Ported C# much higher performance version of LSD Radix Sort of array …

…of objects based on a key, which sped up by 10X. Documented speed comparison in Readme.
DragonSpit · Mar 27, 2019 · de730bf · de730bf
1 parent 69aecd0
commit de730bf
Show file tree

Hide file tree

Showing 3 changed files with 84 additions and 87 deletions.
diff --git a/RadixSort.js b/RadixSort.js
@@ -1,5 +1,4 @@
-// TODO: It may be faster to do shift-mask instead of the current mask-shift, since the mask would then be the same for all bytes (0xff).
-// TODO: Remove extract digit function and just inline the core code. I've seem this improve performance in C#. Not sure how agressively JavaScript inlines small functions.
+// TODO: Remove extract digit function and just inline the core code. I've seen this improve performance in C#. Not sure how agressively JavaScript inlines small functions.
 // TODO: Compare performance versus TimSort for random, pre-sorted and constant, since TimSort is available in JavaScript thru npm
 //       (https://stackoverflow.com/questions/40721767/what-is-the-fastest-way-to-sort-a-largeish-array-of-numbers-in-javascript)
 
@@ -23,52 +22,6 @@ HpcAlgorithms.Sorting = (function()
 		return digit;
 	}
 
-	/**
-	 * Radix Sort (Least Significant Digit - LSD) of unsigned integer array with values up to 32-bits (e.g. 0, 1, 2, ... 2_000_000_000, ...)
-	 * This algorithm is not in-place - i.e. returns a sorted array
-	 * @param  {Array of numbers} inputArray Array of numbers, which must be unsigned integers of values within 32-bits
-	 * @return {Array of numbers} Sorted array of numbers
-	 */
-	var RadixSortLsdUInt32_old = function(inputArray)
-	{
-		if (typeof inputArray.constructor === Array && typeof inputArray[0] === "number") throw new TypeError("Input argument must be an array of unsigned integers");
-		var numberOfBins = 256;
-		var Log2ofPowerOfTwoRadix = 8;
-		var outputArray = new Array(inputArray.length);
-		var count = new Array(numberOfBins);
-		var outputArrayHasResult = false;
-
-		var bitMask = 255;
-		var shiftRightAmount = 0;
-
-		var startOfBin = new Array( numberOfBins );
-
-		while( bitMask != 0 ) // end processing digits when all the mask bits have been processed and shifted out, leaving no bits set in the bitMask
-		{
-			for (var i = 0; i < numberOfBins; i++ )
-				count[ i ] = 0;
-			for (var current = 0; current < inputArray.length; current++ ) // Scan the array and count the number of times each digit value appears - i.e. size of each bin
-				count[ extractDigit( inputArray[ current ], bitMask, shiftRightAmount ) ]++;
-
-			startOfBin[ 0 ] = 0;
-			for( var i = 1; i < numberOfBins; i++ )
-				startOfBin[ i ] = startOfBin[ i - 1 ] + count[ i - 1 ];
-			for ( var current = 0; current < inputArray.length; current++ )
-				outputArray[ startOfBin[ extractDigit( inputArray[ current ], bitMask, shiftRightAmount ) ]++ ] = inputArray[ current ];
-
-			bitMask <<= Log2ofPowerOfTwoRadix;
-			shiftRightAmount += Log2ofPowerOfTwoRadix;
-			outputArrayHasResult = !outputArrayHasResult;
-
-			var tmp = inputArray, inputArray = outputArray, outputArray = tmp; // swap input and output arrays
-		}
-		if ( outputArrayHasResult )
-			for ( var current = 0; current < inputArray.length; current++ ) // copy from output array into the input array
-				inputArray[ current ] = outputArray[ current ];
-
-		return inputArray;
-	}
-
 	var HistogramByteComponents = function(inArray, l, r)
 	{
 	  var numberOfDigits = 4;
@@ -85,14 +38,43 @@ HpcAlgorithms.Sorting = (function()
 	  for (var current = l; current <= r; current++)    // Scan the array and count the number of times each digit value appears - i.e. size of each bin
 	  {
 		var value = inArray[current];
-		count[0][ value &       0xff       ]++;
-		count[1][(value &     0xff00) >>  8]++;
-		count[2][(value &   0xff0000) >> 16]++;
-		count[3][(value & 0xff000000) >> 24]++;
+		count[0][ value        & 0xff]++;
+		count[1][(value >>  8) & 0xff]++;
+		count[2][(value >> 16) & 0xff]++;
+		count[3][(value >> 24) & 0xff]++;
 	  }
 	  return count;
 	}
 
