L-tgray / DispatchCPP

A header-only lib to easily dispatch C++ code across N threads.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

DispatchCPP

A header-only library for asynchronously dispatching C++ code both in parallel and serially. This easy-to-use library allows a variable (zero, or more than zero) amount of arguments to your parallel/serial work you dispatch.

tl;dr:

  1. Clone this repo,
  2. Run make install in it, or simply copy ./src/DispatchCPP into your project's includes/headers folder,
  3. Then compile your project with at least c++11.

Full, compilable examples at the bottom:

To install:

  1. $ git clone https://github.com/L-tgray/DispatchCPP.git
  2. $ cd DispatchCPP
  3. $ make install

This will install DispatchCPP into /usr/local/include.

As an initial recap, to use this async library, you simply:

  1. Create a function to be executed by the queue for each dispatch. This is where you declare the return value and parameters (if any).
// Example of a function returning void, and having zero parameters.
QueueFunction<void> * pFuncVoid = new QueueFunction<void>(
    []() {                           // Main function.
        printf("Hello, world!\n");
    },
    nullptr,                         // No pre function specified (optional, defaults to nullptr).
    nullptr                          // No post function specified (optional, defaults to nullptr).
);

// Example of a function returning void, taking in two int arguments.
QueueFunction<void, int, int> * pFuncAdd = new QueueFunction<void, int, int>(
    [](int x, int y) {               // Main function.
        printf("%d + %d = %d\n", x, y, x + y);
    },
    nullptr,                         // No pre function specified (optional, defaults to nullptr).
    nullptr                          // No post function specified (optional, defaults to nullptr).
);
  1. Create a work queue with our queue functions, with a specific number of threads and params.
// Create our void work queue with 1 thread.
Queue<void> * pQueueVoid = new Queue<void>(pFuncVoid, 1);

// Create our work queue for performing addition with 1 thread.
Queue<void, int, int> * pQueueAdd = new Queue<void, int, int>(pFuncAdd, 1);
  1. Dispatch work, using the same number of params you specified when creating the queue.
// Dispatch 2 hello world printf's.
pQueueVoid->dispatchWork();
pQueueVoid->dispatchWork();

// Expect to see:
// Hello, world!
// Hello, world!

// If true is provided, this function will block until all work is finished.
pQueueVoid->hasWorkLeft(true);

// --------------------------------------

// Dispatch 3 additions.
pQueueAdd->dispatchWork(1, 2);
pQueueAdd->dispatchWork(3, 4);
pQueueAdd->dispatchWork(5, 6);

// Expect to see:
// 1 + 2 = 3
// 3 + 4 = 7
// 5 + 6 = 11

// If true is provided, this function will block until all work is finished.
pQueueAdd->hasWorkLeft(true);
  1. Clean up after ourselves.
delete(pQueueVoid);
delete(pQueueAdd);
delete(pFuncVoid);
delete(pFuncAdd);

If you only specify a single thread for the queue, the queue will be a serial work queue.

When dispatching work to a queue, your work will be handled by one of the N threads you specified during queue creation, handing off the exact parameters you've passed into the dispatch call, itself.

Simply copy the folder ./src/DispatchCPP into your headers/include directory, and then include this header-only library with one statement: #include <DispatchCPP/DispatchCPP.h>.

Pre/Post Queue Functions

It is possible to specify pre and/or post functions for your QueueFunction. The pre function is ran before your main function, having the same dispatched params passed into it, allowing one to preprocess/filter dispatches. The post function is ran after your main function, having the main function's return value (if it has one) passed into it.

By default, QueueFunction's have no pre or post functions. A main function is required, however.

If a pre function determines the main function should not be run, the default behavior is for the post function to not be run, either.

This can be changed, however, by manipulating the QUEUE_FUNCTION_ENABLE_POST_FUNC_CALL_WHEN_MAIN_NOT_INVOKED define in DispatchCPP/QueueFunction.h at the top. Even when defined, though, there is no indication given to the post function as to whether the main func was ran or not.

Pre Functions

Pre functions make it possible to screen/preprocess incoming dispatches, and this function returns a bool to indicate whether the queue's main function should be executed or not. All parameters from the main function will be passed into the pre function.

