grid: Fix out-of-bounds access in grid_library_counter_add

src/grid/common/grid_library.c

@@ -108,6 +108,7 @@ grid_library_config grid_library_get_config(void) { return config; }
 void grid_library_counter_add(const int lp, const enum grid_backend backend,
                               const enum grid_library_kernel kernel,
                               const int increment) {
+  assert(lp >= 0);
   const int back = backend - GRID_BACKEND_REF;
   const int idx = back * 4 * 20 + kernel * 20 + imin(lp, 19);
   const int ithread = omp_get_thread_num();

src/grid/gpu/grid_gpu_collocate.cu

@@ -145,20 +145,20 @@ void grid_gpu_collocate_one_grid_level(
     const cudaStream_t stream, const double *pab_blocks_dev, double *grid_dev,
     int *lp_diff) {
 
-  const int ntasks = last_task - first_task + 1;
-  if (ntasks == 0) {
-    return; // Nothing to do.
-  }
-
-  init_constant_memory();
-
   // Compute max angular momentum.
   const prepare_ldiffs ldiffs = prepare_get_ldiffs(func);
   *lp_diff = ldiffs.la_max_diff + ldiffs.lb_max_diff; // for reporting stats
   const int la_max = task_list->lmax + ldiffs.la_max_diff;
   const int lb_max = task_list->lmax + ldiffs.lb_max_diff;
   const int lp_max = la_max + lb_max;
 
+  const int ntasks = last_task - first_task + 1;
+  if (ntasks == 0) {
+    return; // Nothing to do and lp_diff already set.
+  }
+
+  init_constant_memory();
+
   // Compute required shared memory.
   // TODO: Currently, cab's indicies run over 0...ncoset[lmax],
   //       however only ncoset(lmin)...ncoset(lmax) are actually needed.

src/grid/gpu/grid_gpu_integrate.cu

@@ -209,13 +209,6 @@ void grid_gpu_integrate_one_grid_level(
     const double *grid_dev, double *hab_blocks_dev, double *forces_dev,
     double *virial_dev, int *lp_diff) {
 
-  const int ntasks = last_task - first_task + 1;
-  if (ntasks == 0) {
-    return; // Nothing to do.
-  }
-
-  init_constant_memory();
-
   // Compute max angular momentum.
   const bool calculate_forces = (forces_dev != NULL);
   const bool calculate_virial = (virial_dev != NULL);
@@ -227,6 +220,13 @@ void grid_gpu_integrate_one_grid_level(
   const int lb_max = task_list->lmax + ldiffs.lb_max_diff;
   const int lp_max = la_max + lb_max;
 
+  const int ntasks = last_task - first_task + 1;
+  if (ntasks == 0) {
+    return; // Nothing to do and lp_diff already set.
+  }
+
+  init_constant_memory();
+
   // Compute required shared memory.
   // TODO: Currently, cab's indicies run over 0...ncoset[lmax],
   //       however only ncoset(lmin)...ncoset(lmax) are actually needed.