Struct HostBuffer 

pub struct HostBuffer { /* private fields */ }

HostBuffer bump-allocates per-draw uniforms and transient vertex data.

Each frame writes into CPU scratch; Self::flush copies touched blocks with one queue.write_buffer each. wgpu stages those writes, so the ring needs no fences. A 3-frame ring of persistent buffers means warm frames create nothing — the cost that matters most on wasm.

Uniforms bind once per block via a dynamic offset. Vertex data (stencil fans, stroke strips) lands in a second family of blocks. All uploads go through the alloc/flush seam so a mapped staging backend can replace this implementation without touching call sites.

Implementations

impl HostBuffer

pub fn new(device: &Device) -> Self

new creates an empty host buffer for device.

Uniform stride is at least the per-draw record size and at least the device’s min_uniform_buffer_offset_alignment.

pub fn bind_group_layout(&self) -> &BindGroupLayout

bind_group_layout returns the group-0 layout for per-draw uniforms.

Binding 0 is a dynamic-offset uniform buffer. Pipeline layouts are built from this layout.

pub fn begin_frame(&mut self)

begin_frame rotates to the next arena and resets its cursors.

Blocks are retained so warm frames never create buffers. Trailing unused blocks from a spike drain after a few idle ring passes so a recurring large mesh does not recreate them every frame.

pub fn flush(&mut self, queue: &Queue) -> (u64, u64)

flush uploads this frame’s used scratch to the GPU.

One write_buffer runs per touched block. Returns (uniform_bytes, vertex_bytes) written, for frame statistics.