Example:

// Our queue function to perform addition, however we only want to perform the addition if both numbers are even.
QueueFunction<void, int, int> * pFuncAdd = new QueueFunction<void, int, int>(
    [](int x, int y) {                          // Main function.
        printf("%d + %d = %d\n", x, y, x + y);
    },
    [](int x, int y) {                          // Pre function.
        return(!(x % 2) && !(y % 2));
    },
    nullptr                                     // No post function specified (optional, defaults to nullptr).
);

Post Functions

Post functions make it possible to analyze the return value of your main function. This function does not return anything (void), and accepts either 0 or 1 parameters. If your main function returns void, this function will have no params. If your main function does return a value, this will be the only incoming parameter.

Example:

// Our queue function to perform addition, however we only want to output the addition if the resulting number is odd.
QueueFunction<int, int, int> * pFuncAdd = new QueueFunction<int, int, int>(
    [](int x, int y) {      // Main function.
        return(x + y);
    },
    nullptr,                // No pre function specified (required, in this case, to specify the post function).
    [](int result) {        // Post function.
        if (result % 2) {
            printf("Result = %d\n", result);
        }
    }
);

Full Example 1

In this example, we parallelize the addition of numbers as well as the storing of each result.

Main.cpp:

#include <stdio.h>
#include "DispatchCPP/DispatchCPP.h"

using namespace DispatchCPP;

int main() {
    // Declare our Queue's function. Specify a pre function to make sure the incoming pResult parameter isn't NULL.
    QueueFunction<void, int, int, int *> * pFuncAdd = new QueueFunction<void, int, int, int *>(
        [](int x, int y, int * pResult) { // Main function.
            *pResult = x + y;
        },
        [](int x, int y, int * pResult) { // Pre function.
            return(pResult != NULL);
        },
        nullptr                           // No post function (optional, defaults to nullptr).
    );

    // Create our Queue using this function, with 2 threads.
    Queue<void, int, int, int *> * pQueue = new Queue<void, int, int, int *>(pFuncAdd, 2);

    // Let's create our data to dispatch in parallel.
    int data[5][3] = {
        { 0, 1, 0 },
        { 2, 3, 0 },
        { 4, 5, 0 },
        { 6, 7, 0 },
        { 8, 9, 0 },
    };

    // Dispatch all our work, now, sending in the first 2 ints as params, and the address of the third to store the result.
    pQueue->dispatchWork(data[0][0], data[0][1], &(data[0][2]));
    pQueue->dispatchWork(data[1][0], data[1][1], &(data[1][2]));
    pQueue->dispatchWork(data[2][0], data[2][1], &(data[2][2]));
    pQueue->dispatchWork(data[3][0], data[3][1], &(data[3][2]));
    pQueue->dispatchWork(data[4][0], data[4][1], &(data[4][2]));

    // Wait for the work to finish, now.
    pQueue->hasWorkLeft(true);

    // Print out all the results.
    printf("[%u] %d + %d = %d\n", 0, data[0][0], data[0][1], data[0][2]);
    printf("[%u] %d + %d = %d\n", 1, data[1][0], data[1][1], data[1][2]);
    printf("[%u] %d + %d = %d\n", 2, data[2][0], data[2][1], data[2][2]);
    printf("[%u] %d + %d = %d\n", 3, data[3][0], data[3][1], data[3][2]);
    printf("[%u] %d + %d = %d\n", 4, data[4][0], data[4][1], data[4][2]);

    // Clean up after ourselves.
    delete(pQueue);
    delete(pFuncAdd);
    return(EXIT_SUCCESS);
}

Make sure you have either installed or copied the src/DispatchCPP folder into the same directory as this Main.cpp file, and compile it with at least c++11 specified:

$ g++ -std=c++17 Main.cpp -o Main.out -lpthread

Output:

$ ./Main.out
[0] 0 + 1 = 1
[1] 2 + 3 = 5
[2] 4 + 5 = 9
[3] 6 + 7 = 13
[4] 8 + 9 = 17

Full Example 1 (Condensed)

This example is the same as the one prior, however we inline many things and loop instead of being explicit with arrays.

