DOCSLIB.ORG

Comparisons and Data Movement About This Lecture Outline the Sort Problem Operations Comparing Primitive Va

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text

Load more

Revised 27/07/03 2 About This Lecture Sorting Operations - Comparisons In this lecture we will learn about the two basic operations performed by Sort and Data Movement Algorithms: comparison and element movement We will also learn how to express them as Cmput 115 - Lecture 8 a number of access operations. Department of Computing Science University of Alberta ©Duane Szafron 2000 Some code in this lecture is based on code from the book: Java Structures by Duane A. Bailey or the companion structure package 3 4 Outline The Sort Problem Comparing Elements Given a collection, with elements that can Moving Elements be compared, put the elements in increasing or decreasing order. 0 1 2 3 4 5 6 7 8 60 30 10 20 40 90 70 80 50 0 1 2 3 4 5 6 7 8 10 20 30 40 50 60 70 80 90 5 6 Operations Comparing Primitive Values ¡ ¡ Given a collection, with elements that can be The sorting algorithms we will consider are based on compared, put the elements in increasing or comparing individual elements. decreasing order. ¡ If the elements are primitive values, we can use the < ¡ We must perform two operations to sort a operator to compare them. collection: ¡ If the elements are objects, we cannot use < – compare elements ¡ However, Java has an interface called Comparable (in – move elements the java.lang package) that defines the method ¡ The time to perform each of these two operations, compareTo(Object object) which can be used to and the number of times we perform each operation, compare objects from classes implementing the is critical to the time it takes to sort a collection. Comparable interface. 1 7 8 Comparing Objects - CompareTo() Designing Classes for Sorting ¡ compareTo(Object object) If we write our sorting algorithms using the – If the receiver is “less” than the argument, it compareTo() method, we can apply them to returns a negative int. collections that hold objects from any classes that implement the Comparable interface. – If the receiver is “equal” to the argument, it returns the zero int. ¡ – If the receiver is “greater” than the argument, it For example, the String and Integer classes implement returns a positive int. Comparable. ¡ Any class designer who wants the objects in a class to be sortable using our algorithms only needs to make the class implement the Comparable interface. 9 10 The Time for Comparing Values The Time for Comparing Objects 1 ¡ ¡ The time to actually compare two primitive values is It takes longer to compare two Java objects than two small (one virtual machine instruction). Java primitive values. ¡ ¡ The comparison time for values is dominated by the To compare objects, the compareTo() method must time it takes to access the two elements being access not only the object, but its internal state as well. compared. ¡ ¡ For example, the next slide shows the Java source For example, the two array accesses take much more code from the library (java.String) to compare two time than the actual comparison in the code: Strings. if (data[i] < data[j]) – The details are not important, just notice that it requires many ¡ Therefore, a comparison of primitive values "costs" accesses to the array that each String uses to store its chars. two data accesses. ¡ The important point is that comparing objects can "cost" many data accesses. 11 12 The Time for Comparing Objects 2 Moving Elements public int compareTo(String anotherString) { ¡ Besides comparing elements, the only other operation int len1 = this.count; that is essential to sorting is moving elements. int len2 = anotherString.count; ¡ int n = Math.min(len1, len2); The exact code for moving elements depends on the char v1[] = this.value; type of collection and the pattern of element char v2[] = anotherString.value; movement, but it consists of a series of data int i = this.offset; accesses. int j = anotherString.offset; ¡ One common form of element movement is an while (n-- != 0) { exchange which is done using a single temporary char c1 = v1[i++]; char c2 = v2[j++]; variable and three assignments. if (c1 != c2) { ¡ This process usually involves four container accesses return c1 - c2; and two local variable accesses. } ¡ } Since the local variable accesses often get mapped to return len1 - len2; registers or cache memory we won't count them. } 2 13 14 Exchange Algorithm (1) Exchange Algorithm (2) ref i ref j ref i ref j ref i ref j ref i ref j temp = c[i]; c[j] = temp; £¥¤§¦¨¡©¦©¤©¤§ ¨ § ¢¡ £¥¤§¦¨¡©¦©¤©¤§ ¨ § data a data b ¢¡ data a data b data a data b data a data b ¤§£ ¨ ¦ ¨ ¨¦¤¤§ ¨ § ¤§£ ¨ ¦¨ ¨ ¨¦©¤¤§ ¨ § temp temp temp temp ref i ref j c[i] = c[j]; ¦¨ ¨¦©¤§¤ ¢¤£ data a data b temp 15 16 Comparison and Movement Times Sorting ¡ To predict the total time for an algorithm, we can We will look at these sorting algorithms add the accesses used