Fixed bug where cache would never be allocated at the specified device_id

src/cache.cuh

@@ -14,6 +14,10 @@ namespace gpucache {
 //            printf("Cache: keySize: %lu, valueSize: %u, dim: %u, capacity: %lu, "
 //            printf("Cache: keySize: %lu, valueSize: %u, dim: %u, capacity: %lu, "
 //                   "maxQueryNum: %u, deviceId: %u\n",
 //                   "maxQueryNum: %u, deviceId: %u\n",
 //                   sizeof(KeyType), value_size, dim, capacity, max_query_num, device_id);
 //                   sizeof(KeyType), value_size, dim, capacity, max_query_num, device_id);
+            int device_count;
+            CUDA_CHECK(cudaGetDeviceCount(&device_count));
+            assert(device_id < device_count && "device_id should less than total device num");
+            CUDA_CHECK(cudaSetDevice(device_id));
             CUDA_CHECK(cudaMalloc((void **) &keys, capacity * key_size));
             CUDA_CHECK(cudaMalloc((void **) &keys, capacity * key_size));
             CUDA_CHECK(cudaMalloc((void **) &values, capacity * value_size));
             CUDA_CHECK(cudaMalloc((void **) &values, capacity * value_size));
             CUDA_CHECK(cudaMalloc((void **) &timestamps, capacity * sizeof(uint8_t)));
             CUDA_CHECK(cudaMalloc((void **) &timestamps, capacity * sizeof(uint8_t)));