US 6,088,511 AGrant
Nested Parallel 2D Delaunay Triangulation Method
Issue Date:2000-07-11
•13 Claims
•15 Drawing Sheets
Abstract
A nested parallel implementation of 2D triangulation method recursively sub-divides processors of a parallel computer into asynchronous processor teams. Each of the teams uses data parallel operations to compute a partitioning of the collection of points distributed to it. When each team has a single processor as a result of the recursive partitioning steps, the processors switch to a serial version of the 2D triangulation method. The nested parallel implementation has two levels of recursion: 1) one to partition a collection of points into two new sets; and 2) a second layer nested in the first to compute convex hulls used to form a border around the two new sets of points. In each layer of recursion the implementation sub-divides processors into teams and assigns a control parallel function to each team. Within each team, the processors perform data parallel operations on the collection of points distributed to the processors in the team.
Metadata
Assignee
- Microsoft Corporation
Inventor
- Jonathan C. Hardwick
Application Information
Application Number:US 1081519
Filing Date:1998-06-30
Priority Date:1998-06-30
Art Unit:273
Classifications
IPC:
G06F 944
Field of Search:
71271734528;36;36423;426;479
Patent Drawings (15 sheets)
Description
Field of the Invention
The invention relates to parallel processing on distributed memory and shared memory parallel computers having multiple processors.
Background of the Invention
It is hard to program parallel computers. Dealing with many processors at the same time, either explicitly or implicitly, makes parallel programs harder to design, analyze, build, and evaluate than their serial counterparts. However, using a fast serial computer to avoid the problems of parallelism is often not enough. There are always problems that are too big, too complex, or whose results are needed too soon.
Ideally, a parallel programming model or language should provide the same advantages we seek in serial languages: portability, efficiency, and ease of expression. However, it is typically impractical to extract parallelism from sequential languages. In addition, previous parallel languages have generally ignored the issue of nested parallelism, where the programmer exposes multiple simultaneous sources of parallelism in an algorithm. Supporting nested parallelism is particularly important for irregular algorithms, which operate on non-uniform data structures (for example, sparse arrays, trees and graphs).
Parallel Languages and Computing Models
The wide range of parallel architectures make it difficult to create a parallel computing model that is portable and efficient across a variety of architectures. Despite shifts in market share and the demise of some manufacturers, users can still choose between tightly-coupled shared-memory multiprocessors such as the SGI Power Challenge, more loosely coupled distributed-memory multicomputers such as the IBM SP2, massively-parallel SIMD machines such as the MasPar
MP-2, vector supercomputers such as the Cray C90, and loosely coupled clusters of workstations such as the DEC SuperCluster. Network topologies are equally diverse, including 2D and 3D meshes on the Intel Paragon and ASCI Red machine, 3D tori on the Cray T3D and T3E, butterfly networks on the IBM SP2, fat trees on the Meiko CS-2, and hypercube networks on the SGI Origin2000. With extra design axes to specify, parallel computers show a much wider range of design choices than do serial machines, with each choosing a different set of tradeoffs in terms of cost, peak processor performance, memory bandwidth, interconnection technology and topology, and programming software.
This tremendous range of parallel architectures has spawned a similar variety of theoretical computational models. Most of these are variants of the original CRCW PRAM model (Concurrent-Read Concurrent-Write Parallel Random Access Machine), and are based on the observation that although the CRCW PRAM is probably the most popular theoretical model amongst parallel algorithm designers, it is also the least likely to ever be efficiently implemented on a real parallel machine. That is, it is easily and efficiently portable to no parallel machines, since it places more demands on the memory system in terms of access costs and capabilities than can be economically supplied by current hardware. The variants handicap the ideal PRAM to resemble a more realistic parallel machine, resulting in the locality-preserving H-PRAM, and various asynchronous, exclusive access, and queued PRAMs. However, none of these models have been widely accepted or implemented.
Parallel models which proceed from machine characteristics and then abstract away details--that is, "bottom-up" designs rather than "top-down"--have been considerably more successful, but tend to be specialized to a particular architectural style. For example, LogP is a low-level model for message-passing machines, while BSP defines a somewhat higher-level model in terms of alternating phases of asynchronous computation and synchronizing communication between processors. Both of these models try to accurately characterize the performance of any message-passing network using just a few parameters, in order to allow a programmer to reason about and predict the behavior of their programs.
However, the two most successful recent ways of expressing parallel programs have been those which are arguably not models at all, being defined purely in terms of a particular language or library, with no higher-level abstractions. Both High Performance Fortran and the Message Passing Interface have been created by committees and specified as standards with substantial input from industry, which has helped their widespread adoption. HPF is a full language that extends sequential Fortran with predefined parallel operations and parallel array layout directives. It is typically used for computationally intensive algorithms that can be expressed in terms of dense arrays. By contrast, MPI is defined only as a library to be used in conjunction with an existing sequential language. It provides a standard message-passing model, and is a superset of previous commercial products and research projects such as PVM and NX. Note that MPI is programmed in a control-parallel style, expressing parallelism through multiple paths of control, whereas HPF uses a data-parallel style, calling parallel operations from a single thread of control.
Nested and Irregular Parallelism
Neither HPF or MPI provide direct support for nested parallelism or irregular algorithms. For example, consider the quicksort algorithm set forth below. The irregularity comes from the fact that the two subproblems that quicksort creates are typically not of the same size; that is, the divide-and-conquer algorithm is unbalanced.
Although it was originally written to describe a serial algorithm, the pseudocode shown above contains both data-parallel and control-parallel operations. Comparing the elements of the sequence S to the pivot element a, and selecting the elements for the new subsequences S.sub.1, S.sub.2, and S.sub.3 are inherently data-parallel operations. Meanwhile, recursing on S.sub.1 and S.sub.3 can be implemented as a control-parallel operation by performing two recursive calls in parallel on two different processors.
Note that a simple data-parallel quicksort (such as one written in HPF) cannot exploit the control parallelism that is available in this algorithm, while a simple control-parallel quicksort (such as one written in a sequential language and MPI) cannot exploit the data parallelism that is available. For example, a simple control-parallel divide-and-conquer implementation would initially put the entire problem onto a single processor, leaving the rest of the processors unused. At the first divide step, one of the subproblems would be passed to another processor. At the second divide step, a total of four processors would be involved, and so on. The parallelism achieved by this algorithm is proportional to the number of threads of control, which is greatest at the end of the algorithm. By contrast, a data-parallel divide-and-conquer quicksort would serialize the recursive applications of the function, executing one at a time over all of the processors. The parallelism achieved by this algorithm is proportional to the size of the subproblem being operated on at any instant, which is greatest at the beginning of the algorithm. Towards the end of the algorithm there will be fewer data elements in a particular function application than there are processors, and so some processors will remain idle.
By simultaneously exposing both nested sources of parallelism, a nested parallel implementation of quicksort can achieve parallelism proportional to the total data size throughout the algorithm, rather than only achieving full parallelism at either the beginning (in data parallelism) or the end (in control parallelism) of the algorithm. The benefits of a nested parallel implementation are illustrated in more detail in Hardwick, Jonathan, C., Practical Parallel Divide-and-Conquer Algorithms, PhD Thesis, Carnegie Mellon University, December 1997, CMU-CS-97-197.
Divide-and-Conquer Algorithms
Divide-and-conquer algorithms solve a problem by splitting it into smaller, easier-to-solve parts, solving the subproblems, and then combining the results of the subproblems into a result for the overall problem. The subproblems typically have the same nature as the overall problem (for example, sorting a list of numbers), and hence can be solved by a recursive application of the same algorithm. A base case is needed to terminate the recursion. Note that a divide-and-conquer algorithm is inherently dynamic, in that we do not know all of the subtasks in advance.
Divide-and-conquer has been taught and studied extensively as a programming paradigm, and can be found in any algorithm textbook. In addition, many of the most efficient and widely used computer algorithms are divide-and-conquer in nature. Examples from various fields of computer science include algorithms for sorting, such as merge-sort and quicksort, for computational geometry problems such as convex hull and closest pair, for graph theory problems such as traveling salesman and separators for VLSI layout, and for numerical problems such as matrix multiplication and fast Fourier transforms.
The subproblems that are generated by a divide-and-conquer algorithm can typically be solved independently. This independence allows the subproblems to be solved simultaneously, and hence divide-and-conquer algorithms have long been recognized as possessing a potential source of control parallelism. Additionally, all of the previously described algorithms can be implemented in terms of data-parallel operations over collection-oriented data types such as sets or sequences, and hence, data parallelism can be exploited in their implementation. However, there is still a need to define a model in which to express this parallelism. Previous models have included fork-join parallelism, and the use of processor groups, but both of these models have severe limitations. In the fork-join model available parallelism and the maximum problem size is greatly limited, while group-parallel languages have been limited in their portability, performance, and/or ability to handle irregular divide-and-conquer algorithms.
Summary of the Invention
The invention provides a method for implementing a parallel two-dimensional Delaunay triangulation on a parallel computer. The method is implemented in a nested parallel program that recursively partitions a collection of 2D points into small subsets of points and then applies a 2D triangulation to each subset. The recursive calls to partition the collection of points are distributed to teams of processors in a parallel computer. Within the teams, the processors perform data parallel operations to partition the collection of points into subsets.
An implementation of the nested parallel 2D Delaunay triangulation method uses a parallel programming model called the team parallel model. Significant aspects of the team parallel model include: 1) recursive subdivision of the processors of a parallel machine into asynchronous processor teams to match the runtime behavior of a recursive divide-and-conquer algorithm, where the number of processors in each team are selected based on computational cost of each recursive function call; 2) switching to efficient serial code (either user supplied or converted from the parallel code by a preprocessor/compiler) when the subdivision reaches a single processor; 3) a collection oriented data type that supports data parallel operations within processor teams; and 4) dynamic load balancing of the processing workload among processors.
The team parallel implementation uses asynchronous processor teams to exploit control parallelism of two layers of recursion: an outer layer representing the main program; and an inner layer representing a convex hull routine. The processors within each team perform data parallel operations on collection oriented data types.
In the outer layer, the main program partitions a collection of points into two smaller problems. Specifically, the main program finds the median of the points along an axis, projects the points on a parabaloid centered on the median line, finds the lower convex hull of the points projected into a vertical plane through the median line, and then uses the lower convex hull to form a dividing path through the points. To form two new sets of points, the main program uses the dividing path to create two new borders enclosing two new sets of points. The main program then recurses on each of the two new sets of points.
In the inner layer, a recursive convex hull routine computes the lower convex hull of a collection of points. This layer recurses on irregular problem sizes since it divides the problem into unequal sets of points.
When recursive calls of the main program proceed to the point where each team has a single processor, each processor switches to a serial version of the 2D triangulation program. This version is a separate program supplied by the user, although it could be generated from the parallel version. At this stage, the processors use the dynamic load balancing of the team parallel system to send function calls to idle processors. Both layers of recursion take advantage of a lazy collection oriented data type to minimize inter-processor communication.
While the specific implementation has layers of recursion, the nested parallel approach is sufficiently general and flexible to allow arbitrary combinations of control and data parallel operations. Thus, it is possible to have more or fewer layers of recursion to enhance performance for a particular application and hardware platform.
Additional features of the invention will become apparent with reference to the following detailed description and accompanying drawings. The description begins with a discussion of the nested parallel programming system used to implement the parallel triangulation and then describes an implementation in detail.
Brief Description of the Drawings
FIG. 1 is a block diagram illustrating a parallel computer in which the memory is physically and logically distributed.
FIG. 2 is a block diagram illustrating a parallel computer in which the memory is physically distributed and logically shared.
FIG. 3 is a block diagram illustrating a parallel computer in which the memory is physically and logically shared.
FIG. 4 is a block diagram illustrating the Machiavelli system, an implementation of the team parallel model.
FIGS. 5A-E are diagrams illustrating examples of the use of a lazy collection oriented data type in a nested parallel program.
FIG. 6 is a diagram illustrating the operation of a processor seeking to offload its workload in Machiavelli's dynamic load balancing system.
FIG. 7 is a diagram illustrating the message processing functions of a manager processor in Machiavelli's dynamic load balancing system.
FIG. 8 is a diagram illustrating the message processing functions of an idle processor in Machiavelli's dynamic load balancing system.
FIG. 9 is a diagram illustrating an example of dynamic load balancing in the Machiavelli system.
FIG. 10 is a diagram conceptually illustrating the nested parallelism and irregular problem sizes within an implementation of a 2D Delaunay triangulation.
FIGS. 11A-C are diagrams graphically illustrating the 2D triangulation. FIG. 11A illustrates a view of a collection of points in a 2D plane, with a solid dark line representing the median, and the thicker white line representing the result of a lower convex hull projected into the plane. FIG. 11B represents the points projected onto a parabaloid centered on a point on the median line. The path of the points in the lower convex hull is shown in a dark line. FIG. 12B shows the horizontal projection of the points into a vertical plane through the median line.
FIG. 12 is a flow diagram illustrating the implementation of the 2D triangulation using the team parallel model.
FIG. 13 is a flow diagram illustrating the cases used to merge a new border of points with an old border of points in the implementation of the 2D triangulation method shown in FIG. 12.
FIG. 14 is a flow diagram illustrating the operation of the split function in Machiavelli.
Detailed Description
1. Introduction
1.1 The Team Parallel Model
The team parallel model uses data parallelism within teams of processors acting in a control-parallel manner. The basic data structures are distributed across the processors of a team, and are accessible via data-parallel primitives. The divide stage of a divide-and-conquer algorithm is then implemented by dividing the current team of processors into two or more subteams, distributing the relevant data structures between them, and then recursing independently within each subteam. This natural mixing of data parallelism and control parallelism allows the team parallel model to easily handle nested parallelism.