for comparison and the accesses used for movement in the algorithm. – Selection sort ¡ If the container holds primitive values each – Insertion sort comparison requires two accesses, but if the – Merge sort container holds objects the number of accesses – Quick sort may be harder to compute. ¡ If the algorithm uses exchanges, each exchange requires four accesses, but if the algorithm uses a different style of data movement, the number of accesses may be harder to compute. Revised 27/07/03 18 About This Lecture In this lecture we will review the sorting Sorting - Selection Sort algorithm called Selection Sort which was presented in CMPUT 114. We will analyze the time and space complexity of a standard implementation Cmput 115 - Lecture 9 for sorting arrays. Department of Computing Science University of Alberta ©Duane Szafron 2000 Some code in this lecture is based on code from the book: Java Structures by Duane A. Bailey or the companion structure package 3 19 20 Selection Sort Algorithm Outline Input: anArray – array of Comparable objects Selection Sort Algorithm n – sort the elements in positions 0…(n-1) Selection Sort – Implementation for Arrays Output: Time and Space Complexity of Selection a sorted array [0..(n-1)] Sort Idea: We first pick up the largest item and store it at the end, and then pick up the 2nd largest item and put it at the 2nd last, … Algorithm For last =(n-1),(n-2),…,1 – Find the largest element in positions 0…last – exchange it with the element at the last index 21 22 Selection Sort Code - Arrays Method - getMaxIndex() - Arrays public static void selectionSort(Comparable anArray[], public static int getMaxIndex(Comparable anArray[], int n) { int last) { // pre: 0 <= n <= anArray.length // pre: 0 <= last < anArray.length // post: values in anArray positions 0…..(n-1) // post: return the index of the max value in // positions 0…..(last-1) of the given array // are in ascending order int maxIndex; //index of largest object int maxIndex; //index of largest object int index; //index of inspected element int last; //index of last unsorted element maxIndex = last; for (index = last - 1; index >= 0; index--) { for (last = n-1; last > 0; last--) { if (anArray[index].compareTo(anArray[maxIndex]) > 0) maxIndex = getMaxIndex(anArray, last); maxIndex = index; swap(anArray, last, maxIndex); } } // we could check to see if maxIndex != last return maxIndex } // and only swap in this case. } code based on Bailey pg. 82 code based on Bailey pg. 82 23 24 Counting Comparison Accesses Selection Sort Code (recursion) public static void selectionSort(Comparab lanArray[], How many comparison accesses are required for a selection int n) sort of n elements ? { All the comparisons are done in getMaxIndex. // pre: 0 <= n <= anArray.length // post: objects in anArray positions 0…..(n-1) getMaxIndex’s loop does one comparison per iteration, and // are in ascending order iterates “last” times, so each call to getMaxindex does “last” comparisons. int maxIndex; //index of largest object The loop body in the sort method is executed (n-1) times with if ( n > 1 ) { “last” taking on the values n-1, n-2, … 1. getMaxIndex is called maxIndex = getMaxIndex(anArray, n); once with each value of “last”. swap(anArray, n, maxIndex); The total number of comparisons is: selectionSort(anArray,n-1); (n-1)+(n-2) + … + 1= (1+2+ … n)-n = [n(n+1)/2] - n } Since each comparison requires two accesses there are: } n(n+1) - 2n = n2 - n = O(n2) comparison accesses. 4 25 26 Counting Move Accesses Time Complexity of Selection Sort ¡ How many move accesses are required for a The number of comparisons and moves is selection sort of n elements? independent of the data (the initial order of ¡ The only time we do a move is in a reference exchange (swap) which required 4 accesses. elements doesn’t matter). ¡ The sort method executes swap() once on each Therefore, the best, average and worst iteration of its loop. The loop iterates n-1 times. case time complexities of the Selection ¡ The total number of move accesses is: Sort are all O(n2). 4*(n-1) = O(n). ¡ Since the number of comparison accesses is O(n2), the move accesses are insignificant. ¡ In total the code does O(n2) accesses. 27 Revised 27/07/03 Space Complexity of Selection Sort Besides the collection itself, the only extra storage for this sort is the single temp Sorting - Insertion Sort reference used in the swap method. Therefore, the space complexity of Selection Sort is O(n). Cmput 115 - Lecture 10 Department of Computing Science University of Alberta ©Duane Szafron 2000 Some code in this lecture is based on code from the book: Java Structures by Duane A. Bailey or the companion structure package 29 30 About This Lecture Outline In this lecture we will learn about a sorting The Insertion Sort Algorithm algorithm called the Insertion Sort.
Recommended publications
  • The Analysis and Synthesis of a Parallel Sorting Engine Redacted for Privacy Abstract Approv, John M

