Struct GpuExecutor 

pub struct GpuExecutor { /* private fields */ }

A high-performance WebGPU compute execution context with dynamic buffer pooling and pipeline caching.

Implementations

impl GpuExecutor

pub fn new() -> Option<Self>

Discovers and initialises the default GPU adapter and device.

Returns None if no compatible GPU hardware backend is available.

pub fn execute_batch( &mut self, shader_src: &str, n_rows: usize, vars_cols: &[&[f64]], out: &mut [f64], ) -> Result<(), String>

Dispatches a high-throughput parallel compute job to the GPU using the compiled WGSL shader source.

Performs zero-allocation float array conversions at the CPU-GPU memory boundary using a pre-allocated conversion scratchpad, maintaining optimal cache locality.

Panics

This function may panic if:

• Internal WGPU calls fail unexpectedly (e.g., unwrap() on create_compute_pipeline).
• A compute pipeline cannot be created.
• A bind group cannot be created.
• A command encoder cannot be created.
• Submission of commands to the queue fails.
• pollster::block_on encounters an error when polling the device.
• bytemuck::cast_slice fails (indicating incorrect memory layout assumptions).

Errors

Returns a String describing the error if execution fails due to:

• Failure to fetch the compute pipeline from the cache.
• Uninitialized output or staging buffers.
• Staging buffer mapping fails or the GPU channel disconnects.