There are three important additions to this basic approach. First, in the case when the algorithm is unbalanced, the subteams of processors may not be equally sized--that is, the division may not always be into equal pieces. In this case, the team approach chooses the sizes of the subteams in relation to the cost of their subtasks. However, this passive load-balancing method is often not enough, because the number of processors is typically much smaller than the problem size, and hence granularity effects mean that some processors will end up with an unfair share of the problem when the teams have recursed down to the point where each consists of a single processor. Thus, the second important addition is an active load-balancing system. The approach described here uses function-shipping to farm out computation to idle processors. This can be seen as a form of remote procedure call. Finally, when the team consists of a single processor that processor can use efficient sequential code, either by calling a version of the parallel algorithm compiled for a single processor or by using a completely different sequential algorithm supplied by the programmer.
1.2 Overview of the Components of an Implementation
Machiavelli is a particular implementation of the team parallel model. It is presented as an extension to the C programming language, although any sequential programming language could be used. The basic data-parallel data structure is a vector, which is distributed across all the processors of the current team. Vectors can be formed from any of the basic C datatypes, and from any user-defined datatypes. Machiavelli supplies a variety of basic parallel operations that can be applied to vectors (scans, reductions, permutations, appends, etc), and in addition allows the user to construct simple data-parallel operations of their own. A special syntax is used to denote recursion in a divide-and-conquer algorithm.
Machiavelli is implemented using a preprocessor that translates the language extensions into C plus calls to MPI. The use of MPI ensures the greatest possible portability across both distributed-memory machines and shared-memory machines, but again, any other method of communicating between processors could be used, such as PVM. Data-parallel operations in Machiavelli are translated into loops over sections of a vector local to each processor; more complex parallel operations use MPI function calls in addition to local operations. The recursion syntax is translated into code that computes team sizes, subdivides the processors into teams, redistributes the arguments to the appropriate subteams, and then recurses in a smaller team. User-defined datatypes are automatically mapped into MPI datatypes, allowing them to be used in any Machiavelli operation (for example, sending a vector of user-defined point structures instead of sending two vectors of x and y coordinates). This also allows function arguments and results to be transmitted between processors, allowing the use of a function-shippingactive load-balancing system. Additionally, Machiavelli supports both serial compilation of parallel functions (removing the MPI constructs), and the overriding of these default functions with user-defined serial functions that can implement a more efficient algorithm.
Though Machiavelli's preprocessor converts nested parallel programs into C language code and calls to MPI, it is also possible to implement a preprocessor for the team parallel model to convert a nested parallel language into other well known sequential, imperative programming languages. In general, imperative programming code is comprised of a list of program statements that tell a processor which operations to perform on specified data stored in a computer's memory. Imperative programming code specifies the flow of control through program statements executed in a sequential order. Imperative programming languages typically provide abstractions of the processor's instruction set and memory to hide the details of specific machine instructions and memory usage from the programmer. For instance, an imperative language supplies data types as an abstraction of the data in memory. An imperative language also supplies operators that enable the programmer to create expressions or program statements as an abstraction for the instructions of the processor.
Some examples of imperative programming languages include C++, Java, FORTRAN, Ada, Pascal, Modula-2, etc. The preprocessor may be implemented to convert a nested parallel program to code in a variety of known sequential programming languages, whether they strictly follow the imperative programming model or not. For instance, many object-oriented languages such as C++ are at least in part based on an imperative programming model. Thus, it is possible to design the preprocessor for an object-oriented language as well.
2. Overview of Parallel Computers
The team parallel model is designed to be implemented on a MIMD (multiple instruction, multiple data) architecture. Since a MIMD architecture is a very general parallel architecture, encompassing a wide variety of parallel machines, the team parallel model is portable across a wide variety of parallel processing computers, including both distributed memory and shared memory machines.
To illustrate the broad scope of the MIMD architecture, consider the alternative parallel architecture called SIMD (Single Instruction Multiple Data). In a SIMD architecture, the parallel processors each execute a single instruction stream on their own part of the data. The processors operate in parallel by simultaneously executing the same instruction stream. The processors operate synchronously such that each one is in lock step with the others. Finally, the data of the single instruction stream is spread across the processors such that each operates on its own part of the data. In contrast, the processors in a MIMD architecture execute independently on potentially distinct instruction streams.
In terms of the physical and logical organization of memory, a MIMD architecture encompasses three classes of parallel computers: 1) parallel computers for which memory is physically and logically distributed; 2) parallel computers for which memory is physically distributed and logically shared; and 3) parallel computers for which memory is physically and logically shared. The implementation of the team parallel model assumes a parallel architecture with distributed memory. Since a distributed memory implementation can be executed in a shared memory architecture, the implementation applies to both distributed and shared memory architectures.
FIGS. 1-3 illustrate examples of three parallel computing architectures. FIG. 1 is a block diagram illustrating a distributed memory computer in which the memory is physically and logically distributed. This distributed memory architecture includes a number of processors 20, 22, 24, each with a corresponding memory 26, 28, 30. The processors execute code and operate on data stored in their respective memories. To communicate, the processors pass messages to each other via the interconnect 32. The interconnect may be a custom switch, such as a crossbar, or a commodity network, such as FDDI, ATM, or Ethernet. Examples of this type of parallel computer include the Alpha Cluster from Digital Equipment Corporation and theSP2 from IBM Corporation. The Alpha Cluster is comprised of Alpha workstations interconnected via a FDDI switch. The SP2 is a supercomputer comprised of RS workstations interconnected via an SP2 switch from IBM.
FIG. 2 is a block diagram illustrating a distributed memory computer in which the memory is physically distributed, yet logically shared. The memory is logically shared in the sense that each processor can access another processor's memory without involving that processor. Like the distributed memory architecture of FIG. 1, this architecture includes a number of processors 40, 42, 44, each with a corresponding memory 46, 48, 50. However, the processor-memory nodes are interconnected via a controller (e.g., 52, 54, 56) that enables each processor to access another's memory directly. Each of the processor-memory nodes are interconnected via an interconnect 58. An example of this type of parallel computer is a Cray T3D, which is comprised of Alpha nodes interconnected with a T3D interconnect that supports a form of logical shared memory access called SHMEM.
FIG. 3 is a block diagram illustrating an example of a shared memory architecture in which the memory is physically and logically shared. In this type of architecture, the processors 60, 62, 64 are interconnected via a bus 66. Each of the processors can access each other's data in the shared memory 68. An example of this type of computer is an SGI Challenge from Silicon Graphics Incorporated.
In general, the team-parallel model assumes that we have many more items of data than processors--that is, N>>P. The model also assumes that it is many times faster for a processor to perform a computation using local registers than it is to read or write a word to or from local memory, which in turn is many times faster than initiating communication to a different processor.
The Machiavelli language is designed to be easily portable to all of the architectures shown in FIGS. 1-3. The Machiavelli preprocessor converts program code written in the team parallel style into C and MPI functions. Since MPI is implemented on each of the architectures of FIGS. 1-3, Machiavelli is easily portable to all of them. In each of the architectures, the underlying implementation of the MPI functions may differ, but this is of no concern since a Machiavelli program uses the MPI implementation of the specific architecture to which is compiled. In the Cray T3D, for example, the MPI implementation is based on the SHMEM interface library. In the IBM SP2, the MPI implementation is based on the IBM MPL library. While Machiavelli is specifically based on MPI, it is possible to use other message passing libraries. One alternative, for example, is the PVM message passing library.
3. The Divide and Conquer Programming Strategy
3.1 Overview of Divide and Conquer Programs
A divide-and-conquer programming strategy solves a large instance of a problem by breaking it into smaller instances of the same problem, and using the solutions of these to solve the original problem. This strategy is a variant of the more general top-down programming strategy, but is distinguished by the fact that the subproblems are instances of the same problem. Divide-and-conquer algorithms can therefore be expressed recursively, applying the same algorithm to the subproblems as to the original problem. As with any recursive algorithm, a divide-and-conquer problem needs a base case to terminate the recursion. Typically this base case tests whether the problem is small enough to be solved by a direct method. For example, in quicksort the base case is reached when there are 0 or 1 elements in the list. At this point the list is sorted, so to solve the problem at the base case, the algorithm just returns the input list.
Apart from the base case, a divide-and-conquer problem also needs a divide phase, to break the problem up into subproblems, and a combine phase, to combine the solutions of the subproblems into a solution to the original problem. As an example, quicksort's divide phase breaks the original list into three lists--containing elements less than, equal to, and greater than the pivot--and its combine phase appends the two sorted subsolutions on either side of the list containing equal elements.
This structure of a base case, direct solver, divide phase, and combine phase can be generalized into a template (or skeleton) for divide-and-conquer algorithms. Pseudocode for such a template is shown in the pseudocode listing below. The literature contains many variations of this basic template.
Using this basic template as a reference, there are now several axes along which to differentiate divide-and-conquer algorithms, and in particular their implementation on parallel architectures. Sections 3.1.1 to 3.1.8 will describe these axes. Four of these axes--branching factor, balance, data-dependence of divide function, and sequentiality--have been previously described in the theoretical literature. However, the remaining three--data parallelism, embarrassing divisibility, and data-dependence of size function--have not been widely discussed, although they are important from an implementation perspective.
3.1.1 Branching Factor
The branching factor of a divide-and-conquer algorithm is the number of subproblems into which a problem is divided. This is the most obvious way of classifying divide-and-conquer algorithms, and has also been referred to as the degree of recursion. For true divide-and-conquer algorithms the branching factor must be two or more, since otherwise the problem is not being divided.
3.1.2 Balance
A divide-and-conquer algorithm is balanced if it divides the initial problem into equally-sized subproblems. This has typically been defined only for the case where the sizes of the subproblems sum to the size of the initial problem, for example in a binary divide-and-conquer algorithm dividing a problem of size N into two subproblems of size N/2.
3.1.3 Near Balance
A particular instantiation of a balanced divide-and-conquer algorithm is near-balanced if it cannot be mapped in a balanced fashion onto the underlying machine at run time. Near-balance is another argument for supporting some form of load balancing, as it can occur even in a balanced divide-and-conquer algorithm. This can happen for two reasons. The first is that the problem size is not a multiple of a power of the branching factor of the algorithm. For example, a near-balanced problem of size 17 would be divided into subproblems of sizes 9 and 8 by a binary divide-and-conquer algorithm. At worst, balanced models with no load-balancing must pad their inputs to the next higher power of the branching factor (i.e., to 32 in this case). This can result in a slowdown of at most the branching factor.
The second reason is that, even if the input is padded to the correct length, it may not be possible to evenly map the tree structure of the algorithm onto the underlying processors. For example, in the absence of load-balancing a balanced binary divide-and-conquer problem on a twelve-processor SGI Power Challenge could efficiently use at most eight of the processors. Again, this can result in a slowdown of at most the branching factor.
3.1.4 Embarrassing Divisibility
A balanced divide-and-conquer algorithm is embarassingly divisible if the divide step can be performed in constant time. This is the most restrictive form of a balanced divide-and-conquer algorithm, and is a divide-and-conquer case of the class of "embarassingly parallel" problems in which no (or very little) inter-processor communication is necessary. In practice, embarassingly divisible algorithms are those in which the problem can be treated immediately as two or more subproblems, and hence no extra data movement is necessary in the divide step. For the particular case of an embarassingly divisible binary divide-and-conquer algorithm, Kumaran and Quinn coined the term left-right algorithm (since the initial input data is merely treated as left and right halves) and restricted their model to this class of algorithms (See Santhosh Kumaran and Michael J. Quinn, Divide-and-conquer programming on MIMD computers. In Proceedings of the 9.sup.th International Parallel Processing Symposium, pages 734-741. IEEE, April 1995). Examples of embarassingly divisible algorithms include dot product and matrix multiplication of balanced matrices.
3.1.5 Data Dependence of Divide Function
A divide-and-conquer algorithm has a data-dependent divide function if the relative sizes of subproblems are dependent in some way on the input data. This subclass accounts for the bulk of unbalanced divide-and-conquer algorithms. For example, a quicksort algorithm can choose an arbitrary pivot element with which to partition the problem into subproblems, resulting in an unbalanced algorithm with a data-dependent divide function. Alternatively, it can use the median element, resulting in a balanced algorithm with a divide function that is independent of the data.
3.1.6 Data Dependence of Size Function
An unbalanced divide-and-conquer algorithm has a data-dependent size function if the total size of the subproblems is dependent in some way on the input data. This definition should be contrasted with that of the data-dependent divide function, in which the total amount of data at a particular level is fixed, but its partitioning is not. This category includes algorithms that either add or discards elements based on the input data. In practice, algorithms that discard data, in an effort to further prune the problem size are more common.
3.1.7 Control Parallelism or Sequentiality
A divide-and-conquer algorithm is sequential if the subproblems must be executed in a certain order. Ordering occurs when the result of one subtask is needed for the computation of another subtask. The ordering of subproblems in sequential divide-and-conquer algorithms eliminates the possibility of achieving control parallelism through executing two or more subtasks at once, and hence, any speed-up must be achieved through data parallelism.
3.1.8 Data Parallelism
A divide-and-conquer algorithm is data parallel if the test, divide, and combine operations do not contain any serial bottlenecks. As well as the control parallelism inherent in the recursive step, it is also advantageous to exploit data parallelism in the test, divide and combine phases. Again, if this parallelism is not present, the possible speedup of the algorithm is severely restricted. Data parallelism is almost always present in the divide stage, since division is typically structure-based (for example, the two halves of a matrix in matrix multiplication) or value-based (for example, elements less than and greater than the pivot in quicksort), both of which can be trivially implemented in a data-parallel fashion.
