File: HistogramBase.cs
The percentile lookup iterates nested i / j loops over BucketCount × SubBucketCount. Since CountsArrayIndex(bucketIndex, subBucketIndex) maps this to a flat array, and the counts array is already flat in LongHistogram, the entire scan can be done as a single flat loop:
public long GetValueAtPercentile(double percentile)
{
var countAtPercentile = Math.Max(
(long)((Math.Min(percentile, 100.0) / 100.0 * TotalCount) + 0.5), 1);
long runningCount = 0;
for (int i = 0; i < CountsArrayLength; i++)
{
runningCount += GetCountAtIndex(i);
if (runningCount >= countAtPercentile)
return HighestEquivalentValue(ValueFromIndex(i));
}
throw new ArgumentOutOfRangeException(...);
}
The nested loop computes GetCountAt(i, j) → GetCountAtIndex(CountsArrayIndex(i, j)) anyway — the flat version eliminates the CountsArrayIndex call per iteration and improves cache locality by accessing memory sequentially.
File:
HistogramBase.csThe percentile lookup iterates nested
i / jloops overBucketCount × SubBucketCount. SinceCountsArrayIndex(bucketIndex, subBucketIndex)maps this to a flat array, and the counts array is already flat inLongHistogram, the entire scan can be done as a single flat loop:The nested loop computes
GetCountAt(i, j)→GetCountAtIndex(CountsArrayIndex(i, j))anyway — the flat version eliminates theCountsArrayIndexcall per iteration and improves cache locality by accessing memory sequentially.