Main.cpp:

#include <stdio.h>
#include "DispatchCPP/DispatchCPP.h"

using namespace DispatchCPP;

int main() {
    // Create our Queue inline, creating the QueueFunction within our call to the Queue's constructor.
    Queue<void, int, int, int *> * pQueue = new Queue<void, int, int, int *>(
        // Declare our Queue's function. Specify a pre function to make sure the incoming pResult parameter isn't NULL.
        new QueueFunction<void, int, int, int *>(
            [](int x, int y, int * pResult) { // Main function.
                *pResult = x + y;
            },
            [](int x, int y, int * pResult) { // Pre function.
                return(pResult != NULL);
            },
            nullptr                           // No post function (optional, defaults to nullptr).
        ),
        2,   // This queue will have 2 threads.
        true // Passing in true indicates for the Queue to deallocate the `QueueFunction *` passed in during creation.
    );

    // Let's create our data to dispatch in parallel.
    int data[5][3] = {
        { 0, 1, 0 },
        { 2, 3, 0 },
        { 4, 5, 0 },
        { 6, 7, 0 },
        { 8, 9, 0 },
    };

    // Dispatch all our work, now, sending in the first 2 ints as params, and the address of the third to store the result.
    for (unsigned int index = 0; index < 5; ++index) {
        pQueue->dispatchWork(data[index][0], data[index][1], &(data[index][2]));
    }

    // Wait for the work to finish, now.
    pQueue->hasWorkLeft(true);

    // Print out all the results.
    for (unsigned int index = 0; index < 5; ++index) {
        printf("[%u] %d + %d = %d\n", index, data[index][0], data[index][1], data[index][2]);
    }

    // Clean up after ourselves.
    delete(pQueue);
    return(EXIT_SUCCESS);
}

Make sure you have either installed or copied the src/DispatchCPP folder into the same directory as this Main.cpp file, and compile it with at least c++11 specified:

$ g++ -std=c++17 Main.cpp -o Main.out -lpthread

Output:

$ ./Main.out
[0] 0 + 1 = 1
[1] 2 + 3 = 5
[2] 4 + 5 = 9
[3] 6 + 7 = 13
[4] 8 + 9 = 17

Full Example 2 (Capturing and Using Variables)

In this example, instead of passing in variables to each dispatch to give the function context, we can simply capture variables with the lambda, itself.

Main.cpp:

#include <stdio.h>
#include "DispatchCPP/DispatchCPP.h"

using namespace DispatchCPP;

int main() {
    // Let's create our data to dispatch in parallel.
    int data[5][3] = {
        { 0, 1, 0 },
        { 2, 3, 0 },
        { 4, 5, 0 },
        { 6, 7, 0 },
        { 8, 9, 0 },
    };

    // Construct our Queue, inlining the creation of the Queue's function.
    Queue<void, int> * pQueue = new Queue<void, int>(
        // Create our QueueFunction, capturing the data array.
        new QueueFunction<void, int>(
            [&data](int index) {                 // Main function.
                data[index][2] = data[index][0] + data[index][1];
            },
            nullptr,                             // No pre function (optional, defaults to nullptr).
            nullptr                              // No post function (optional, defaults to nullptr).
        ),
        2,       // Use 2 threads to dispatch work from this queue.                            
        true     // Deallocate the QueueFunction passed into this Queue constructor.
    );

    // Dispatch all our work, now, sending in each of the indices to compute.
    for (unsigned int index = 0; index < 5; ++index) {
        pQueue->dispatchWork(index);
    }

    // Wait for the work to finish, now.
    pQueue->hasWorkLeft(true);

    // Print out all the results.
    for (unsigned int index = 0; index < 5; ++index) {
        printf("[%u] %d + %d = %d\n", index, data[index][0], data[index][1], data[index][2]);
    }

    // Clean up after ourselves.
    delete(pQueue);
    return(EXIT_SUCCESS);
}

Make sure you have either installed or copied the src/DispatchCPP folder into the same directory as this Main.cpp file, and compile it with at least c++11 specified:

$ g++ -std=c++17 Main.cpp -o Main.out -lpthread

Output:

$ ./Main.out
[0] 0 + 1 = 1
[1] 2 + 3 = 5
[2] 4 + 5 = 9
[3] 6 + 7 = 13
[4] 8 + 9 = 17

Full Example 3 (Serializing Work Done In Parallel)

In this example, we execute many add instructions in parallel, however we want to serialize the printing each of these results with a serial queue.

