Illegal memory access for SparseIndices.Sort with TBB backend

[pca006132]

The sort function will perform illegal memory access when dealing with large meshes and with TBB backend enabled. This can be checked by enabling the address sanitizer.

Build instruction:

mkdir -p build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DMANIFOLD_PAR="TBB" -DCMAKE_CXX_FLAGS="-g -fsanitize=address" -DMANIFOLD_USE_CUDA=off -DMANIFOLD_EXPORT=on ..

Run the following test:

TEST(Boolean, LargeSpheres) {
  auto a = Manifold::Sphere(100, 2000);
  auto b = a + a.Translate({30, 0, 0});
  a.GetMesh();
}

[pca006132]

Note that thrust have quite a lot of undefined behavior if you enable the UB sanitizer...
Not sure if this is related to it. I will find some time to try to reproduce it with a pure thrust program.

[pca006132]

Thinking about this, maybe we should use single threaded sort for now until the root cause of this is fixed.

[elalish]

Perhaps so; any luck with the Thrust-only repro? That sounds like a bug Nvidia should take pretty seriously.

[pca006132]

Ah, it seems that the issue disappeared after updating thrust. Forgot to rerun the test after thrust update. Closing for now.