Related
I need assistance to resolve an error in the following code:
#include <iostream>
#include <mpi.h>
using namespace std;
//matrix in two dimension in memory!!
int main(int argc, char** argv)
{
const int WIDTH = 100;
const int HEIGHT = 100;
int id, P;
double tempValue = 0;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &P);
MPI_Comm_rank(MPI_COMM_WORLD, &id);
double A[WIDTH][HEIGHT];
double x[HEIGHT], b[WIDTH];
int upperBound, lowerBound = 0;
// Master controls worksharing..
if (id == 0)
{
// Init A & x
for (int i = 0; i < WIDTH; i++)
for (int j = 0; j < HEIGHT; j++)
A[i][j] = 1;
for (int j = 0; j < HEIGHT; j++)
x[j] = 2;
// Send to each node its portion of A to be processed
int portionSize = WIDTH / P;
for (int i = 0; i < P; i++)
{
lowerBound = i * portionSize;
upperBound = (i + 1) * portionSize;
// let the last node process the remainder
if (i == (P - 1))
upperBound += (HEIGHT - portionSize * P);
if (i > 0)// Do not send to master node!!
{
// Send to node i the lower & upper bounds the A portion
//and complete vector x
MPI_Send(&lowerBound, 1, MPI_INT, i, 0, MPI_COMM_WORLD);
MPI_Send(&upperBound, 1, MPI_INT, i, 0, MPI_COMM_WORLD);
MPI_Send(&A[lowerBound][0], (upperBound - lowerBound) * HEIGHT,
MPI_DOUBLE, i, 0, MPI_COMM_WORLD);
MPI_Send(&x[0], HEIGHT, MPI_DOUBLE, i, 0, MPI_COMM_WORLD);
}
}
// master perform part of the job...
for (int i = 0; i < portionSize; i++)
{
tempValue = 0;
for (int j = 0; j < HEIGHT; j++)
tempValue += A[i][j] * x[j];
b[i] = tempValue;
}
//Get the results in order, each node would send their boundaries and data part
for (int i = 1; i < P; i++)
{
MPI_Recv(&lowerBound, 1, MPI_INT, i, 0, MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
MPI_Recv(&upperBound, 1, MPI_INT, i, 0, MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
MPI_Recv(&P[lowerBound], (upperBound - lowerBound), MPI_DOUBLE, i, 0,
MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
}
// Print the first 2 values to check..
cout << "b[0]=" << b[0] << " b[Width-1]=" << b[WIDTH - 1] << endl;
}
else // the rest of the workers do their parts
{
//Receive the inputs
MPI_Recv(&lowerBound, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
MPI_Recv(&upperBound, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
MPI_Recv(&A[lowerBound][0], (upperBound - lowerBound) * WIDTH, MPI_DOUBLE, 0, 0,
MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
MPI_Recv(&x, HEIGHT, MPI_DOUBLE, 0, 0, MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
cout << "Node:" << id << " Received from:" << lowerBound << " to " << upperBound - 1
<< endl;
double* result = new double[upperBound - lowerBound];
//Do the job
for (int i = lowerBound, resultCounter = 0; i < upperBound; i++, resultCounter++)
{
tempValue = 0;
for (int j = 0; j < HEIGHT; j++)
tempValue += A[i][j] * x[j];
result[resultCounter] = tempValue;
}
//send the results
MPI_Send(&lowerBound, 1, MPI_INT, 0, 0, MPI_COMM_WORLD);
MPI_Send(&upperBound, 1, MPI_INT, 0, 0, MPI_COMM_WORLD);
MPI_Send(&result[0], upperBound - lowerBound, MPI_DOUBLE, 0, 0, MPI_COMM_WORLD);
delete[] result;
}
MPI_Finalize();
return 0;
}
When I compile the code in Microsoft Visual Studio 2019, I get this error message:
Error (active) E0142 expression must have pointer-to-object type ConsoleApplication9 C:\Users\m_swe\Desktop\Assignments\Assignments\PrjP2P\MatMPI\MatMPI\Source.cpp 59
Error C2109 subscript requires array or pointer type ConsoleApplication9 C:\Users\m_swe\Desktop\Assignments\Assignments\PrjP2P\MatMPI\MatMPI\Source.cpp 59
I think the problem is on line: 59
MPI_Recv(&P[lowerBound], (upperBound - lowerBound), MPI_DOUBLE, i, 0,
MPI_Recv takes in a pointer to a buffer (the first argument) where you are going to receive and store the incoming data. In this case it could be in some variable which you can define inside the for loop, as:
int receivedValues[ WIDTH * HEIGHT ];
for (int i = 1; i < P; i++)
{
MPI_Recv(&lowerBound, 1, MPI_INT, i, 0, MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
MPI_Recv(&upperBound, 1, MPI_INT, i, 0, MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
MPI_Recv(&receivedValues[0], (upperBound - lowerBound), MPI_DOUBLE, i, 0,
MPI_COMM_WORLD, MPI_STATUSES_IGNORE);
// do your computation here with receivedValues
}
}
.Hello, i have a problem with my c++ code. I'm trying to make a parallel implementation from my sequential code for sobel operator using OpenCV.