Main.cpp:

#include <stdio.h>
#include "DispatchCPP/DispatchCPP.h"

using namespace DispatchCPP;

int main() {
    // Let's create our data to dispatch in parallel.
    int data[5][3] = {
        { 0, 1, 0 },
        { 2, 3, 0 },
        { 4, 5, 0 },
        { 6, 7, 0 },
        { 8, 9, 0 },
    };

    // Construct our serial Queue, used to serialize calls to printf.
    Queue<void, int> * pQueuePrintf = new Queue<void, int>(
        // Create our QueueFunction, capturing the data array.
        new QueueFunction<void, int>(
            [&data](int index) {                 // Main function.
                printf("[%d] %d + %d = %d\n", index, data[index][0], data[index][1], data[index][2]);
            },
            nullptr,                             // No pre function (optional, defaults to nullptr).
            nullptr                              // No post function (optional, defaults to nullptr).
        ),
        1,       // Use only a single thread to dispatch work from this queue.
        true     // Deallocate the QueueFunction passed into this Queue constructor.
    );

    // Construct our parallel Queue to be used for addition.
    Queue<void, int> * pQueueAdd = new Queue<void, int>(
        // Create our QueueFunction, capturing the data array and the serial Queue.
        new QueueFunction<void, int>(
            [&data, pQueuePrintf](int index) {   // Main function.
                // Perform the addition.
                data[index][2] = data[index][0] + data[index][1];

                // Dispatch the printf to a serialized queue, so only one result is printed at a time.
                pQueuePrintf->dispatchWork(index);
            },
            nullptr,                             // No pre function (optional, defaults to nullptr).
            nullptr                              // No post function (optional, defaults to nullptr).
        ),
        4,       // Use 4 threads to dispatch work from this queue.
        true     // Deallocate the QueueFunction passed into this Queue constructor.
    );

    // Dispatch all our work, now, sending in each of the indices to compute.
    for (unsigned int index = 0; index < 5; ++index) {
        pQueueAdd->dispatchWork(index);
    }

    // Wait for the work to finish, now.
    pQueueAdd->hasWorkLeft(true);    // Wait for all the additions to be dispatched, first.
    pQueuePrintf->hasWorkLeft(true); // Now wait for all the printf's to be dispatched.

    // Clean up after ourselves.
    delete(pQueuePrintf);
    delete(pQueueAdd);
    return(EXIT_SUCCESS);
}

Make sure you have either installed or copied the src/DispatchCPP folder into the same directory as this Main.cpp file, and compile it with at least c++11 specified:

$ g++ -std=c++17 Main.cpp -o Main.out -lpthread

Possible output:

$ ./Main.out
[0] 0 + 1 = 1
[2] 4 + 5 = 9
[1] 2 + 3 = 5
[4] 8 + 9 = 17
[3] 6 + 7 = 13

Full Example 4 (Thread Init/Close Funcs)

If you need thread-specific objects or data created, there are thread init and thread close functions available to QueueFunctions. These allow you specify a lambda which gets executed at thread init, as well as a lambda that gets executed at thread close.

Main.cpp:

#include <stdio.h>
#include <vector>
#include <map>
#include "DispatchCPP/DispatchCPP.h"

using namespace std;
using namespace DispatchCPP;

int main() {
    // Declare our thread-specific map of data, which gets initialized right after the thread's created.
    map<QueueTID, vector<int>> threadData = map<QueueTID, vector<int>>();

    // Construct our parallel Queue to be used for generating arrays of random numbers.
    Queue<void> * pQueueRand = new Queue<void>(
        // Create our QueueFunction, capturing the data array and the serial Queue.
        new QueueFunction<void>(
            [&threadData]() {                    // Main function.
                // Grab our thread ID. 
                QueueTID tid = QueueThread::TID();

                // Generate a random number, and add it to this thread's vector of rand's.
                threadData[tid].push_back(rand() % 100);
            },  
            nullptr,                             // No pre function (optional, defaults to nullptr).
            nullptr,                             // No post function (optional, defaults to nullptr).
            [&threadData]() {                    // Init function.
                // Grab our thread ID. 
                QueueTID tid = QueueThread::TID();

