Lecture 10: Interleaved SUMMA Assignment

Scaffold Head

#include <stdlib.h> #include <stdio.h> #include <iostream> #include <string.h> #include <math.h> #include "mpi.h" void readInput(int rows, int cols, float *data){ // Open the file pointer /*FILE* fp = fopen(file.c_str(), "rb"); // Check if the file exists if(fp == NULL){ std::cout << "ERROR: File " << file << " could not be opened" << std::endl; MPI::COMM_WORLD.Abort(1); } for(int i=0; i<rows*cols; i++){ if(!fscanf(fp, "%f", &data[i])){ std::cout << "ERROR: Not enough values in file " << file << std::endl; MPI::COMM_WORLD.Abort(1); } }*/ // checkerboard for(int i=0; i<rows; i++) for(int j=0; j<cols; j++) data[i*cols+j] = (i+j) % 3; } void printOutput(int rows, int cols, float *data){ FILE *fp = fopen("outSUMMA.txt", "wb"); // Check if the file was opened if(fp == NULL){ std::cout << "ERROR: Output file outSUMMA.txt could not be opened" << std::endl; MPI::COMM_WORLD.Abort(1); } for(int i=0; i<rows; i++){ for(int j=0; j<cols; j++) fprintf(fp, "%lf ", data[i*cols+j]); fprintf(fp, "\n"); } fclose(fp); } int main (int argc, char *argv[]){ // Initialize MPI MPI_Init(NULL, NULL); int numP, myId; MPI_Comm_size(MPI_COMM_WORLD, &numP); MPI_Comm_rank(MPI_COMM_WORLD, &myId); int m = 9; //atoi(argv[1]); int k = 9; //atoi(argv[2]); int n = 9; //atoi(argv[3]); int gridDim = sqrt(numP); // Check if a square grid could be created if(gridDim*gridDim != numP){ // Only the first process prints the output message if(!myId) std::cout << "ERROR: The number of processes must be square" << std::endl; MPI_Abort(MPI_COMM_WORLD, 1); } if((m%gridDim) || (n%gridDim) || (k%gridDim)){ // Only the first process prints the output message if(!myId) std::cout << "ERROR: 'm', 'k' and 'n' must be multiple of sqrt(numP)" << std::endl; MPI_Abort(MPI_COMM_WORLD, 1); } if((m < 1) || (n < 1) || (k<1)){ // Only the first process prints the output message if(!myId) std::cout << "ERROR: 'm', 'k' and 'n' must be higher than 0" << std::endl; MPI_Abort(MPI_COMM_WORLD, 1); } float *A; float *B; float *C; // Only one process reads the data from the files and broadcasts the data if(!myId){ A = new float[m*k]; readInput(m, k, A); B = new float[k*n]; readInput(k, n, B); C = new float[m*n]; } // The computation is divided by 2D blocks int blockRowsA = m/gridDim; int blockRowsB = k/gridDim; int blockColsB = n/gridDim; // Create the datatypes of the blocks MPI::Datatype blockAType = MPI::FLOAT.Create_vector(blockRowsA, blockRowsB, k); MPI::Datatype blockBType = MPI::FLOAT.Create_vector(blockRowsB, blockColsB, n); MPI::Datatype blockCType = MPI::FLOAT.Create_vector(blockRowsA, blockColsB, n); blockAType.Commit(); blockBType.Commit(); blockCType.Commit(); float* myA = new float[blockRowsA*blockRowsB]; float* myB = new float[blockRowsB*blockColsB]; float* myC = new float[blockRowsA*blockColsB](); // double buffering float* buffA[2]; float* buffB[2]; buffA[0] = new float[blockRowsA*blockRowsB]; buffA[1] = new float[blockRowsA*blockRowsB]; buffB[0] = new float[blockRowsB*blockColsB]; buffB[1] = new float[blockRowsB*blockColsB]; MPI_Request row_req[2]; MPI_Request col_req[2]; // Measure the current time MPI_Barrier(MPI_COMM_WORLD); double start = MPI::Wtime(); MPI_Request req; // Scatter A and B if(!myId){ for(int i=0; i<gridDim; i++){ for(int j=0; j<gridDim; j++){ MPI_Isend(A+i*blockRowsA*k+j*blockRowsB, 1, blockAType, i*gridDim+j, 0, MPI_COMM_WORLD, &req); MPI_Isend(B+i*blockRowsB*n+j*blockColsB, 1, blockBType, i*gridDim+j, 0, MPI_COMM_WORLD, &req); } } } MPI_Recv(myA, blockRowsA*blockRowsB, MPI::FLOAT, 0, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE); MPI_Recv(myB, blockRowsB*blockColsB, MPI::FLOAT, 0, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE); // Create the communicators MPI::Intercomm rowComm = MPI::COMM_WORLD.Split(myId/gridDim, myId%gridDim); MPI::Intercomm colComm = MPI::COMM_WORLD.Split(myId%gridDim, myId/gridDim);

Scaffold Foot

// Only process 0 writes // Gather the final matrix to the memory of process 0 if(!myId){ for(int i=0; i<blockRowsA; i++) memcpy(&C[i*n], &myC[i*blockColsB], blockColsB*sizeof(float)); for(int i=0; i<gridDim; i++) for(int j=0; j<gridDim; j++) if(i || j) MPI::COMM_WORLD.Recv(&C[i*blockRowsA*n+j*blockColsB], 1, blockCType, i*gridDim+j, 0); } else MPI::COMM_WORLD.Send(myC, blockRowsA*blockColsB, MPI::FLOAT, 0, 0); // Measure the current time double end = MPI::Wtime(); if(!myId){ std::cout << "# Time with " << numP << " processes: " << end-start << " seconds" << std::endl; for (int i = 0; i < m; i++) for (int j = 0; j < n; j++) std::cout << C[i*m+j] << (j+1 == n ? "\n" : " "); std::cout << std::endl; printOutput(m, n, C); delete [] A; delete [] B; delete [] C; } MPI::COMM_WORLD.Barrier(); delete [] myA; delete [] myB; delete [] myC; delete [] buffA[0]; delete [] buffA[1]; delete [] buffB[0]; delete [] buffB[1]; // Terminate MPI MPI::Finalize(); return 0; }

Start time:

Mo 22 Okt 2018 10:51:00

End time:

Mo 01 Apr 2019 12:00:00

General test timeout:

10.0 seconds

Tests

15 6 6 15 6 6 15 6 6
6 15 6 6 15 6 6 15 6
6 6 15 6 6 15 6 6 15
15 6 6 15 6 6 15 6 6
6 15 6 6 15 6 6 15 6
6 6 15 6 6 15 6 6 15
15 6 6 15 6 6 15 6 6
6 15 6 6 15 6 6 15 6
6 6 15 6 6 15 6 6 15