[Feature] Storage buffers

[vmarcella]

Overview

Add storage buffer support for read-write GPU memory used by compute shaders.

Current State

No response

Scope

Goals:

• Storage buffer creation with read/write usage
• Binding to compute and render pipelines
• CPU readback support
• Dynamic resizing

Non-Goals:

• Atomic operations
• Indirect buffers

Proposed API

pub struct StorageBuffer {
  buffer: wgpu::Buffer,
  size: u64,
}

pub struct StorageBufferBuilder {
  size: u64,
  usage: StorageBufferUsage,
  initial_data: Option<Vec<u8>>,
}

#[derive(Clone, Copy)]
pub struct StorageBufferUsage {
  pub read: bool,
  pub write: bool,
  pub cpu_readable: bool,
}

impl StorageBufferBuilder {
  pub fn new(size: u64) -> Self;
  pub fn with_usage(self, usage: StorageBufferUsage) -> Self;
  pub fn with_data<T: bytemuck::Pod>(self, data: &[T]) -> Self;
  pub fn build(self, gpu: &Gpu) -> StorageBuffer;
}

impl StorageBuffer {
  pub fn write<T: bytemuck::Pod>(&self, gpu: &Gpu, data: &[T]);
  pub fn read<T: bytemuck::Pod>(&self, gpu: &Gpu) -> Vec<T>;
  pub fn size(&self) -> u64;
  pub fn as_binding(&self) -> BindingResource;
}

Acceptance Criteria

• Storage buffer creates with specified size
• Data writes from CPU
• Data reads back to CPU
• Binds to compute pipeline
• Example with data processing

Affected Crates

lambda-rs, lambda-rs-platform

Notes

• Read-back requires staging buffer and map_async
• STORAGE usage flag required