4. Implementation of Team Approach
4.1 Overview of Implementation
Team parallelism is designed to support all of the characteristics of the divide-and-conquer algorithms discussed in section 3. There are four main features of the team parallel model:
1. Asynchronous subdividable teams of processors.
2. A collection-oriented data type supporting data-parallel operations within teams.
3. Efficient serial code executing on single processors.
4. An active load-balancing system.
By combining all of the features, the team parallel model can be used to develop efficient implementations of a wide range of divide-and-conquer algorithms, including both balanced and unbalanced examples. The next sections define each of these features and their relevance to divide-and-conquer algorithms, and discuss their implications, concentrating in particular on message-passing distributed memory machines.
4.1.1 Teams of Processors
As its name suggests, team parallelism uses teams of processors. These are independent and distinct subsets of processors. Teams can divide into two or more subteams, and merge with sibling subteams to reform their original parent team. This matches the behavior of a divide-and-conquer algorithm, with one subproblem being assigned per subteam.
Sibling teams run asynchronously with respect to each other, with no communication or synchronization involved. This matches the independence of recursive calls in a divide-and-conquer algorithm. Communication between teams happens only when teams are split or merged.
The use of smaller and smaller teams has performance advantages for implementations of team parallelism. First, assuming that the subdivision of teams is done on a locality-preserving basis, the smaller subteams will have greater network locality than their parent team. For most interconnection network topologies, more bisection bandwidth is available in smaller subsections of the network than is available across the network as a whole, and latency may also be lower due to fewer hops between processors. For example, achievable point-to-point bandwidth on the IBM SP2 falls from 34 MB/s on 8 processors, to 24 MB/s on 32 processors, and to 22 MB/s on 64 processors (See William Gropp, Tuning MPI program for peak performance. http://www.mcs.anl.gov/mpi/tutorials/perf/, 1996). Also, collective communication constructs in a message-passing layer typically also have a dependency on the number of processors involved. For example, barriers, reductions and scans are typically implemented as a virtual tree of processors, resulting in a latency of O(log P), while the latency of all-to-all communication constructs has a term proportional to P, corresponding to the point-to-point messages on which the construct is built.
In addition, the fact that the teams run asynchronously with respect to one another can reduce peak inter-processor bandwidth requirements. If the teams were to operate synchronously with each other, as well as within themselves, then data-parallel operations would execute in lockstep across all processors. For operations involving communication, this would result in total bandwidth requirements proportional to the total number of processors. However, if the teams run asynchronously with respect to each other, and are operating on an unbalanced algorithm, then it is less likely that all of the teams will be executing a data-parallel operation involving communication at the same instant in time. This is particularly important on machines where the network is a single shared resource, such as a bus on a shared-memory machine.
Since there is no communication or synchronization between teams, all data that is required for a particular function call of a divide-and-conquer algorithm must be transferred to the appropriate subteam before execution can begin. Thus, the division of processors among subteams is also accompanied by a redistribution of data among processors. This is a specialization of the general team-parallel model to the case of message-passing machines, since on a shared-memory machine no redistribution would be necessary.
The choice of how to balance workload across processors (in this case, choosing the subteams such that the time for subsequent data-parallel operations within each team is minimized) while minimizing interprocessor communication (in this case, choosing the subteams such that the minimum time is spent redistributing data) has been proven to be NP-complete. Therefore, most realistic systems use heuristics. For divide-and-conquer algorithms, simply maximizing the network locality of processors within subteams is a reasonable choice, even at the cost of increased time to redistribute the subteams. The intuitive reason is that the locality of a team will in turn affect the locality of all future subteams that it creates, and this network locality will affect both the time for subsequent data-parallel operations and the time for redistributing data between future subteams.
4.1.2 Collection-oriented Data Type
Within a team, computation is performed in a data-parallel fashion, thereby supporting any parallelism present in the test, divide and merge phases of a divide-and-conquer algorithm. Conceptually, this can be thought of as strictly synchronous with all processors executing in lockstep, although in practice this is not required of the implementation. An implementation of the team-parallel programming model must therefore supply a collection-oriented distributed data type, and a set of data-parallel operations operating on this data type that are capable of expressing the most common forms of divide and merge operations.
The Machiavelli implementation of the team parallel model use vectors (one-dimensional arrays) as the primary distributed data type, since these have well-established semantics and are appropriate for many divide-and-conquer algorithms. However, other collection-oriented data types may be used as well.
The implementation also assumes the existence of at least the following collective communication operations, which operate within the context of a team of p processors:
Barrier No processor can continue past a barrier until all processors have arrived at it.
Broadcast One processor broadcasts m elements of data to all other processors.
Reduce Given an associative binary operator e and m elements of data on each processor, return to all processors m results. The i.sup.th result is computed by combining the i.sup.th element on each processor using .sym..
Scan Given an associative binary operator x and m elements of data on each processor, return to all processors m results. The i.sup.th result on processor j is computed by combining the i.sup.th element from the first j-l processors using x. This is also known as the "parallel prefix" operation.
Gather Given m elements of data on each processor, return to all processors the total of m.times.p elements of data.
All-to-all communication Given m.times.p elements of data on each processor, exchange m elements with every other processor.
Personalized all-to-all communication Exchange an arbitrary number of elements with every other processor.
All of these can be constructed from simple point-to-point sends and receives on a message-passing machine. However, their abstraction as high level operations exposes many more opportunities for architecture-specific optimizations. Recent communication libraries, such as MPI and BSPLib, have selected a similar set of primitives, strengthening the claim that these are a necessary and sufficient set of collective communication operations.
4.1.3 Efficient Serial Code
In the team parallel model, when a team of processors running data-parallel code has recursed down to the point at which the team only contains a single processor, it switches to a serial implementation of the program. At this point, all of the parallel constructs reduce to purely local operations. Similarly, the parallel team-splitting operation is replaced by a conventional sequential construct with two (or more) recursive calls. When the recursion has finished, the processors return to the parallel code on their way back up the call tree. Using specialized sequential code is faster than simply running the standard team-parallel code on one processor: even if all parallel calls reduce to null operations (for example, a processor sending a message to itself on a distributed-memory system), the overhead of the function calls can be avoided by removing them completely. Two versions of each function are therefore required: one to run on a team of processors, built on top of parallel communication functions, and a second specialized to run on a single processor, using purely local loops.
This might seem like a minor performance gain, but in the team parallel model, most of the computational time is expected to be spent in sequential code. For a divide-and-conquer problem of size n on P processors, the expected height of the algorithmic tree is log n, while the expected height of our team recursion tree is log P. Since n>>P, the height of the algorithm tree is much greater than that of the recursion tree. Thus, the bulk of the algorithm's time will be spent executing serial code on single processors.
The team parallel model also allows the serial version of the divide-and-conquer algorithm to be replaced by user-supplied serial code. This code may implement the same algorithm using techniques that have the same complexity but lower constants. For example, serial quicksort could be implemented using the standard technique of moving a pointer in from each end of the data, and exchanging items as necessary. Alternatively, the serial version may use a completely different sorting algorithm that has a lower complexity (for example, radix sort).
Allowing the programmer to supply specialized serial code in this way implies that the collection-oriented data types must be accessible from serial code with little performance penalty, in order to allow arguments and results to be passed between the serial and parallel code. For the distributed vector type chosen for Machiavelli, this is particularly easy, since on one processor the block distribution of a vector is represented by a simple C array.
4.1.4 Load Balancing
For unbalanced problems, the relative sizes of the subteams is chosen to approximate the relative work involved in solving the subproblems. This is a simple form of passive load balancing. However, due to the difference in grain size between the problem size and the machine size, it is at best approximate, and a more aggressive form of load balancing must be provided to deal with imbalances that the team approach cannot handle. The particular form of load balancing is left to the implementation, but should be transparent to the programmer.
As a demonstration of why an additional form of load balancing is necessary, consider the following pathological case, where we can show that one processor is left with essentially all of the work.
For simplicity, assume that work complexity of the algorithm being implemented is linear, so that the processor teams are being divided according to the sizes of the subproblems. Consider a divide-and-conquer algorithm on P processors and a problem of size n. The worst case of load balancing occurs when the divide stage results in a division of n-1 elements (or units of work), and 1 element. Assuming that team sizes are rounded up, these two subproblems will be assigned to subteams consisting of P-1 processors and 1 processor, respectively. Now assume that the same thing happens for the subproblem of size n-1 being processed on P-1 processors; it gets divided into subproblems of size n-2 and 1. Taken to its pathological conclusion, in the worst case we have P-1 processors each being assigned 1 unit of work and 1 processor being assigned n+1-P units of work (we also have a very unbalanced recursion tree, of depth P instead of the expected O(log P)). For n>>P, this results in one processor being left with essentially all of the work.
Of course, if we assume n>>P then an efficient implementation should not hand one unit of work to one processor, since the costs of transferring the work and receiving the result would outweigh the cost of just doing the work locally. There is a practical minimum problem size below which it is not worth subdividing a team and transferring a very small problem to a subteam of one processor. Instead, below this grain size it is faster to process the two recursive calls (one very large, and one very small) serially on one team of processors. However, this optimization only reduces the amount of work that a single processor can end up with by a linear factor. For example, if the smallest grain size happens to be half of the work that a processor would expect to do "on average" (that is, n/2P), then by applying the same logic as in the previous paragraph, we can see that P-1 processors will each have n/2P data, and one processor will have the remaining n-(P-1) (n/2P) data, or approximately half of the data.
As noted above, most of the algorithm is spent in serial code, and hence load-balancing efforts should be concentrated there. The Machiavelli implementation uses a function shipping approach to load balancing. This takes advantage of the independence of the recursive calls in a divide-and-conquer algorithm, which allows us to execute one or more of the calls on a different processor, in a manner similar to that of a specialized remote procedure call. As an example, an overloaded processor running serial code in a binary divide-and-conquer algorithm can ship the arguments for one recursive branch of the algorithm to an idle processor, recurse on the second branch, and then wait for the helping processor to return the results of the first branch. If there is more than one idle processor, they can be assigned future recursive calls, either by the original processor or by one of the helping processors. Thus, all of the processors could ultimately be brought into play to load-balance a single function call. The details of determining when a processor is overloaded, and finding an idle processor, depend on the particular implementation.
4.1 Summary
The above sections have defined a team parallel programming model, which is designed to allow the efficient parallel expression of a wider range of divide-and-conquer algorithms than has previously been possible.
Specifically, team parallelism uses teams of processors to match the run-time behavior of divide-and-conquer algorithms, and can fully exploit the data parallelism within a team and the control parallelism between them. The teams are matched to the subproblem sizes and run asynchronously with respect to each other. Most computation is done using serial code on individual processors, which eliminates parallel overheads. Active load-balancing is used to cope with any remaining irregular nature of the problem; a function-shipping approach is sufficient because of the independent nature of the recursive calls to a divide-and-conquer algorithm.
4.2 The Team Parallel System
This section describes the Machiavelli system, an implementation of the team parallel model for distributed-memory parallel machines.
Machiavelli uses vectors as its basic collection-oriented data structure, and employs MPI as its parallelization mechanism. To the user, Machiavelli appears as three main extensions to the C language: the vector data structure and associated functions; a data-parallel construct that allows direct creation of vectors; and a control-parallel construct that expresses the recursion in a divide-and-conquer algorithm. The data-parallel and control-parallel constructs are translated into standard C and calls to the MPI library by a preprocessor.
The remainder of this section is arranged as follows. The next section provides an overview of the Machiavelli system, and the extensions to C that it implements. The following three sections then describe particular extensions and how they are implemented, namely vectors, predefined vector functions, and the data-parallel construct. After that come three sections on implementation details that are generally hidden from the programmer: teams, divide-and-conquer recursion, and load balancing.
4.2.1 Overview of Machiavelli
Perhaps the simplest way to give an overview of Machiavelli is with an example. The following pseudocode shows quicksort written in Machiavelli; compare this to the description of quicksort on page 3.
1. The Machiavelli function length (vector) returns the length of a vector. Here it is used both to test for the base case, and to select the middle element of the vector, for use as the pivot.
2. The Machiavelli function get (vector, index) extracts an element of a vector. Here it is used to extract the pivot element.
3. The apply-to-each operator {expr: elt in vec.vertline.cond} is used for data-parallel expressions. It is read as "in parallel, for each element elt in the vector vec such that the condition cond holds, return the expression expr". Here the apply-to-each operator is being used to select out the elements less than, equal to, and greater than the pivot, to form three new vectors.
4. The Machiavelli function, limited by C's type system, is specialized for a particular type (doubles in this case).
5. In Machiavelli, a vector is specified by prepending the name of a type with vec.sub.--.
6. In Machiavelli, vectors are explicitly freed as reflected in the free function call.
7. Rather than allowing the application of parallel functions inside an apply-to-all, Machiavelli uses an explicit split syntax to express control parallelism. This is specialized for the recursion in a divide-and-conquer function.
FIG. 4 is a block diagram illustrating the Machiavelli system, an implementation of the team parallel model. To produce efficient code from this syntax, the Machiavelli system uses three main components. First, a preprocessor 402 translates the syntactic extensions into pure C and MPI, and produces a parallel and serial version of each user-defined function. Second, a collection of predefined operations is specialized by the preprocessor for the types supplied by the user at compile time (for example, doubles in the above example). Third, a run-time system handles team operations and load balancing. A standard C compiler 404 compiles the code generated by the preprocessor 402, and a standard linker 406 links the compiled code against an MPI library 408 and the Machiavelli run-time library 410.