My actual idea is scatter a picture using 2d buffer, make the sobel operation to averaged_rows*cols. and then make the gathering When i have sent the averaged_rows and every rank receive it, i try to use MPI_Scatter and this execution error appears:
sent to 1
sent to 2
sent to 3
recieved by 1
recieved by 2
recieved by 3
[roronoasins-GL552VW:3245] *** An error occurred in MPI_Scatter
[roronoasins-GL552VW:3245] *** reported by process [1759117313,1]
[roronoasins-GL552VW:3245] *** on communicator MPI_COMM_WORLD
[roronoasins-GL552VW:3245] *** MPI_ERR_TRUNCATE: message truncated
[roronoasins-GL552VW:3245] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[roronoasins-GL552VW:3245] *** and potentially your MPI job)
[roronoasins-GL552VW:03239] 2 more processes have sent help message help-mpi-errors.txt / mpi_errors_are_fatal
[roronoasins-GL552VW:03239] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages
What I actually do is scatter pic buffer, bcast the picture to the rest of ranks and the gathering.
MPI_Scatter(pic, cols*rows_av, MPI_INT, picAux, cols*rows_av, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Bcast (pic3, cols*rows, MPI_INT, 0, MPI_COMM_WORLD);
int ip_gx, ip_gy, sum;
for(int y = ip*pic_struct[2]; y < (ip+1)*pic_struct[2] -1; y++){
for(int x = 1; x < pic_struct[1]- 1; x++){
int gx = x_gradient(pic3, x, y);
int gy = y_gradient(pic3, x, y);
int sum = abs(gx) + abs(gy);
sum = sum > 255 ? 255:sum;
sum = sum < 0 ? 0 : sum;
picAux[y][x] = sum;
}
}
MPI_Gather(picAux, cols*rows_av, MPI_INT, pic, cols*rows_av, MPI_INT, 0, MPI_COMM_WORLD);
I'd like to know what is happening with the Scatter function, i thought that i could scatter single picture pieces to the rest ranks to calculate sobel, maybe i'm wrong.
My code is here if u want to check it. Thanks for your time.
// > compile with mpic++ mpi_sobel.cpp -o mpi_sobel `pkg-config --libs opencv` -fopenmp -lstdc++
#include <iostream>
#include <cmath>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <omp.h>
#include <mpi.h>
using namespace std;
using namespace cv;
Mat src, dst;
/*
Computes the x component of the gradient vector
at a given point in a image.
