Beagleboard: Execute fragment shader on the GPU causes significant drop in performance

The numbers in each row correspond to the following primitive type:

Image+None | Rectangle+None | RoundedRect+None | Ellipse+None | Path+None | Text+None

-Dprism.tess : use tessellated triangles, no complex shader computation

-Dprism.cacheshapes : use shape caching, no complex shader computation

Both methods will get us 2X boost in fps for 50 elements, and 3X boost in fps for 250 elements in FXTester.

baseline 50
27.64 | 15.94 | 13.70 | 16.05 | 32.57 | 41.87
baseline 50 + -Dprism.tess=true
27.50 | 31.41 | 30.52 | 31.89 | 31.46 | 41.91
baseline 50 + -Dprism.cacheshapes=true
27.47 | 32.47 | 32.35 | 32.57 | 32.52 | 41.84

baseline 250
10.19 | 4.11 | 3.36 | 4.17 | 13.10 | 24.74
baseline 250 + -Dprism.tess=true
10.19 | 12.80 | 12.27 | 13.56 | 13.01 | 24.86
baseline 250 + -Dprism.cacheshapes=true
10.18 | 13.14 | 13.15 | 13.12 | 13.12 | 24.80