+	var HistogramByteComponentsAndKeyArray = function(inArray, l, r, getKey)
+	{
+		var numberOfDigits = 4;
+		var numberOfBins = 256;
+		var inKeys = new Array(inArray.length);
+
+		var count = new Array(numberOfDigits);
+		for (var d = 0; d < numberOfDigits; d++)
+		{
+			count[d] = new Array(numberOfBins);
+			for (var b = 0; b < numberOfBins; b++)
+			count[d][b] = 0;
+		}
+
+		for (var current = l; current <= r; current++)    // Scan the array and count the number of times each digit value appears - i.e. size of each bin
+		{
+			var value = getKey(inArray[current]);
+			inKeys[current] = value;
+			count[0][ value        & 0xff]++;
+			count[1][(value >>  8) & 0xff]++;
+			count[2][(value >> 16) & 0xff]++;
+			count[3][(value >> 24) & 0xff]++;
+		}
+		return {
+			count: count,
+			inKeys: inKeys
+		};
+	}
+
 	/**
 	 * Radix Sort (Least Significant Digit - LSD) of unsigned integer array with values up to 32-bits (e.g. 0, 1, 2, ... 2_000_000_000, ...)
 	 * This algorithm is not in-place - i.e. returns a sorted array
@@ -147,39 +129,56 @@ HpcAlgorithms.Sorting = (function()
 	 * @param  {function} getKey function to extract and return a numeric key from the user defined type/class to sort on
 	 * @return {Array of numbers} Sorted array of a user defined type
 	 */
-	var RadixSortLsdUdtUInt32 = function(inputArray, getKey)
-	{
-		var numberOfBins = 256;
-		var Log2ofPowerOfTwoRadix = 8;
-		var OutputArray = new Array(inputArray.length);
-		var count = new Array(numberOfBins);
-		var OutputArrayHasResult = false;
-		var bitMask = 255;
+	function RadixSortLsdUdtUInt32(inputArray, getKey) {
+		var numberOfBitsPerDigit = 8;
+		var numberOfBins = 1 << numberOfBitsPerDigit;
+		var numberOfDigits = 4;
+		var outputArray   = new Array(inputArray.length);
+		var outSortedKeys = new Array(inputArray.length);
+		var outputArrayHasResult = false;
+		var bitMask = numberOfBins - 1;
 		var shiftRightAmount = 0;
-		var startOfBin = new Array( numberOfBins );
-
+		var d = 0;
+
+		var retValue = HistogramByteComponentsAndKeyArray(inputArray, 0, inputArray.length - 1, getKey);
+		var count  = retValue.count;
+		var inKeys = retValue.inKeys;
+
+		var startOfBin = new Array(numberOfDigits);
+		for (d = 0; d < numberOfDigits; d++)
+		{
+		  startOfBin[d] = new Array(numberOfBins);
+		  startOfBin[d][0] = 0;
+		  for (var b = 1; b < numberOfBins; b++ )
+			startOfBin[d][b] = startOfBin[d][b - 1] + count[d][b - 1];
+		}
+
+		d = 0;
 		while( bitMask != 0 )    // end processing digits when all the mask bits have been processed and shifted out, leaving no bits set in the bitMask
 		{
-			for (var i = 0; i < numberOfBins; i++ )
-				count[ i ] = 0;
-			for (var current = 0; current < inputArray.length; current++ )    // Scan the array and count the number of times each digit value appears - i.e. size of each bin
-				count[ extractDigit( getKey(inputArray[ current ]), bitMask, shiftRightAmount ) ]++;
-			startOfBin[ 0 ] = 0;
-			for( var i = 1; i < numberOfBins; i++ )
-				startOfBin[ i ] = startOfBin[ i - 1 ] + count[ i - 1 ];
-			for ( var current = 0; current < inputArray.length; current++ )
-				OutputArray[ startOfBin[ extractDigit( getKey(inputArray[ current ]), bitMask, shiftRightAmount ) ]++ ] = inputArray[ current ];
-			bitMask <<= Log2ofPowerOfTwoRadix;
-			shiftRightAmount += Log2ofPowerOfTwoRadix;
-			OutputArrayHasResult = !OutputArrayHasResult;
-			var tmp = inputArray, inputArray = OutputArray, OutputArray = tmp;    // swap input and output arrays
+		  var startOfBinLoc = startOfBin[d];
+
+		  for (var current = 0; current < inputArray.length; current++)
+		  {
+			var endOfBinIndex = (inKeys[current] & bitMask) >> shiftRightAmount;
+			var index = startOfBinLoc[endOfBinIndex];
+			outputArray[  index] = inputArray[current];
+			outSortedKeys[index] = inKeys[    current];
+			startOfBinLoc[endOfBinIndex]++;
+		  }
+
+			bitMask <<= numberOfBitsPerDigit;
+			shiftRightAmount += numberOfBitsPerDigit;
+			outputArrayHasResult = !outputArrayHasResult;
+			d++;
+
+			var tmp = inputArray, inputArray = outputArray, outputArray = tmp;        // swap input and output arrays
+			var tmpKeys = inKeys;  inKeys = outSortedKeys;  outSortedKeys = tmpKeys;  // swap input and output key arrays
 		}
-		if ( OutputArrayHasResult )
-			for ( var current = 0; current < inputArray.length; current++ )    // copy from output array into the input array
-				inputArray[ current ] = OutputArray[ current ];
-		return inputArray;
-	}
 