    The Analysis and Synthesis of a Parallel Sorting Engine Redacted for Privacy Abstract Approv, John M

    AN ABSTRACT OF THE THESIS OF Byoungchul Ahn for the degree of Doctor of Philosophy in Electrical and Computer Engineering, presented on May 3. 1989. Title: The Analysis and Synthesis of a Parallel Sorting Engine Redacted for Privacy Abstract approv, John M. Murray / Thisthesisisconcerned withthe development of a unique parallelsort-mergesystemsuitablefor implementationinVLSI. Two new sorting subsystems, a high performance VLSI sorter and a four-waymerger,werealsorealizedduringthedevelopment process. In addition, the analysis of several existing parallel sorting architectures and algorithms was carried out. Algorithmic time complexity, VLSI processor performance, and chiparearequirementsfortheexistingsortingsystemswere evaluated.The rebound sorting algorithm was determined to be the mostefficientamongthoseconsidered. The reboundsorter algorithm was implementedinhardware asasystolicarraywith external expansion capability. The second phase of the research involved analyzing several parallel merge algorithms andtheirbuffer management schemes. The dominant considerations for this phase of the research were the achievement of minimum VLSI chiparea,design complexity, and logicdelay. Itwasdeterminedthattheproposedmerger architecture could be implemented inseveral ways. Selecting the appropriate microarchitecture for the merger, given the constraints of chip area and performance, was the major problem.The tradeoffs associated with this process are outlined. Finally,apipelinedsort-merge system was implementedin VLSI by combining a rebound sorter
  • Parallel Sorting Algorithms + Topic Overview

    Parallel Sorting Algorithms + Topic Overview

    + Design of Parallel Algorithms Parallel Sorting Algorithms + Topic Overview n Issues in Sorting on Parallel Computers n Sorting Networks n Bubble Sort and its Variants n Quicksort n Bucket and Sample Sort n Other Sorting Algorithms + Sorting: Overview n One of the most commonly used and well-studied kernels. n Sorting can be comparison-based or noncomparison-based. n The fundamental operation of comparison-based sorting is compare-exchange. n The lower bound on any comparison-based sort of n numbers is Θ(nlog n) . n We focus here on comparison-based sorting algorithms. + Sorting: Basics What is a parallel sorted sequence? Where are the input and output lists stored? n We assume that the input and output lists are distributed. n The sorted list is partitioned with the property that each partitioned list is sorted and each element in processor Pi's list is less than that in Pj's list if i < j. + Sorting: Parallel Compare Exchange Operation A parallel compare-exchange operation. Processes Pi and Pj send their elements to each other. Process Pi keeps min{ai,aj}, and Pj keeps max{ai, aj}. + Sorting: Basics What is the parallel counterpart to a sequential comparator? n If each processor has one element, the compare exchange operation stores the smaller element at the processor with smaller id. This can be done in ts + tw time. n If we have more than one element per processor, we call this operation a compare split. Assume each of two processors have n/p elements. n After the compare-split operation, the smaller n/p elements are at processor Pi and the larger n/p elements at Pj, where i < j.
  • View Publication