In a distributed memory parallel machine, the binary version of the program is loaded into and executed from each node of the machine. Specifically, each processor in the machine executes a version of the program from its associated memory. In a shared memory machine, each processor executes the binary version from shared memory.
4.2.2 Vectors
Machiavelli is built around the vector as its basic data structure. A vector is a dynamically-created ordered collection of values, similar to an array in sequential C, but is distributed across a team of processors. After being declared as a C variable, a vector is created when it is first assigned to. It is then only valid within the creating team. Vectors have reference semantics: thus, assigning one vector to another will result in the second vector sharing the same values as the first. To copy the values an explicit data-parallel operation must be used (see the section on data parallel operations below). A vector is also strongly typed: it cannot be "cast" to a vector of another type. Again, an explicit data-parallel operation must be used to copy and cast the values into a new vector.
Vectors can be created for any of the basic types, and for any types that the user defines. Accessing the fields of vectors of user-defined types is done using the standard C "." operator. For example, given a user-defined type containing floating point fields x and y, and a vector points of such types, a vector of the product of the fields could be computed using:
A vector is represented on every processor within its team by a structure containing its length, the number of elements that are currently on this processor, a pointer to a block of memory containing those elements, and a flag indicating whether the vector is unbalanced or not.
Machiavelli normally uses a simple block distribution for vectors. This corresponds to the case of a vector balanced across all the processors within a team; the unbalanced case is discussed in Section 4.3.2 entitled Unbalanced Vectors below. Thus, for a vector of size n on P processors, the first processor in the team has the first n/P elements, the second processor has the next n/P elements, and so on. When the vector size n is not an exact multiple of the number of processors P, the last processor will have fewer than n/P elements on it. In the extreme case of n=1 then P-1 processors will have no elements.
Given this distribution, it is easy to construct efficient functions of the vector length and team size that compute the maximum number of elements per processor (used to allocate space), the number of elements on a given processor, and the processor and offset on that processor that a specific index maps to. This last function is critical for performing irregular data-transfer operations, such as sends or fetches of elements.
To allow vectors of user-defined types to be manipulated using MPI operations (for example, to extract an element from such a vector) Machavelli uses MPI's derived datatype functionality. This encodes all the relevant information about a C datatype in a single type descriptor, which can then be used by MPI communication functions to manipulate variables and buffers of the matching type. For every new type used in a vector, the preprocessor therefore generates initialization code to define a matching type descriptor. For example, the pseudocode listing below shows a point structure defined by the user, and the function generated by the preprocessor to create the appropriate type descriptor for MPI.
To operate on the vector types, Machiavelli supplies a range of basic data-parallel vector functions. In choosing which functions to support a trade-off must be made between simplicity (providing a small set of functions that can be implemented easily and efficiently) and generality (providing a larger set of primitives that abstract out more high-level operations).
The vector functions can be divided into four basic types: reductions, scans, vector reordering, and vector manipulation. Each of these function types are described further below.
4.2.3 Reductions
A reduction operation on a vector returns the (scalar) result of combining all elements in the vector using a binary associative operator .sym.. Machiavelli supports the reduction operations shown in Table1 below. The operations reduce.sub.-- min.sub.-- index and reduce.sub.-- max.sub.-- index extend the basic definition of reduction, in that they return the (integer) index of the minimum or maximum element, rather than the element itself.
The implementation of reduction operations is very straightforward. Every processor performs a loop over its own section of the vector, accumulating results into a variable. They then combine the accumulated local results in an MPI.sub.-- Allreduce operation, which returns a global result to all of the processors. The preprocessor generates reduction functions specialized for a particular type and operation as necessary. As an example, the following pseudocode listing shows the specialization forreduce.sub.-- min.sub.-- double, which returns the minimum element of a vector of doubles. The use of the team structure passed in the first argument will be explained in more detail in Section 4.4.
4.2.4 Scans
A scan, or parallel prefix, operation can be thought of as a generalized reduction. Take a vector v of length n, containing elements v.sub.0, v.sub.1, v.sub.2, . . . , and an associative binary operator x with an identity value of id. A scan of v returns a vector of the same length n, where the element v.sub.i has the value id xv.sub.0 xv.sub.1 x . . . xv.sub.i-1. Note that this is the "exclusive" scan operation; the inclusive scan operation does not use the identity value, and instead sets the value of v.sub.1 to v.sub.0 xv.sub.1 x . . . xv.sub.i-1. Machiavelli supplies a slightly smaller range of scans than of reductions, as shown in Table 2, because there is no equivalent of the maximum and minimum index operations.
Scans are only slightly more difficult to implement than reductions. Again, every processor performs a loop over its own section of the vector, accumulating a local result. The processors then combine their local results using an MPI.sub.-- Scan operation, which returns an intermediate scan value to each processor. A second local loop is then performed, combining this scan value with the original source values in the vector to create the result vector. There is an additional complication in that MPI provides an inclusive scan rather than an exclusive one. For operations with a computable inverse (for example, sum) the exclusive scan can be computed by applying the inverse operation to the inclusive result and the local intermediate result. For operations without a computable inverse (for example, min), the inclusive scan is computed and the results are then shifted one processor to the "right" using MPI.sub.-- Sendrecv. As an example, the following pseudocode shows scan.sub.-- sum.sub.-- int, which returns the exclusive prefix sum of a vector of integers.
4.2.5 Vector Reordering
There are two basic vector reordering functions, send and fetch, which transfer source elements to a destination vector according to an index vector. In addition, the function pack, which is used to redistribute the data in an unbalanced vector can be seen as a specialized form of the send function. These are the most complicated Machiavelli functions, but they can be implemented using only one call to MPI.sub.-- Alltoall to setup the communication, and one call to MPI.sub.-- Alltoallv to actually perform the data transfer. send (vec.sub.-- source, vec.sub.-- indices, vec.sub.-- dest)
send is an indexed vector write operation. It sends the values from the source vector vec.sub.-- source to the positions specified by the index vector vec.sub.-- indices in the destination vector vec.sub.-- dest, so that vec.sub.-- dest[vec.sub.-- indices[i]]=vec.sub.-- source[i].
This is implemented using the following operations on each processor. For simplicity, assume that there are P processors, that the vectors are of length n, and that there are exactly n/P elements on each processor.
1. Create two arrays, num.sub.-- to.sub.-- send[P] and num.sub.-- to.sub.-- recv[]. These will be used to store the number of elements to be sent to and received from every other processor, respectively.
2. Iterate over this processor's n/P local elements of vec.sub.-- indices. For every index element i, compute the processor q that it maps to, and increment num.sub.-- to.sub.-- send[q]. Each processor now knows how many elements it will send to every other processor.
3. Exchange num.sub.-- to.sub.-- send[P] with every other processor using MPI.sub.-- Alltoall(). The result of this is num.sub.-- to.sub.-- recv[P], the number of elements to receive from every other processor.
4. Allocate a data array data.sub.-- to.sub.-- send[n/P] and an index array indices.sub.-- to.sub.-- send[n/P]. These will be used to buffer data and indices before sending to other processors. Similarly, allocate a data array data.sub.-- to.sub.-- recv[n/P] and an index array indices.sub.-- to.sub.-- recv[n/P]. Conceptually, the data and index arrays are allocated and indexed separately, although in practice they can be allocated as an array of structures to improve locality.
5. Perform a plus-scan over num.sub.-- to.sub.-- recv[] and num.sub.-- to.sub.-- send[], resulting in arrays of offsets send.sub.-- ptr[P] and recv.sub.-- ptr[P]. These offsets will act as pointers into the data.sub.-- and indices.sub.-- arrays.
6. Iterate over this processor's n/P local elements of vec.sub.-- indices[]. For each element vec.sub.-- indices[i], compute the processor q and offset o that it maps to. Fetch and increment the current pointer, ptr=send.sub.-- ptr[q]++. Copy vec.sub.-- source[i] to data.sub.-- to.sub.-- send[ptr], and copy o to indices.sub.-- to.sub.-- send[ptr].
7. Call MPI.sub.-- Alltoallv(). Send data from data.sub.-- to.sub.-- send[] according to the counts num.sub.-- to.sub.-- send[], and receive into data.sub.-- to.sub.-- recv[] according to the counts num.sub.-- to.sub.-- recv[]. Do the same for indices.sub.-- to.sub.-- send[].
8. Iterate over data.sub.-- to.sub.-- recv[] and indices.sub.-- to.sub.-- recv[], performing the vec.sub.-- dest[indices.sub.-- to.sub.-- recv[i]]=data.sub.-- to.sub.-- recv[i].
Note that steps 1-3 and 5 are independent of the particular data type being sent. They are therefore abstracted out into library functions. The remaining steps are type-dependent, and are generated as a function by the preprocessor for every type that is the subject of a send.
fetch (vec.sub.-- source, vec.sub.-- indices, vec.sub.-- dest)
fetch is an indexed vector read operation. It fetches data values from the source vec.sub.-- source (from the positions specified by the index vector vec.sub.-- indices) and stores them in the destination vector vec.sub.-- dest, so that vec.sub.-- dest[i] =vec.sub.-- source[vec.sub.-- indices[i]].
Obviously, this could be implemented using two send operations--one to transfer the indices of the requested data items to the processors that hold the data, and a second to transfer the data back to the requesting processors. However, by combining them into a single function some redundant actions can be removed, since we know ahead of time how many items to send and receive in the second transfer. Again, for simplicity assume that there are P processors, that all the vectors are of length n, and that there are exactly n/P elements on each processor.
1. Create two arrays, num.sub.-- to.sub.-- request[P] and num.sub.-- to.sub.-- recv[P]. These will be used to store the number of requests to be sent to every other processor, and the number of requests to be received by every other processor, respectively.
2. Iterate over this processor's n/P local elements of vec.sub.-- indices. For every index element i, compute the processor q that it maps to, and increment num.sub.-- to.sub.-- request[q]. Each processor now knows how many elements it will request from every other processor.
3. Exchange num to request[] with every other processor using MPI.sub.-- Alltoall (). The result of this is num.sub.-- to.sub.-- recv[], the number of requests to receive from every other processor (which is the same as the number of elements to send).
4. Allocate an index array indices.sub.-- to.sub.-- request[n/P]. This will be used to buffer indices to request before sending to other processors. Similarly, allocate an index array indices.sub.-- to.sub.-- recv[n/P].
5. Perform a plus-scan over num.sub.-- to.sub.-- request[] and num.sub.-- to.sub.-- recv[], resulting in arrays of offsets request.sub.-- ptr[P] and recv.sub.-- ptr[P]. These offsets will act as pointers into the indices.sub.-- to.sub.-- request[] and indices.sub.-- to.sub.-- recv[] arrays.
6. Allocate an index array requested.sub.-- index[n/P]. This will store the index in a received data buffer that we will eventually fetch the data from.
7. Iterate over this processor's n/P local elements of vec.sub.-- indices[]. For each element vec.sub.-- indices[i], compute the processor q and offset o that it maps to.
Fetch and increment the current pointer, ptr=request.sub.-- ptr[q]++. Copy o to indices.sub.-- to.sub.-- request[ptr].
8. Call MPI.sub.-- Alltoallv(). Send data from request.sub.-- to.sub.-- send[] according to the element counts in num.sub.-- to.sub.-- request[], and receive into request.sub.-- to.sub.-- recv[] according to the element counts in num.sub.-- to.sub.-- recv[].
9. Allocate data arrays data.sub.-- to.sub.-- send[n/P] and data.sub.-- to.sub.-- recv[n/P].
10. Iterate over request.sub.-- to.sub.-- recv[], extracting each offset in turn, fetching the requested element, and storing it in the data buffer, data.sub.-- to.sub.-- send[i]=vec.sub.-- dest[request.sub.-- to.sub.-- recv[i]].
11. Call MPI.sub.-- Alltoallv (). Send data from data.sub.-- to.sub.-- send[ ] according to the element counts in num.sub.-- to.sub.-- recv[], and receive into data.sub.-- to.sub.-- recv[] according to the counts in num.sub.-- to.sub.-- request[ ].
12. Iterate over data.sub.-- to.sub.-- recv[] and requested.sub.-- index[], performing the operation vec.sub.-- dest[i]=data.sub.-- to.sub.-- recv[requested.sub.-- index[i]].
Again, steps 1-8 are independent of the particular data type being requested, and abstracted out into library functions, while the remaining steps are generated as a function by the preprocessor for every type that is the subject of a fetch.
pack (vec.sub.-- source)
pack redistributes the data in an unbalanced vector so that it has the block distribution property described in Section 4.2.2. An unbalanced vector (that is, one that does not obey this property, but instead has an arbitrary amount of data on each processor) can be formed either by an apply-to-each operator with a conditional (see Section 4.3) or by appending the results of recursive function calls (see Section 4.5). The pack function is normally called as part of another Machiavelli operation.
pack is simpler than send since we will send contiguous blocks of elements between processors, rather than sequences of elements with the appropriate offsets to store them in.
1. Exchange vec.sub.-- source.nelt.sub.-- here with every other processor using MPI.sub.-- Alltoall (). The result of this is num.sub.-- on.sub.-- each[P].
2. Perform a plus-scan across num.sub.-- on.sub.-- each[P] into first.sub.-- on.sub.-- each[P]. The final result of the plus scan is the total length n of the vector.
3. From n, compute the number of elements per processor in the final block distribution, final.sub.-- on.sub.-- each[P], and allocate a receiving array data.sub.-- to.sub.-- recv[n/P].
