Direct access, filters etc

[Kzinti]

[fast_rx]

[fzammetti]

[jaguard]

[Kzinti]

[gamefreaks]

If you are modifying the whole surface, for very fast performance, rather than calling Lock/Unlock, you might want to try modifying the PF sources and add a function to return a BackBuffer pointer.
Calling Lock/Unlock copies the display to a tempery buffer, and back again on calling Unlock.
The memory copying alone is a performance crippler!

Hope this helps.

[Dave Johnston]

[jaguard]

[Guest]

[quote="jaguard]Well, its simple. We get 3x3 mask, multiply each on coefficients, sum them all, and return value to the middle element(i[1,1] zero-bazed).
The simplest filter is blur -
[1/9, 1/9, 1/9]
[1/9, 1/9, 1/9]
[1/9, 1/9, 1/9]

[/quote]

Woah your doing how many divides and multiplies per pixel?? Im guessing this is where the huge performance hit is comming from.

[jaguard]

[Guest]

I really dont see the difference unless your using that matrix to represent something else.

Any constant multiplied or divided into a matrix has to do operate on element. Unless all those values are the same number and then you would only have to do it once.