[Rev 02] RFR: 8238954: Improve performance of tiled snapshot rendering

[Ambarish Rapte]

On Sat, 14 Mar 2020 13:25:00 GMT, Frederic Thevenet <github.com+7450507+fthevenet at openjdk.org> wrote:

>> modules/javafx.graphics/src/main/java/com/sun/javafx/tk/quantum/QuantumToolkit.java line 1600:
>> 
>>> 1599:                                 }
>>> 1600:                             }
>>> 1601:                         }
>> 
>> The `if (exactHeightDivFound.get())` and `else` block here look effectively same. Could you please take a re-look or
>> explain what am I missing here.
>
> Functionally, both side of the condition actually do the exact same thing, but the side effects of which dimension is
> processed in the inner loop sometime have measurable consequences in terms of performances; I've added the following
> comment to the code to better explain this:
>   // In order to minimize the number of time we have to resize the underlying
>   // surface for capturing a tile, choose a dimension that has an exact divider
>   // (if any) to be processed in the inner most loop.
>   // E.g. looping on width then height in the example bellow requires four
>   // surface resizing, whereas the opposite requires only two:
>   //
>   //       for (w;;)                    for (h;;)
>   //           for(h;;)                     for(w;;)
>   //       -----------------           -----------------
>   //       |       |       |           |       |       |
>   //       |   1   |   3   |           |   1   |   2   |
>   //    h  |       |       |        h  |       |       |
>   //       -----------------           -----------------
>   //       |   2   |   4   |           |   3   |   4   |
>   //       -----------------           -----------------
>   //               w                           w

Than for the additional comment, this will reduce the number of textures created by below call in `renderTile()`
`com.sun.prism.RTTexture rt = tileImg.getRT(w, h, rf);`

-------------

PR: https://git.openjdk.java.net/jfx/pull/112