Optimize VDS operations with r-tree

doxygen/examples/tables/propertyLists.dox

@@ -616,6 +616,10 @@ encoding for object names.</td>
 <td>#H5Pset_virtual_view/#H5Pget_virtual_view</td>
 <td>Sets/gets the view of the virtual dataset (VDS) to include or exclude missing mapped elements.</td>
 </tr>
+<tr>
+<td>#H5Pset_virtual_spatial_tree/#H5Pget_virtual_spatial_tree</td>
+<td>Sets/gets the flag to use spatial trees when searching many VDS mappings</td>
+</tr>
 </table>
 //! [dapl_table]
  *

fortran/src/H5Pff.F90

@@ -4365,6 +4365,96 @@ END FUNCTION h5pget_chunk_cache_c
 
   END SUBROUTINE h5pget_chunk_cache_f
 
+!>
+!! \ingroup FH5P
+!!
+!! \brief Retrieves the flag for whether to use/not use a spatial tree
+!! during mapping operations on a Virtual Dataset. The default value is true.
+!!
+!! Use of a spatial tree will accelerate the process of searching through mappings
+!! to determine which contain intersections with the user's selection region.
+!! With the tree disabled, all mappings will simply be iterated through and
+!! checked directly.
+!!
+!! Certain workflows may find that tree creation overhead outweighs the time saved
+!! on reads. In this case, disabling this property will lead to a performance improvement,
+!! though it is expected that almost all cases will benefit from the tree on net.
+!!
+!! \param dapl_id  Target dataset access property list identifier.
+!! \param use_tree Value of the setting.
+!! \param hdferr   \fortran_error
+!!
+!! See C API: @ref H5Pget_virtual_spatial_tree()
+!!
+  SUBROUTINE h5pget_virtual_spatial_tree_f(dapl_id, use_tree, hdferr)
+    IMPLICIT NONE
+    INTEGER(HID_T) , INTENT(IN)              :: dapl_id
+    LOGICAL        , INTENT(OUT)             :: use_tree
+    INTEGER        , INTENT(OUT)             :: hdferr
+    LOGICAL(C_BOOL) :: c_use_tree
+
+    INTERFACE
+        INTEGER(C_INT) FUNCTION H5Pget_virtual_spatial_tree_c(dapl_id, use_tree) &
+             BIND(C, NAME='H5Pget_virtual_spatial_tree')
+          IMPORT :: C_INT, HID_T, C_BOOL
+          IMPLICIT NONE
+          INTEGER(HID_T), INTENT(IN), VALUE :: dapl_id
+          LOGICAL(C_BOOL), INTENT(OUT) :: use_tree
+        END FUNCTION H5Pget_virtual_spatial_tree_c
+    END INTERFACE
+
+    hdferr = INT(H5Pget_virtual_spatial_tree_c(dapl_id, c_use_tree))
+
+    ! Transfer value of C C_BOOL type to Fortran LOGICAL
+    use_tree = c_use_tree
+
+    END SUBROUTINE h5pget_virtual_spatial_tree_f
+
+!>
+!! \ingroup FH5P
+!!
+!! \brief Sets the dapl to use/not use a spatial tree
+!! during mapping operations on a Virtual Dataset. The default value is true.
+!!
+!! Use of a spatial tree will accelerate the process of searching through mappings
+!! to determine which contain intersections with the user's selection region.
+!! With the tree disabled, all mappings will simply be iterated through and
+!! checked directly.
+!!
+!! Certain workflows may find that tree creation overhead outweighs the time saved
+!! on reads. In this case, disabling this property will lead to a performance improvement,
+!! though it is expected that almost all cases will benefit from the tree on net.
+!!
+!! \param dapl_id  Target dataset access property list identifier.
+!! \param use_tree Value of the setting.
+!! \param hdferr   \fortran_error
+!!
+!! See C API: @ref H5Pset_virtual_spatial_tree()
+!!
+  SUBROUTINE h5pset_virtual_spatial_tree_f(dapl_id, use_tree, hdferr)
+    IMPLICIT NONE
+    INTEGER(HID_T) , INTENT(IN)              :: dapl_id
+    LOGICAL        , INTENT(IN)              :: use_tree
+    INTEGER        , INTENT(OUT)             :: hdferr
+    LOGICAL(C_BOOL) :: c_use_tree
+
+    INTERFACE
+        INTEGER FUNCTION h5pset_virtual_spatial_tree_c(dapl_id, use_tree) &
+             BIND(C, NAME='H5Pset_virtual_spatial_tree')
+          IMPORT :: HID_T, C_BOOL
+          IMPLICIT NONE
+          INTEGER(HID_T), INTENT(IN), VALUE :: dapl_id
+          LOGICAL(C_BOOL), INTENT(IN), VALUE :: use_tree
+        END FUNCTION h5pset_virtual_spatial_tree_c
+    END INTERFACE
+
+    ! Transfer value of Fortran LOGICAL to C C_BOOL type
+    c_use_tree = use_tree
+
+    hdferr = INT(h5pset_virtual_spatial_tree_c(dapl_id, c_use_tree))
+
+  END SUBROUTINE h5pset_virtual_spatial_tree_f
+
 #ifdef H5_DOXYGEN
 !>
 !! \ingroup FH5P