    View Publication

    Patience is a Virtue: Revisiting Merge and Sort on Modern Processors Badrish Chandramouli and Jonathan Goldstein Microsoft Research {badrishc, jongold}@microsoft.com ABSTRACT In particular, the vast quantities of almost sorted log-based data The vast quantities of log-based data appearing in data centers has appearing in data centers has generated this interest. In these generated an interest in sorting almost-sorted datasets. We revisit scenarios, data is collected from many servers, and brought the problem of sorting and merging data in main memory, and show together either immediately, or periodically (e.g. every minute), that a long-forgotten technique called Patience Sort can, with some and stored in a log. The log is then typically sorted, sometimes in key modifications, be made competitive with today’s best multiple ways, according to the types of questions being asked. If comparison-based sorting techniques for both random and almost those questions are temporal in nature [7][17][18], it is required that sorted data. Patience sort consists of two phases: the creation of the log be sorted on time. A widely-used technique for sorting sorted runs, and the merging of these runs. Through a combination almost sorted data is Timsort [8], which works by finding of algorithmic and architectural innovations, we dramatically contiguous runs of increasing or decreasing value in the dataset. improve Patience sort for both random and almost-ordered data. Of Our investigation has resulted in some surprising discoveries about particular interest is a new technique called ping-pong merge for a mostly-ignored 50-year-old sorting technique called Patience merging sorted runs in main memory.
  • From Merge Sort to Timsort Nicolas Auger, Cyril Nicaud, Carine Pivoteau

    From Merge Sort to Timsort Nicolas Auger, Cyril Nicaud, Carine Pivoteau

    Merge Strategies: from Merge Sort to TimSort Nicolas Auger, Cyril Nicaud, Carine Pivoteau To cite this version: Nicolas Auger, Cyril Nicaud, Carine Pivoteau. Merge Strategies: from Merge Sort to TimSort. 2015. hal-01212839v2 HAL Id: hal-01212839 https://hal-upec-upem.archives-ouvertes.fr/hal-01212839v2 Preprint submitted on 9 Dec 2015 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Merge Strategies: from Merge Sort to TimSort Nicolas Auger, Cyril Nicaud, and Carine Pivoteau Universit´eParis-Est, LIGM (UMR 8049), F77454 Marne-la-Vall´ee,France fauger,nicaud,[email protected] Abstract. The introduction of TimSort as the standard algorithm for sorting in Java and Python questions the generally accepted idea that merge algorithms are not competitive for sorting in practice. In an at- tempt to better understand TimSort algorithm, we define a framework to study the merging cost of sorting algorithms that relies on merges of monotonic subsequences of the input. We design a simpler yet competi- tive algorithm in the spirit of TimSort based on the same kind of ideas. As a benefit, our framework allows to establish the announced running time of TimSort, that is, O(n log n).
  • An Efficient Implementation of Batcher's Odd-Even Merge

