compileIt for revolution?

[Alex Tweedly]

I sent this message Saturday, and it just came back to me as bounced 
(because the list-server was off-line).
So here it is again - with an addendum ...

>Geoff Canyon wrote:
>
>  
>
>>On Jun 24, 2005, at 4:26 AM, Jon wrote:
>>
>>    
>>
>>>With all due respect, Jim, if you are trying to do even simple math  
>>>on a large array of numbers (like computing a histogram of image  
>>>data), Rev is simply too slow to use.  It has nothing to do with  
>>>data structures: it has to do with slow pCode trying to run tight  
>>>loops.
>>>      
>>>
>>As stated previously, I cannot accept this statement/conclusion  
>>without seeing the code in question. There are simply too many ways  
>>to write slow code to assume that Rev can't do a task based solely on  
>>a very general description -- unless the description involves real  
>>time first-person gaming ;-)
>>    
>>
>
>Well, the problem Jon mentioned (histogram of image data) is pretty well 
>defined. To be usable, you need to do it for reasonable size pictures in 
>less than 1-2 secs. (where "reasonable" is defined to be whatever your camera takes, since 
>that's the data you're most likely to want a histogram of, in 
>preparation for controlling contrast, colour balance, etc.)
>
>Naive code to do it is something like
>  
>
>>  put 0 into count
>>  repeat for each char c in the imageData of img "Image"
>>    switch count mod 4
>>    case 0
>>      break
>>    case 1
>>      add 1 to tRed[chartonum(c)]
>>      break
>>    case 2
>>      add 1 tGreen[chartonum(c)]
>>      break
>>    case 3
>>      add 1 to tBlue[chartonum(c)]
>>--      put -1 into count
>>      break
>>    default
>>    put "error " && count into msg
>>    end switch
>>    add 1 to count
>>  end repeat
>>  displayHistogram tRed, tGreen, tBlue
>>    
>>
>
>For me, on my fairly puny laptop, that takes around 175 seconds, so I 
>only need a 300x speedup.
>  
>
And with an equally naive C external, it takes about 400 ms - about a 
quarter of which is formatting the result back suitable for Rev, and 
then Rev reformatting it suitable for my displayHistogram function.

I tried some other Transcript approaches (I think the fastest one I 
found was
 replacing the unused bytes by 'cr's
 sorting lines by char 1 of each   (remembering to set caseSensitive to 
true first !!)
 scanning through the lines, counting as you go until char 1 changes,
            then putting that count into the appropriate array element
 and repeat the sort + scan for char 2 and char 3 in turn

This brought the time needed for a small photo (800x600) from 14 secs 
down to 7; unfortunately on a large photo that took more than 5 minutes, 
and increased my Windows Virtual Memory twice. But even on the small 
photo the time taken is still enough to make it unusable.
 

-- 
Alex Tweedly       http://www.tweedly.net

-------------- next part --------------
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.323 / Virus Database: 267.8.1/28 - Release Date: 24/06/2005