Faster exit from _cull_canvas_item if alpha is zero

[miv391]

This is a tiny optimization for RendererCanvasCull::_cull_canvas_item(). There are a few checks in the beginning of that function which exit early if the item is not visible. Later in the function there is a check if modulate.a < 0.007 and if yes, that also means exit. Between the other checks and this alpha check some calculations are being made. Unless there are some very sneaky side effects, those calculations seem to be unnecessary if the modulate.a < 0.007 causes exit.

This PR moves modulate.a < 0.007 to be made right after other visibility checks.

My test project has 10000 sprites total, half of them (5000) have modulate.a = 0 and thus they are not visible. Results of my Visual Studio release build:

	fps
master	310
this PR	340

I don't believe having a huge amount of sprites with alpha = 0 is a common use case. but this should not make any use case slower either.

[miv391]

Anyone interested in reviewing this PR? The change is very small.

[clayjohn]

The only concern I have is for the children nodes, Its possible for a node to have a module of 0, but for its children to be drawn. With this change, the children would not inherit the parent material or be sorted.

Perhaps the branches with side effects could also be moved up so they still happen before this check?

[miv391]

I moved all code with side effects to be executed before making the alpha check. Compared to the original code, only the material owner stuff and the alpha test itself was moved.

[clayjohn]

This makes sense and should be safe since the parts that have moved don't rely on the parts that now happen after the check.

The one thing that maaaaay cause issues is the line immediately below this change Rect2 rect = ci->get_rect(); as it updates the bounding rect for the canvas item. With this change the bounding rect will no longer be calculated when the item is hidden from having a low modulate.

I think this is fine because get_rect() caches the rect. So even if its not generated here, it is guaranteed to be generated before it is accessed.

I'm not suggesting it get moved for safety because I suspect that most of the performance benefit of this change comes from not re-calculating the rect. And for the above reasons I think it is safe not to call it in this situation.

[AThousandShips]

Suggested change

	Color modulate(ci->modulate.r * p_modulate.r, ci->modulate.g * p_modulate.g, ci->modulate.b * p_modulate.b, ci->modulate.a * p_modulate.a);
	Color modulate = ci->modulate * p_modulate;

There's already an operator for this, let's improve readability here while we're at it

[miv391]

Fixed.

[Repiteo]

Thanks!