I need to send array pieces to all processes using MPI_Scatter then to get sum of all elements. Where should I initialize array then to scatter it? In root rank?
If I initialize array on root rank then other ranks dont get their data. Otherway I can initialize array for everyone (out of if(rank == root)...else), but it means, that I create array several times.
#include <stdio.h>
#include <stdlib.h>
#include <mpi.h>
#include <iostream>
#include <time.h>
using namespace std;
int main(int argc, char* argv[])
{
int size;
int rank;
srand(time(NULL));
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
int arr_size = size * 2;
int block = arr_size / (size);
int* B = new int[block];
if (rank == 0)
{
int* A = new int[arr_size];
cout << "generated array: " << endl;
for (int i = 0; i < arr_size; i++)
{
A[i] = rand() % 100;
cout << A[i] << " ";
}
cout << endl;
MPI_Scatter(A, block, MPI_INT, B, block, MPI_INT, 0, MPI_COMM_WORLD);
}
cout << "process " << rank << " received: " << endl;
for (int i = 0; i < block; i++)
{
cout << B[i] << " ";
}
cout << endl;
int local_sum = 0;
for (int i = 0; i < block; i++)
{
local_sum += B[i];
}
cout << "sum in process " << rank << " = " << local_sum << endl;
cout << endl;
int global_sum;
MPI_Reduce(&local_sum, &global_sum, 1, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD);
if (rank == 0)
{
cout << "sum = " << global_sum << endl;
}
MPI_Finalize();
return 0;
}
I get something like this (only root rank got its data):
process 1 received:
process 3 received:
-842150451 -842150451
-842150451 -842150451
sum in process 1 = -1684300902
sum in process 3 = -1684300902
process 2 received:
-842150451 -842150451
sum in process 2 = -1684300902
process 0 received:
4 9
sum in process 0 = 13
sum = -757935397
MPI_Scatter() is a collective operation and must hence be invoked by all the ranks.
Declare int *A = NULL; on all ranks and only allocate and populate on rank zero.
int* A = NULL;
int* B = new int[block];
if (rank == 0)
{
A = new int[arr_size];
cout << "generated array: " << endl;
for (int i = 0; i < arr_size; i++)
{
A[i] = rand() % 100;
cout << A[i] << " ";
}
cout << endl;
}
MPI_Scatter(A, block, MPI_INT, B, block, MPI_INT, 0, MPI_COMM_WORLD);
Related
//How do I count numbers in the input vector that is greater and lesser than input number X?
#include <iostream>
#include
#include
using namespace std;
int main()
{
int i, size;
float verdi = 0;
vector<float> minvektor;
cout << "input vectorsize " << endl;
cin >> size;
cout << "Input vector-numbers, use enten space or enter to space/divide numbers. " << endl;
for (i = 0; i < size; i++)
{
cin >> verdi;
minvektor.push_back(verdi);
}
for (i = 0; i < size; i++)
{
cout << minvektor[i] << " ";
}
// -------------------------------------------------
cout << "\ninput number X to compare to vector: " << endl;
float X;
cin >> X;
//count numbers in a vector that is greater than X
//output ex: 5 numbers in a vector that is greater than number X
}
in an R script I source a cpp file to make some calculations. In that R script, a function defined in the cpp file is called and a matrix and an integer is provided. After a few rounds through the loop it gives the error "Not a matrix" (in line of code resid = (x(_,j) - x(_,i))*(x(_,j) - x(_,i));), even though for the rounds before it worked.
R script:
## all together
# rm(list=ls())
library(RcppArmadillo)
library(Rcpp)
sourceCpp("~/test.cpp",verbose = FALSE)
cat("start loop")
for(n in c(45:46)){
cat("\n", n, "\n")
p_m <- matrix(data=rnorm(n^2,1,1),nrow = n, ncol=n)
print(class(p_m))
print(some_function(p_m,nosamples=10))
}
cat("finished")
I start this R script via the command line. R version R-4.1.0. In R-Studio it crashes with a fatal error.