4. Allocate two arrays, num.sub.-- to.sub.-- recv[P] and num.sub.-- to.sub.-- send[P].
5. Iterate over final.sub.-- on.sub.-- each[P], computing which processor(s) will contribute data for each destination processor in turn. If this processor will be receiving, update num.sub.-- to.sub.-- recv[ ]. If this processor will be sending, update num.sub.-- to.sub.-- send[].
6. Call MPI.sub.-- Alltoallv(), sending data from vec.sub.-- source.data[] according to the element counts in num.sub.-- to.sub.-- send[], and receiving into data.sub.-- to.sub.-- recv[] according to the element counts in num.sub.-- to.sub.-- recv.
7. Free the old data storage in vec.sub.-- source.data[ ] and replace it with data.sub.-- to.sub.-- recv[].
4.2.6 Vector manipulation
Machiavelli also supplies seven functions that manipulate vectors in various ways. Most of these have simple implementations. All but length and free are specialized for the particular type of vector being manipulated.
free (vector)
Frees the memory associated with vector vector.
new.sub.-- vec (n)
Returns a vector of length n. This is translated in the parallel and serial code to calls to the underlying Machiavelli functions alloc.sub.-- vec.sub.-- type and alloc.sub.-- vec.sub.-- type.sub.-- serial, respectively.
length (vector)
Returns the length of vector vector. This simply returns the length field of the vector structure, that is, vector.length.
get (vector, index)
Returns the value of the element of vector vector at index index. Using the length of vector, every processor computes the processor and offset that index maps to. The processors then perform a collective MPI.sub.-- Broadcast operation, where the processor that holds the value contributes the result. As an example, the following code listing shows the parallel implementation of get for a user-defined point type.
set (vector, index, value)
Sets the element at index index of vector vector to the value value. Again, every processor computes the processor and offset that index maps to. The processor that holds the element then sets its value. As an example, the following code listing shows the parallel implementation of set for a vector of characters.
index (length, start, increment)
Returns a vector of length length, containing the numbers start, start+increment, start+2* increment, . . . This is implemented as a purely local loop on each processor, and is specialized for each numeric type. As an example, the following code listing shows the implementation of index for integer vectors.
distribute (length, value)
Returns a vector of length length, containing the value value in each element. This is defined for any user-defined type, as well as for the basic C types. Again, it is implemented with a purely local loop on each processor. As an example, the following code listing shows the parallel implementation of distribute for double-precision floating-point vectors.
vector (scalar)
Returns a single-element vector containing the variable scalar. This is equivalent to dist (1, scalar), and is provided merely as a convenient shorthand.
replicate (vector, n)
Given a vector vector of length m, and an integer n, returns a vector of length m.times.n, containing n copies of vector. This is converted into a doubly-nested loop in serial code as shown in the following code listing, and into a sequence of n operations in parallel code.
append (vector, vector [,vector])
Appends two or more vectors together, returning the result of their concatenation as a new vector. This is implemented as successive calls to a variant of the pack function. Here it is used to redistribute elements of a vector that is spread equally among the processors to a smaller subset of processors, each representing a particular section of a longer vector. The Machiavelli preprocessor converts an n-argument append to n successive calls to pack, each to a different portion of the result vector. As an example, the following code listing shows the implementation of append for three integer vectors.
The next four functions (odd, even, interleave, and transpose) can all be constructed from send combined with other primitives. However, providing direct functions allows for a more efficient implementation by removing the need for the use of generalized indexing. That is, each of the four functions preserves some property in its result that allows us to precompute the addresses to send blocks of elements to, rather than being forced to compute the address for every element, as in send.
even (vector, n)
odd (vector, n)
Given a vector, and an integer n, even returns the vector composed of the even-numbered blocks of elements of length n from vector. Thus, even (foo, 3) returns the elements 0, 1, 2, 6, 7, 8, 12, 13, 14, . . . of vector foo. odd does the same, but for the odd-numbered blocks of elements. The length of vector is assumed to be an exact multiple of twice the block-size n. As an example, the next code listing shows the serial implementation of even for a user-defined pair type. The parallel implementations of odd and even simply discard the even and odd elements respectively, returning an unbalanced vector. Note that the use of generalized odd and even primitives (rather than simply basic single-element odd and even) allows them to be used for other purposes. For example, even (bar, length (bar)/2) returns the first half of vector bar.
interleave (vec1, vec2, n)
Given two vectors vec1 and vec2, and an integer n, it returns the vector composed of the first n elements from vec1, followed by the first n elements from vec2, followed by the second n elements from vec1, and so on. As such, it does the opposite of even and odd. Again, the use of a generalized interleave primitive allows it to be used for other purposes. For example, given two m.times.n matrices A and B, interleave (A, B, n) returns the 2 m.times.n matrix whose rows consist of the appended rows of A and B. The lengths of vec1 and vec2 are assumed to be the same, and an exact multiple of the blocksize n.
transpose (vector, m, n)
Given a vector vector which represents an n.times.m matrix, returns the vector representing the transposed n.times.m matrix
4.3 Data-Parallel Operations
For general data-parallel computation, Machiavelli uses the apply-to-each operator, which has the following syntax:
expr is any expression (without side-effects) that can be the right-hand side of an assignment in C. elt is an iteration variable over a vector vec. The iteration variable is local to the body of the apply-to-each operator. There may be more than one vector, but they must have the same length.cond is any expression without side-effects that can be a conditional in C.
The effect of this construct is to iterate over the source vector(s), testing whether the condition is true for each element, and if so evaluating the expression and writing the result to the result vector.
4.3.1 Implementation
The Machiavelli preprocessor converts an apply-to-each operation without a conditional into a purely local loop on each processor, iterating over the source vectors and writing the resultant expression for each element into the destination vector. The absence of synchronization between processors explains why the expressions within an apply-to-each cannot rely on side effects; any such effects would be per-processor, rather than global across the machine. In general, this means that C's pre- and post-operations to increment and decrement variables cannot be used inside an apply-to-each. As an example, the following code listing shows a simple data-parallel operation and the resulting code.
The independence of loop expressions also enables the Machiavelli preprocessor to perform loop fusion on adjacent apply-to-each operations that are iterating across the same vectors.
4.3.2 Unbalanced vectors
If a conditional is used in an apply-to-each, then the per-processor pieces of the destination vector may not have the same length as the pieces of source vector(s). The result is that we are left with an unbalanced vector; that is, one in which the amount of data per processor is not fixed. This is marked as such using an "unbalanced" flag in its vector structure. As an example, the following code listing shows the parallel implementation of an apply-to-each with a simple conditional.
An unbalanced vector can be balanced (that is, its data can be evenly redistributed) across the processors using a pack function, as described above. The advantage of not balancing a vector is that by not calling pack we avoid two MPI collective operations, one of which transfers a small and fixed amount of information between processors (MPI.sub.-- Allgather) while the other may transfer a large and varying amount of data (MPI.sub.-- Alltoallv).
Given an unbalanced vector, we can still perform many operations on it in its unbalanced state. In particular, reductions, scans, and apply-to-each operations (including those with conditionals) that operate on a single vector are all oblivious to whether their input vector is balanced or unbalanced, since they merely loop over the number of elements present on each processor. Given the underlying assumption that local operations are much cheaper than transferring data between processors, it is likely that the time saved by avoiding data movement in this way outweighs the time lost in subsequent operations caused by not balancing the data across the processors, and hence resulting in all other processors waiting for the processor with the most data. Machiavelli only performs a pack when required, but allows the user to manually insert additional pack operations.
The remaining Machiavelli operations that operate on vectors all require their input vectors to be packed before they can proceed. The implementations of get, set, send, fetch are therefore extended with a simple conditional that tests the "unbalanced" flag of their input vector structures, and performs a pack on any that are unbalanced. These operations share the common need to quickly compute the processor and offset that a specific vector index maps to; a balanced block distribution can satisfy this need. Apply-to-each operations on multiple vectors are also extended with test-and-pack, although in this case the requirement is to assure that vectors being iterated across in the same loop share the same distribution. Two further optimizations that are possible when using unbalanced vectors are discussed in the next section.
4.4 Teams
Machiavelli uses teams to express control-parallel behavior between data-parallel sections of code, and in particular to represent the recursive branches of a divide-and-conquer algorithm. A team is a collection of processors, and acts as the context for all functions on vectors within it. A vector is distributed across the processors of its owning team, and can only be operated on by data-parallel functions within that team. Teams are divided when a divide-and-conquer algorithm makes recursive calls, and merged when the code returns from the calls. Otherwise, teams are mutually independent, and do not communicate or synchronize with each other. However, unless the programmer wants to bypass the preprocessor and gain direct access to the underlying team functions, the existence of teams is effectively hidden.
4.4.1 Implementation
A team is represented by the MPI concept of a communicator. Specifically, a communicator describes a specific collection of processors, and when passed to an MPI communication function, restricts the "visible universe" of that communication function to the processors present in the communicator.
The internal representation of a team consists of a structure containing the MPI communicator, the number of processors in the team, and the rank of this processor in the team. All processors begin in a "global" team, which is then subdivided by divide-and-conquer algorithms to form smaller teams. The preprocessor adds a pointer to the current team structure as an extra argument to every parallel function, as was seen in the implementations of Machiavelli functions above. In this way, the subdivision of teams on the way down a recursion tree, and their merging together on the way up the tree, is naturally encoded in the passing of smaller teams as arguments to recursive calls, and reverting to the parent teams when returning from a call.
4.5 Divide-and-conquer Recursion
Machiavelli uses the following syntax:
to represent the act of performing divide-and-conquer function calls. n is the result returned by invoking the function function the argument list arg.sub.n. Team parallelism is implemented by dividing the current team into one subteam per function call, sending the arguments to the subteams, recursing within the subteams, and then fetching the results from the subteams. Each of these steps will be described. Note that in the Machiavelli implementation func must be the same for every call in a given split, although this is not a requirement of the team parallel model
4.5.1 Computing team sizes
Before subdividing a team into two or more subteams, we need to know how many processors to allocate to each team. For Machiavelli's approximate load balancing of teams, the subteam sizes are chosen according to the relative amount of expected work that the subtasks are expected to require. This is computed at runtime by calling an auxiliary cost function defined for each divide-and-conquer function. The cost function takes the same arguments as the divide-and-conquer function, but returns as a result an integer representing the relative amount of work that those arguments will require. By default, the preprocessor generates a cost function that returns the size of the first vector in the argument list (that is, it assumes that the cost will be a linear function of the first vector argument). This can be overridden for a particular divide-and-conquer function divconq by defining a cost function divconc.sub.-- qcost. The following code listing shows a simple cost function for quicksort, which has an expected cost of O(n log n).
The results of a cost function have no units, since they are merely compared to each other. The actual act of subdividing a team is performed with the MPI.sub.-- Comm.sub.-- split function which, when given a flag declaring which new subteam this processor should join, creates the appropriate MPI communicator.
4.5.2 Transferring arguments and results
Having created the subteams, we must redistribute any vector arguments to the respective subteams (scalar arguments are already available on each processor, since we are programming in an SPMD style). The task is to transfer each vector to a smaller subset of processors. This can be accomplished with a specialized form of the pack function; all that is necessary is to change the computation of the number of elements per processor for the destination vector, and to supply a "processor offset" that serves as the starting point for the subset of processors. However, there are two optimizations that can be made to reduce the number of collective operations.
First, the redistribution function can accept unbalanced vectors as arguments, just as the original pack function can. This is particularly important for divide-and-conquer functions, where the arguments to recursive calls may be computed using a conditional in an apply-to-each, which results in unbalanced vectors. Without the ability to redistribute these unbalanced vectors, the number of collective communication steps would be doubled (first to balance the vector across the original team, and then to redistribute the balanced vector to a smaller subteam).
Second, the redistribution function can use a single call to MPI.sub.-- Alltoallv to redistribute a vector to each of the subteams. Consider the recursion in quicksort:
les and grt are being supplied as the argument to recursive function calls that will take place in different subteams. Since these subteams are disjoint, a given processor will send data from either les or grt to any other given processor, but never from both. We can therefore give MPI.sub.-- Alltoallv the appropriate pointer for the data to send to each of the other processors, sending from les to processors in the first subteam and from grt to processors in the second subteam. Thus, only one call to MPI.sub.-- Alltoallv is needed for each of the vector arguments to a function.
After the recursive function call, we merge teams again, and must now extract a result vector from each subteam, and redistribute it across the original (and larger) team. This can be accomplished by simply setting the "unbalanced" flag of each result vector, and relying on later operations that need the vector to be in a balanced form to redistribute it across the new, larger team. This can be seen as a reverse form of the "don't pack until we recurse" optimization that was outlined above--now, we don't expand a vector until we need to.
To illustrate these features, we will use the simplified version of quicksort shown below, which partitions the input into two vectors (containing those elements less than or equal to the pivot, and greater than the pivot) instead of three:
As an example, consider the algorithm sorting the following set of numbers from 0 to 15:
[3 8 0 4 12 10 5 9 13 7 11 15 2 14 1 6]
on four processors. FIG. 5A shows an overview of the recursion tree for this example. FIGS. 5B and 5C show the effects with and without both the "don't pack before split" optimization, and FIGS. 5C and 5E show the effects with and without the "postpone append till top level" optimization.
