Currently wgpu backend is using the default implementation using mask_fill for clamp operators. This issue ticket is to improve the performance using wgpu specific and more optimal kernel for clamp, including clamp_min, clamp_max.