pub fn gemm_compute_shader() -> Vec<u32>Expand description
Generate an OpenCL SPIR-V compute kernel for GEMM: C = alpha * A * B + beta * C.
Naive reference (one thread per output element, row-major f32 layout).
Kernel parameters: (CrossWorkgroup float* A, CrossWorkgroup float* B, CrossWorkgroup float* C, uint m, uint n, uint k, float alpha, float beta).