FIG. 5A shows an example of how this version of quicksort travels down and then back up the recursion tree. As it travels down the recursion tree (levels 502 to 506), quicksort splits the set of numbers into smaller and smaller left and right parts. As it travels up the recursion tree (levels 508 to 512), quicksort combines the results. On each level traveling down the recursion tree, the middle element is chosen as the pivot, and two new subproblems are formed containing the elements less than or equal to, and greater than, the pivot. In this example, four processors are being used, so when four subproblems have been formed the individual processors run a serial version of the algorithm on their subproblem (step 506). They then return to the parallel code, which appends the subresults as it returns up the tree.
FIGS. 5B and 5C show in more detail the layout of the vectors used on the way down the recursion tree, while FIGS. 5D and 5E show the layout of the vectors used on the way up the recursion tree.
FIG. 5B shows the layout without the use of lazy vectors and the ability of the redistribution function to handle lazy input. The boxes show the data elements held on each processor, where a horizontal line represents a given vector. The contiguous segments of numbers represent the portions of a vector stored locally for one of the four processors. The enclosing ovals show the division of processors between teams. Initially, all processors are in one team. After the partitioning step, the resulting vectors are unbalanced (522). They are then rebalanced so that for each vector, each processor (except the last one, in this case) has an equal number of elements (see oval 524).
In particular, quicksort picks "9" as the pivot element at level 522, and the two conditional apply-to-each expressions create two new vectors: leq, containing the elements less than or equal to the pivot (a total of 10 elements), and grt, containing the elements greater than the pivot (the remaining 6 elements). As can be seen, these vectors are unbalanced. The runtime system now uses two pack operations to make these vectors "balanced" as shown in oval 524 (that is, with a constant number of elements per processor, except for the last one).
Note that if the system did not support lazy vectors, this rebalancing step would be part of the partitioning operation. If the system did support lazy vectors, but the redistribution function required balanced input, the redistribution function would call the pack function.
After the vectors are rebalanced, the split step then subdivides the team into two new teams, and calls the redistribution function to redistribute the vectors to the two teams (526). The data is redistributed so that each team gets one (balanced) vector as shown in level 526.
The code then recurses in each new team, creating unbalanced vectors after the partitioning step (528). In this example, the two teams pick "5" and "13" as their new pivot elements, and create two new sets of vectors. Again, two pack operations are required to balance the vectors on their respective teams (530).
The two teams recurse again, creating four individual teams and redistributing the data so that each new team gets one of the four vectors shown at level 532. At this stage the subteams each contain a single processor, so the parallel code now calls the serial code instead of recursing again, resulting in the data on each processor being sorted (534).
FIG. 5C shows the effect of extending the redistribution function to handle lazy input. In particular, FIG. 5C shows how the interprocessor communication for rebalancing the vectors before a split as in FIG. 5B is eliminated since Machiavelli's implementation of the split step supports lazy input. The steps in FIG. 5C are initially the same as in FIG. 5B. Note that levels 540 and 542 in FIG. 5B are the same as levels 520 and 522 in FIG. 5C.
However, in contrast to FIG. 5B, the vectors are left in their unbalanced state after the partitioning step (542). The split step again subdivides the team and calls the redistribution function, which combines the balancing and redistribution of data into a single step. When the algorithm recurses, subdividing into two teams of two processors (see level 544 to 546), the unbalanced vectors are redistributed directly to their new teams, saving two intermediate pack operations. Again, when each of these two teams split into two new teams of one processor each (level 546 to 548), the two recursions now redistribute the unbalanced vectors directly to the four new teams, saving two intermediate pack operations each. This combination of rebalancing and redistributing the vectors into a single step saves an unnecessary data rearrangement step in each team on each level of recursion, and thus reduces expensive interprocessor communication.
FIG. 5C ends similarly to FIG. 5B. When each team only has one processor, the parallel code on each processor invokes the serial code, which sorts the local elements of the vector (550). When the four individual processors finish the serial code, each contains a sorted portion of the result vector as shown.
FIG. 5D shows the return from the recursion (when we are appending subresults together) in the case where lazy vectors are supported but not the optimization of postponing an append operation until the top level. Initially all the processors are in individual teams (560). First, they return to the parent teams from which the subteams were formed. Thanks to lazy vectors, no redistribution of data is necessary. Instead, the left and right vectors are created as unbalanced. Depending on the processor, the local data for these vectors will either be empty, or be the entire result of the subteam (562). Thus, no data elements are moved, either between processors or on processors (the data pointer in the representation of the left and right vectors will be either null or will point to the data of the result vector).
Next, these vectors are appended together within each team, creating a new balanced result vector (564). The general append operation requires two pack operations to redistribute the data between the processors to form the balanced vectors shown at level 564.
The processors then return from another level of recursion, to the original top-level team (566). The results of the two teams in turn become the left and right vectors of the top-level team as shown in level 566. Again, the new left and right vectors are left unbalanced, and no data movement is required until the final append step to create the final sorted result (568). In the final append step, these two unbalanced vectors are appended with a further pack to form the final result.
FIG. 5E shows the effect of postponing the append step until the top level of the recursion tree. In effect, we are completely eliminating the append step, and leaving the overall result in an unbalanced state. Again, the return up the recursion tree starts with the four processors returning from serial code (580). As before, the subteams return to the parent teams, and left and right are created in an unbalanced state, pointing to the data from result (582).
The append step now does much the same thing, creating a new result vector whose per-processor data pointer shares the data of left and right (584). However, the append step is now replaced by a local renaming operation that just chooses whether to use the elements from the left or right vector as the result vector on this processor. No interprocessor communication is needed to rearrange data, since the vector is left in an unbalanced state.
At the next level of recursion, the results again become the left and right vectors of the top-level team (586). Another renaming step creates an unbalanced result vector (588). As with all lazy vectors, this will only be balanced if subsequent operations on it require it to be in a balanced state. Note that there is no data movement whatsoever between processors in the steps shown; we are taking advantage of the fact that the split function puts the subproblem of smaller numbers on the "left" and the subproblem of larger numbers on the "right", and thus their results will be naturally ordered.
If any subsequent operations need the result vector to be balanced, a single pack operation will be required to redistribute the elements on the processors from the state shown at level 588 in FIG. 5E to the state shown in level 568 in FIG. 5D.
As illustrated in the quicksort example above, the preprocessor for the Machiavelli system treats the Append operation immediately following a Split call as a special case. This special case implementation of the Append function avoids the inter-processor communication associated with the normal Append function. In the general Append operation, the preprocessor generates one Pack operation per source vector, to reshuffle the data across the processors as shown in 4.2.6. For example, given two equal-size source vectors to be appended together, which are spread across four processors, the run-time code performs the following steps:
create a new vector twice as big (i.e. twice as many elements per processor);
pack the first vector from all four processors into the result vector in the first two processors; and
pack the second vector from all four processors into the result vector in the last two processors.
The Machiavelli implementation uses the Pack code to implement the general Append function because this code is sufficiently general to handle a variety of redistribution functions. To clarify, the Machiavelli implementation uses the Pack code to redistribute data elements among processors in the following cases:
1) to redistribute or balance elements of a vector among processors, and
2) as part of the Append function to redistribute elements from more processors to fewer processors.
It is important to note that the Pack code used in the Machiavelli system represents only one possible way of implementing functions for redistributing data among processors. It is a design choice to use Pack code in the Append function. It is also possible to implement distinct functions for re-balancing an unbalanced vector (e.g., a Rebalance or Redistribute function) and for distributing input vectors from more processors to fewer processors (e.g., a Subappend function). The names chosen for the functions are not critical to the invention.
The Machiavelli preprocessor reduces inter-processor communication associated with returns from split calls by treating an Append after a Split as a special case. If the preprocessor determines that the results of a Split call are immediately used as input to an Append function, then it can replace the normal Append function with the "renaming" code described in connection with FIG. 5E. This special case code swaps pointers around so that the vector resulting from the append step on each processor points to the data that was just returned from the Split call. This approach eliminates inter-processor communication that would otherwise result using normal Append operations after Split functions while traveling back up the recursion tree.
Note that the quicksort example makes the assumption that processors are assigned to subteams in numerically increasing order, e.g., processors 0-1 were used for the left part of the split in the example above, and processors 2-3 were used for the right part of the split. Obviously if the order were swapped so that processors 2-3 received the left and processors 0-1 received the right, then the "renaming" append code would give a back-to-front answer. The Machiavelli team-splitting code makes sure that this assumption is always true.
The Machiavelli preprocessor implements the "renaming" optimization by looking one line ahead of a split call to determine if the next line includes an Append operation to join the results of the split call into a result vector. An alternative implementation of the preprocessor could use data-flow analysis to determine if this property holds in the more general case. In particular, standard data-flow analysis used in compilers could be used to find candidates for the renaming optimization in cases where there are additional statements between a Split call and an Append on the results of the Split call, or where vectors derived from the result of a Split call are being appended. As an example of the latter case, consider a program that includes code to add 1 to every element in the left vector, and 2 to every element in the right, creating two new unbalanced vectors, and then an Append function to append the two new vectors. Data-flow analysis could be used to determine that the renaming optimization could be applied to the two new vectors.
An alternative implementation would use another flag in the structure representing a vector. If set, this flag indicates that the vector is not only unbalanced, but is the result of a split. It also records which subteam the vector came from. The flag is cleared after any operation that balances a vector. The flag is set for all vectors returned from a subteam, e.g., returned from a divide-and-conquer function. In contrast to the implementation where the preprocessor always generates code to call the append function when it sees "append()" in the input code, this approach makes the append function adaptive at runtime based on the state of the input vector. At runtime, the Append function determines whether all of its input vectors are the results of a split, and whether they are in the correct order (e.g. the first vector came from subteam 0, the second from subteam 1, etc). If so, it performs the renaming optimization. Otherwise, it redistributes the data as before. This approach enables a preprocessor to get the same result as achieved through data flow analysis without overly complicating the preprocessor code. In this particular application, it avoids unnecessary inter-processor communication in cases where the Append function operates on the results of the Split call or data derived from these results without using data-flow analysis before run-time.
4.5.3 Serial code
The Machiavelli preprocessor generates two versions of each user-defined and inbuilt function. The first version, as described in previous sections, uses MPI in parallel constructs, and team-parallelism in recursive calls. The second version is specialized for single processors with purely local data (that is, a team size of one). In this version, apply-to-each constructs reduce to simple loops, as do the predefined vector operations, and the team-based recursion is replaced with simple recursive calls. As was previously discussed, this results in much more efficient code. For example, the following code listing shows the serial implementation of a fetch function, which can be compared to the 12-step description of the parallel equivalent described above.
Where more efficient serial algorithms are available, they can be used in place of the serial versions of parallel algorithms that Machiavelli compiles. Specifically, the user can force the default serial version of a parallel function to be overridden by defining a function whose name matches that of the parallel function but has the added suffix ".sub.-- serial". For example, the following code listing shows a more efficient serial implementation of quicksort supplied by the user:
MPI functions can also be used within Machiavelli code, in cases where the primitives provided are too restrictive. As explained above, all user-defined types have equivalent MPI types constructed by the Machiavelli preprocessor, and these MPI types can be used to transfer instances of those types in MPI calls. The fields of a vector and of the current team structure (always available as the pointer tm in parallel code) are also accessible, as shown in the following code listing.
Putting all these steps together, the code generated by the preprocessor for the recursion in the parallel version of quicksort is shown below.
4.6 Load Balancing
As mentioned above, the team-parallel model restricts load balancing to individual processors because that is where most of the time is spent when n>>P. We are thus trying to cope with the situation where some processors finish first and are idle. Our goal is to ship function invocations from working processors to idle processors, and then return the result to the originating processor.
The implementation of Machiavelli's load-balancing system is restricted by the capabilities of MPI, and by our desire to include as few parallel constructs as possible in the execution of sequential code on single processors. In the absence of threads or the ability to interrupt another processor, an idle processor is unable steal work from another processor, because the "victim" would have to be involved in receiving the steal message and sending the data. Therefore, the working processors must request help from idle processors. The request cannot use a broadcast operation because MPI broadcasts are a collective operation that all processors must be involved in, and therefore other working processors would block the progress of a request for help. Similarly, the working processor cannot use a randomized algorithm to ask one of its neighbors for help (in the hope that it picks an idle processor), because the neighbor might also be working, which would cause the request to block. Thus, to determine whether another processor is idle (and hence can respond quickly to our request) the run-time code dedicates one processor to keeping track of each processor's status. This processor acts as the manager of active load balancing.
The load-balancing process employs a three-way hand-shaking protocol among the processor requesting help, the manager, and the idle processor. FIGS. 6, 7, and 8 are flow diagrams illustrating the operation of a processor seeking help, the manager, and the idle processor, respectively. The preprocessor inserts a load-balance test function into the sequential version of every divide-and-conquer function. This test determines whether to ask the manager for help with one or more of the recursive function calls. Decision block 602 in FIG. 6 represents the test executed to determine whether to make a request for help. If this test is satisfied, a short message is sent to the manager, containing the total size of the arguments to the function call (604). If the test is not satisfied, the requesting processor continues with sequential execution of the function (606).
For binary divide-and-conquer algorithms, the requesting processor then blocks waiting for a reply (606). There is no point in postponing the test for a reply until after the requesting processor has finished the second recursive call, since at that point it would be faster to process the first recursive call locally than to send the arguments to a remote processor, wait for it to finish, and then receive the results. For divide-and-conquer algorithms with a branching factor greater than two, the requesting processor can proceed with another recursive call while waiting for the reply (shown as an optional step 608).
FIG. 7 illustrates how the manager processor manages dynamic load balancing. The manager sits in a message-driven loop, maintaining a list of idle processors and waiting for messages to arrive. The other processors send the manager messages when requesting help (see message 706 in FIG. 7) and when they become idle (see message 708 in FIG. 7).