fortran/src/hdf5_fortrandll.def.in

@@ -420,6 +420,8 @@ H5P_mp_H5PGET_VIRTUAL_VSPACE_F
 H5P_mp_H5PGET_VIRTUAL_SRCSPACE_F
 H5P_mp_H5PGET_VIRTUAL_FILENAME_F
 H5P_mp_H5PGET_VIRTUAL_DSETNAME_F
+H5P_mp_H5PGET_VIRTUAL_SPATIAL_TREE_F
+H5P_mp_H5PSET_VIRTUAL_SPATIAL_TREE_F
 H5P_mp_H5PGET_DSET_NO_ATTRS_HINT_F
 H5P_mp_H5PSET_DSET_NO_ATTRS_HINT_F
 H5P_mp_H5PSET_VOL_F

fortran/test/tH5P.F90

@@ -777,8 +777,10 @@ SUBROUTINE test_misc_properties(total_error)
   INTEGER, INTENT(INOUT) :: total_error
 
   INTEGER(hid_t) :: fapl_id = -1 !  Local fapl
+  INTEGER(hid_t) :: dapl_id = -1 !  Local dapl
   LOGICAL :: use_file_locking           ! (H5Pset/get_file_locking_f)
   LOGICAL :: ignore_disabled_locks      ! (H5Pset/get_file_locking_f)
+  LOGICAL :: use_spatial_tree           ! (H5Pset/get_dset_use_spatial_tree_f)
   INTEGER :: error
 
   ! Create a default fapl
@@ -826,6 +828,39 @@ SUBROUTINE test_misc_properties(total_error)
   CALL H5Pclose_f(fapl_id, error)
   CALL check("H5Pclose_f", error, total_error)
 
+  ! Create a dataset access property list
+  CALL H5Pcreate_f(H5P_DATASET_ACCESS_F, dapl_id, error)
+  CALL check("H5Pcreate_f", error, total_error)
+
+  ! Test H5Pset/get_virtual_spatial_tree_f
+  ! true value
+  use_spatial_tree = .TRUE.
+  CALL h5pset_virtual_spatial_tree_f(dapl_id, use_spatial_tree, error)
+  CALL check("h5pset_virtual_spatial_tree_f", error, total_error)
+  use_spatial_tree = .FALSE.
+  CALL h5pget_virtual_spatial_tree_f(dapl_id, use_spatial_tree, error)
+  CALL check("h5pget_virtual_spatial_tree_f", error, total_error)
+  if(use_spatial_tree .neqv. .TRUE.) then
+    total_error = total_error + 1
+    write(*,*) "Got wrong use_spatial_tree flag from h5pget_virtual_spatial_tree_f"
+  endif
+
+  ! false value
+  use_spatial_tree = .FALSE.
+  CALL h5pset_virtual_spatial_tree_f(dapl_id, use_spatial_tree, error)
+  CALL check("h5pset_virtual_spatial_tree_f", error, total_error)
+  use_spatial_tree = .TRUE.
+  CALL h5pget_virtual_spatial_tree_f(dapl_id, use_spatial_tree, error)
+  CALL check("h5pget_virtual_spatial_tree_f", error, total_error)
+  if(use_spatial_tree .neqv. .FALSE.) then
+    total_error = total_error + 1
+    write(*,*) "Got wrong use_spatial_tree flag from h5pget_virtual_spatial_tree_f"
+  endif
+
+  ! Close the dapl
+  CALL H5Pclose_f(dapl_id, error)
+  CALL check("H5Pclose_f", error, total_error)
+
 END SUBROUTINE test_misc_properties
 
