[QST] It seems high-level api don't support selecting which GPU to run

I created a compress manager and compressed my data by

CascadedManager nvcomp_manager{chunk_size , nvcompBatchedCascadedDefaultOpts, resource->stream, device_id};
CompressionConfig comp_config = nvcomp_manager.configure_compression(resource->data_length
   resource->resizeDeviceCompMemory(comp_config.max_compressed_buffer_size);
nvcomp_manager.compress(resource->data_pointer, resource->device_comp_result, comp_config);
CHECK_ERROR(cudaStreamSynchronize(resource->stream), __FILE__, __LINE__);

I have passed device id to select which GPU to run compress but it seems always run on GPU 0. Is there something wrong with the way I use it?