Remove unused variables and code in NPB kernels about dataracebench HOT 3 CLOSED

I have renamed the file name from DRB173-threadprivate-orig-no.c to DRB173-threadprivate3-orig-no.c. We already have DRB085-threadprivate-orig-no.c with the same name and DRB091-threadprivate2-orig-no.c. You might like to do this once you are done with Big table result and create a new PR.

from dataracebench.

As per my understanding of the code, I have simplified as below. There are a few questions, @chunhualiao please comment if my understanding is correct. x is a Threadprivate. That said, if we are using private(i,j,k), it will always result in j and k value as 0. The x[0] = -1.0 only for tid=0. For the rest of the tids it will be 0.0. The final j, k-values are quite indeterministic here. And more of it, it is not a correct execution.

//sample simplification to test - not correct

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

static double x[20];
#pragma omp threadprivate(x)

int main(){
  int i;
  double j,k;

  #pragma omp parallel for default(shared) private(i,j,k)
  for (i = 0; i < 20; i++){
    x[i] = -1.0;
    j = x[0];
    k = i+0.05;
  }
  printf ("%f %f\n", j, k);

  return 0;
}

To have a data race free code, we can have something like below:

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

static double x[20];
#pragma omp threadprivate(x)

int main(){
  int i;
  double j,k;

  #pragma omp parallel for default(shared)
  for (i = 0; i < 20; i++){
    x[i] = -1.0;
    if(omp_get_thread_num()==0){
      j = x[0];
    }
    if(omp_get_thread_num()==0){
      k = i+0.05;
    }
  }

  printf ("%f %f\n", j, k);

  return 0;
}

Please share your thoughts. or else can use copyprivate, but that we already have; one such kernel.

from dataracebench.

We need @Yaying-llnl-summer to write down in the comment area about which variables in the benchmarks are exposing the tools' problems. Based on the information, we can then remove unnecessary variables.

@Yaying-llnl-summer The tests you are adding need

1. better commenting

2. removing unnecessary variables and statements

3. justification about the new ones are not just duplicating existing programs. The new ones exposing tools' new problems.

from dataracebench.