 Martin Fouilleul
|
e9ecd9d521
|
[mtl canvas] Fix artifacts on tiger that was due to epsilon comparison for cubics equation in draw kernel. Removed that (now do <= 0), but then we need to weed out false negatives for solid triangles. So we compute if the triangle is solid in the triangle kernel and store it in the mg_triangle_data struct for now.
|
2023-03-22 15:02:04 +01:00 |
|
Martin Fouilleul
|
0d8bc824a2
|
[mtl canvas] bucket triangles into per-shape tile queues, then gather into tile arrays. This somewhat improves perf for perf_text test and avoids an awful degradation due to sorting in tiger test (tiger now runs at ~20fps, which is still much slower than it could be, but the triangle coverage method we're using just incurs too much cross product computations...)
|
2023-03-21 10:22:37 +01:00 |
|
Martin Fouilleul
|
a4ef58f2d8
|
[mtl canvas] Tiling per rectangles into per-shape tile queues (linked lists of triangle indices) to avoid sorting
|
2023-03-20 16:46:12 +01:00 |
|
Martin Fouilleul
|
92f4909d63
|
[mtl canvas] testing the metal canvas rendering the ghostscript tiger, and acknowledging it's painfully slow
|
2023-03-18 14:35:51 +01:00 |