Expose Viewport Dimensions Through `fvdb.viz` API

[swahtz]

Expose Viewport Dimensions Through fvdb.viz API

Problem

The image2d.slang overlay shader in nanovdb-editor uses 1:1 pixel mapping, so any overlay image must match the viewport's exact resolution for correct alignment. Currently the viewport dimensions (updated if window resizes) are only available as stack-local variables inside the render loop. They need to be exposed through the Python API so consumers can render overlays at the correct resolution.

Architecture

flowchart LR
    subgraph NanovdbEditor ["nanovdb-editor"]
        RenderLoop["show() render loop"] -->|"atomic store"| ImplAtomics["EditorImpl atomics"]
        ImplAtomics -->|"read"| EditorAPI["pnanovdb_editor_t API"]
    end
    subgraph FvdbCore ["fvdb-core"]
        EditorAPI --> Viewer["Viewer.cpp getters"]
        Viewer --> Pybind["ViewerBinding.cpp"]
        Pybind --> ScenePy["_scene.py properties"]
    end

Loading

Changes: nanovdb-editor

1. Add atomic viewport dimension members to EditorImpl

In editor/Editor.h, add two atomic members to pnanovdb_editor_impl_t (alongside the existing show_active atomic, around line 82):

std::atomic<pnanovdb_int32_t> viewport_width{0};
std::atomic<pnanovdb_int32_t> viewport_height{0};

2. Store dimensions in the render loop

In editor/Editor.cpp, store the viewport dimensions at two points:

a) After the initial resolution is determined from INI/defaults (around line 452), so values are available before the first render frame:

editor->impl->viewport_width.store(image_width, std::memory_order_relaxed);
editor->impl->viewport_height.store(image_height, std::memory_order_relaxed);

b) After get_camera_view_proj updates image_width/image_height each frame (line 642):

imgui_window_iface->get_camera_view_proj(imgui_window, &image_width, &image_height, &view, &projection);
editor->impl->viewport_width.store(image_width, std::memory_order_relaxed);
editor->impl->viewport_height.store(image_height, std::memory_order_relaxed);

3. Expose through the pnanovdb_editor_t interface

In nanovdb_editor/putil/Editor.h, add function pointers to the pnanovdb_editor_t interface for reading the viewport dimensions, following the pattern of existing getters like get_resolved_port. The implementations read from the atomics added in step 1.

Changes: fvdb-core

4. Add C++ getters to the Viewer class

In src/fvdb/detail/viewer/Viewer.h, declare read-only getters:

int viewportWidth(const std::string &scene_name);
int viewportHeight(const std::string &scene_name);

In src/fvdb/detail/viewer/Viewer.cpp, implement by calling the new nanovdb-editor API from step 3. These do NOT need getCamera() since viewport dimensions are not part of the camera struct. No setters -- viewport size is controlled by the browser window. scene_name is kept for API consistency even though the viewport is scene-independent.

5. Add pybind11 bindings

In src/python/ViewerBinding.cpp, add bindings following the camera_fov pattern:

.def("viewport_width",
     &fvdb::detail::viewer::Viewer::viewportWidth,
     py::arg("scene_name"),
     "Get the viewport render width in pixels")
.def("viewport_height",
     &fvdb::detail::viewer::Viewer::viewportHeight,
     py::arg("scene_name"),
     "Get the viewport render height in pixels")

6. Add Python properties to _scene.py

In fvdb/viz/_scene.py, add read-only properties (after camera_fov):

@property
def viewport_width(self) -> int:
    """Return the viewport render width in pixels."""
    server = _get_viewer_server_cpp()
    return server.viewport_width(self._name)

@property
def viewport_height(self) -> int:
    """Return the viewport render height in pixels."""
    server = _get_viewer_server_cpp()
    return server.viewport_height(self._name)

Thread safety

memory_order_relaxed is sufficient -- we only need eventual visibility, not ordering guarantees. The Python API reading a slightly stale value for one frame is acceptable. This matches the existing pattern used by resolved_port and show_active in pnanovdb_editor_impl_t.