Module shader 

WGSL shader source generation for common compute kernels.

Each function returns a complete, self-contained WGSL source string suitable for passing to device.create_shader_module().

Functions

attention_wgsl

Generate a WGSL compute shader for scaled dot-product attention.

batched_gemm_wgsl

Generate WGSL source for a batched (strided) GEMM kernel.

binary_wgsl

Generate WGSL source for an element-wise binary operation.

conv2d_wgsl

Generate a WGSL compute shader for 2D convolution in NCHW format.

elementwise_wgsl

Generate WGSL source for an element-wise unary operation.

gemm_wgsl

Generate WGSL source for a tiled GEMM kernel: C = alpha * A * B + beta * C.

gemm_wgsl_f16

Generate WGSL source for a tiled GEMM kernel using FP16 storage.

reduction_final_wgsl

Generate WGSL for the final scalar reduction of partial sums.

reduction_wgsl

Generate WGSL source for a parallel workgroup-level reduction.