Greatly Improve Performance of Precomputation #71
Conversation
These changes greatly improve the performance of precomputing the map data, as well as laying the groundwork for a vastly improved can_move function.
|
You can run it for yourself on your own local device. Note that for development purposes, you may want to add some sorting to the keys, otherwise you will get a different result for data.js every time you run let keys = Object.keys(G.maps);
keys.sort();
let new_maps = {};
for (let x of keys) {
new_maps[x] = G.maps[x];
}
G.maps = new_maps;Inserting this short snippet before the loop in |
|
Hello, thanks for this PR, can you explain in details what you did, why and the measurements you made to determine the scale of the speed improvement, please? |
Reusing expensive computations (such as a binary search result) and overall improvements to the functions used resulted in a significant increase in performance. The way I measured performance was simply by running |
|
@Telokis On my machine, precomputing all the maps would take 127975ms, or 128 seconds. With the performance improvements, It went down to 72218ms, or 72 seconds. |
|
I felt that even though this was a performance improvement in one of the lesser-used parts of the code (map precomputation), since it was an improvement on the |
|
Ok, I understand, thank you. |
|
@Telokis It produces exactly the same results. |
|
I'm a little unsure of the naming of the functions |
|
I ran a precompute without this PR, figured I'd post the numbers, will post another one for comparison once I check out the, do note I'm getting different timings each time I run it |
These changes greatly improve the performance of precomputing the map data, as well as laying the groundwork for a vastly improved can_move function.