    An Efficient Implementation of Batcher's Odd-Even Merge

    254 IEEE TRANSACTIONS ON COMPUTERS, VOL. c-32, NO. 3, MARCH 1983 An Efficient Implementation of Batcher's Odd- Even Merge Algorithm and Its Application in Parallel Sorting Schemes MANOJ KUMAR, MEMBER, IEEE, AND DANIEL S. HIRSCHBERG Abstract-An algorithm is presented to merge two subfiles ofsize mitted from one processor to another only through this inter- n/2 each, stored in the left and the right halves of a linearly connected connection pattern. The processors connected directly by the processor array, in 3n /2 route steps and log n compare-exchange A steps. This algorithm is extended to merge two horizontally adjacent interconnection pattern will be referred as neighbors. pro- subfiles of size m X n/2 each, stored in an m X n mesh-connected cessor can communicate with its neighbor with a route in- processor array in row-major order, in m + 2n route steps and log mn struction which executes in t. time. The processors also have compare-exchange steps. These algorithms are faster than their a compare-exchange instruction which compares the contents counterparts proposed so far. of any two ofeach processor's internal registers and places the Next, an algorithm is presented to merge two vertically aligned This executes subfiles, stored in a mesh-connected processor array in row-major smaller of them in a specified register. instruction order. Finally, a sorting scheme is proposed that requires 11 n route in t, time. steps and 2 log2 n compare-exchange steps to sort n2 elements stored Illiac IV has a similar architecture [1]. The processor at in an n X n mesh-connected processor array.
  • Sequential Merge Sort 2 Let's Make Mergesort Parallel

    Sequential Merge Sort 2 Let's Make Mergesort Parallel

    CSE341T/CSE549T 09/15/2014 Merge Sort Lecture 6 Today we are going to turn to the classic problem of sorting. Recall that given an array of numbers, a sorting algorithm permutes this array so that they are arranged in an increasing order. You are already familiar with basics of the sorting algorithm we are learning today, but we will see how to make it parallel today. 1 Reminder: Sequential Merge Sort As a reminder, here is the pseudo-code for sequential Merge Sort. MergeSort(A; n) 1 if n = 1 2 then return A 3 Divide A into two Aleft and Aright each of size n=2 0 4 Aleft =MERGESORT(Aleft; n=2) 0 5 Aright =MERGESORT(Aright; n=2) 6 Merge the two halves into A0 7 return A0 The running time of the merge procedure is Θ(n). The overall running time (work) of the entire computation is W (n) = 2W (n=2) + Θ(n) = Θ(n lg n). 2 Let’s Make Mergesort Parallel The most obvious thing to do to make the merge sort algorithm parallel is to make the recursive calls in parallel. 1 MergeSort(A; n) 1 if n = 1 2 then return A 3 Divide A into two Aleft and Aright each of size n=2 0 4 Aleft spawn MERGESORT(Aleft; n=2) 0 5 Aright MERGESORT(Aright; n=2) 6 sync 7 Merge the two halves into A0 8 return A0 The work of the algorithm remains unchanged. What is the span? The recurrence is SMergeSort(n) = SMergeSort(n=2) + Smerge(n) Since we are merging the arrays sequentially, the span of the merge call is Θ(n) and the recurrence solves to SMergeSort(n) = Θ(n).
  • An Efficient Methodology to Sort Large Volume of Data

    An Efficient Methodology to Sort Large Volume of Data

    INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 03, MARCH 2020 ISSN 2277-8616 An Efficient Methodology to Sort Large Volume of Data S.Bharathiraja, G.Suganya, M.Premalatha, R.Kumar, Sakkaravarthi Ramanathan Abstract—Sorting is a basic data processing technique that is used in all day-day applications. To cope up with technological advancement and extensive increase in data acquisition and storage, Sorting requires improvement to minimize time taken for processing, response time and space required for processing. Various sorting techniques have been proposed by researchers but the applicability of those techniques for large volume of data is not assured. The main focus of this work is to propose a new sorting technique titled Neutral Sort, to reduce the time taken for sorting and decrease the response time for a large volume of data. Neutral Sort is designed as an enhancement to Merge Sort. The advantages and disadvantages of existing techniques in terms of their performance, efficiency and throughput are discussed and the comparative study shows that Neutral sort drastically reduces time taken for sorting and hence reduces the response time. Index Terms—Chunking, Banding, Sorting, Efficiency, Merge sort, Bigdata Sorting, Neutral Sort. —————————— —————————— 1 INTRODUCTION N data science, ordering of elements is very much useful in divided chunks may already be in sorted manner and hence I various applications and data base querying. Different doesn’t need further split, we have proposed a modified approaches have been proposed by researchers for Merge sort algorithm titled “Neutral sort” to reduce time performing the sorting operation effectively based on the type complexity and hence to reduce response time.
  • Parallel Techniques