returns gradient in the x direction
| 1 0 -1 |
Gx = | 2 0 -2 |
| 1 0 -1 |
*/
int x_gradient(int** image, int x, int y)
{
return image[y-1][x-1] +
2*image[y][x-1] +
image[y+1][x-1] -
image[y-1][x+1] -
2*image[y][x+1] -
image[y+1][x+1];
}
/*
Computes the y component of the gradient vector
at a given point in a image
returns gradient in the y direction
| 1 2 1 |
Gy = | 0 0 0 |
|-1 -2 -1 |
*/
int y_gradient(int** image, int x, int y)
{
return image[y+1][x-1] +
2*image[y+1][x] +
image[y+1][x+1] -
image[y-1][x-1] -
2*image[y-1][x] -
image[y-1][x+1];
}
int main(int argc, char** argv)
{
string picture;
if (argc == 2) {
picture = argv[1];
src = imread(argv[1], CV_LOAD_IMAGE_GRAYSCALE);
}
else {
picture = "input/logan.jpg";
src = imread(picture.c_str(), CV_LOAD_IMAGE_GRAYSCALE);
}
if( !src.data )
{ return -1; }
dst.create(src.rows, src.cols, src.type());
int rows_av, rows_extra;
Size s = src.size();
int rows = s.height;
int cols = s.width;
int pic[rows][cols];int picAux[rows][cols];
int ** pic3;
pic3 = new int*[rows];
for(int y = 0; y < rows; y++)
pic3[y] = new int[cols];
int pic_struct[3], pic_struct_recv[3];
int np, ip;
double start_time = omp_get_wtime();
if (MPI_Init(&argc, &argv) != MPI_SUCCESS){
exit(1);
}
MPI_Comm_size(MPI_COMM_WORLD, &np);
MPI_Comm_rank(MPI_COMM_WORLD, &ip);
MPI_Status status;
if(ip==0)
{
for(int y = 0; y < rows ; y++)
for(int x = 0; x < cols; x++)
{
pic3[y][x] = src.at<uchar>(y,x);
pic[y][x] = 0;
picAux[y][x] = 0;
}
src.release();
rows_av = rows/np;
//cols_av = cols/np;
pic_struct[0] = rows;
pic_struct[1] = cols;
pic_struct[2] = rows_av;
//pic_struct[3] = cols_av:
for(int i=1; i < np; i++)
{
//rows = (i <= rows_extra) ? rows_av+1 : rows_av;
pic_struct[0] = rows;
MPI_Send(&pic_struct, sizeof(pic_struct), MPI_BYTE, i, 0, MPI_COMM_WORLD);
cout << "sent to " << i << endl;
}
}else{//ip
MPI_Recv(&pic_struct, sizeof(pic_struct), MPI_BYTE, 0, 0, MPI_COMM_WORLD, &status);
cout << "recieved by " << ip << endl;
}
MPI_Scatter(pic, cols*rows_av, MPI_INT, picAux, cols*rows_av, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Bcast (pic3, cols*rows, MPI_INT, 0, MPI_COMM_WORLD);
cout << "bcast" << endl;
//MPI_Barrier(MPI_COMM_WORLD);
int ip_gx, ip_gy, sum;
for(int y = ip*pic_struct[2]; y < (ip+1)*pic_struct[2] -1; y++){
for(int x = 1; x < pic_struct[1]- 1; x++){
ip_gx = x_gradient(src, x, y);
ip_gy = y_gradient(src, x, y);
sum = abs(ip_gx) + abs(ip_gy);
sum = sum > 255 ? 255:sum;
sum = sum < 0 ? 0 : sum;
picAux[y][x] = sum;
}
}
MPI_Gather(picAux, cols*rows_av, MPI_INT, pic, cols*rows_av, MPI_INT, 0, MPI_COMM_WORLD);
cout << "gather" << endl;
MPI_Finalize();
if(!ip)
{
double time = omp_get_wtime() - start_time;
for( int i = 0 ; i < rows ; i++ )
{
delete [] pic3[i] ;
delete [] pic3 ;
}
cout << "Number of processes: " << np << endl;
cout << "Rows, Cols: " << rows << " " << cols << endl;
cout << "Rows, Cols(Division): " << rows_av << ", " << cols << endl << endl;
cout << "Processing time: " << time << endl;
for(int i=0; i < 6 ; i++) picture.erase(picture.begin());
for(int i=0; i < 4 ; i++) picture.pop_back();
picture.insert(0,"output/");
picture += "-sobel.jpg";
for(int y = 0; y < rows; y++)
for(int x = 0; x < cols; x++)
dst.at<uchar>(y,x) = pic[y][x];
if(imwrite(picture.c_str(), dst)) cout << "Picture correctly saved as " << picture << endl;
else cout << "\nError has occurred being saved." << endl;
}
return 0;
}
Update: I forgot rows_av in ranks != 0 and pic3 sending is fixed. I've packed src in contiguous buffer and it is right in each rank.
