pub fn gpu_batch_adjoint(
gpu: &GpuTrainingContext,
unitaries_flat: &[Complex],
d_rho_flat: &[Complex],
batch_count: usize,
) -> Vec<Complex>Expand description
GPU-accelerated batched adjoint for QUG backward pass.
Computes ∂L/∂ρ_before = U† · ∂L/∂ρ_after · U for all positions in one batch.