    Parallel Techniques

    Algorithms and Applications Evaluating Algorithm Cost Processor-time product or cost of a computation can be defined as: Cost = execution time * total number of processors used Cost of a sequential computation simply its execution time, ts. Cost of a parallel computation is tp * n. Parallel execution time, tp, is given by ts/S(n). Cost-Optimal Parallel Algorithm One in which the cost to solve a problem on a multiprocessor is proportional to the cost (i.e., execution time) on a single processor system. Can be used to compare algorithms. Parallel Algorithm Time Complexity Can derive the time complexity of a parallel algorithm in a similar manner as for a sequential algorithm by counting the steps in the algorithm (worst case) . Following from the definition of cost-optimal algorithm But this does not take into account communication overhead. In textbook, calculated computation and communication separately. Sorting Algorithms - Potential Speedup O(nlogn) optimal for any sequential sorting algorithm without using special properties of the numbers. Best we can expect based upon a sequential sorting algorithm using n processors is Has been obtained but the constant hidden in the order notation extremely large. Also an algorithm exists for an n-processor hypercube using random operations. But, in general, a realistic O(logn) algorithm with n processors not be easy to achieve. Sorting Algorithms Reviewed Rank sort (to show that an non-optimal sequential algorithm may in fact be a good parallel algorithm) Compare and exchange operations (to show the effect of duplicated operations can lead to erroneous results) Bubble sort and odd-even transposition sort Two dimensional sorting - Shearsort (with use of transposition) Parallel Mergesort Parallel Quicksort Odd-even Mergesort Bitonic Mergesort Rank Sort The number of numbers that are smaller than each selected number is counted.
  • Mergesort, Quicksort Oct

    Mergesort, Quicksort Oct

    COMP 250 Fall 2017 13 - Recursive algorithms 3: mergesort, quicksort Oct. 6, 2017 Mergesort In lecture 7, we saw three algorithms for sorting a list of n items. We saw that, in the worst case, all of these algorithm required O(n2) operations. Such algorithms will be unacceptably slow if n is large. To make this claim more concrete, consider that if n = 220 ≈ 106 i.e one million, then n2 ≈ 1012. How long would it take a program to run that many instructions? Typical processors run at about 109 basic operations per second (i.e. GHz). So a problem that takes in the order of 1012 operations would require thousands of seconds of processing time. ( Having a multicore machine with say 4 processors only can speed things up by a factor of 4 { max { which doesn't change the argument here. ) Today we consider an alternative sorting algorithm that is much faster that these earlier O(n2) algorithms. This algorithm is called mergesort. Here is the idea. If the list has just one number (n = 1), then do nothing. Otherwise, partition the list of n elements into two lists of size about n=2 elements each, sort the two individual lists (recursively, using mergesort), and then merge the two sorted lists. For example, suppose we have a list < 8; 10; 3; 11; 6; 1; 9; 7; 13; 2; 5; 4; 12 > : We partition it into two lists < 8; 10; 3; 11; 6; 1 > < 9; 7; 13; 2; 5; 4; 12 > : and sort these (by applying mergesort recursively): < 1; 3; 6; 8; 10; 11 > < 2; 4; 5; 7; 9; 12; 13 > : Then, we merge these two lists to get < 1; 2; 3; 4; 5; 6; 7; 8; 9; 10; 11; 12; 13 > : Here is pseudocode for the algorithm.
  • Parallel Implementation of Sorting Algorithms

