[OpenJDK Rasterizer] Marlin renderer contribution for review

[Jim Graham]

The only "gotcha" with this technique is that the rendering inner loop 
should really use "ceil(x,y - 0.5)" to properly compute the shape 
insideness rule and that would mean that using an integer cast would 
require negative values to get the ceil() effect...

			...jim

On 3/19/15 10:21 AM, Laurent Bourgès wrote:
> Jim,
>
>>> I use FastMath mainly to handle boundaries and coordinates can be either
>>> positive or negative ... so I do not have the choice to deal with
>>> negative values !
>>
>>
>> That's probably true for normalization then.  For the rendering part, we could carefully constrain our values so that they are always positive or negative since we have a target clip to render relative to.
>
> Good catch: I will test using int cast in the renderer code : only
> positive !
>
> Laurent
>