 !-------------------------------------------------------------------------

java/src/hdf/hdf5lib/H5.java

@@ -10675,6 +10675,59 @@ public synchronized static native void H5Pset_virtual_prefix(long dapl_id, Strin
     public synchronized static native void H5Pset_efile_prefix(long dapl_id, String prefix)
         throws HDF5LibraryException, NullPointerException;
 
+    /**
+     * @ingroup JH5P
+     *
+     * H5Pget_virtual_spatial_tree accesses the flag for whether to use/not use a spatial tree
+     * during mapping operations on a Virtual Dataset. The default value is true.
+     *
+     * Use of a spatial tree will accelerate the process of searching through mappings
+     * to determine which contain intersections with the user's selection region.
+     * With the tree disabled, all mappings will simply be iterated through and
+     * checked directly.
+     *
+     * Certain workflows may find that tree creation overhead outweighs the time saved
+     * on reads. In this case, disabling this property will lead to a performance improvement,
+     * though it is expected that almost all cases will benefit from the tree on net.
+     *
+     * @param dapl_id
+     *            IN: Dataset access property list
+     *
+     * @return true if the given dapl is set to use a spatial tree, false if not.
+     *
+     * @exception HDF5LibraryException
+     *            Error from the HDF5 Library.
+     **/
+    public synchronized static native boolean H5Pget_virtual_spatial_tree(long dapl_id)
+        throws HDF5LibraryException;
+
+    /**
+     * @ingroup JH5P
+     *
+     * H5Pset_virtual_spatial_tree sets the dapl to use/not use a spatial tree
+     * during mapping operations on a Virtual Dataset. The default value is true.
+     *
+     * Use of a spatial tree will accelerate the process of searching through mappings
+     * to determine which contain intersections with the user's selection region.
+     * With the tree disabled, all mappings will simply be iterated through and
+     * checked directly.
+     *
+     * Certain workflows may find that tree creation overhead outweighs the time saved
+     * on reads. In this case, disabling this property will lead to a performance improvement,
+     * though it is expected that almost all cases will benefit from the tree on net.
+     *
+     * @param dapl_id
+     *            IN: Dataset access property list
+     *
+     * @param use_tree
+     *            IN: the use_tree flag setting
+     *
+     * @exception HDF5LibraryException
+     *            Error from the HDF5 Library.
+     **/
+    public synchronized static native void H5Pset_virtual_spatial_tree(long dapl_id, boolean use_tree)
+        throws HDF5LibraryException;
+
     // public synchronized static native void H5Pset_append_flush(long plist_id, int ndims, long[] boundary,
     // H5D_append_cb func, H5D_append_t udata) throws HDF5LibraryException;
 

java/src/jni/h5pDAPLImp.c

@@ -312,6 +312,51 @@ H5D_append_cb(hid_t dataset_id, hsize_t *cur_dims, void *cb_data)
     return (herr_t)status;
 } /* end H5D_append_cb */
 
+/*
+ * Class:     hdf_hdf5lib_H5
+ * Method:    H5Pset_virtual_spatial_tree
+ * Signature: (JZ)V
+ */
+JNIEXPORT void JNICALL
+Java_hdf_hdf5lib_H5_H5Pset_1virtual_1spatial_1tree(JNIEnv *env, jclass clss, jlong dapl_id, jboolean use_tree)
+{
+    bool   use_tree_val;
+    herr_t retVal = FAIL;
+
+    UNUSED(clss);
+
+    use_tree_val = (JNI_TRUE == use_tree) ? true : false;
+
+    if ((retVal = H5Pset_virtual_spatial_tree((hid_t)dapl_id, (bool)use_tree_val)) < 0)
+        H5_LIBRARY_ERROR(ENVONLY);
+
+done:
+    return;
+} /* end Java_hdf_hdf5lib_H5_H5Pset_1virtual_1spatial_1tree */
+
+/*
+ * Class:     hdf_hdf5lib_H5
+ * Method:    H5Pget_virtual_spatial_tree
+ * Signature: (J)Z
+ */
+JNIEXPORT jboolean JNICALL
+Java_hdf_hdf5lib_H5_H5Pget_1virtual_1spatial_1tree(JNIEnv *env, jclass clss, jlong dapl_id)
+{
+    bool     use_tree = false;
+    jboolean bval     = JNI_FALSE;
+
+    UNUSED(clss);
+
+    if (H5Pget_virtual_spatial_tree((hid_t)dapl_id, (bool *)&use_tree) < 0)
+        H5_LIBRARY_ERROR(ENVONLY);
+
+    if (use_tree == true)
+        bval = JNI_TRUE;
+
+done:
+    return bval;
+} /* end Java_hdf_hdf5lib_H5_H5Pget_1virtual_1spatial_1tree */
+
 #ifdef __cplusplus
 } /* end extern "C" */
 #endif /* __cplusplus */