The cpp file:
// [[Rcpp::depends(RcppArmadillo)]]
#include <RcppArmadillo.h>
using namespace Rcpp;
// [[Rcpp::export]]
NumericVector some_function(NumericMatrix x,int nosamples) {
int ncol = x.ncol();
NumericVector out2(nosamples);
int loops;
int loops2;
double result=0;
NumericVector::iterator it;
double acc = 0;
NumericVector resid(ncol);
NumericVector out(ncol*(ncol-1)/2);
loops2=0;
std::cout << nosamples << std::endl;
std::cout << (ncol-1) << std::endl;
std::cout << ncol*(ncol-1)/2 << std::endl;
while(loops2 < (nosamples)){
std::cout << "loops2:" << std::endl;
std::cout << loops2 << std::endl;
loops=0;
int i;
int j;
for(j=0;j<(ncol-1);++j){
std::cout << " j: " << j << std::endl;
for (i = (j+1); i < (ncol); ++i) {
std::cout << " i: " << i << std::endl;
resid = (x(_,j) - x(_,i))*(x(_,j) - x(_,i)); //here it stops
std::cout << " i: " << i << std::endl;
for(int ii=0; ii<ncol;++ii){
acc += resid[i];
}
result=sqrt(acc);
loops += 1;
out[loops] = result;
std::cout << " i: " << i << std::endl;
}
}
std::cout << "loops:" << std::endl;
std::cout << loops << std::endl;
out = out[out > 0];
it = std::min_element(out.begin(), out.end());
out= *it;
std::cout << out << std::endl;
loops2 += 1;
out2[loops2]=out[0];
}
std::cout << "cpp finished" << std::endl;
return(out2);
}
Can someone explain what the problem is about?
Thanks and kind regards
Edit
I adapted some things in the cpp file (shown below) and the error disappeared. First I thought, everything is fine. But when I increase the number of loops, another problem occurs: the function breaks, but no error is shown. It breaks after loop number 543 ("loop2: 543"). At least it does the same in each round of the while loop with the same data.
I adapted the R-script and the ccp file to make this problem (at least on my machine) reproducible.
I know this code seems to be somehow meaningless, but it is part of a bigger program and I wanted to give here a minimum example.
The R script:
## all together
# rm(list=ls())
library(RcppArmadillo)
library(Rcpp)
sourceCpp("~/test.cpp",verbose = FALSE)
cat("start loop")
for(n in c(100:101)){
cat("\n", n, "\n")
p_m <- matrix(data=rnorm(n^2,1,1),nrow = n, ncol=n)
print(class(p_m))
print(some_function(p_m,nosamples=800))
}
cat("finished")
The cpp file:
// [[Rcpp::depends(RcppArmadillo)]]
// [[Rcpp::depends(RcppEigen)]]
#include <RcppArmadillo.h>
#include <RcppEigen.h>
using namespace Rcpp;
using Eigen::Map;
using Eigen::VectorXd;
typedef Map<VectorXd> MapVecd;
// [[Rcpp::export]]
NumericVector some_function(NumericMatrix x,int nosamples) {
int ncol = x.ncol();
NumericVector out(ncol*(ncol-1)/2);
NumericVector out2(nosamples);
NumericVector out3(ncol*(ncol-1)/2);
NumericVector resid(ncol);
int loops;
int loops2;
double result=0;
double acc = 0;
int show_cout=0;
loops2=0;
std::cout << nosamples << std::endl;
std::cout << (ncol-1) << std::endl;
std::cout << ncol*(ncol-1)/2 << std::endl;
while(loops2 < (nosamples)){
std::cout << "loops2:" << loops2 << std::endl;
loops=0;
int i;
int j;
for(j=0;j<(ncol-1);++j){
// std::cout << " j: " << j << std::endl;
for (i = (j+1); i < (ncol); ++i) {
if(show_cout==1){
std::cout << " i: " << i << std::endl;
}
resid = (x(_,j) - x(_,i))*(x(_,j) - x(_,i));
if(show_cout==1){
std::cout << " i: " << i << std::endl;
}
for(int ii=0; ii<ncol;++ii){
acc += resid[ii];
}
result=sqrt(acc);
loops += 1;
out[loops] = result;
if(show_cout==1){
std::cout << " i: " << i << std::endl;
}
}
}
// std::cout << "loops:" << loops << std::endl;
//
out = out[out > 0];
const MapVecd xy(as<MapVecd>(out));
out3=xy.minCoeff();
out2[loops2]=out3[0];
loops2 += 1;
}
std::cout << "cpp finished" << std::endl;
return(out2);
}
Two things here:
Use out[loops++] = result; instead of loops += 1; out[loops] = result; because you were starting at 1, and probably accessing the last element outside of the range of this vector.