updated code here: https://pastebin.com/jPV9mGFW
I have noticed that into the 3/4 dark there is noise, with this new issue i dont know if gathering is the problem now or i am doing the operations with number_process*rows/total_processes wrong.
MPI_Scatter(pic, cols*rows_av, MPI_UNSIGNED_CHAR, picAux, cols*rows_av, MPI_UNSIGNED_CHAR, 0, MPI_COMM_WORLD);
int ip_gx, ip_gy, sum;
for(int y = ip*rows_av+1; y < (ip+1)*rows_av-1; y++){
for(int x = 1; x < cols ; x++){
ip_gx = x_gradient(src, x, y);
ip_gy = y_gradient(src, x, y);
sum = abs(ip_gx) + abs(ip_gy);
sum = sum > 255 ? 255:sum;
sum = sum < 0 ? 0 : sum;
picAux[y][x] = sum;
//picAux[y*rows_av+x] = sum;
}
}
MPI_Gather(picAux, cols*rows_av, MPI_UNSIGNED_CHAR, pic, cols*rows_av, MPI_UNSIGNED_CHAR, 0, MPI_COMM_WORLD);
Loop updated and image is full calculated now but i cant use images bigger than 2048x1536.
for(int y = 1; y < rows_av-1; y++){
for(int x = 1; x < cols ; x++){
ip_gx = x_gradient(src, x, ip*rows_av+y);
ip_gy = y_gradient(src, x, ip*rows_av+y);
sum = abs(ip_gx) + abs(ip_gy);
sum = sum > 255 ? 255:sum;
sum = sum < 0 ? 0 : sum;
picAux[y*cols+x] = sum;
}
}
How could i send larger images than 2048x1536?
--------------------------------------------------------------------------
mpirun noticed that process rank 2 with PID 0 on node roronoasins-GL552VW exited on signal 11 (Segmentation fault).
--------------------------------------------------------------------------
Images size issue was the stack limited size. with ulimit -s unlimited works fine but im now working to improve memory efficiency. Last code will be updated in the pastebin code above.
I am on Ubuntu 16.04, GCC 5.4, latest OpenCV. Suppose I have a vector of double
std::vector<std::vector<double>> vecvecdouble;
vecvecdouble.resize(3, std::vector<double>(3, 0));
for (int j = 0; j < 3; j++){
for (int i = 0; i < 3; i++){
if (i == 0){
vecvecdouble[i][j] = 1;
vecvecdouble[i][j] = 1;
}
if (i == 1){
vecvecdouble[i][j] = 2;
vecvecdouble[i][j] = 2;
}
if (i == 1 && j == 0){
std::cout << vecvecdouble[i - 1][j] << std::endl;
std::cout << vecvecdouble[i][j] << std::endl;
std::cout << vecvecdouble[i + 1][j] << std::endl;
}
}
}
It prints
1
2
0
as expected. However, if I do the same thing with OpenCV cv::mat
std::vector<std::vector<cv::Mat>> vecvecmat;
vecvecmat.resize(
3, std::vector<cv::Mat>(3, cv::Mat(4, 4, CV_64FC1, cv::Scalar(0.0))));
for (int j = 0; j < 3; j++){
for (int i = 0; i < 3; i++){
if (i == 0){
vecvecmat[i][j].at<double>(0, 0) = 1;
vecvecmat[i][j].at<double>(0, 1) = 1;
}
if (i == 1){
vecvecmat[i][j].at<double>(0, 0) = 2;
vecvecmat[i][j].at<double>(0, 1) = 2;
}
if (i == 1 && j == 0){
std::cout << vecvecmat[i - 1][j] << std::endl;
std::cout << vecvecmat[i][j] << std::endl;
std::cout << vecvecmat[i + 1][j] << std::endl;
}
}
}
It prints
