Gabriel Wang

> How to break complex tasks into simpler ones? > 1. Let the 2D GPU do the "transform with mask and opa" and save the result to a temporary frame-buffer...

> If I see it correctly the need of task groups is derived from the issue mentioned in the previous point (handle complex operation in a suboptimal way) and handle...

> Anyway, it's not trivial. The takeaway for me is that a simple pe->caps & task->cap test is not enough because there are a lot of factors to consider. Probably...

> For the dependency question: I wonder if there could be a point when we malloc a temporary frame buffer to "restart" the dependency chain on an area. E.g. while...

> However, I think rounded rectangle should not be "mask + fill" as some hardware are capable of drawing rounded rectangles directly. The hardware still can fallback to "generate mask...

> I'm still not sure about it. Imagine that there is a complex task group, e.g. A + B + C + D. There is no PE which can do...

@kisvegabor As we mentioned before the PE will provide a capability description. And in the capability matching, PE will provides a score together with a less optimal solution. For example,...

> can you mention an example task group? The 1st floor provided an example. If without a task group, I have no idea how to handle the capability matching in...

> > fill: coords, color/gradient, radius border: coords, width, color/gradient, radius shadow: coords, rectangle radius, blur radius, spread, offset arc: coords, width, color(/gradient?), start/end angle, line cap, (bg image?) image:...

> So, coords is for coordinates. I think the rest is quite self-explanatory. E.g. fill is a struct with > > coordinate gradient radius mask (see blow) opa fields. I...