Consolidation with CDI-PIO (develop)

Generally, this is !11 (merged) rebased onto the develop branch.

This is the first develop-based version accepted by the Buildbot of ICON: https://buildbot.dkrz.de/grid?branch=m300488/cdi-pio-2-integration

tests/deco2d_model.c

@@ -55,10 +55,6 @@ modelRegionCompute(double region[], int nlev, int nlat, int nlon, const int chun
                         (double) ((j + js + k + ks) % nlat) / (double) (nlat - 1), mscale, mrscale);
 }
 
-#ifdef USE_MPI
-static void findPartition2D(int npart[2], int num_parts);
-#endif
-
 void
 modelRun(struct model_config setup, MPI_Comm comm)
 {
@@ -376,62 +372,6 @@ modelRun(struct model_config setup, MPI_Comm comm)
   Free(filename);
 }
 
-#ifdef USE_MPI
-static void
-findPartition2D(int npart[2], int num_parts)
-{
-  const uint64_t rscale = 256;
-  uint32_t *factors = NULL;
-  xassert(num_parts > 0);
-  int numFactors = PPM_prime_factorization_32((uint32_t) num_parts, &factors);
-  /* try to distribute prime factors on dimensions to get
-   * approx. 2 times as many parts in x dim than y dim */
-  const uint64_t optimumRatio = rscale * 2;
-  npart[0] = num_parts, npart[1] = 1;
-  uint_fast32_t npart_attempt[2];
-  uint64_t bestRatio = (uint64_t) num_parts * rscale, bestDiff = (uint64_t) llabs((long long) (bestRatio - optimumRatio));
-  /* test all assignments of factors to dimensions, starting with
-   * only one assigned to x dim (omitting 0 because that would
-   * always give npart[1] > npart[0] */
-  for (int assign2X = 1; assign2X <= numFactors; ++assign2X)
-    {
-      uint_fast32_t pattern = (UINT32_C(1) << assign2X) - 1, lastPattern = pattern << (numFactors - assign2X);
-      do
-        {
-          npart_attempt[0] = 1;
-          npart_attempt[1] = 1;
-          /* loop over all factors */
-          for (uint_fast32_t i = 0; i < (uint_fast32_t) numFactors; ++i)
-            {
-              uint_fast32_t dim_idx = (pattern >> i) & 1;
-              npart_attempt[dim_idx] *= factors[i];
-            }
-          uint64_t ratio = ((uint64_t) npart_attempt[0] * rscale) / (uint64_t) npart_attempt[1];
-          uint64_t diff = (uint64_t) llabs((long long) (ratio - optimumRatio));
-          if (diff < bestDiff)
-            {
-              npart[0] = (int) npart_attempt[0];
-              npart[1] = (int) npart_attempt[1];
-              bestDiff = diff;
-              bestRatio = ratio;
-            }
-          {
-            uint_fast32_t t;
-#if HAVE_DECL___BUILTIN_CTZ
-            t = pattern | (pattern - 1);
-            pattern = (t + 1) | (((~t & -~t) - 1) >> (__builtin_ctz((unsigned) pattern) + 1));
-#else
-            t = (pattern | (pattern - 1)) + 1;
-            pattern = t | ((((t & -t) / (pattern & -pattern)) >> 1) - 1);
-#endif
-          }
-        }
-      while (pattern <= lastPattern);
-    }
-  Free(factors);
-}
-#endif
-
 /*
  * Local Variables:
  * c-file-style: "Java"