Use
for(int ii=0; ii<ncol;++ii){ double eps = x(ii, j) - x(ii, i); acc += eps * eps; }
instead of relying on this resid vector.
I'm exploring MPI in C++ and I wanted to parallelize the creation of a picture of the Mandelbrot set. I'm using the ppm format. Each processor builds its part and sends it back to the main process that receives it as MPI_CHAR. This is the code:
#include "mpi.h"
#include <iostream>
#include <string>
#include <fstream>
#include <complex>
using namespace std;
int mandelbrot(int x, int y, int width, int height, int max) {
complex<float> point((float) (y - height/2.0) * 4.0/width, (float) (x - width/2.0) * 4.0/width);
complex<float> z(0, 0);
unsigned int iteration = 0;
while (abs(z) < 4 && iteration < max) {
z = z * z + point;
iteration++;
}
return iteration;
}
int main(int argc, char **argv) {
int numprocs;
int myid;
int buff_size = 404270; // 200x200
char buff[buff_size];
int i;
MPI_Status stat;
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&numprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&myid);
int width = 200, height = 200, max_iter = 1000;
if (myid == 0) {
ofstream image("mandel.ppm");
image << "P3\n" << width << " " << height << " 255\n";
for(i=1; i < numprocs; i++) {
MPI_Probe(i, 0, MPI_COMM_WORLD, &stat);
int length;
MPI_Get_count(&stat, MPI_CHAR, &length);
MPI_Recv(buff, length, MPI_CHAR, i, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
image << buff;
}
} else {
stringstream ss;
// proc rank: 1, 2, ..., n
int part = height/(numprocs-1), start = (myid - 1) * part, end = part * myid;
printf("%d -> %d\n", start, end);
for (int row = start; row < end; row++) {
for (int col = 0; col < width; col++) {
int iteration = mandelbrot(row, col, width, height, max_iter);
if (row == start) ss << 255 << ' ' << 255 << ' ' << 255 << "\n";
else if (iteration < max_iter) ss << iteration * 255 << ' ' << iteration * 20 << ' ' << iteration * 5 << "\n";
else ss << 0 << ' ' << 0 << ' ' << 0 << "\n";
}
}
printf("\n sizeof = %d\n", ss.str().length());
MPI_Send(ss.str().c_str(), ss.str().length(), MPI_CHAR, 0, 0, MPI_COMM_WORLD);
}
MPI_Finalize();
return 0;
}
Code compilation:
$ mpic++ -std=c++0x mpi.mandel.cpp -o mpi.mandel
Running with 3 processes (process main + process rank 1 and 2)
$ mpirun -np 3 ./mpi.mandel
Resulting ppm pictures when running with 3, 4, and 5 process:
It seems that the point-to-point communication of sending-receiving is mixing the results when more than 3 processes try to send the MPI_CHAR elements to the main process. How can avoid this behavior?
