Investigate replacing grid internal container
bpglaser opened this issue · comments
A simpler internal grid representation would make transitioning to multi-thread much simpler.
Attempts were made to replace grid internals with alternatives. None proved fruitful. Thus, with the current carver algorithm, a 2d vec is the most viable.
Stats for subtracting 5 from castle width:
- Hashmap 16.64s
- Vec 20.45s
- Vec<Vec<T>> 1.90s
Given the large number of reallocation due to shifts in the grid, it might be worthwhile to use a linked list container.
Changing the internal structure led to massive improvements in performance:
| command | grid-rotation-rework | master | % faster |
|---|---|---|---|
| -w -100 | 3.689447611 | 4.769283138 | 22.64% |
| -h -100 | 5.455471693 | 13.48595331 | 59.55% |
| -w -100 -h -100 | 8.971742094 | 17.9943384 | 50.14% |
| -w +100 | 4.620084357 | 5.897690078 | 21.66% |
| -h +100 | 6.80669775 | 21.28925174 | 68.03% |
| -w +100 -h +100 | 11.74770414 | 31.4032533 | 62.59% |