ui: optimize _map_line_to_polygon

[deanlee]

Optimized _map_line_to_polygon with pairwise projection, minimized reshaping, and generating the triangle strip directly without intermediate Python loops.

before (run selfdrive/ui/tests/profile_onroad.py ):

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
     3606   34.184    0.009   34.184    0.009 {built-in method raylib._raylib_cffi.EndDrawing}
    10192    2.191    0.000    3.419    0.000 openpilot/selfdrive/ui/onroad/model_renderer.py:337(_map_line_to_polygon)
   306052    1.841    0.000   41.869    0.000 openpilot/.venv/lib/python3.11/site-packages/pyray/__init__.py:47(wrapped_func)
     2396    1.737    0.001    1.737    0.001 {built-in method raylib._raylib_cffi.UpdateTexture}
   116750    1.467    0.000    2.146    0.000 openpilot/.venv/lib/python3.11/site-packages/pyray/__init__.py:117(func)
    71401    1.030    0.000    1.030    0.000 {built-in method numpy.array}
    25137    0.965    0.000    1.754    0.000 openpilot/system/ui/lib/shader_polygon.py:199(triangulate)
    28730    0.757    0.000    0.757    0.000 {built-in method raylib._raylib_cffi.EndShaderMode}

after:

 ncalls  tottime  percall  cumtime  percall filename:lineno(function)
   3606   35.384    0.010   35.384    0.010 {built-in method raylib._raylib_cffi.EndDrawing}
  10192    1.987    0.000    2.768    0.000 openpilot/selfdrive/ui/onroad/model_renderer.py:335(_map_line_to_polygon)
 306052    1.925    0.000   43.395    0.000 openpilot/.venv/lib/python3.11/site-packages/pyray/__init__.py:47(wrapped_func)
   2396    1.844    0.001    1.844    0.001 {built-in method raylib._raylib_cffi.UpdateTexture}
 116750    1.543    0.000    2.256    0.000 openpilot/.venv/lib/python3.11/site-packages/pyray/__init__.py:117(func)
  28730    0.782    0.000    0.782    0.000 {built-in method raylib._raylib_cffi.EndShaderMode}

Profiler comparison (before → after):

Function	Time (s) Before	Time (s) After	Improvement
_map_line_to_polygon	3.419	2.768	~19% faster
shader_polygon.triangulate	1.754	—	eliminated

This refactor keeps rendering output identical while reducing CPU overhead and simplifying the data flow between model rendering and shader polygon generation.

[github-actions]

raylib UI Preview

All Screenshots

[adeebshihadeh]

FYI @deanlee, part of the motivation for moving to the new graphics stack is the dream of kicking the UI entirely off the GPU.

This will need lots of optimization work if you want to help with that :)

We should be able to move GPU things off in small PRs one by one.

[deanlee]

Not sure rlsw aligns with that direction: https://github.com/raysan5/raylib/blob/master/src/external/rlsw.h

UI perf still needs some cleanup first.

[adeebshihadeh]

Oh we might not want to use that at all. The important thing was ditching Qt and Weston.