Wrong answer with omp simd directive and c++ class

[khsa1]

Hi, I am tried to add the "#pragma omp simd" directive to a loop but get an incorrect answer. I have a simple reproducer below:

#include <iostream>

class Matrix {
public:
    double *data = nullptr;
    Matrix() {}
};


int main(int argc,char *argv[])
{
    int rows = 3, cols = 3;
    Matrix m;
    m.data = new double[rows*cols];

    for(int i = 0; i < rows*cols; i++) m.data[i] = 1.0;
#pragma omp simd collapse(2)
    for(int j = 0; j < cols; j++)
        for(int i = 0; i < rows; i++)
            m.data[j*cols+i] = 2.0;

    for(int i = 0; i < rows*cols; i++)
        std::cout << "m[" << i << "] = " << m.data[i] << std::endl;

    return 0;
}

Which I build with:

icpc -qopenmp reproducer.cpp

This gives the following output:

m[0] = 2
m[1] = 2
m[2] = 2
m[3] = 1
m[4] = 1
m[5] = 1
m[6] = 1
m[7] = 1
m[8] = 2

I found that if I remove the simd directive on line 17 or the constructor on line 6 then I get the expected result of 2 for all elements of m.

Does anyone have an idea of why I get incorrect results with my code? Am I doing something wrong or missing any necessary directives/declarations?

Thanks,
Sam

[SantoshY_Intel · Accepted solution]

Hi,

Thanks for reaching out to us.

We were able to reproduce your issue on a Linux machine using the Intel classic C++ compiler(icpc) as shown in the below screenshot:

But, we can see from the below screenshots that it is working fine with the latest Intel oneAPI C++ compilers(icpx & dpcpp).

As it is working fine with Intel oneAPI C++ compilers, could you please try compiling with icpx or dpcpp compiler instead of Intel classic C++ compiler(icpc)?

Also, Intel Classic Compiler will enter "Legacy Product Support" mode, signaling the end of regular updates. Please refer to the article below for more details.

https://www.intel.com/content/www/us/en/developer/articles/technical/adoption-of-llvm-complete-icx.html

Thanks & Regards,

Santosh

[khsa1]

Thank you, that fixed that issue but I am still getting a wrong answer in my application code. I've narrowed down the problem to another issue for which I have created a reproducer.

There are 3 files:

1. main.cpp:

#include "helper.hpp"

#pragma omp declare simd
void increment_n_samefile(int *Count_N) {
    Count_N[0]++;
}

int main(int argc,char *argv[])
{
    int rows = 10, cols = 10;
    int Count_N[2] = {0};

#pragma omp simd collapse(2)
    for(int j = 0; j < cols; j++)
        for(int i = 0; i < rows; i++)
            increment_n_samefile(Count_N);

#pragma omp simd collapse(2)
    for(int j = 0; j < cols; j++)
        for(int i = 0; i < rows; i++)
            increment_n_otherfile(Count_N);


    printf("Count_N: {%d, %d}\n", Count_N[0], Count_N[1]);

    return 0;
}

2. helper.cpp:

#include "helper.hpp"

#pragma omp declare simd
void increment_n_otherfile(int *Count_N) {
    Count_N[1]++;
}

3. helper.hpp:

#include <omp.h>
#include <time.h>
#include <syslog.h>
#include <unistd.h>
#include <iostream>

#pragma omp declare simd
void increment_n_otherfile(int *Count_N);

I build with:

icpx -c -O2 --std=c++11 -qopenmp -g   main.cpp -o main.o
icpx -c -O2 --std=c++11 -qopenmp -g   helper.cpp -o helper.o
icpx -O2 --std=c++11 -qopenmp -g  -o miniapp.exe main.o helper.o

 which produces the following output:

Count_N: {100, 50}

I would expect 100 for both elements of the array, but the increment that is defined in a separate file is only incremented 50 times. Any idea why this is happening? Also, will this even work in a more complex code where the variable is incremented inside a conditional or should I expect race conditions when multiple iterations are executed at once in a vectorized code?

Thank you again for your help,

Sam

[SantoshY_Intel]

Hi,

Thanks for providing the reproducer code samples and the steps to reproduce your issue.

We can expect race conditions when multiple iterations are executed at once in a vectorized code.

So, please use the "#pragma omp atomic" in your helper.cpp code as shown below:

#include "helper.hpp"
#pragma omp declare simd
void increment_n_otherfile(int *Count_N) {
#pragma omp atomic
    Count_N[1]++;
}

Now running the program using the icpx compiler gives you the correct output as shown in the below screenshot.

If this resolves your issue, make sure to accept this as a solution. This would help others with a similar issue.

Thanks & Regards,

Santosh

[khsa1]

Thanks for you help.

-Sam

[SantoshY_Intel]

Hi,

Thanks for accepting our solution. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Thanks & Regards,

Santosh