c++ getline file output change when is copied to string - c++

Question

i have a file that each line of it have a row of matrix and in each row, double numbers placed with space between them, this file include a lot of these matrices with a empty line between them
now i have two different versions of code:
1- single thread read from file with getline(file, readLine) and directly process readLine, split it and use stod to make double numbers and make matrix
#include <fstream>
#include <string>
using namespace std;
void decomposeSerial(double *A, long n)
{
long i, j, k;
for (k = 0; k < n; k++) {
for (j = k + 1; j < n; j++)
A[k*n + j] = A[k*n + j] / A[k*n + k];
for (i = k + 1; i < n; i++)
for (j = k + 1; j < n; j++)
A[i*n + j] = A[i*n + j] - A[i*n + k] * A[k*n + j];
}
}
void main() {
const string inFilePath = ".\\data_in\\file.txt";
const string outFilePath = ".\\data_out\\file.txt";
ifstream inFile(inFilePath);
ofstream outFile(outFilePath);
int n;
int matrixLine = 0;
double * matrix = NULL;
string readLine;
while (getline(inFile, readLine)) {
if (!readLine.empty()) {
if (matrixLine == 0) {
n = 0;
string temp = readLine;
size_t pos = 0;
while ((pos = temp.find(" ")) != string::npos) {
temp.erase(0, pos + 1);
n++;
}
matrix = (double *)malloc(sizeof(double) * n * n);
}
size_t pos = 0;
string token;
int i = 0;
while ((pos = readLine.find(" ")) != string::npos) {
token = readLine.substr(0, pos);
matrix[matrixLine * n + i] = stod(token);
readLine.erase(0, pos + 1);
i++;
}
matrixLine++;
if (matrixLine == n) {
decomposeSerial(matrix, n);
double det = 1;
for (long o = 0; o < n; o++) {
det *= matrix[o * n + o];
}
outFile << det << "\n";
}
}
else {
matrixLine = 0;
}
}
inFile.close();
outFile.close();
}
http://codeshare.io/5enk9x
2- single thread read from file with getline(file, readLine) and append readLine to an element of a string array dedicated for this matrix, and after this, in parallel, each thread get one of these elements and go through the same process to make matrix
#include <fstream>
#include <string>
#include <omp.h>
using namespace std;
double det[1000];
string input[1000];
int ns[1000];
void computation(double* src, int n, int l)
{
long i, j, k;
for (k = 0; k < n; k++) {
for (j = k + 1; j < n; j++)
src[k*n + j] = src[k*n + j] / src[k*n + k];
for (i = k + 1; i < n; i++)
for (j = k + 1; j < n; j++)
src[i*n + j] = src[i*n + j] - src[i*n + k] * src[k*n + j];
}
double res = 1;
for (int j = 0; j < n; j++) {
res *= src[j*n + j];
}
det[l] = res;
}
void main() {
const string inFilePath = ".\\data_in\\file.txt";
const string outFilePath = ".\\data_out\\file.txt";
ifstream inFile(inFilePath);
int matrixCount = 0;
bool inMatrix = false;
string readLine;
int dim = 0;
while (getline(inFile, readLine)) {
dim++;
if (readLine.empty()) {
ns[matrixCount] = dim - 1;
dim = 0;
inMatrix = false;
matrixCount++;
}
else {
if (inMatrix == false) {
inMatrix = true;
input[matrixCount] = readLine;
}
else {
input[matrixCount] += readLine;
}
}
}
ns[matrixCount] = dim;
matrixCount++;
inFile.close();
#pragma omp parallel
{
#pragma omp for schedule(dynamic)
for (int i = 0; i < matrixCount; i++) {
string matrixStr = input[i];
int n = ns[i];
double * matrix = (double *)malloc(sizeof(double) * n * n);
size_t pos = 0;
string token;
int k = 0;
while ((pos = matrixStr.find(" ")) != string::npos) {
token = matrixStr.substr(0, pos);
matrix[k] = stod(token);
matrixStr.erase(0, pos + 1);
k++;
}
computation(matrix, n, i);
free(matrix);
}
}
ofstream outFile(outFilePath);
for (int i = 0; i < matrixCount; i++) {
outFile << det[i] << "\n";
}
outFile.close();
}
http://codeshare.io/ad83Yy
but incredibly, second code work much much slower to make matrices
when i print readLine what comes from getline func with printf("%s", readLine) it prints weird chars, anyway i got that when i append readLine to string array element, these weired chars change on the console and i guess that's why i get slower performance as functions line str.find(" ") or stod(str) work better with first weird ones comparing to second ones
if you think the same, you may suggest a way to prevent char changing in appending

Answer 1

These kinds of performance issues can't be reasoned. You need to use a profiler and measure which parts of the code take up how much time.
To start with I would make both codes more similar. There are a lot of differences that could confuse the issue (for once version 1 has a giant memory leak).