    Parallel Implementation of Sorting Algorithms

    IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 4, No 3, July 2012 ISSN (Online): 1694-0814 www.IJCSI.org 164 Parallel Implementation of Sorting Algorithms Malika Dawra1, and Priti2 1Department of Computer Science & Applications M.D. University, Rohtak-124001, Haryana, India 2Department of Computer Science & Applications M.D. University, Rohtak-124001, Haryana, India Abstract Usage of memory and other computer resources is also a factor in classifying the A sorting algorithm is an efficient algorithm, which sorting algorithms. perform an important task that puts elements of a list in a certain order or arrange a collection of elements into a Recursion: some algorithms are either particular order. Sorting problem has attracted a great recursive or non recursive, while others deal of research because efficient sorting is important to may be both. optimize the use of other algorithms such as binary search. This paper presents a new algorithm that will Whether or not they are a comparison sort split large array in sub parts and then all sub parts are examines the data only by comparing two processed in parallel using existing sorting algorithms elements with a comparison operator. and finally outcome would be merged. To process sub There are several sorting algorithms available to parts in parallel multithreading has been introduced. sort the elements of an array. Some of the sorting algorithms are: Keywords: sorting, algorithm, splitting, merging, multithreading. Bubble sort 1. Introduction In the bubble sort, as elements are sorted they gradually “bubble” (or rise) to their proper location Sorting is the process of putting data in order; either in the array.
  • Strategies for Stable Merge Sorting

    Strategies for Stable Merge Sorting

    Strategies for Stable Merge Sorting Sam Buss∗ Alexander Knop∗ Abstract relative order of pairs of entries in the input list. We introduce new stable natural merge sort algorithms, Information-theoretic considerations imply that any called 2-merge sort and α-merge sort. We prove upper and comparison-based sorting algorithm must make at least lower bounds for several merge sort algorithms, including log2(n!) ≈ n log2 n comparisons in the worst case. How- Timsort, Shiver's sort, α-stack sorts, and our new 2-merge ever, in many practical applications, the input is fre- and α-merge sorts. The upper and lower bounds have quently already partially sorted. There are many adap- the forms c · n log m and c · n log n for inputs of length n tive sort algorithms which will detect this and run faster comprising m runs. For Timsort, we prove a lower bound of on inputs which are already partially sorted. Natu- (1:5−o(1))n log n. For 2-merge sort, we prove optimal upper ral merge sorts are adaptive in this sense: they detect and lower bounds of approximately (1:089 ± o(1))n log m. sorted sublists (called \runs") in the input, and thereby We state similar asymptotically matching upper and lower reduce the cost of merging sublists. One very popular bounds for α-merge sort, when ' < α < 2, where ' is the stable natural merge sort is the eponymous Timsort of golden ratio. Tim Peters [25]. Timsort is extensively used, as it is in- Our bounds are in terms of merge cost; this upper cluded in Python, in the Java standard library, in GNU bounds the number of comparisons and accurately models Octave, and in the Android operating system.
  • Sorting with Gpus: a Survey Dmitri I

    Sorting with Gpus: a Survey Dmitri I

    1 Sorting with GPUs: A Survey Dmitri I. Arkhipov, Di Wu, Keqin Li, and Amelia C. Regan Abstract—Sorting is a fundamental operation in computer Our key findings are the following: science and is a bottleneck in many important fields. Sorting • Effective parallel sorting algorithms must use the faster is critical to database applications, online search and indexing, biomedical computing, and many other applications. The explo- access on-chip memory as much and as often as possible sive growth in computational power and availability of GPU as a substitute to global memory operations. coprocessors has allowed sort operations on GPUs to be done • Algorithmic improvements that used on-chip memory much faster than any equivalently priced CPU. Current trends and made threads work more evenly seemed to be more in GPU computing shows that this explosive growth in GPU effective than those that simply encoded sorts as primitive capabilities is likely to continue for some time. As such, there is a need to develop algorithms to effectively harness the power of GPU operations. GPUs for crucial applications such as sorting. • Communication and synchronization should be done at points specified by the hardware. Index Terms—GPU, sorting, SIMD, parallel algorithms. • Which GPU primitives (scan and 1-bit scatter in partic- ular) are used makes a big difference. Some primitive I. INTRODUCTION implementations were simply more efficient than others, In this survey, we explore the problem of sorting on GPUs. and some exhibit a greater degree of fine grained paral- As GPUs allow for effective hardware level parallelism, re- lelism than others.