[MultiThreading] Parallel BVH narrow phase

[alxbilger]

Based on #2043

A new component is introduced: ParallelBVHNarrowPhase. As the name suggests, it is the parallel implementation of BVHNarrowPhase.

Details

The broad phase provides a list of pairs of collision models. In the initial BVHNarrowPhase component, the list was processed sequentially. In ParallelBVHNarrowPhase each pair is processed in a task, which is given to a thread pool.

Note that the traversal of the bounding volume hierarchy is not parallel. It is still single threaded.

Due to race conditions, some operations must be done prior to the parallel tasks:

1. Detection output creation. This operation is very time consuming (see [SofaKernel] Switch from map_ptr_stable_compare to std::unordered_map #2047). If [SofaKernel] Switch from map_ptr_stable_compare to std::unordered_map #2047 is merged, we can expect better results.
2. Topology initialization.

Results

The tested scene is applications/plugins/MultiThreading/examples/ParallelBruteForceBroadPhase.scn during 500 time steps.

The narrow phase duration has been measured with ParallelBVHNarrowPhase compared to BVHNarrowPhase.

BVHNarrowPhase                   3.5  ms
BVHNarrowPhase + PR#2047         2.36 ms
ParallelBVHNarrowPhase           2.1  ms
ParallelBVHNarrowPhase + PR#2047 0.97 ms

---

By submitting this pull request, I acknowledge that
I have read, understand, and agree SOFA Developer Certificate of Origin (DCO).

---

Reviewers will merge this pull-request only if

• it builds with SUCCESS for all platforms on the CI.
• it does not generate new warnings.
• it does not generate new unit test failures.
• it does not generate new scene test failures.
• it does not break API compatibility.
• it is more than 1 week old (or has fast-merge label).

[alxbilger]

[ci-build][with-all-tests]