                // Initialize our thread's vector.
                threadData[tid] = vector<int>();

                // Let the user know what's up. 
                printf("Thread %llu starting...\n", (unsigned long long int) tid);
            },  
            []() {                               // Close function.
                // Grab our thread ID. 
                QueueTID tid = QueueThread::TID();

                // Let the user know what's up. 
                printf("Thread %llu stopping...\n", (unsigned long long int) tid);
            }   
        ),  
        4,    // Use 4 threads to dispatch work from this queue.
        true  // Deallocate the QueueFunction passed into this Queue constructor.
    );  

    // Dispatch our work, informing the thread pool to generate 50 random numbers across the 4 threads.
    for (unsigned int index = 0; index < 50; ++index) {
        pQueueRand->dispatchWork();
    }   

    // Wait for the work to finish, now.
    pQueueRand->hasWorkLeft(true);

    // Iterate over all entries in the map.
    for (auto const& entry : threadData) {
        // Grab each part of the map.
        QueueTID    entryThread = entry.first;
        vector<int> entryVector = entry.second;

        // Print it all out.
        printf("Thread %llu numbers =>", (unsigned long long int) entryThread);
        for (unsigned int index = 0; index < ((unsigned int) entryVector.size()); ++index) {
            printf(" %d", entryVector[index]);
        }   
        printf("\n");
    }   

    // Clean up after ourselves.
    delete(pQueueRand);
    return(EXIT_SUCCESS);
}

Make sure you have either installed or copied the src/DispatchCPP folder into the same directory as this Main.cpp file, and compile it with at least c++11 specified:

$ g++ -std=c++17 Main.cpp -o Main.out -lpthread

Possible output:

$ ./Main.out
Thread 123145504194560 starting...
Thread 123145504731136 starting...
Thread 123145505267712 starting...
Thread 123145505804288 starting...
Thread 123145504194560 numbers => 7 78 92 16 35 26 79 45 28
Thread 123145504731136 numbers => 49 73 58 30 72 65 42 3 27 40 12 3 57 33 99 10 33
Thread 123145505267712 numbers => 23 9 40 87 69 67 49 21 67 93 36 85 91 94
Thread 123145505804288 numbers => 44 29 9 60 78 97 12 79 72 57
Thread 123145504194560 stopping...
Thread 123145505804288 stopping...
Thread 123145504731136 stopping...
Thread 123145505267712 stopping...

Full Example 5 (Sorting N Vectors: Manual vs DispachCPP vs Pure PThreads)

Main.cpp:

#include <stdio.h>
#include <algorithm>
#include <chrono>
#include <queue>
#include <vector>
#include "DispatchCPP/DispatchCPP.h"

using namespace std;
using namespace DispatchCPP;

void initVectors(vector<vector<int>> * pAllVectorsManual,
                 vector<vector<int>> * pAllVectorsParallel,
                 vector<vector<int>> * pAllVectorsPThreads,
                 unsigned int          numVectors,
                 unsigned int          vectorSize) {
    // Initialize all our vectors.
    for (unsigned int index = 0; index < numVectors; ++index) {
        // Create our vector for each.
        vector<int> tempVectorManual   = vector<int>();
        vector<int> tempVectorParallel = vector<int>();
        vector<int> tempVectorPThreads = vector<int>();

        // Preallocate space.
        tempVectorManual.reserve(vectorSize);
        tempVectorParallel.reserve(vectorSize);
        tempVectorPThreads.reserve(vectorSize);

        // For each vector, we insert 1000 numbers.
        for (unsigned int subIndex = 0; subIndex < vectorSize; ++subIndex) {
            int randNum = rand();
            tempVectorManual.push_back(randNum);
            tempVectorParallel.push_back(randNum);
            tempVectorPThreads.push_back(randNum);
        }   

