[squeak-dev] Quadrangle >> exampleInViewer baby graphics bug
Lawson English
lenglish5 at cox.net
Sun Apr 25 11:01:33 UTC 2010
Lawson English wrote:
> Igor Stasenko wrote:
>> On 24 April 2010 10:55, Lawson English <lenglish5 at cox.net> wrote:
>>
>>> Lawson English wrote:
>>>
>>>> Thanks. I had gotten that far with John Dougan's help and tried to
>>>> save
>>>> back to the repository but it wouldn't accept since I lack a
>>>> password. I
>>>> converted all references of Floats to Fractions just to see if
>>>> there was an
>>>> easy way to give it higher precision but it slowed down to < Apple
>>>> ][ speeds
>>>> when I did that.
>>>>
>>>>
>>>> Obviously the naive way isn't going to work. I gotta think there's
>>>> some
>>>> glitch with what I did, or incompatibility with the algorithm and
>>>> Fractions....
>>>>
>>>>
>>>>
>>> Or maybe things like x^100/y^100 takes a rather long time to
>>> calculate...
>>>
>>>
>>
>> you can optimize, rather than multiplying x*x*...*x 100 times in a row,
>>
>> use a power of two numerics.
>> Given that:
>> x^y = (x^a)*(x^b)
>> where y = a+b
>>
>> so,
>>
>> x^100 = (x^64)*(x^32)*(x^4)
>>
>> then, you can reuse an intermediate results of computation i.e:
>>
>> x2 := x*x.
>> x4 := x2*x2.
>> x8 := x4*x4.
>> x16 := x8*x8.
>> x32 := x16*x16.
>> x64 := x32*x32.
>>
>> result := x64*x32*x4.
>>
>> so, totally 8 multiplications, instead of 100.
>>
>>
>
my bad. The algorithm for the M set is recursive, 100 timesRepeat: [z:=
z squared]
so you're not talking z^100 but z^2^100. which is why I truncated it to
1000 decimals...
> where the real and imaginary were already set as scaledDecimal: 1000
>
> so a test after each new run would give 7 levels of color for z^128
> and run 12 times faster.
>
> only 1.4 seconds per pixel. Getting close... ;-)
>
>
> Lawson
>
>
>
>
>
>
>
>
More information about the Squeak-dev
mailing list
|