Differential D13383 Diff 36062 libs/pigment/compositeops/KoCompositeOpFunctions.h

Changeset View

Standalone View

View Options

libs/pigment/compositeops/KoCompositeOpFunctions.h

Context not available.
453template<class T> 453​template<class T>
454inline T cfLightenOnly(T src, T dst) { return qMax(src, dst); } 454​inline T cfLightenOnly(T src, T dst) { return qMax(src, dst); }
455 455​
456​template<class T>
457​inline T cfGlow(T src, T dst) {
458​ using namespace Arithmetic;
459​ // see http://www.pegtop.net/delphi/articles/blendmodes/quadratic.htm for formulas of Quadratic Blending Modes like Glow, Reflect, Freeze, and Heat
460​
461​ if(dst == unitValue<T>()) {
dkazakovUnsubmitted
Not Done

I would also add to make the code a bit faster:

if (src == zeroValue<T>()) {
    return zeroValue<T>();
}
dkazakov: I would also add to make the code a bit faster: ``` if (src == zeroValue<T>()) { return…
ReptorianAuthorUnsubmitted
Not Done

I decided to remove this as there was issues with the rendering result that leads to undesirable result like 0, 1, 1 among many areas. Now, I followed your suggestion of (Code()) {return;} for heat and freeze during the meantime.

Reptorian: I decided to remove this as there was issues with the rendering result that leads to…
dkazakovUnsubmitted
Not Done

How can it cause flipping problems? If I understand the formula correctly

return clamp<T>(div(mul(src, src), inv(dst)));

when src is null, then the result will also be null:

clamp((0*0) / inv(dst)) -> 0 for any dst

Or I don't understand something :)

dkazakov: How can it cause flipping problems? If I understand the formula correctly ``` return clamp<T>…
462​ return unitValue<T>();
463​ }
464​
dkazakovUnsubmitted
Not Done

Instead of direct multiplication and division operations (and pow() function), one should use functions from Arithmetic namespace here. Otherwise, the values may become denormalized (e.g. int8 colorspaces are normalized in range 0...255 instead of standard 0...1.0). E.e. this line should look like:

return clamp<T>(div(mul(src, src), inv(dst)));
dkazakov: Instead of direct multiplication and division operations (and pow() function), one should use…
465​ return clamp<T>(div(mul(src, src), inv(dst)));
466​}
467​
468​template<class T>
469​inline T cfReflect(T src, T dst) {
470​ using namespace Arithmetic;
471​
dkazakovUnsubmitted
Done

This check is not needed here, because there is already a check in cfGlow. More than that, it is incorrect, when dst is a unit value, the result may have any value in range 0...1.

dkazakov: This check is not needed here, because there is already a check in `cfGlow`. More than that, it…
472​ return (cfGlow(dst,src));
473​}
474​
dkazakovUnsubmitted
Done

Second clamp is not needed here

dkazakov: Second `clamp` is not needed here
475​template<class T>
476​inline T cfHeat(T src, T dst) {
477​ using namespace Arithmetic;
478​ // Heat, and Freeze only works properly on 8-bit images. It does not work properly on any other color depth. For now, if Heat and Freeze are proven useful for 8-bit painting, then there should be some way of solving this issue.
479​
480​ if(dst == zeroValue<T>()) {
481​ return zeroValue<T>();
482​ }
483​
484​ if(src == unitValue<T>()) {
485​ return unitValue<T>();
486​ }
487​
488​ return inv(clamp<T>(div(mul(inv(src), inv(src)),dst)));
489​}
490​
491​template<class T>
492​inline T cfFreeze(T src, T dst) {
493​ using namespace Arithmetic;
494​
495​ return clamp<T>(cfHeat(dst,src));
496​}
497​
498​
456#endif // KOCOMPOSITEOP_FUNCTIONS_H_ 499​#endif // KOCOMPOSITEOP_FUNCTIONS_H_
Context not available.