        // Add these vectors to each of their vectors.
        pAllVectorsManual->push_back(tempVectorManual);
        pAllVectorsParallel->push_back(tempVectorParallel);
        pAllVectorsPThreads->push_back(tempVectorPThreads);
    }   
}

void sortVectorsManual(vector<vector<int>> * pAllVectors) {
    // Iterate over all vectors we've been given.
    for (unsigned int index = 0; index < ((unsigned int) pAllVectors->size()); ++index) {
        // Grab the current vector and sort it.
        vector<int> * pVector = &((*pAllVectors)[index]);
        sort(pVector->begin(), pVector->end());
    }   
}

void sortVectorsParallel(vector<vector<int>> * pAllVectors, unsigned int numThreads) {
    // Declare our Queue, with our QueueFunction inlined within it.
    Queue<void, vector<int> *> newQueue( 
        new QueueFunction<void, vector<int> *>( 
            [](vector<int> * pVector) {
                // Sort the incoming vector.
                sort(pVector->begin(), pVector->end());
            }   
        ),  
        numThreads, // Use the number of threads we've been told to use.
        true        // Deallocate the QueueFunction we passed into this constructor.
    );  

    // Dispatch all our work, now.
    for (unsigned int index = 0; index < ((unsigned int) pAllVectors->size()); ++index) {
        // Grab the current vector and dispatch it to be sorted.
        vector<int> * pVector = &((*pAllVectors)[index]);
        newQueue.dispatchWork(pVector);
    }   

    // Wait for all the sorting to finish.
    newQueue.hasWorkLeft(true);
}

// Declare our structure for each thread to use.
typedef struct __THREAD_STRUCT__ {
    pthread_mutex_t      * pWorkMutex;
    queue<vector<int> *> * pWork;
    volatile bool          shouldQuit;
} threadStruct, * pThreadStruct;

void * sortVectorsPThreads_Func(void * pContext) {
    // Cast ourselves a pointer to a thread object.
    pThreadStruct pSelf = ((pThreadStruct) pContext);

    // Iterate until we're told to stop.
    while (!(pSelf->shouldQuit)) {
        // Grab the lock, and fetch our work to be dispatched.
        pthread_mutex_lock(pSelf->pWorkMutex);
        vector<int> * pNewWork = ((vector<int> *) NULL);
        if (pSelf->pWork->size() > 0) {
            pNewWork = pSelf->pWork->front();
            pSelf->pWork->pop();
        }
        pthread_mutex_unlock(pSelf->pWorkMutex);

        // Sort this vector, now.
        if (pNewWork) {
            sort(pNewWork->begin(), pNewWork->end());

        // Otherwise just wait until we have to exit.
        } else {
            usleep(5);
        }
    }

    // Always return NULL.
    return((void *) NULL);
}

void sortVectorsPThreads(vector<vector<int>> * pAllVectors, unsigned int numThreads) {
    // Initialize our thread variables to be used.
    pthread_mutex_t      threadMutex;
    queue<vector<int> *> threadWork = queue<vector<int> *>();
    
    // Initialize the mutex.
    pthread_mutex_init(&threadMutex, NULL);

    // Initialize our thread info structure.
    threadStruct threadInfo = {
        &threadMutex,
        &threadWork,
        false
    };

    // Grab the lock on the mutex before creating the threads, halting them until we dipatch work.
    pthread_mutex_lock(&threadMutex);

    // Create our threads, now.
    pthread_t * pAllThreads = (pthread_t *) calloc(numThreads, sizeof(pthread_t));
    for (unsigned int index = 0; index < numThreads; ++index) {
        pthread_create(&(pAllThreads[index]), NULL, sortVectorsPThreads_Func, (void *) &threadInfo);
    }

    // Dispatch all our work, now.
    for (unsigned int index = 0; index < ((unsigned int) pAllVectors->size()); ++index) {
        threadWork.push(&((*pAllVectors)[index]));
    }

    // Let go of the lock on the mutex, letting all of the threads take off.
    pthread_mutex_unlock(&threadMutex);

    // Monitor until all work has been dispatched.
    while (true) {
        // Give up the CPU for a bit.
        usleep(5);

        // Determine if we have any work left.
        bool allWorkDispatched = false;
        pthread_mutex_lock(&threadMutex);
        allWorkDispatched = (threadWork.size() == 0);
        pthread_mutex_unlock(&threadMutex);

        // Have we dispatched all work yet?
        if (allWorkDispatched) {
            threadInfo.shouldQuit = true;
            break;
        }
    }

