list, a message that vanish
without trace. Now I am resending parts of that message. The rest of the
message is in the forum...

Peter Read wrote:
> One note of caution regarding the use of the "repeat for each" loop,
> whilst you will get a loop iteration for every value in the collection
> (fldhexa3 in your example), you are not guaranteed the order in
> which they will occur. This doesn't matter in a lot of cases but
> does matter when the sequence is significant.  In the case of your
> example I believe sequence is critical, otherwise the pixels might
> appear to be scrambled!

In my own setup, the control structure "repeat for each" is caching
the content of the working variable. This means that we could not
modify the content of this variable while the control structure is
working with it.

After reading your message, now I understand why this happens.
If developers are allowed to modify a variable used by a control
structure like "repeat for  each" then we could get unexpected
result as you describe: elements are processed in an out of order
sequence.

Ben Rubinstein wrote in response to Peter Reid:
> Are you sure? My understanding has always been that
> chunk items, e.g.
> repeat for each [ byte | char | word | item | line]  in 
> will always be sequential (indeed that's why this structure is fast)
> - it's only when dealing with hashed arrays that the sequence is
> not reliable, i.e.
> repeat for each key  in 
> repeat for each element  in 
> Do you have experience to the contrary?

As I wrote earlier in this same message, Peter could have found
some way to change the working variable in a repeat for each
control structure and the result is that elements are processed
with an out of order sequence.

By the way, after reading all comments and recommendations
about the atkinson dither algorithm, I wrote a new handler that
is a lot faster that my first handler, but the result STILL is not
visually identical to that handler.

My educated guess is that Atkinson dither algorithm requires
numbers with more than 2 numbers after the decimal point.
That is: 243.643578 instead of just 243.64
Later this week, I will try to discover the minimal number of
decimal numbers that this algorithm requires.

Please, check this new handler in the forum and suggest how to make
this faster and more precise:

https://forums.livecode.com/viewtopic.php?f=10=159078=b46bc56428cb5b0f48d2c6520c107c47#p159078

Have a nice week!

Al
___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Re: Atkinson dither algorithm & 'for each' loop

[Richard Gaskin via use-livecode]

Peter Reid wrote:

> One note of caution regarding the use of the "repeat for each" loop,
> whilst you will get a loop iteration for every value in the collection
> (fldhexa3 in your example), you are not guaranteed the order in which
> they will occur.

Maybe I misunderstand, but are you thinking of arrays there?

With string containers a "repeat for each" expression should parse from 
beginning to end, sequentially.


Any exception to that would be prohibitively unpredictable, and should 
be reported as a bug.



> The following adjusted loop guarantees the sequence at the expense of
> speed:
>
>   put 1 into i
>   repeat for each word theWord in fldhexa3
>  put word i of fldhexa3 into theWord

The speed lost there appears to be the result of a redundancy, and a 
particularly expensive one:


We love the convenience of chunk expressions, but in loops we want to 
remain mindful of "  of " because satisfying 
such expressions will require the engine to start from the beginning of 
the container, examine every character and counting delimiters, until it 
reaches the number of such delimiters specified in "".


So the "repeat" line above parses the chunk value into theWord, but then 
the next line ignores that that's already happened and reassigns the 
same theWord value using an expression that requires then engine to 
count words from the beginning of fldhexa3 each time through the loop.



With this specific algo I believe there may be further benefits in using 
a chunk type other than word (based on a a hunch derived from word 
chunks being parsed more flexibly than items or lines), and perhaps not 
converting the binary data to hex at all, instead parsing bytes directly 
with a "step 4" option in the loop to keep track of the four components 
that define each pixel.


--
 Richard Gaskin
 Fourth World Systems
 Software Design and Development for the Desktop, Mobile, and the Web
 
 ambassa...@fourthworld.comhttp://www.FourthWorld.com

___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Re: Atkinson dither algorithm & 'for each' loop

[Bob Sneidar via use-livecode]

That is correct Ben. It's not the repeat for each that is unreliable (probably 
a bad word to use here) but it is arrays which do not retain the sequence of 
key/values in the order they were put in. 

To get around this, when possible use numbered keys, then:

put  the keys of aMyArray into tKeyList
sort the lines of tKeyList numeric ascending
repeat for each line tKey in tKeyList
...

