pub fn launch_kernel<R: Runtime, Alg: Algorithm>(
client: &ComputeClient<R>,
input: &MatmulInputHandleRef<'_, R>,
out_grad: &MatmulInputHandleRef<'_, R>,
weight_grad: &TensorHandleRef<'_, R>,
problem: ConvolutionProblem,
blueprint_strategy: &BlueprintStrategy<RuntimeArgs, Alg::Routine>,
dtypes: MatmulElems,
) -> Result<(), ConvSetupError>