[2, 2, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0]
[2, 2, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0]
[2, 2, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0]
which is completely unexpected because I am expecting it to print
[1, 1, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0]
[2, 2, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0]
[0, 0, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0;
0, 0, 0, 0]
However, if I don't try to resize the vector in one single line and go through two for loops it actually returns the expected result
std::vector<std::vector<cv::Mat>> vecvecmat;
vecvecmat.resize(3);
for (int i = 0; i < 3; i++){
for (int j = 0; j < 3; j++){
cv::Mat mymat = cv::Mat(4, 4, CV_64FC1, cv::Scalar(0.0));
vecvecmat[i].push_back(mymat);
}
}
for (int j = 0; j < 3; j++){
for (int i = 0; i < 3; i++){
if (i == 0){
vecvecmat[i][j].at<double>(0, 0) = 1;
vecvecmat[i][j].at<double>(0, 1) = 1;
}
if (i == 1){
vecvecmat[i][j].at<double>(0, 0) = 2;
vecvecmat[i][j].at<double>(0, 1) = 2;
}
if (i == 1 && j == 0){
std::cout << vecvecmat[i - 1][j] << std::endl;
std::cout << vecvecmat[i][j] << std::endl;
std::cout << vecvecmat[i + 1][j] << std::endl;
}
}
}
What is wrong with the line?
vecvecmat.resize(3, std::vector<cv::Mat>(3, cv::Mat(4, 4, CV_64FC1, cv::Scalar(0.0))));
The documentation for the copy constructor for cv::Mat says (emphasis mine):
Parameters
m Array that (as a whole or partly) is assigned to the constructed matrix. No data is copied by these constructors. Instead, the header pointing to m data or its sub-array is constructed and associated with it. The reference counter, if any, is incremented. So, when you modify the matrix formed using such a constructor, you also modify the corresponding elements of m. If you want to have an independent copy of the sub-array, use Mat::clone().
The initial matrix that you construct in your call to std::vector::resize is being copied into each element of the vector with this copy constructor*, so they are all pointing to the same data. When you modify one matrix, you modify all of them.
* or maybe with operator=, which does the same thing (I'm not sure which, but it doesn't affect the result)
I suggest initializing your vector like this:
std::vector<std::vector<cv::Mat>> vecvecmat;
vecvecmat.resize(3, std::vector<cv::Mat>());
for(auto& v : vecvecmat)
{
for(std::size_t i = 0; i < 3; ++i)
{
v.emplace_back(4, 4, CV_64FC1, cv::Scalar(0.0));
}
}
I am trying to do some MPI parallel work. I am able to run this on any number of processors. The issue is that each processor will take one job, execute it and send it back then the program is done. I want to be able to send a processor another job when it has finished. Im not sure how to implement this. Basically I am trying to send each core 10 jobs.
