Iterating w/ speed in mind about grid_map

[pfankhauser]

Here's also an updated version of your example for the fastest execution that I have been able to achieve yet:

grid_map::Matrix& data_layer_one = my_grid_map.get("layer_one");
grid_map::Matrix& data_layer_two = my_grid_map.get("layer_two");
for (grid_map::GridMapIterator iterator(my_grid_map); !iterator.isPastEnd(); ++iterator) {
  const int i = iterator.getLinearIndex();
  // do something with data_layer_one(i);
  // do something with data_layer_two(i);
}

Here we make use of the linear index access of Eigen matrices to maximize performance.

[pfankhauser]

Thanks for your comments. You are right that the performance can be increased when locally storing a reference or copy of the data layer. There's actually a node that runs a performance comparison:

rosrun grid_map_demos iterator_benchmark

In this example, we iterate through a grid map and copy the data from one layer into another if it is bigger in value. For a map of size 5000 x 5000 cells (25 mio. cells), the performance can be improved from the convenient usage of the iterator to the optimized usage from 1350 ms to 135 ms.

I have updated the documentation to highlight this here: https://github.com/ethz-asl/grid_map#iterators

Does this clarify your question?

[stonier]

Does this clarify your question?

Yes, and thanks for the pointer about the getLinearIndex().

You might like to add a comment to the iterators demo program - many programmers (myself included in this situation) tend to go directly to code before documents when it comes to discovering how to do something specific.

[pfankhauser]

Yes, good idea, thanks.