+		return outputArrayHasResult ? outputArray : inputArray;
+	}
+
     return {
         //someProperty: 'prop value',
         RadixSortLsdUInt32: RadixSortLsdUInt32,

diff --git a/package.json b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "hpc-algorithms",
-  "version": "1.0.4",
+  "version": "1.0.5",
   "description": "High Performance Computing Algorithms",
   "main": "RadixSort.js",
   "scripts": {

diff --git a/readme.md b/readme.md
@@ -8,11 +8,9 @@ Let us know what other algorithms could use acceleration or improvement.
 Faster and Better Algorithms, starting with high performance sorting:
 - LSD Radix Sort for unsigned integer arrays. 20-30X faster than JavaScript's built-in array sort for arrays less than 35 Million.
 5-10X faster for arrays greater than 35 Million. This sort algorithm is not in-place, returning a new sorted array.
-Discussion, benchmarks and example in https://duvanenko.tech.blog/2017/06/15/faster-sorting-in-javascript/
-- LSD Radix Sort for arrays of user defined classes by an unsigned integer key. This is a stable sort,
-while JavaScript built-in is not stable.
-Discussion, benchmarks and example in https://duvanenko.tech.blog/2017/07/10/sorting-arrays-of-objects-in-javascript-with-radix-sort/
-
-Even higher performance coming soon for both of these algorithms, with support for more data types.
+Discussion, benchmarks and usage in https://duvanenko.tech.blog/2017/06/15/faster-sorting-in-javascript/
+- LSD Radix Sort for arrays of objects by an unsigned integer key. 15X faster than JavaScript's built-in .sort().
+This is a stable sort, while JavaScript built-in is not stable. This sort algorithm is not in-place, returning a new sorted array.
+Discussion, benchmarks and usage in https://duvanenko.tech.blog/2017/07/10/sorting-arrays-of-objects-in-javascript-with-radix-sort/
 
 If you have a specific needs for higher performance algorithms, let us know.