Bob S


> On Oct 12, 2017, at 04:48 , Ben Rubinstein via use-livecode 
>  wrote:
> 
> Hi Peter,
> 
>> One note of caution regarding the use of the "repeat for each" loop, whilst 
>> you will get a loop iteration for every value in the collection
>> (fldhexa3 in your example), you are not guaranteed the order in which they
>> will occur.
> 
> Are you sure? My understanding has always been that chunk items, e.g.
>   repeat for each [ byte | char | word | item | line]  in 
> 
> will always be sequential (indeed that's why this structure is fast) - it's 
> only when dealing with hashed arrays that the sequence is not reliable, i.e.
>   repeat for each key  in 
>   repeat for each element  in 
> 
> Do you have experience to the contrary?
> 
> best regards,
> 
> Ben


___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Re: Atkinson dither algorithm & 'for each' loop

[Ben Rubinstein via use-livecode]

Hi Peter,

One note of caution regarding the use of the "repeat for each" loop, 
whilst you will get a loop iteration for every value in the collection

(fldhexa3 in your example), you are not guaranteed the order in which they
will occur.


Are you sure? My understanding has always been that chunk items, e.g.
repeat for each [ byte | char | word | item | line]  in 

will always be sequential (indeed that's why this structure is fast) - it's 
only when dealing with hashed arrays that the sequence is not reliable, i.e.

repeat for each key  in 
repeat for each element  in 

Do you have experience to the contrary?

best regards,

Ben

On 12/10/2017 12:05, Peter Reid via use-livecode wrote:

One note of caution regarding the use of the "repeat for each" loop, whilst you 
will get a loop iteration for every value in the collection (fldhexa3 in your example), 
you are not guaranteed the order in which they will occur.  This doesn't matter in a lot 
of cases but does matter when the sequence is significant.  In the case of your example I 
believe sequence is critical, otherwise the pixels might appear to be scrambled!

The following adjusted loop guarantees the sequence at the expense of speed:

   put 1 into i
   repeat for each word theWord in fldhexa3
  put word i of fldhexa3 into theWord
  put 00 & theword & theword & theword after tVar2
  add 1 to i
   end repeat

The original "improved" loop reduces the run-time to 25%.  However, the "modified 
improved" loop only manages to reduce the original run-time to 50%.

The suggested loop above takes advantage of the "for each" mechanism to produce a set of 
iterations very rapidly but gets slowed by the need to guarantee sequence. I wonder whether the LC 
engine could impose strict sequence more effectively with a variant of the "for each" 
loop such as

   repeat for each sequenced word x in theCollection
  ...
   end repeat

My own tests, comparing the speed of the 4 common repeat loops, imply that the current "for each" form is hugely faster than the 
others.  I tested "repeat for each...", "repeat while...", "repeat until...", "repeat with..." and 
a simulated "repeat for each sequenced..." forms using a simple loop body that added lines of text one after another, e.g.

   put empty into tData
   repeat with i = 1 to tMaxI
 put line i of tList & return after tData
   end repeat

I ran this test for 250,000 iterations for each type of loop, which produced 
the following timings:

   Starting test for 250,000 iterations...
   repeat for each... 0 mins 0 secs 111 millisecs
   repeat while... 0 mins 30 secs 569 millisecs
   repeat until... 0 mins 30 secs 379 millisecs
   repeat with... 0 mins 30 secs 341 millisecs
   repeat for each seq... 0 mins 30 secs 524 millisecs

As you can see, in this test the "repeat for each..." form was approx. 275 times faster than the 
other forms.  Also the simulated "repeat for each sequenced..." form was no faster than the other 
forms.  This shows how variable the speed will be with the simulated "repeat for each 
sequenced...", depending on the details of the loop body.

If there was a "repeat for each sequenced..." form of loop in LC, any speed-up 
could be very beneficial even if the amount of speed-up was only 10 times faster!

Cheers

Peter
--
Peter Reid
Loughborough, UK


On 9 Oct 2017, at 10:18am, use-livecode-requ...@lists.runrev.com wrote:

Message: 12
Date: Sat, 7 Oct 2017 15:53:44 +0200
From: Malte Pfaff-Brill 
To: use-livecode@lists.runrev.com
Subject: Re: Atkinson dither algorithm
Message-ID: <42023b36-0a4e-4251-bb0c-9cd46de55...@derbrill.de>
Content-Type: text/plain; charset=us-ascii

Hi Al,

I already posted on the forums, but for completeness also here:

a lot can be done by replacing repeat with with repeat for each where you can.