if (myRank == 0) {
int numCores = MPI::COMM_WORLD.Get_size();
for (int rank = 1; rank < numCores; rank++) {
MPI::COMM_WORLD.Send(&yarray[0], imax, MPI::DOUBLE, rank, 0);
MPI::COMM_WORLD.Send(&tarray[0], imax, MPI::DOUBLE, rank, 0);
MPI::COMM_WORLD.Recv(&ans, imax, MPI::DOUBLE, MPI::ANY_SOURCE, MPI_ANY_TAG, mystatus);
answers[counter] = ans;
counter++;
}
}
else
{
MPI::COMM_WORLD.Recv(&yarray1, imax, MPI::DOUBLE, MPI::ANY_SOURCE, MPI_ANY_TAG, mystatus);
MPI::COMM_WORLD.Recv(&tarray1, imax, MPI::DOUBLE, MPI::ANY_SOURCE, MPI_ANY_TAG, mystatus);
double floor = 0.5, ceiling = 3.5, range = (ceiling - floor);
double rnd = floor + double((range * rand()) / (RAND_MAX + 1.0));
yarray [0] = rnd;
yarray1 [0] = rnd;
double temp = 0;
for (int k = 0; k < imax; k++) {
tarray1[k+1] = tarray1[k] + h;
yarray1[k+1] = yarray1[k] + h * (2 * yarray1[k] - 2 * tarray1[k] * tarray1[k] - 3);
}
temp = yarray1[int(imax)];
//cout << "Rank = " << myRank << " Solution = " << temp << endl;
MPI::COMM_WORLD.Send(&temp, 1, MPI::DOUBLE, 0, 0);
}
Update: within in myrank == 0
while(counter != jobs){
MPI::COMM_WORLD.Recv(&ans, imax, MPI::DOUBLE, MPI::ANY_SOURCE, MPI_ANY_TAG, mystatus);
answers[counter] = ans;
counter++;
}
You need to have some sort of feedback from rank 0 to the other ranks. After the other ranks return their work to rank 0, they should receive a new message back that tells them either their next job or that there is no more work to be completed. The ranks should continue looping until there is no more work to be done.
What I am trying to acheive in this simplified code is:
2 types of processes (root, and children, ids/rank = 10 and 0-9 respectively)
init:
root will listen to children "completed"
children will listen to root notification when all has completed
while there is no winner (not all done yet):
children will have 20% chance they will be done (and notify root they are done)
root will check that all are done
if all done: send notification to children of "winner"
I have code like:
int numprocs, id, arr[10], winner = -1;
bool stop = false;
MPI_Request reqs[10], winnerNotification;
MPI_Init(NULL, NULL);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &id);
for (int half = 0; half < 1; half++) {
for (int round = 0; round < 1; round++) {
if (id == 10) { // root
// keeps track of who has "completed"
fill_n(arr, 10, -1);
for (int i = 0; i < 10; i++) {
MPI_Irecv(&arr[i], 1, MPI_INT, i, 0, MPI_COMM_WORLD, &reqs[i]);
}
} else if (id < 10) { // children
// listen to root of winner notification/indication to stop
MPI_Irecv(&winner, 1, MPI_INT, 10, 1, MPI_COMM_WORLD, &winnerNotification);
}
while (winner == -1) {
//cout << id << " is in loop" << endl;
if (id < 10 && !stop && ((rand() % 10) + 1) < 3) {
// children has 20% chance to stop (finish work)
MPI_Send(&id, 1, MPI_INT, 10, 0, MPI_COMM_WORLD);
cout << id << " sending to root" << endl;
stop = true;
} else if (id == 10) {
// root checks number of children completed
int numDone = 0;
for (int i = 0; i < 10; i++) {
if (arr[i] >= 0) {
//cout << "root knows that " << i << " has completed" << endl;
numDone++;
}
}
cout << "numDone = " << numDone << endl;
// if all done, send notification to players to stop
if (numDone == 10) {
winner = 1;
for (int i = 0; i < 10; i++) {
MPI_Send(&winner, 1, MPI_INT, i, 1, MPI_COMM_WORLD);
}
cout << "root sent notification of winner" << endl;
}
}
}
}
}
MPI_Finalize();
Output from debugging couts look like: problem seems to be root is not receiving all childrens notification that they are completed?
2 sending to root
3 sending to root
0 sending to root
4 sending to root
1 sending to root
8 sending to root
9 sending to root
numDone = 1
numDone = 1
... // many numDone = 1, but why 1 only?
7 sending to root
...
I thought perhaps I can't receive into an array: but I tried
if (id == 1) {
int x = 60;
MPI_Send(&x, 1, MPI_INT, 0, 0, MPI_COMM_WORLD);
} else if (id == 0) {
MPI_Recv(&arr[1], 1, MPI_INT, 1, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
cout << id << " recieved " << arr[1] << endl;
}
Which works.