It works when creating the buffer buff with the same length as the receiving message:
.
.
for (int i=1; i < numprocs; i++) {
MPI_Probe(i, 0, MPI_COMM_WORLD, &stat);
int length;
MPI_Get_count(&stat, MPI_CHAR, &length);
printf("\nfrom %d <<-- %d (stat.source=%d) Receiving %d chars\n", myid, i, stat.MPI_SOURCE, length);
char buff[length + 1];
MPI_Recv(buff, length, MPI_CHAR, i, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
buff[length] = '\0';
image << buff;
}
.
.
Thus, we don't need anymore the declaration at the beginning int buff_size = 404270; neither char buff[buff_size];
I made a prim's algorithm but whenever i try to use the code it give me the same matrix back. In general it isn't minimizing. Can anyone check the code and let me know why it isn't minimizing my matrix
#include <iostream>
#include <stdlib.h>
#include <time.h>
#include <vector>
#include <list>
#include <cstdlib>
#include <iomanip>
#include <limits.h>
int minKey(int n,int key[], bool mst[])
{
// Initialize min value
int min = INT_MAX, min_index;
for (int i = 0; i < n; i++)
if (mst[i] == false && key[i] < min)
min = key[i], min_index = i;
return min_index;
}
void print(int n,int **matrix)
{
for(int i=0; i<n; i++)
{
for(int j=0; j<n; j++) // print the matrix
{
cout << setw(2) << matrix[i][j] << " ";
}
cout << endl;
}
}
int **gen_random_graph(int n)
{
srand(time(0));
int **adj_matrix = new int*[n];
for(int i = 0; i < n; i++)
{
for (int j = i; j < n; j++) //generating a N x N matrix based on the # of vertex input
{
adj_matrix[i] = new int[n];
}
}
for(int u = 0; u < n; u++)
{
for (int v = u; v < n; v++) //decide whether it has an edge or not
{
bool edgeOrNot = rand() % 2;
adj_matrix[u][v] = adj_matrix[v][u] = edgeOrNot;
cout << u << " " << v << " " << adj_matrix[u][v] << endl;
if(adj_matrix[u][v] == true)
{
adj_matrix[v][u] = true;
if(u == v) //We can't have i = j in an undirected graph
{
adj_matrix[u][v] = -1;
}
cout << u << " " << v << " " << adj_matrix[u][v] << endl;
}
else
{
adj_matrix[v][u] = adj_matrix[u][v] = -1;
cout << u << " " << v << " " << adj_matrix[u][v] << "else" << endl;
}
}
}
for(int i = 0; i < n; i++)
{
for(int j = i; j < n; j++) //create the N x N with edges and sets the weight between the edge randomly
{
if(adj_matrix[i][j] == true)
{
int weight = rand() % 10 + 1;
adj_matrix[i][j] = adj_matrix[j][i] = weight;
cout << " ( " << i << "," << j << " ) " << "weight: " << adj_matrix[i][j] << endl;
}
}
}
print(n,adj_matrix);
return (adj_matrix);
}
void solve_mst_prim_matrix(int n, int **matrix)
{
int parent[n]; // Array to store constructed MST
int key[n]; // Key values used to pick minimum weight edge in cut
bool mstSet[n]; // To represent set of vertices not yet included in MST
// Initialize all keys as INFINITE
for (int i = 0; i < n; i++)
{
key[i] = INT_MAX, mstSet[i] = false;
}
// Always include first 1st vertex in MST.
