Thanks Sean! I'll optimize some of my other for(;;) loops as well...
On Mon, Jun 15, 2015 at 3:59 PM, Sean Luke <[log in to unmask]> wrote:
> MASON actually has a MutableDouble.
>
> java/mason/sim/util/MutableDouble.java
>
> You could use that. Also, don't use for(Object object : objects) -- it translates to building an Iterator, and at least as of Java 7 Iterators are still quite slow.
>
> I'd do this:
>
> final public void SparseGrid2DMultiply(SparseGrid2D theGrid, double multiplier) {
> Bag objects = theGrid.getAllObjects();
> int len = objects.size();
> for(int i = 0; i < len; i++)
> {
> MutableDouble d = (MutableDouble)(objects.get(i));
> d.val *= multiplier;
> }
> }
>
>
> On Jun 15, 2015, at 1:02 PM, Sadat Chowdhury <[log in to unmask]> wrote:
>
>> I still have a lot of files to change, and have not actually run/tested with this new method — maybe it would throw the ConcurrentModificationException, I don’t know yet.
>>
>> I am going to look into MutableDouble — it makes more sense as far as accuracy goes. Thanks!
>>
>>
>>> On Jun 15, 2015, at 12:58 PM, Ernesto Carrella <[log in to unmask]> wrote:
>>>
>>> You could try populating your SparseGrid2D with either an AtomicDouble or a MutableDouble from commons-lang.
>>> That way you can modify them without pulling them out of the grid and putting them back in.
>>>
>>> I am quite surprised it doesn't throw a ConcurrentModification exception since you are effectively looping through a list of objects that you are also modifying (with remove() and setLocation()). I suppose MASON passes a safe copy in that bag.
>>>
>>> On Mon, Jun 15, 2015 at 5:23 PM Sadat Chowdhury <[log in to unmask]> wrote:
>>> Hi All,
>>>
>>> When I increased the 2D world size of my system, there was a significant slowdown of the simulation. After I investigated using a profiler, it turned out to be due to methods that scan through 2D grids that in my system are defined as DoubleGrid2D.
>>>
>>> I have started converting those DoubleGrid2D into SparseGrid2D — because it fits the description perfectly: the world can be potentially unbounded and most times there will be fewer actual objects in the world.
>>>
>>> While it was relatively easy to convert all the parts that was setting and accessing objects in the grid — I am at a point where I have a lot of DoubleGrid2D’s that used full-scanning functions like .lowerBound() and .multiply(). Supposing I have a SparseGrid2D that has only one Double object per coordinate, what would be the most efficient (fast) way to implement a function similar to multiply() ?
>>>
>>> This is what I have so far, but I am just wondering if it can be done more efficiently (faster)?
>>>
>>> final public void SparseGrid2DMultiply(SparseGrid2D theGrid,
>>> double multiplier) {
>>> /*
>>> * theGrid is assumed to contain one and only one Double in a given
>>> * coordinate
>>> */
>>> Bag objects = theGrid.getAllObjects();
>>> for (Object object : objects) {
>>> double val = (Double) object;
>>> if (val != 0.0) {
>>> val = val * multiplier;
>>> Int2D location = theGrid.getObjectLocation(object);
>>> theGrid.remove(object);
>>> theGrid.setObjectLocation(val, location);
>>> }
>>> }
>>>
>>> }
>>>
>>> Any thoughts/comments is greatly appreciated,
>>> Sadat.
|