small change

spconv/pool.py

@@ -60,8 +60,6 @@ class SparseMaxPool(SparseModule):
         indices = input.indices
         spatial_shape = input.spatial_shape
         batch_size = input.batch_size
-        torch.cuda.synchronize()
-        t = time.time()
         if not self.subm:
             out_spatial_shape = ops.get_conv_output_size(
                 spatial_shape, self.kernel_size, self.stride, self.padding,
@@ -75,9 +73,6 @@ class SparseMaxPool(SparseModule):
         out_features = Fsp.indice_maxpool(features, indice_pairs.to(device),
                                           indice_pairs_num.to(device),
                                           outids.shape[0])
-        torch.cuda.synchronize()
-
-        print("maxpool", spatial_shape, time.time() - t)
         out_tensor = spconv.SparseConvTensor(out_features, outids,
                                              out_spatial_shape, batch_size)
         out_tensor.indice_dict = input.indice_dict

test/benchmark.py

@@ -56,6 +56,7 @@ class Net(nn.Module):
             spconv.SubMConv3d(256, 256, 3, bias=False, indice_key="c6"),
         )
         max_batch_size = 1
+        # grid (dense map) is used for indice generation. use pre-allocated grid can run faster.
         self.grid = torch.full([max_batch_size, *shape], -1, dtype=torch.int32).cuda()
         # self.grid = None
         self.shape = shape