compiler: Add optional pass for runtime stability check

[FabioLuporini]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 87.50000% with 11 lines in your changes are missing coverage. Please review.

Project coverage is 86.66%. Comparing base (32308f4) to head (aaea94a).

Files	Patch %	Lines
devito/passes/iet/errors.py	89.36%	2 Missing and 3 partials ⚠️
devito/operator/operator.py	63.63%	3 Missing and 1 partial ⚠️
devito/core/operator.py	50.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2327   +/-   ##
=======================================
  Coverage   86.66%   86.66%           
=======================================
  Files         229      231    +2     
  Lines       43176    43260   +84     
  Branches     8004     8022   +18     
=======================================
+ Hits        37417    37490   +73     
- Misses       5062     5068    +6     
- Partials      697      702    +5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mloubout]

Maybe finitecheck it is likely someone would use energy since it used for various appplications

[mloubout]

Overkill, just f.sub the Abs(Pow won't change if it's NaN or not

[FabioLuporini]

OK

[FabioLuporini]

OK I've tweaked things

new generated code:

    /* Stability check */
    if ((time)%(100) == 0)
    {
      int check0 = is_finite0(x_M,x_m,y_M,y_m,nthreads,u_vec);
      if (!check0)
      {
        return 100;
      }
    }

and

static int is_finite0(const int x_M, const int x_m, const int y_M, const int y_m, const int nthreads, struct dataobj *restrict u_vec)
{
  float (*restrict u)[u_vec->size[1]][u_vec->size[2]] __attribute__ ((aligned (64))) = (float (*)[u_vec->size[1]][u_vec->size[2]]) u_vec->data;

  float accumulator = 0.0F;
  #pragma omp parallel num_threads(nthreads)
  {
    #pragma omp for collapse(2) schedule(static,1) reduction(+:accumulator)
    for (int x = x_m; x <= x_M; x += 1)
    {
      for (int y = y_m; y <= y_M; y += 1)
      {
        accumulator += u[0][x + 2][y + 2];
      }
    }
  }
  return isfinite(accumulator);
}

[FabioLuporini]

also added another tiny test

[EdCaunt]

Looks good

[mloubout]

Nitpicking: Might be cases where there is no stepping dim and only time dime, but can be for later if someone runs into it (unlikely).