This implements most of the std::vector interface, with some minor elisions of
infrequently used methods such as "insert", and some STL bookkeeping.
Performance is favorable compared to std::vector, and, in some cases, somewhat
better than naive code to manually manage a malloc'd dynamic array.
BM_FlatArray_Push: 100 iterations in 72889721 nanos: 728897.21 nanos/it
BM_FlatArray_Fill: 100 iterations in 2700746 nanos: 27007.46 nanos/it
BM_FlatArray_Memset: 100 iterations in 2687810 nanos: 26878.10 nanos/it
BM_rcVector_Push: 100 iterations in 51114316 nanos: 511143.16 nanos/it
BM_rcVector_PushPreallocated: 100 iterations in 9532106 nanos: 95321.06 nanos/it
BM_rcVector_Assign: 100 iterations in 2371534 nanos: 23715.34 nanos/it
BM_rcVector_AssignIndices: 100 iterations in 2573052 nanos: 25730.52 nanos/it
BM_rcVector_Resize: 100 iterations in 2455761 nanos: 24557.61 nanos/it
BM_stdvector_Push: 100 iterations in 71739167 nanos: 717391.67 nanos/it
BM_stdvector_PushPreallocated: 100 iterations in 15723737 nanos: 157237.37 nanos/it
BM_stdvector_Assign: 100 iterations in 2564985 nanos: 25649.85 nanos/it
BM_stdvector_AssignIndices: 100 iterations in 5174923 nanos: 51749.23 nanos/it
BM_stdvector_Resize: 100 iterations in 2520054 nanos: 25200.54 nanos/it
Remove dstReg and dstDist from rcBuildRegions entirely -- just use temporary scratch space in expandRegions.
This saves a considerable amount of time when the number of spans is large. A sample map tried has 16M spans, and only writes at most 4000 of them.
These are by no means complete coverage of the functions they test, but are at least a start. For consistency, all floating point comparisons use Approx(), regardless of whether they need to or not.
Removed unnecessary C++11 unique_ptr stuff, now that rcHeightfield has a constructor and destructor.
* Added virtual qualifier to a few destructors in subclasses.
* Removed a few empty destructors that didn't need to be explicitly defined.
* Fixed a few typos
* Removed unnecessary empty virtual destructors in some classes with parent's who derrived from a class with a virtual destructor
When the detail mesh sampling sees an unset height it previously tried
to recover by looking for heights of neighbors. When the heightfield was
not eroded this recovery would fail since the contour simplification
could mean that the nearest valid cell was further away than a neighbor.
We now walk adjacent cells in a spiral to find a height and do this
depending on the max simplification error.
Fix#153
This changes the detail sampling to find height data by using the flood
fill method for polygons that were merged because of a border vertex
removal. This could potentially cause overlapping polygons to receive
the same region IDs which would later make the fast-path sampling in
getHeightData sample the wrong heights.
Also includes optimizations for this path and getHeightData itself.
Previously the seeding was responsible for 25% of the total time spent
in getHeightData in these cases. We now use the direction and prefer
moving directly towards the center of the polygon. This makes the
overhead of seedArrayWithPolyCenter virtually nonexistent (~2%
according to profiling).
Fix#146
I'm not sure if C++ vars are scoped into {} blocks, but in ported code that is an issue. It's easier to have vars names unique. Does not harm anyone I guess? ;-)
- do circumCircle calculations relative to first vertex for adde
precision
- improved the results of triangulateHull(), produces less degen
triangles
- fixed case with contour merging when outline is null
