Skip to content

Conversation

@YangLiuWillow
Copy link
Collaborator

  • primal_module_parallel.rs works when there is only one dual_node in each unit. Likely due to the different global_time in different unit being used when storing Obstacle in priority queue of each unit. Perhaps it would help to assign each edge a pointer to the global_time in its respective unit.
  • dual_module_parallel.rs seems to work fine.
  • After I changed edge and vertex accessing to pointers, primal_module_serial.rs and dual_module_pq.rs no longer works with plugin_single_hair.rs. I have yet to figure out why. It is likely due to how Obstacle is hashed. For now, when I run plugin_single_hair_debug_1 test, it generates way too manydual_nodes (like nearly 1k).
  • generate animation for primal_module_parallel.rs: run the test cases in primal_module_parallel.rs. E.g. for defect_vertices = [7, 21, 28], run cargo test primal_module_parallel_tentative_test_2 -- --nocapture.

杨柳 and others added 30 commits May 27, 2024 17:23
… of subgraph mismatch, but the original mwpf solver seems unstable
…ual_module_parallel.rs seems to run fine now, need further optimization using rayon and pointers
杨柳 and others added 21 commits August 22, 2024 20:55
… rayon, need to test for more complicated cases for non 0ns resolve time
…vertices again in before calling , because rn after fuse, the nodes do not grow into other units
…ort conflict when combining two units, e.g. check if boundary vertex overlay with vertices of invalid_subgraph
… a single dual_node in a unit. It is likely due to the different global_time in different units. Maybe consider assigning each edge a pointer to global_time.
…one dual_node, the is_valid() function is responsible for this. We need to check the logic of is_valid(). Line 866-868 in fn compute_maximum_update_length()
…ore for loops, especially in parallel_solve_step_callback
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants