Parallelization of BubbleSort Open-MP C ++

I'm trying to paralyze the bubblesort algorithm using Open-MP and C ++, and the parallelization strategy is the pipe. The following code, to me, makes sense, but does not work, and only sorts the sub-vectors (Input Size) / (N) with N being the number of threads. Would not the "critical" clause cause the threads to be organized and return the ordered array? I tried to use the "atomic" clause instead of "critical", but the compiler returns me the error error: invalid form of ‘#pragma omp atomic’ before ‘;’ token Example : vector 9 4 19 9 5 4 5 6 20 18, with 3 threads, generates vector 4 9 9 4 5 5 19 6 18 20.

#include <iostream>
#include <fstream>
#include <omp.h>
#include <string.h>
#include <stdlib.h>
using namespace std;

int *leArq(int numeros,ifstream &ifs)
{
  int *vetor;
  char caractere[255];  
  if(ifs.good())
  {   
    vetor = new int [numeros];
    for(int i =0;i<numeros;i++)
    {
      ifs>>vetor[i];
      //cout<<vetor[i]<<endl;
    }
  }
  return vetor;
}

void bubbleparalelo(int *a,int tam,int nT)
{   
  int fatia = tam/nT;
  #pragma omp parallel num_threads(nT)
  {
    int indice = omp_get_thread_num();
    int temp;   
    for(int i=0; i<=tam; i++)
    {
      for(int j=indice*fatia; j<(indice+1)*fatia; j++)
      {
        if(a[j]>a[j+1])
        {   
          temp=a[j];
          #pragma omp atomic
            a[j]=a[j+1];
            a[j+1]=temp;
        }
      }
    }
    printf("Thread numero: %d\n\n", indice);
  }
  if (tam <= 30)
  {
    for(int i=0;i<tam;i++)
      printf("Vetor: %d\n", a[i]);
  }
  return;
}

int main(int argc, char **argv)
{   
   int *a;
   int tam;
   ifstream ifs;
   ifs.open(argv[1]);
   int nThreads = atoi(argv[2]);
   ifs>>tam;
   printf("Lendo o arquivo: %s\n", argv[1]);
   a = leArq(tam,ifs);
   bubbleparalelo(a,tam,nThreads);

   return 0;
}