When the manager receives a request for help (706), it checks the list for idle processors (710). If the list contains no idle processors, the manager responds to the requesting processor with a "no help available" message (714). Otherwise, it removes an idle processor from its list (716), and sends the idle processor a message instructing it to help the requesting processor with a function call of the reported size (718).
FIG. 8 shows the operation of an idle processor. Like the manager, the idle processor sits in a message-driven loop, which it enters after finishing the sequential phase of its divide-and-conquer program. In its message loop, the idle processor blocks waiting for a message. On receiving the message from the manager (806), it sets up appropriate receive buffers for the argument (808), and then sends the requesting processor an acknowledgement message signaling its readiness (810). This three-way handshaking protocol guarantees that the receiver has allocated buffers before the arguments are sent, which can result in faster performance from some MPI implementations by avoiding system buffering. It also avoids possible race conditions in the communication between the three processors.
Returning again to FIG. 6, the requesting processor proceeds depending on how the manager responds to its request for help. If the requesting processor receives a "no help available" message from the master (610), it continues with sequential execution (612). Otherwise, it receives an acknowledgement message from an idle processor (614), at which point it sends the arguments of a first recursive function call (616), continues with the second recursive function call (618), and then waits to receive the result of the first function call (620). FIG. 8 shows how the idle processor processes the recursive call it received from the requesting processor. The idle processor receives the function arguments via a reply message (812) to its acknowledge message from the requesting processor. In response to this message, the idle processor invokes the function on the function's arguments (814), and sends the results back to the requesting processor (816). It then notifies the manager that it is once again idle (818), and waits for another message by returning to its message loop.
FIG. 7 shows how the manager responds to messages from another processor indicating that it is idle. When the manager receives an idle notification, it adds the idle processor to the idle processor list (720). It then checks whether all of the processors are idle (722). When all processors are idle, the manager sends every processor a message that causes it to exit its message-driven loop and return to parallel code (724). In response to this message, each of the processors returns to parallel code as shown in FIG. 8 (see message 820, and step 822 in FIG. 8).
To further illustrate dynamic load balancing, FIG. 9 shows an example of load balancing between two worker processors and a manager. In the diagram, time runs vertically downwards. Each worker processor executes a "parent" function call that in turn makes two "child" function calls. The first worker asks for help with its first child call from the manager but does not get it because the second worker is still busy. After the first worker has finished, the second worker asks for help. As the first worker is idle, it is instructed by the manager to help the second worker, which ships over its first child call and proceeds with its second.
Clearly, every processor should not make a request for help before every serial recursive call, because this would result in the manager being flooded with messages towards the leaves of the recursion tree. As noted above, there is a minimum problem size below which it is not worth asking for help, because the time to send the arguments and receive the results is greater than the time to compute the result locally. The time required to send the arguments and receive the results provides a threshold (i.e. a lower bound), which can be compared to the result of the cost function described above: if the cost function for a particular recursive call (that is, the expected time taken by the call) returns a result less than the threshold, the processor does not request help.
The threshold therefore acts as a tuning function for the load-balancing system. Since it is dependent on many factors, including the algorithm, architecture, MPI implementation, problem size, machine size, and input data, it can be supplied as either a compile-time or run-time parameter. Reasonable approximations of this value for each divide-and-conquer function can be found using a routine that measures the time taken to send and receive the arguments and results for a function of size n between two processors, and that also measures the time taken to perform the function. The routine then adjusts n up or down appropriately, until the two times are approximately equal. This provides a rough estimate of the problem size below which it is not worth asking for help.
One possible example of a cost function used to approximate the computational cost of a recursive function call is the function quicksort.sub.-- cost in section 4.5.1 above. This cost function provides a cost estimate for a recursive function call quicksort, which can be compared against the threshold value. In this case, the threshold value is a cost estimate for sending the arguments to another processor and receiving the results.
Note that most divide-and-conquer functions, which have monotonically decreasing subproblem sizes, act as self-throttling systems using this load-balancing approach. Initially, request traffic is low, because the subproblems being worked on are large, and hence there is a large time delay between any two requests from one processor. Similarly, there is little request traffic at the end of the algorithm, because all the subproblems have costs below the cutoff limit. Requests are made when they can best be filled, towards the middle of an unbalanced algorithm, when some processors have finished but others are still working.
To be able to ship the arguments and results of function calls between processors, they are converted into MPI messages. For every divide-and-conquer function, the preprocessor therefore generates two auxiliary functions. One, given the same arguments as the divide-and-conquer function, wraps them up into messages and sends them to a specified processor. The other has the same return type as the divide-and-conquer function; when called, it receives the result as a message from a specified processor.
Scalar arguments are copied into a single untyped buffer before sending. This enables the runtime system to send all of the scalar arguments to a function in a single message, incurring only the overhead of a single message. However, vector arguments are sent as individual messages. The reason for this is that we expect the vectors to be comparatively long, and their transmission time to therefore be dominated by bandwidth instead of latency. The act of copying separate vectors into a single buffer for sending would require more additional time than the latency of the extra messages.
The act of transmitting function arguments to a processor, and receiving results from it, is effectively a remote procedure call.
5. Delaunay Triangulation
The team parallel model can be used to implement efficient divide-and-conquer programs with irregular subproblem sizes. To illustrate the capabilities of Machiavelli, the team parallel system has been used to implement a parallel 2D Delaunay triangulation program. This implementation has been demonstrated to be portable and to provide a better speed-up over efficient serial code than previous parallel implementations.
5.1 Definition of 2D Delaunay Triangulation
A Delaunay triangulation in the point set R.sup.2 is the unique triangulation of a set S of points such that there are no elements of S within the circumcircle of any triangle. Finding a Delaunay triangulation--or its dual, the Voronoi diagram--is an important problem in many domains, including pattern recognition, terrain modelling, and mesh generation for the solution of partial differential equations. Delaunay triangulations and Voronoi diagrams are among the most widely-studied structures in computational geometry, and have appeared in many other fields under different names, including domains of action in crystallography, Wigner-Seitz zones in metallurgy, Thiessen polygons in geography, and Blum's transforms in biology.
5.2 The Program
A pseudocode listing of an implementation of the Delaunay Triangulation is provided below (see Guy E. Blelloch, Gary L. Miller, and Dafna Talmor, Developing a practical projection-based Delaunay algorithm In Proceedings of the 12.sup.th Annual Symposium on Computational Geometry. ACM, May 1996).
Algorithm DELAUNAY (P,B)
Input: P, a set of points in R.sup.2,
B, a set of Delaunay edges of P which is the border of a region in R.sup.2 containing P.
Output: The set of Delaunay triangles of P which are contained within B.
Method:
1. If all the points in P are on the boundary B, return END.sub.-- GAME(B).
2. Find the point q that is the median along the x axis of all internal points (points in P and not on the boundary). Let L be the line x=q.sub.x.
3 Let P'={(p.sub.y -q.sub.y .parallel.p-q.parallel..sup.2).vertline.(p.sub.x,p.sub.y).epsilon.P}. These points are derived from projecting the points P onto a 3D paraboloid centered at q, and then projecting them onto the vertical plane through the line L.
4. Let H=LOWER.sub.-- CONVEX.sub.-- HULL (P'). H is a path of Delaunay edges of the set P. Let P.sub.H be the set of points the path H consists of, and H be the path H traversed in the opposite direction.
5. Create two subproblems:
B.sup.L =BORDERMERGE(B, H)
B.sup.R =BORDERMERGE(B, H)
p.sup.L ={p.epsilon.P.vertline.p is left of L}U {p'.epsilon.P.sub.H .vertline.p' contributed to B.sup.L }
p.sup.R ={p.epsilon.P.vertline.p is right of L}U {p'.epsilon.P.sub.H .vertline.p' contributed to B.sup.R }
6. Return DELAUNAY (P.sup.L,B.sup.L)U DELAIJNAY (P.sup.R,B.sup.R)
This triangulation program is an example of how the team parallel model can be used to implement a nested parallel algorithm with an irregular sub-problem size. As shown in FIG. 10, the program has two layers of recursion: an outer triangulation layer (1002) and an inner convex hull layer (1004). Both layers use a divide and conquer approach. The boxes of the outer layer (e.g., 1006-1018) represent how the main recursively splits the problem into two sub-problems of roughly equal size. Each of the recursive cells has an inner layer of recursion as reflected by the lines within the boxes. Each of these lines represents a recursive function call of the inner convex hull layer as reflected in expanded box 1020. Note that the recursive function calls in the inner layer have irregular problem sizes (e.g., 1022 and 1024).
Each subproblem is determined by a region R, which is the union of a collection of Delaunay triangles. This region is represented by the polygonal border B of the region, composed of Delaunay edges, and the set of points P of the region, composed of internal points and points on the border. At each recursive call, the region is divided into two using a median line cut of the internal points. The set of internal points is subdivided into those to the left and to the right of the median line. The polygonal border is subdivided using a new path of Delaunay edges that corresponds to the median line: the new path separates Delaunay triangles whose circumcenter is to the left of the median line, from those whose circumcenter is to the right of the median line. Once the new path is found, the new border of Delaunay edges for each subproblem is determined by merging the old border with the new path, in the Border.sub.-- Merge subroutine. Some of the internal points may appear in the new path, and may become border points of the new subproblems. Since it is using a median cut, the algorithm guarantees that the number of internal points is reduced by a factor of at least two at each call.
The new separating path of Delaunay edges is a lower convex hull of a simple transformation of the current point set. To obtain this path H, the program projects the points onto a paraboloid whose center is on the median line L, then projects the points horizontally onto a vertical plane whose intersection with the x-y plane is L. FIG. 11A shows an example of the median line L and the path H. FIG. 11B shows an example of the points projected onto the paraboloid, and FIG. 11C shows the paraboloid points projected onto the vertical plane.
The two-dimensional lower convex hull of those projected points is the required new border path H. This divide-and-conquer method can proceed as long as the subproblem contains internal points. Once the subproblem has no more internal points, it is a set of (possibly pinched) cycles of Delaunay edges. There may be some missing Delaunay edges between border points that still have to be found. To do that, the program moves to the End.sub.-- Game routine.
5.3 Implementation
As implemented in Machiavelli, the parallel program is used only as a coarse partitioner, subdividing the problem into pieces small enough to be solved on a single processor using the sequential Triangle described in Jonathan Richard Shewchuk, "Triangle: Engineering a 2D Quality Mesh Generator and Delaunay Triangulator" in Ming C. Lin and Dinesh Manocha, editors, Applied Computational Geometry: Towards Geometric Engineering, vol. 1148 of Lecture Notes in Computer Science, pp. 203-222, Springer-Verlag, May 1996.
The following sections describe several implementation decisions and optimizations that affect the performance of the final program, including choosing the data structures, improving the performance of specific substeps, and using a "lazy append" optimization. Most of the optimizations reduce or eliminate interprocessor communication.
5.3.1 Data structures
The basic data structure used by the code is a point, represented using two double-precision floating-point values for the x and y coordinates, and two integers, one serving as a unique global identifier and the other as a communication index within team phases of the algorithm. The points are stored in balanced Machiavelli vectors. To describe the relationship between points in a border, the code uses corners. A corner is a triplet of points corresponding to two segments in a path. Comers are not balanced across the processors as points are, but rather are stored in unbalanced vectors on the same processor as their "middle" point. The other two points are replicated in the corner structure, removing the need for a global communication step when operating on them. In particular, the structure of corners and their unbalanced replication allows the complicated border merge step to operate on purely local data on each processor. Finally, an additional vector of indices I (effectively, pointers) links the points in P with any corners that they are the middle point of in the borders B and H. As an alternative to the corner structure, the border structure could be implemented using pointers to link each point with a neighboring point in the border. However, this approach has the disadvantage that it would involve more inter-processor communication.
Given these data structures, the following sections describe the implementation and optimization of each of the phases of the algorithm in turn.
FIGS. 12A and 12B show the operation of the program in flow diagram form. The following sections describing the program are annotated with references to corresponding steps in the flow diagram.
5.3.2 Identifying internal points
The outer recursive layer begins by identifying the internal points in the input point set P (1202). Finding local internal points (those in P but not on the boundary B) is accomplished with a simple data-parallel operation across P and I that identifies points with no corresponding corner in B. If there are no more internal points (1204), the program invokes the End-Game routine (1206). If internal points remain, these local points are rebalanced across the current team to create a new vector of points, using a single call to pack (1208).
5.3.3 Finding the median
The next step in the outer layer is to call a median routine (1210) to find a median of the new vector of points. There are a variety of alternative parallel and serial algorithms for finding a median given a collection of points.
Initially a parallel version of the quickmedian algorithm was used to find the median of the internal points along the x or y axis. (See C.A.R. Hoare. "Algorithm 63 (partition) and Algorithm 65 (find). Communications of the ACM, 4(7): 321-322, 1961.) This algorithm is singly-recursive, redistributing a subset of the data among the processors on each step, which results in a high communication overhead. It proved to be significantly faster to replace this with a median-of-medians algorithm, in which each processor first uses a serial quickmedian to compute the median of its local data, shares this local median with the other processors in a collective communication step, and finally computes the median of all the local medians. This median of medians routine is represented as step 1210 in FIG. 12A. The result of the median of medians approach is not guaranteed to be the exact median, but in practice it is sufficiently good for load-balancing purposes. Although it is possible to construct input sets that would cause pathological behavior because of this modification, a simple randomization of the input data before use makes this highly unlikely in practice. Overall, the modification increased the speed of the Delaunay triangulation algorithm for the data sets and machine sizes studied.