    // Join back each thread, now.
    for (unsigned int index = 0; index < numThreads; ++index) {
        pthread_t * pCurrentThread = &(pAllThreads[index]);
        pthread_join(*pCurrentThread, NULL);
    }
    pthread_mutex_destroy(&threadMutex);
    free(pAllThreads);
}

int main() {
    // Unbuffered output.
    setbuf(stdout, NULL);

    // Declare the number of vectors and vector size to use, as well as number of threads.
    unsigned int numVectors = 1000;
    unsigned int vectorSize = 100000;
    unsigned int numThreads = 4;

    // Declare our vector of vectors.
    vector<vector<int>> allVectorsManual   = vector<vector<int>>();
    vector<vector<int>> allVectorsParallel = vector<vector<int>>();
    vector<vector<int>> allVectorsPThreads = vector<vector<int>>();

    // Initialize numVector's number of vectors, with each vector having vectorSize number of
    // entries. All vectors will have the same exact entries in them, in the same exact order.
    printf("Initializing %u vectors, each with %u entries...", numVectors, vectorSize);
    initVectors(&allVectorsManual, &allVectorsParallel, &allVectorsPThreads, numVectors, vectorSize);
    printf("done!\n");

    // --- Sort all vectors manually. --------------------------------------------
    printf("[MANUAL]      Sorting %u vectors (each w/%u entries) with 1 thread... ", numVectors, vectorSize);
    auto beforeManual = chrono::high_resolution_clock::now();
    sortVectorsManual(&allVectorsManual);
    auto afterManual = chrono::high_resolution_clock::now();
    double manualMS = ((double) chrono::duration_cast<chrono::microseconds>(afterManual - beforeManual).count()) / 1000.0f;
    printf("%.2f milliseconds\n", manualMS);
    // ---------------------------------------------------------------------------

    // --- Sort all vectors in parallel with numThreads' worth of  threads. ------
    printf("[DispatchCPP] Sorting %u vectors (each w/%u entries) with %u threads...", numVectors, vectorSize, numThreads);
    auto beforeParallel = chrono::high_resolution_clock::now();
    sortVectorsParallel(&allVectorsParallel, numThreads);
    auto afterParallel = chrono::high_resolution_clock::now();
    double parallelMS = ((double) chrono::duration_cast<chrono::microseconds>(afterParallel - beforeParallel).count()) / 1000.0f;
    printf("%.2f milliseconds (%.2fx speedup)\n", parallelMS, manualMS / parallelMS);
    // ---------------------------------------------------------------------------

    // --- Sort all vectors in parallel with pthreads, using numThreads' threads -
    printf("[pthreads]    Sorting %u vectors (each w/%u entries) with %u threads...", numVectors, vectorSize, numThreads);
    auto beforePThread = chrono::high_resolution_clock::now();
    sortVectorsPThreads(&allVectorsPThreads, numThreads);
    auto afterPThread = chrono::high_resolution_clock::now();
    double pthreadMS = ((double) chrono::duration_cast<chrono::microseconds>(afterPThread - beforePThread).count()) / 1000.0f;
    printf("%.2f milliseconds (%.2fx speedup)\n", pthreadMS, manualMS / pthreadMS);
    // ---------------------------------------------------------------------------
    // Clean up after ourselves.
    return(EXIT_SUCCESS);
}

Make sure you have either installed or copied the src/DispatchCPP folder into the same directory as this Main.cpp file, and compile it with at least c++11 specified:

$ g++ -std=c++17 Main.cpp -o Main.out -lpthread

Possible output:

$ ./Main.out
[MANUAL]      Sorting 1000 vectors (each w/100000 entries) with 1 thread... 4604.19 milliseconds
[DispatchCPP] Sorting 1000 vectors (each w/100000 entries) with 4 threads...1219.81 milliseconds (3.77x speedup)
[pthreads]    Sorting 1000 vectors (each w/100000 entries) with 4 threads...1217.23 milliseconds (3.78x speedup)

About

A header-only lib to easily dispatch C++ code across N threads.

License:MIT License


Languages

Language:C++ 93.2%Language:C 3.9%Language:Makefile 2.9%