Add threshold for r-tree usage

Author: mattjala

doxygen/examples/tables/propertyLists.dox

@@ -618,7 +618,7 @@ encoding for object names.</td>
 </tr>
 <tr>
 <td>#H5Pset_dset_use_spatial_tree/#H5Pget_dset_use_spatial_tree</td>
-<td>Sets/gets the flag to use spatial trees when searching VDS mappings</td>
+<td>Sets/gets the flag to use spatial trees when searching many VDS mappings</td>
 </tr>
 </table>
 //! [dapl_table]

fortran/src/H5Pff.F90

@@ -4404,7 +4404,7 @@ END SUBROUTINE h5pget_dset_use_spatial_tree_f
 !! \ingroup FH5P
 !!
 !! \brief Sets the value of the "use spatial tree" flag which enables spatial tree
-!!        construction and usage for VDS mapping searches.
+!!        construction and usage for large VDS mapping searches.
 !!
 !! \param dapl_id  Target dataset access property list identifier.
 !! \param use_tree Value of the setting.

release_docs/CHANGELOG.md

@@ -408,10 +408,10 @@ HDF5 release, platforms tested, and known problems in this release.
    to iterate through every source dataset's dataspace and check for an intersection
    with the user-selected region for a read/write in the virtual dataset.
 
-   Virtual datasets now use an r-tree (defined in H5RT.c) to perform a spatial search.
-   This allows the dataspaces that intersect the user-selection to be computed with,
-   in most cases, vastly fewer costly intersection checks, improving the speed of VDS
-   read/write operations.
+   Virtual datasets with many mappings now use an r-tree (defined in H5RT.c) to
+   perform a spatial search. This allows the dataspaces that intersect the
+   user-selection to be computed with, in most cases, much fewer intersection checks,
+   improving the speed of VDS read/write operations.
 
    Virtual datasets will use the r-tree by default, since the majority of use cases,
    should see improvements from use of the tree. However, because some workflows may

src/H5Dprivate.h

@@ -125,6 +125,9 @@
 /* Default virtual dataset list size */
 #define H5D_VIRTUAL_DEF_LIST_SIZE 8
 
+/* Threshold for use of a tree for VDS mappings */
+#define H5D_VIRTUAL_TREE_THRESHOLD 50
+
 #ifdef H5D_MODULE
 #define H5D_OBJ_ID(D) (((H5D_obj_create_t *)(D))->dcpl_id)
 #else /* H5D_MODULE */

src/H5Dvirtual.c

@@ -3088,7 +3088,7 @@ H5D__virtual_pre_io(H5D_dset_io_info_t *dset_info, H5O_storage_virtual_t *storag
     if (H5P_get(dapl_plist, H5D_ACS_USE_TREE_NAME, &tree_enabled) < 0)
         HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, FAIL, "can't get virtual use tree flag");
 
-    if (tree_enabled && !storage->tree) {
+    if (tree_enabled && !storage->tree && storage->list_nused >= H5D_VIRTUAL_TREE_THRESHOLD) {
         int rank = 0;
         assert(!storage->is_in_tree);
         /* Get the rank of the dataset */

src/H5RT.c

@@ -335,7 +335,7 @@ H5RT__bulk_load(H5RT_node_t *node, int rank, H5RT_leaf_t *leaves, size_t count,
             assert(child_leaf_start + leaves_left <= leaves + count);
 
             /* Allocate this child node */
-            if (NULL == (node->children.nodes[i] = H5FL_CALLOC(H5RT_node_t)))
+            if (NULL == (node->children.nodes[i] = H5FL_MALLOC(H5RT_node_t)))
                 HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "failed to allocate memory for R-tree node");
 
             child_leaf_count = (leaves_left < slab_size) ? leaves_left : slab_size;
@@ -386,8 +386,7 @@ H5RT_create(int rank, H5RT_leaf_t *leaves, size_t count)
     if (count == 0)
         HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, NULL, "r-tree must have at least one leaf");
 
-    /* TBD: May replace with malloc for optimization */
-    if (NULL == (rtree = H5FL_CALLOC(H5RT_t)))
+    if (NULL == (rtree = H5FL_MALLOC(H5RT_t)))
         HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, NULL, "failed to allocate memory for R-tree");
 
     rtree->rank    = rank;
@@ -445,7 +444,7 @@ H5RT__search_recurse(H5RT_node_t *node, int rank, hsize_t min[], hsize_t max[],
 
             if (H5RT__leaves_intersect(rank, min, max, curr_min, curr_max)) {
                 /* We found an intersecting leaf, create a result structure for it */
-                if (NULL == (new_result = H5FL_CALLOC(H5RT_result_t)))
+                if (NULL == (new_result = H5FL_MALLOC(H5RT_result_t)))
                     HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "failed to allocate result structure");
 
                 new_result->leaf = curr_leaf;