Struct GpuInfoGeometry 

pub struct GpuInfoGeometry { /* private fields */ }

Information-geometry operations with GPU-ready infrastructure.

The public 0.20 surface is a correctness-first CPU baseline after WebGPU context creation. Amari-Chentsov batches, Fisher metrics, and Bregman/KL-style divergences are validated and computed with amari-info-geom semantics while shader-backed tensor/fisher/divergence pipelines remain private restoration candidates.

This preserves public API stability without claiming unvalidated GPU kernels.

Implementations

impl GpuInfoGeometry

pub async fn new() -> Result<Self, GpuError>

Initialize GPU context for information geometry operations

pub async fn new_with_device_preference( device_type: &str, ) -> Result<Self, GpuError>

Create with specific device preference for edge computing

pub async fn amari_chentsov_tensor( &self, x: &Multivector<3, 0, 0>, y: &Multivector<3, 0, 0>, z: &Multivector<3, 0, 0>, ) -> Result<f64, GpuError>

Compute single Amari-Chentsov tensor (CPU fallback for small operations)

pub async fn amari_chentsov_tensor_batch( &self, x_batch: &[Multivector<3, 0, 0>], y_batch: &[Multivector<3, 0, 0>], z_batch: &[Multivector<3, 0, 0>], ) -> Result<Vec<f64>, GpuError>

Batch compute Amari-Chentsov tensors with CPU-baseline semantics.

All three batches must have equal length. The current public path uses amari-info-geom::amari_chentsov_tensor for correctness; the shader path remains private until hardware parity is validated.

pub async fn amari_chentsov_tensor_from_typed_arrays( &self, flat_data: &[f64], batch_size: usize, ) -> Result<Vec<f64>, GpuError>

Compute tensor batch from TypedArray-style flat data.

Each item is [x0,x1,x2, y0,y1,y2, z0,z1,z2] for Cl(3,0,0) vector components. All values must be finite.

pub async fn device_info(&self) -> Result<GpuDeviceInfo, GpuError>

Get device information for edge computing

pub async fn memory_usage(&self) -> Result<u64, GpuError>

Get current memory usage if the backend exposes it.

Portable WebGPU does not expose allocator usage through wgpu, so this returns 0 rather than a fabricated placeholder value.

pub async fn fisher_information_matrix( &self, parameters: &[f64], ) -> Result<GpuFisherMatrix, GpuError>

Compute a Fisher Information Matrix CPU baseline.

Interprets parameters as coordinates of a probability-simplex style point and delegates to amari-info-geom::DuallyFlatManifold’s Fisher metric implementation. Values must be finite and non-negative.

pub async fn bregman_divergence_batch( &self, p_batch: &[Vec<f64>], q_batch: &[Vec<f64>], ) -> Result<Vec<f64>, GpuError>

Batch compute KL-style Bregman divergences on the CPU correctness path.