java/src/jni/h5pDAPLImp.h

@@ -89,6 +89,20 @@ JNIEXPORT void JNICALL Java_hdf_hdf5lib_H5_H5Pset_1virtual_1printf_1gap(JNIEnv *
  */
 JNIEXPORT jlong JNICALL Java_hdf_hdf5lib_H5_H5Pget_1virtual_1printf_1gap(JNIEnv *, jclass, jlong);
 
+/*
+ * Class:     hdf_hdf5lib_H5
+ * Method:    H5Pset_virtual_spatial_tree
+ * Signature: (JZ)V
+ */
+JNIEXPORT void JNICALL Java_hdf_hdf5lib_H5_H5Pset_1virtual_1spatial_1tree(JNIEnv *, jclass, jlong, jboolean);
+
+/*
+ * Class:     hdf_hdf5lib_H5
+ * Method:    H5Pget_virtual_spatial_tree
+ * Signature: (J)Z
+ */
+JNIEXPORT jboolean JNICALL Java_hdf_hdf5lib_H5_H5Pget_1virtual_1spatial_1tree(JNIEnv *, jclass, jlong);
+
 #ifdef __cplusplus
 } /* end extern "C" */
 #endif /* __cplusplus */

release_docs/CHANGELOG.md

@@ -25,6 +25,7 @@ For releases prior to version 2.0.0, please see the release.txt file and for mor
 
 ## Performance Enhancements:
 
+- Up to [2500% faster](https://github.com/HDFGroup/hdf5/blob/develop/release_docs/CHANGELOG.md#rtree) Virtual Dataset read/write operations
 - [30% faster opening](https://github.com/HDFGroup/hdf5/blob/develop/release_docs/CHANGELOG.md#layoutcopydelay) and [25% faster closing](https://github.com/HDFGroup/hdf5/blob/develop/release_docs/CHANGELOG.md#fileformat) of virtual datasets.
 - [Reduced memory overhead](https://github.com/HDFGroup/hdf5/blob/develop/release_docs/CHANGELOG.md#fileformat) via shared name strings and optimized spatial search algorithms for virtual datasets.
 
@@ -461,6 +462,26 @@ Simple example programs showing how to use complex number datatypes have been ad
 
    This layout copy is now delayed until either a user requests the DCPL, or until the start of an operation that needs to read the layout from the DCPL.
 
+### Virtual datasets now use a spatial tree to optimize searches<a name="rtree"></a>
+
+   Virtual dataset operations with many (>1,000) mappings were much slower than
+   corresponding operations on normal datasets. This was due to the need
+   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 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
+   find that the overhead of the tree outweighs the time saved on searches, there is
+   a new Dataset Access Property List (DAPL) property to control use of the spatial tree.
+
+   This property can be set or queried with the new API functions
+   H5Pset_virtual_spatial_tree()/H5Pget_virtual_spatial_tree().
+
 ## Parallel Library
 
 ### Added H5FDsubfiling_get_file_mapping() API function for subfiling VFD

src/H5Dprivate.h

@@ -54,6 +54,7 @@
 #define H5D_ACS_VDS_PREFIX_NAME           "vds_prefix"           /* VDS file prefix */
 #define H5D_ACS_APPEND_FLUSH_NAME         "append_flush"         /* Append flush actions */
 #define H5D_ACS_EFILE_PREFIX_NAME         "external file prefix" /* External file prefix */
+#define H5D_ACS_USE_TREE_NAME             "tree"                 /* Whether to use spatial tree */
 
 /* ======== Data transfer properties ======== */
 #define H5D_XFER_MAX_TEMP_BUF_NAME          "max_temp_buf"        /* Maximum temp buffer size */
@@ -124,6 +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 */