Cache intermediate results automatically

[elalish]

Regarding this comment @pca006132, I did a little test with ./manifold_test --gtest_filter=Samples.Frame -v before and after removing the two cache-forcing lines. I'm seeing at least double the number of Boolean operations when I remove those lines. Is there a way to make the CsgNode aware of when it's being used more than once so it can skip flattening and allow reuse instead? Maybe some kind of ref-counter?

[pca006132]

Yes, we are already using ref-counter so maybe we can do use use_count.

[elalish]

Sounds good; do you want to give it a try?

[pca006132]

diff --git a/src/manifold/src/csg_tree.cpp b/src/manifold/src/csg_tree.cpp
index 9e31137..d11fb5f 100644
--- a/src/manifold/src/csg_tree.cpp
+++ b/src/manifold/src/csg_tree.cpp
@@ -366,7 +366,7 @@ std::vector<std::shared_ptr<CsgNode>> &CsgOpNode::GetChildren(
 
   CsgNodeType op = op_;
   for (auto &child : children_) {
-    if (child->GetNodeType() == op) {
+    if (child->GetNodeType() == op && child.use_count() == 1) {
       auto grandchildren =
           std::dynamic_pointer_cast<CsgOpNode>(child)->GetChildren(finalize);
       int start = children_.size();

Indeed it works, but there is still 2 additional boolean operations, not sure what they are, will check later.

[elalish]

Excellent, I love a simple fix! Looking forward to the PR. 👍

[elalish]

Any more progress on this, or would you like me to take a look?

[pca006132]

Sorry I am a bit busy recently, maybe you can take a look at it first.

[pca006132]

OK I understand what is going on for that 2 additional boolean operations now. The problem is that we are cloning all children when we do CsgOpNode::Transform, so we kind of lose track of how many copies of the nodes are there. A way to mitigate this is to share the child list.