Private, firstprivate and lastprivate clauses

Check the following program:

#include <stdio.h>
#include <unistd.h>
#include <omp.h>

int main(void)
{    
    int var = 100;

    printf("Before parallelism, var's address is %p\n", &var);

    #pragma omp parallel for
    for (int i = 0; i < 4; i++)
    {
        printf("var's address in thread %d is %p\n", omp_get_thread_num(), &var);
        var = i;
    }

    printf("After parallelism, var's address is %p, and value is %d\n", &var, var);
    return 0;
}

Build and run it:

# gcc -fopenmp parallel.c
# ./a.out
Before parallelism, var's address is 0x7ffe993d79fc
var's address in thread 1 is 0x7ffe993d79fc
var's address in thread 2 is 0x7ffe993d79fc
var's address in thread 3 is 0x7ffe993d79fc
var's address in thread 0 is 0x7ffe993d79fc
After parallelism, var's address is 0x7ffe993d79fc, and value is 0
# ./a.out
Before parallelism, var's address is 0x7ffc13c7cd8c
var's address in thread 3 is 0x7ffc13c7cd8c
var's address in thread 2 is 0x7ffc13c7cd8c
var's address in thread 0 is 0x7ffc13c7cd8c
var's address in thread 1 is 0x7ffc13c7cd8c
After parallelism, var's address is 0x7ffc13c7cd8c, and value is 1
# ./a.out
Before parallelism, var's address is 0x7ffd72be3bbc
var's address in thread 3 is 0x7ffd72be3bbc
var's address in thread 2 is 0x7ffd72be3bbc
var's address in thread 0 is 0x7ffd72be3bbc
var's address in thread 1 is 0x7ffd72be3bbc
After parallelism, var's address is 0x7ffd72be3bbc, and value is 1

Every thread prints the same address value of var variable, the var variable is shared in all threads. Because there is no protection of accessing shared variable, the final value of var is undefined.

Use private clause can let every thread get its own copy of shared variable, but the initial value is still unspecified:

#include <stdio.h>
#include <unistd.h>
#include <omp.h>

int main(void)
{    
    int var = 100;

    printf("Before parallelism, var's address is %p\n", &var);

    #pragma omp parallel for private(var)
    for (int i = 0; i < 4; i++)
    {
        printf("var's address in thread %d is %p, value is %d\n", omp_get_thread_num(), &var, var);
        var = i;
    }

    printf("After parallelism, var's address is %p, and value is %d\n", &var, var);
    return 0;
} 

Build and run it:

# gcc -fopenmp parallel.c
# ./a.out
Before parallelism, var's address is 0x7ffde0e4c124
var's address in thread 3 is 0x7f0fe9bf0e20, value is 0
var's address in thread 1 is 0x7f0feabf2e20, value is 0
var's address in thread 2 is 0x7f0fea3f1e20, value is 0
var's address in thread 0 is 0x7ffde0e4c0c0, value is 24
After parallelism, var's address is 0x7ffde0e4c124, and value is 100

From the printed address of var variable in every thread, we can see every thread gets a local copy of var, but the initial value is random. The local var variable in main function is no affected.

firstprivate clause makes a copy from the shared variable:

#include <stdio.h>
#include <unistd.h>
#include <omp.h>

int main(void)
{    
    int var = 100;

    printf("Before parallelism, var's address is %p\n", &var);

    #pragma omp parallel for firstprivate(var)
    for (int i = 0; i < 4; i++)
    {
        printf("var's address in thread %d is %p, value is %d\n", omp_get_thread_num(), &var, var);
        var = i;
    }

    printf("After parallelism, var's address is %p, and value is %d\n", &var, var);
    return 0;
}

Build and run it:

# gcc -fopenmp parallel.c
# ./a.out
Before parallelism, var's address is 0x7ffe3ae46e00
var's address in thread 0 is 0x7ffe3ae46da0, value is 100
var's address in thread 3 is 0x7ff6870ede20, value is 100
var's address in thread 2 is 0x7ff6878eee20, value is 100
var's address in thread 1 is 0x7ff6880efe20, value is 100
After parallelism, var's address is 0x7ffe3ae46e00, and value is 100

The var in every thread is initialized as 100.

lastprivate clause makes the value in last sequential iteration assigned back to the shared variable:

#include <stdio.h>
#include <unistd.h>
#include <omp.h>

int main(void)
{    
    int var = 100;

    printf("Before parallelism, var's address is %p\n", &var);

    #pragma omp parallel for lastprivate(var)
    for (int i = 0; i < 4; i++)
    {
        printf("var's address in thread %d is %p, value is %d\n", omp_get_thread_num(), &var, var);
        var = i;
    }

    printf("After parallelism, var's address is %p, and value is %d\n", &var, var);
    return 0;
}

Build and run it:

# gcc -fopenmp parallel.c
# ./a.out
Before parallelism, var's address is 0x7ffd53ea3fb0
var's address in thread 0 is 0x7ffd53ea3f50, value is 24
var's address in thread 3 is 0x7fd70584fe20, value is 0
var's address in thread 2 is 0x7fd706050e20, value is 0
var's address in thread 1 is 0x7fd706851e20, value is 0
After parallelism, var's address is 0x7ffd53ea3fb0, and value is 3
# ./a.out
Before parallelism, var's address is 0x7fff72604380
var's address in thread 0 is 0x7fff72604320, value is 24
var's address in thread 2 is 0x7f6f3ac49e20, value is 0
var's address in thread 1 is 0x7f6f3b44ae20, value is 0
var's address in thread 3 is 0x7f6f3a448e20, value is 0
After parallelism, var's address is 0x7fff72604380, and value is 3

Since in following loop:

for (int i = 0; i < 4; i++)
{
    ......
}

i = 3 is the last sequential iteration, the shared variable var is always updated to 3.