key[0] = 0; // Make key 0 so that this vertex is picked as first vertex
parent[0] = -1; // First node is always root of MST
// The MST will have n vertices
for (int count = 0; count < n-1; count++)
{
// Pick the minimum key vertex from the set of vertices
// not yet included in MST
int u = minKey(n,key, mstSet);
// Add the picked vertex to the MST Set
mstSet[u] = true;
// Update key value and parent index of the adjacent vertices of
// the picked vertex. Consider only those vertices which are not yet
// included in MST
for (int v = 0; v < n; v++)
// matrix[u][v] is non zero only for adjacent vertices of m
// mstSet[v] is false for vertices not yet included in MST
// Update the key only if matrix[u][v] is smaller than key[v]
if (matrix[u][v] && mstSet[v] == false && matrix[u][v] < key[v])
parent[v] = u, key[v] = matrix[u][v];
}
cout << endl;
print(n,matrix);
}
int main()
{
int N;
cout << "Enter number of vertices" << endl;
cin >> N;
int **matrix = gen_random_graph(N);
solve_mst_prim_matrix(N, matrix);
return 0;
}
Correct me if I'm wrong, but after reading your code, you did not even change any value of **matrix in your solve_mst_prim_matrix function. So it basically prints the same thing..
What is wrong with my code ? I have the error like this.
Unhandled exception at 0x00d21673 in mnozenie_macierzy.exe : 0xC0000005: Access violation writing location 0xcdcdcdcd.
It create the first array and the half to the second. The program multiplies arrays.
Sorry for my English if It isn't correct. I hope you understand me.
#include <iostream>
#include <time.h>
using namespace std;
void losowa_tablica(int **tab1, int **tab2, int a, int b, int c, int d)
{
int i, j;
for(i=0; i<a; i++)
{
cout << endl;
for(j=0; j<b; j++)
{
tab1[i][j]=rand();
cout << "tab1[" << i << "][" << j << "] : \t" << tab1[i][j] << "\t";
}
}
cout << endl;
for(i=0; i<c; i++)
{
cout << endl;
for(j=0; j<d; j++)
{
tab2[i][j]=rand();
cout << "tab2[" << i << "][" << j << "] : \t" << tab2[i][j] << "\t";
}
}
cout << endl << endl;
}
int **mnozenie(int **tab1, int **tab2, int a, int b, int c, int d)
{
int g, suma, i, j;
int **mac=new int*[a];
for(int i=0; i<d; i++)
mac[i]=new int[d];
for(i=0; i<a; i++)
for(j=0; j<d; j++)
{
g=b-1, suma=0;
do
{
suma+=tab1[i][g]*tab2[g][j];
g--;
}while(g!=0);
mac[i][j]=suma;
}
return mac;
}
int main()
{
int a,b,c,d;
cout << "Podaj liczbe wierszy pierwszej macierzy: " << endl;
cin >> a;
cout << "Podaj liczbe kolumn pierwszej macierzy: " << endl;
cin >> b;
cout << "Podaj liczbe wierszy drugiej macierzy: " << endl;
cin >> c;
cout << "Podaj liczbe kolumn drugiej macierzy: " << endl;
cin >> d;
int **tab1=new int*[a];
for(int i=0; i<b; i++)
tab1[i]=new int[b];
int **tab2=new int*[c];
for(int i=0; i<d; i++)
tab2[i]=new int[d];
losowa_tablica(tab1, tab2, a, b, c, d);
if ( b==c )
{
cout << "Mnozenie wykonalne" << endl;
int **mno=mnozenie(tab1, tab2, a, b, c, d);
}
else cout << "Mnozenie niewykonalne" << endl;
system("pause");
}
Your code yields undefined behavior:
int **tab1=new int*[a]; // allocating an array of 'a' elements
for(int i=0; i<b; i++) // if b > a then the next line will eventually yield UB
tab1[i]=new int[b];
int **tab2=new int*[c]; // allocating an array of 'c' elements
for(int i=0; i<d; i++) // if d > c then the next line will eventually yield UB
tab2[i]=new int[d];
int **mac=new int*[a]; // allocating an array of 'a' elements
for(int i=0; i<d; i++) // if d > a then the next line will eventually yield UB
mac[i]=new int[d];
In practice, the above code will most likely perform a memory access violation at some point.