28   const int l = threadIdx.x;  // The output column index of each B^T D B operation  in __launch_bounds__()  local
61 …_index + j * NUM_NODES_OUT + i] * qf_array[qf_index + j] * B_in[b_in_index + j * NUM_NODES_IN + l];  in __launch_bounds__()
66             result *= orients_in[NUM_NODES_IN * e + l] ? -1.0 : 1.0;  in __launch_bounds__()
72 …ndex = e_stride * e + comp_in_stride * comp_in + comp_out_stride * comp_out + NUM_NODES_IN * i + l;  in __launch_bounds__()
76             s_C[NUM_NODES_IN * threadIdx.y + l] = result;  in __launch_bounds__()
78             s_CT[NUM_NODES_IN * i + l] =  in __launch_bounds__()
79 …(l > 0 ? s_C[NUM_NODES_IN * threadIdx.y + l - 1] * curl_orients_in[3 * NUM_NODES_IN * e + 3 * l - …  in __launch_bounds__()
80 …          s_C[NUM_NODES_IN * threadIdx.y + l] * curl_orients_in[3 * NUM_NODES_IN * e + 3 * l + 1] +  in __launch_bounds__()
81 …(l < (NUM_NODES_IN - 1) ? s_C[NUM_NODES_IN * threadIdx.y + l + 1] * curl_orients_in[3 * NUM_NODES_…  in __launch_bounds__()
83             s_CT[NUM_NODES_IN * i + l] = result;  in __launch_bounds__()
90 …ndex = e_stride * e + comp_in_stride * comp_in + comp_out_stride * comp_out + NUM_NODES_IN * i + l;  in __launch_bounds__()
94 …(i > 0 ? s_CT[NUM_NODES_IN * (i - 1) + l] * curl_orients_out[3 * NUM_NODES_OUT * e + 3 * i - 1] : …  in __launch_bounds__()
95                   s_CT[NUM_NODES_IN * i + l] * curl_orients_out[3 * NUM_NODES_OUT * e + 3 * i + 1] +  in __launch_bounds__()
96 …(i < (NUM_NODES_OUT - 1) ? s_CT[NUM_NODES_IN * (i + 1) + l] * curl_orients_out[3 * NUM_NODES_OUT *…  in __launch_bounds__()
98               values_array[val_index] = s_CT[NUM_NODES_IN * i + l];  in __launch_bounds__()