--repeat with i = 1 to the number of words of fldhexa3
   --   put 00 & word i of fldhexa3 & word i of fldhexa3 & word i of 
fldhexa3 after tVar2
   --end repeat

   repeat for each word theWord in fldhexa3
  put 00 & theword & theword & theword after tVar2
   end repeat


A sidenode:

I always use strict compile mode, therefore I added the needed variable 
declarations and noticed you use startTime as a variablename, which is a 
reserved keyword. That is not a good idea.  (I noticed, because I managed to 
freeze liveCode where I fixed only half of the use of startTime. Booom.)

Cheers,

malte



___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode



___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode

Re: Atkinson dither algorithm & 'for each' loop

[Peter Reid via use-livecode]

One note of caution regarding the use of the "repeat for each" loop, whilst you 
will get a loop iteration for every value in the collection (fldhexa3 in your 
example), you are not guaranteed the order in which they will occur.  This 
doesn't matter in a lot of cases but does matter when the sequence is 
significant.  In the case of your example I believe sequence is critical, 
otherwise the pixels might appear to be scrambled!

The following adjusted loop guarantees the sequence at the expense of speed:

  put 1 into i
  repeat for each word theWord in fldhexa3
 put word i of fldhexa3 into theWord
 put 00 & theword & theword & theword after tVar2
 add 1 to i
  end repeat

The original "improved" loop reduces the run-time to 25%.  However, the 
"modified improved" loop only manages to reduce the original run-time to 50%.

The suggested loop above takes advantage of the "for each" mechanism to produce 
a set of iterations very rapidly but gets slowed by the need to guarantee 
sequence. I wonder whether the LC engine could impose strict sequence more 
effectively with a variant of the "for each" loop such as

  repeat for each sequenced word x in theCollection
 ...
  end repeat

My own tests, comparing the speed of the 4 common repeat loops, imply that the 
current "for each" form is hugely faster than the others.  I tested "repeat for 
each...", "repeat while...", "repeat until...", "repeat with..." and a 
simulated "repeat for each sequenced..." forms using a simple loop body that 
added lines of text one after another, e.g.

  put empty into tData
  repeat with i = 1 to tMaxI
put line i of tList & return after tData
  end repeat

I ran this test for 250,000 iterations for each type of loop, which produced 
the following timings:

  Starting test for 250,000 iterations...
  repeat for each... 0 mins 0 secs 111 millisecs
  repeat while... 0 mins 30 secs 569 millisecs
  repeat until... 0 mins 30 secs 379 millisecs
  repeat with... 0 mins 30 secs 341 millisecs
  repeat for each seq... 0 mins 30 secs 524 millisecs

As you can see, in this test the "repeat for each..." form was approx. 275 
times faster than the other forms.  Also the simulated "repeat for each 
sequenced..." form was no faster than the other forms.  This shows how variable 
the speed will be with the simulated "repeat for each sequenced...", depending 
on the details of the loop body.

If there was a "repeat for each sequenced..." form of loop in LC, any speed-up 
could be very beneficial even if the amount of speed-up was only 10 times 
faster!

Cheers

Peter
--
Peter Reid
Loughborough, UK

> On 9 Oct 2017, at 10:18am, use-livecode-requ...@lists.runrev.com wrote:
> 
> Message: 12
> Date: Sat, 7 Oct 2017 15:53:44 +0200
> From: Malte Pfaff-Brill 
> To: use-livecode@lists.runrev.com
> Subject: Re: Atkinson dither algorithm  
> Message-ID: <42023b36-0a4e-4251-bb0c-9cd46de55...@derbrill.de>
> Content-Type: text/plain; charset=us-ascii
> 
> Hi Al,
> 
> I already posted on the forums, but for completeness also here:
> 
> a lot can be done by replacing repeat with with repeat for each where you can.
> 
> --repeat with i = 1 to the number of words of fldhexa3
>   --   put 00 & word i of fldhexa3 & word i of fldhexa3 & word i of 
> fldhexa3 after tVar2
>   --end repeat
> 
>   repeat for each word theWord in fldhexa3
>  put 00 & theword & theword & theword after tVar2
>   end repeat
> 
> 
> A sidenode:
> 
> I always use strict compile mode, therefore I added the needed variable 
> declarations and noticed you use startTime as a variablename, which is a 
> reserved keyword. That is not a good idea.  (I noticed, because I managed to 
> freeze liveCode where I fixed only half of the use of startTime. Booom.)
> 
> Cheers,
> 
> malte


___
use-livecode mailing list
use-livecode@lists.runrev.com
Please visit this url to subscribe, unsubscribe and manage your subscription 
preferences:
http://lists.runrev.com/mailman/listinfo/use-livecode