Module gpu_kernels 

GPU kernel implementations for batch special-function evaluation.

This module houses compute-shader sources and host-side dispatch stubs for two GPU backends:

Sub-module	Backend	Feature flag	Default
wgsl	WebGPU (WGSL shaders)	gpu via scirs2-core/gpu	off
cuda	CUDA / ROCm	cuda_kernels	off

Both backends currently provide stub implementations that return “not available” errors. The actual dispatch logic lives in crate::gpu_dispatch which calls these stubs under DispatchTarget::Gpu and falls back to the CPU rayon path automatically.

Pure Rust policy

The cuda_kernels feature is off by default. CUDA depends on C dynamic libraries (libcuda.so, libcudart.so) which violate the Pure Rust Policy unless feature-gated. WebGPU is Pure Rust (via wgpu) but the host-side integration with scirs2-core/gpu is not yet stabilised.

Modules

cuda

CUDA/ROCm kernel dispatch stubs for batch special-function evaluation.

wgsl

WGSL compute-shader kernels for WebGPU-backed dispatch of batch special functions.

Enums

CudaDispatchError

Error type for CUDA/ROCm dispatch.

WgslDispatchError

Error type for WGSL/WebGPU dispatch.

Constants

BESSEL_J0_PTX_STUB

Inline PTX stub for batch Bessel J₀ evaluation.

BESSEL_J0_WGSL

WGSL compute shader for batch Bessel J₀ evaluation.

ERFC_WGSL

WGSL compute shader for batch erfc evaluation.

ERFINV_WGSL

WGSL compute shader for batch inverse-erf evaluation.

ERF_PTX_STUB

Inline PTX stub for batch erf evaluation.

ERF_WGSL

WGSL compute shader for batch erf evaluation.

GAMMA_PTX_STUB

Inline PTX stub for batch Gamma evaluation on CUDA devices.

GAMMA_WGSL

WGSL compute shader for batch Gamma evaluation (Lanczos g=7 approximation).

LGAMMA_WGSL

WGSL compute shader for batch log-gamma evaluation.

Functions

bessel_j0_batch_cuda

Attempt batch Bessel J₀ evaluation on a CUDA device.

bessel_j0_batch_wgpu

Stub: returns WgslDispatchError::GpuNotAvailable when wgpu_kernels is off.

erf_batch_cuda

Attempt batch erf evaluation on a CUDA device.

erf_batch_wgpu

Stub: returns WgslDispatchError::GpuNotAvailable when wgpu_kernels is off.

erfc_batch_wgpu

Stub: returns WgslDispatchError::GpuNotAvailable when wgpu_kernels is off.

erfinv_batch_wgpu

Stub: returns WgslDispatchError::GpuNotAvailable when wgpu_kernels is off.

gamma_batch_cuda

Attempt batch Gamma evaluation on a CUDA device.

gamma_batch_wgpu

Stub: returns WgslDispatchError::GpuNotAvailable when wgpu_kernels is off.

lgamma_batch_wgpu

Stub: returns WgslDispatchError::GpuNotAvailable when wgpu_kernels is off.