5.3.4 Project onto a parabola
Next, each processor in the team projects the points onto a parabaloid (1212). Again, this is a purely local step, involving a simple data-parallel operation on each processor to create the new vector of points.
5.3.5 Finding the lower convex hull
Each processor now invokes the convex hull routine (1214). The subtask of finding the lower 2D convex hull of the projected inner points of the problem was shown by Blelloch et al to be the major source of floating-point operations within the original algorithm, and it is therefore worthy of serious study. For the Machiavelli implementation, as in the original algorithm, a simple version of quickhull was originally used. (See Franco P. Preparata and Michael Ian Shamos. Computational Geometry: An Introduction. Texts and Monographs in Computer Science. Springer-Verlag, 1985.) Quickhull is itself divide-and-conquer in nature, and is implemented as such using recursive calls to the Machiavelli toolkit. A sample code listing of quickhulls provided below:
The basic quickhull algorithm is fast on uniform point distributions, but tends to pick extreme "pivot" points when operating on very non-uniform point distributions, resulting in a poor division of data and a consequent lack of progress. Chan et al describe a variant that uses the pairing and pruning of points to guarantee that recursive calls have at most 3/4 of the original points. Experimentally, pairing a random selection of .sqroot.n points was found to give better performance when used as a substep of the Delaunay triangulation algorithm than pairing all n points (see Timothy M. Y. Chan, Jack Snoeyink, and Chee-Keng Yap. Output-sensitive construction of polytopes in four dimensions and clipped Voronoi diagrams in three. In Proceedings of the 6.sup.th Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 282-291, 1995.).
As with the median-of-medians approach and the use of quickhull itself, the effects of receiving non-optimal results from an algorithm substep (that is, using a sampling of the points rather than all of the points) are more than offset by the decrease in running time of the substep.
5.3.6 Creating Border Structures
The next step is to convert the result of the convex hull routine to a new border structure. The result of the quickhull function is a vector of indices of the convex hull of the projected points. The intermediate results of the quickhull function are left in an unbalanced state on the way back up the recursive call tree, and are rebalanced with a single call to pack at the top level. This approach of deferring the pack function eliminates log P levels of all-to-all communication, as in the quicksort example discussed in section 4.5.2.
Calls to fetch are then used to fetch the points themselves, and the corners that they "own". These calls to fetch are used to convert the vector representing the convex hull to a vector of corners (1216).
5.3.7 Creating the subproblems:
Having found the hull and hence the dividing path H, the program now merges the current border B with H, creating two new borders B.sup.L and B.sup.R and partitioning the points into P.sup.L and P.sup.R (1218). The merge phase requires line orientation tests to decide how to merge corners. The merge phase involves only local operations because the corner structures are replicated in each processor that they overlap.
FIG. 13 shows the six cases used in the line orientation test for merging the new and old borders. The old border is in thick lines, and the partitioning path is shown in dotted lines. The convention in the drawings (and the program) is that the interior lies to the left of the border when proceeding in the direction of the arrows. The resulting two new borders are in thin lines, with the new left border marked with double arcs, and the new right border with a single arc. The program checks these six cases to determine how to merge new and old borders.
Although the pseudocode listed above shows two calls to the border merge function (one for each direction of the new dividing path), in practice it is faster to make a single pass, creating both new borders and point sets at the same time.
5.3.8 Invoking the Split Function
FIG. 14 illustrates the operation of the split function to separate the problem into two subproblems. Before run-time, the preprocessor converts the program written in Machiavelli to C code and calls to MPI. With each instance of the split function, the preprocessor inserts a cost function into the code to evaluate the computational cost of each subproblem. The cost function approximates the computational cost of the main routine using the expression nlogn where n is the size of the input vector, namely, the collection of points. The preprocessor also inserts code to calculate the team sizes based on the results of the cost functions for each subproblem. Continuing the same example, the team sizes can be calculated as (1/p)(nlogn)(nlogn+mlogm) and (1/p)(mlogm)(nlogn+mlogm), where p is the number of processors in the current team being subdivided and m is the size of the other input vector.
The first two steps (1402, 1404) of FIG. 14 illustrate the initial run-time steps of the code that the preprocessor generates for the split function. The code generated for the split function creates a data structure to represent the new teams and allocates a result vector to hold the results of the new team. The split function code then redistributes the input vectors to the appropriate processors in the two new sub-teams (1406). Each of the processors in the new teams determines which processor team it belongs to (1408) and then determines whether the size of its team is one (1410). If the size is one, the processor executes the serial version of the main program (1412). The switch to the serial version terminates the recursion of the main program, and as such, represents the end game (1414). If the team size is greater than one, the parallel version of the program recurses back to the beginning of FIG. 12.
5.3.9 End game
Since we are using the parallel algorithm as a coarse partitioner, the end game is replaced with a serial Delaunay triangulation program. The serial version is also called if the parallel version recurses down to a team size of one before the end game is reached. Note, in an alternative implementation, it is possible to use a parallel 2D triangulation as the end game if the end game is reached before the team size is one. In other words, the parallel version may be adapted to provide a triangulation in addition to merely serving as a coarse partitioner.
The current implementation uses the version of Dwyer's algorithm that is implemented in the Triangle mesh generation package by Shewchuk, which has performance comparable to that of the original code by Dwyer. Since the input format for Triangle differs from that used by the Machiavelli code, conversion steps are necessary before and after calling it. These translate between the pointer-based format of Triangle, which is optimized for sequential code, and the indexed format with triplet replication used by the parallel code. No changes are necessary to the source code of Triangle.
Upon termination of the serial version of the program, the code generated for the split function returns up the recursion tree. The result that each processor provides is a vector of triangles. The parallel program travels back up the tree and forms a result vector from the left and right vector of triangles at each level of recursion. In Machiavelli, this operation is expressed as result=append (left, right) where left and right are the left and right vectors of triangles. Using the lazy vector optimization illustrated in FIG. 5E, the append operation can be deferred until the top of the tree is reached or eliminated entirely. The implementation of the parallel 2D triangulation uses the lazy append optimization to eliminate inter-processor communication. Specifically, the preprocessor adds code to mark all vectors of triangles as lazy and removes intermediate append operations. No inter-processor communication is required since the result vector remains in a lazy state with each of its elements being associated with the processor that provided them at the bottom of the recursion tree.
6. Conclusion
In view of the many possible implementations of the invention, it should be recognized that the implementation described above is only an example of the invention and should not be taken as a limitation on the scope of the invention. Rather, the scope of the invention is defined by the following claims. I therefore claim as my invention all that comes within the scope and spirit of these claims.
Claims
I claim:
1. A parallel two-dimensional triangulation method comprising: executing a recursive main program by subdividing processors in a parallel machine into teams, executing the main program in a control parallel manner such that each team executes a call to the main program in parallel; wherein the main program includes the following steps: taking as input a data structure representing a collection of points in two-dimensional space; in a data parallel operation across each of the processors in the team, finding internal points in the collection of points; when no internal points remain, returning a collection of triangles: when internal points remain, performing the following steps: a) computing a median of the collection of points; b) projecting the collection of points onto a parabola; c) finding a lower convex hull of the projected points by distributing the points among the processors in the team and executing a parallel convex hull routine within the team, including subdividing the processors in the team into sub-teams, with one sub-team per parallel function call; d) creating two borders for the collection of points; and e) invoking the main program on two new collections of points defined by the two borders.
2. The method of claim 1 further including: when a team size includes only one processor, executing a serial program version of the two-dimensional triangulation.
3. The method of claim 1 further including: performing dynamic load balancing among processors in a team by determining when computational cost of executing a routine on a processor exceeds a threshold and sending the arguments of the routine to an idle processor for execution of the routine.
4. The method of claim 1 wherein the borders of step d) are formed using solely local operations on the processors.
5. The method of claim 4 wherein the borders are stored in a collection oriented data structure whose elements are capable of being distributed unevenly across the processors such that no inter-processor communication is needed to form the borders in step d).
6. The method of claim 1 wherein the parallel convex hull routine of step c) makes recursive function calls to partition the projected points into subsets of points while descending a recursion tree resulting in a convex hull structure that is unevenly distributed among the processors at leaves of the recursion tree, and the parallel convex hull routine defers balancing the convex hull structure until returning to a top level of the recursion tree.
7. The method of claim 1 further including: for each subdividing step of the main program, distributing the two new collection of points to each of the processors in new teams associated with the new collection of points; when a team size includes only one processor, executing a serial program version of the two-dimensional triangulation such that each processor locally computes a 2D triangulation on a sub-set of the collection of points that has been distributed to the processor; and providing a result of the main program as an unbalanced collection oriented data type distributed across each of the processors, with each processor providing the 2D-triangulation of the sub-set of points that has been distributed to the processor.
8. A parallel 2D triangulation method for execution on a parallel computer with two or more processors, the method comprising: with a team of processors executing in parallel, partitioning a collection of 2D points into two new collections of 2D points using data parallel operations on elements of the collection of 2D points distributed among the processors in the team; creating sub-teams of the team of processors to further partition each of the new collections of 2D points; redistributing the points in the two new collections to the processors in the sub-teams; repeating the partitioning, creating and redistributing steps until the size of a sub-team is one processor; and invoking a serial 2D triangulation on the points distributed to the single processor sub-teams.
9. The method of claim 8 wherein the partitioning step includes two layers of recursion, including an outer recursive routine for finding a median through the collection of 2D points, and an inner recursive routine for finding a convex hull used to compute borders around the new collections of 2D points.
10. The method of claim 9 wherein inner recursive routine includes: with a team of processors executing in parallel, partitioning a collection of 2D points into two new collections of 2D points using data parallel operations on portions of the collection of 2D points distributed among the processors in the team; creating sub-teams of the team of processors to further partition each of the new collections of 2D points; redistributing the points in the two new collections to the processors in the sub-teams; repeating the partitioning, creating and redistributing steps until the size of a sub-team is one processor; and invoking a serial convex hull routine on the points distributed to the single processor sub-teams.
11. A computer readable medium having instructions for performing the steps of claim 8.
12. A computer readable medium comprising: a nested parallel program comprising a recursive function for partitioning a collection of 2D points into two or more subsets of points and distributing each subset to a separate team of processors in a parallel computer; wherein calls to the recursive function are executable in a control parallel manner on each of the separate teams of processors, and the processors within each team execute the recursive function using data parallel operations on data elements in the subset distributed to the team; and a 2D triangulation program for computing a 2D triangulation of a subset of points distributed to a processor.
13. The computer readable medium of claim 12 wherein the 2D triangulation is invoked when the recursive function subdivides the processors into teams such that a team has only one processor.
Patent Citations (5)
Non-Patent Literature (13)
- Jonathan C. Hardwick, "An Efficient Implementation of Nested Data Parallelism for Irregular Divide-and-Conquer Algorithms," First International Workshop on High-Level Programming Models and Supportive Environments, Apr. 1996.
- Guy E. Blelloch, Jonathan C. Hardwick, Jay Sipelstein, Marco Zagha, and Siddhartha Chatterjee, "Implementation of a Portable Nested Data-Parallel Language," Journal of Parallel and Distributed Computing, vol. 21, pp. 4-14, 1994.
- Stephen T. Barnard, "PMRSB: Parallel Multilevel Recursive Spectral Bisection," Proceedings of Supercomputing, 1995.
- Eugene D. Brooks III, "PCP: A Paradigm Which Spans Uniprocessor, SMP and MPP Architectures," SC '95 Poster Presentation, Jun. 14, 1995.
- Soumen Chakrabarti, James Demmel, Katherine Yelick, "Modeling the Benefits of Mixed Data and Task Parallelism," In Proceedings of the 7.sup.th Annual ACM Symposium on Parallel Algorithms and Architectures, Jul., 1995.
- Tom Axford, "The Divide-and-Conquer Paradigm as a Basis for Parallel Language Design," Advances in Parallel Algorithms, Blackwell, 1992.
- Thomas J. Sheffler and Siddhartha Chatterjee, "An Object-Oriented Approach to Nested Data Parallelism," In Proceedings of the 5.sup.th Symposium on the Frontiers of Massively Parallel Computation, IEEE, Feb., 1995.
- Daniel W. Palmer, Jan. F Prins and Stephen Westfold, "Work-Efficient Nested Data-Parallelism," In Proceedings of the 5.sup.th Symposium on the Frontiers of Massively Parallel Computation, IEEE, pp. 186-193, Feb., 1995.
- Manuel M.T. Chakravarty, Friedrich Wilhelm Schroer, Martin Simons, "V-Nested Parallelism in C," In Proceedings of the Working Conference on Massively Parallel Programming Models, IEEE, Computer Society Press, 1995.
- Jay Sipelstein, "Data Representation Optimizations for Collection-Oriented Languages," Ph.D. Thesis Proposal, Ver. 1.0, for School of Computer Science, Carnegie Mellon University,, May 12, 1992.
- Jonathan Richard Shewchuk, "Triangle: Engineering a 2D Quality Mesh Generator and Delaunay Triangulator," In First Workshop on Applied Computational Geometry, pp. 124-133, ACM, May, 1996.
- Guy E. Blelloch, Gary L. Miller, Dafna Talmor, Developing a Practical Projection-Based Parallel Delaunay Algorithm, In Proceedings of the 12.sup.th Annual Symposium on Computational Geometry, ACM, May, 1996.
- Timothy M.Y. Chan, Jack Snoeyink, Chee-Keng Yong, "Output-Sensitive Construction of Polytopes in Four Dimensions and Clipped Voronoi Diagrams in Three," In Proceedings of the 6.sup.th Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 282-291, 1995.