UPDATE
This seems to be resolved if I add a MPI_Barrier(MPI_COMM_WORLD) before the end of the while loop, but why? Even if the processes run out of sync, eventually, children will send to root that they have completed and root should "listen" to that and process accordingly? What seems to be happening is root keeps running, hogging up all resources for children to execute at all? Or whats happening here?
UPDATE 2: some children not getting notification from root
Ok now the problem that root does not receive children's notification that they have completed by #MichaelSh's answer, I focus on children not receiving from parent. Here's a code that reproduces that problem:
int numprocs, id, arr[10], winner = -1;
bool stop = false;
MPI_Request reqs[10], winnerNotification;
MPI_Init(NULL, NULL);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &id);
srand(time(NULL) + id);
if (id < 10) {
MPI_Irecv(&winner, 1, MPI_INT, 10, 0, MPI_COMM_WORLD, &winnerNotification);
}
MPI_Barrier(MPI_COMM_WORLD);
while (winner == -1) {
cout << id << " is in loop ..." << endl;
if (id == 10) {
if (((rand() % 10) + 1) < 2) {
winner = 2;
for (int i = 0; i < 10; i++) {
MPI_Send(&winner, 1, MPI_INT, i, 0, MPI_COMM_WORLD);
}
cout << "winner notifications sent" << endl;
}
}
}
cout << id << " b4 MPI_Finalize. winner is " << winner << endl;
MPI_Finalize();
Output looks like:
# 1 run
winner notifications sent
10 b4 MPI_Finalize. winner is 2
9 b4 MPI_Finalize. winner is 2
0 b4 MPI_Finalize. winner is 2
# another run
winner notifications sent
10 b4 MPI_Finalize. winner is 2
8 b4 MPI_Finalize. winner is 2
Notice some processes doesnt seem to get the notification from the parent? Why is that, MPI_Wait for child processes will just hang them? So how do I resolve this?
Also
All MPI_Barrier does in your case -- it waits for child responses to complete. Please check my answer for a better solution
If I dont do this, I suppose each child response will just take few ms? So even if I dont wait/barrier, I'd expect the receive to still happen soon after the send? Unless processes end up hogging resources and other processes does not run?
Please try this block of code (error checking omitted for simplicity):
...
// root checks number of children completed
int numDone = 0;
MPI_Status statuses[10];
MPI_Waitall(10, reqs, statuses);
for (int i = 0; i < 10; i++) {
...
Edit A better solution:
Each child initiates root winner notification receipt and sends its notification to the root.
Root initiates winner notification receipt to the array and goes into wait for all notifications to be received, and then sends winner's id to children.
Insert this code below after for (int round = 0; round < 1; round++)
if (id == 10)
{ // root
// keeps track of who has "completed"
memset(arr, -1, sizeof(arr));
for (int i = 0; i < 10; i++)
{
MPI_Irecv(&arr[i], 1, MPI_INT, i, 0, MPI_COMM_WORLD, &reqs[i]);
}
}
else if (id < 10)
{ // children
// listen to root of winner notification/indication to stop
MPI_Irecv(&winner, 1, MPI_INT, 10, 1, MPI_COMM_WORLD, &winnerNotification);
}
if (id < 10)
{
while(((rand() % 10) + 1) < 3) ;
// children has 20% chance to stop (finish work)
MPI_Send(&id, 1, MPI_INT, 10, 0, MPI_COMM_WORLD);
std::cout << id << " sending to root" << std::endl;
// receive winner notification
MPI_Status status;
MPI_Wait(&winnerNotification, &status);
// Process winner notification
}
else if (id == 10)
{
MPI_Status statuses[10];
MPI_Waitall(10, reqs, statuses);
// if all done, send notification to players to stop
{
winner = 1;
for (int i = 0; i < 10; i++)
{
MPI_Send(&winner, 1, MPI_INT, i, 1, MPI_COMM_WORLD);
}
std::cout << "root sent notification of winner" << std::endl;
}
}