[svn-r27197] Improve performance of "printf" VDS with selections in the file space - the

library no longer iterates over all source datasets in the printf mapping, it
now only looks at the ones that could be involved in I/O (the ones whose bounds
overlap with the selection in the unlimited dimension).

Tested: ummon

4 files changed, 97 insertions, 20 deletions

diff --git a/src/H5Dvirtual.c b/src/H5Dvirtual.c
index bfd4e0c..f6a16ee 100644
--- a/src/H5Dvirtual.c
+++ b/src/H5Dvirtual.c
@@ -1493,6 +1493,10 @@ H5D__virtual_pre_io(H5D_io_info_t *io_info,
     const H5S_t *mem_space, hsize_t *tot_nelmts)
 {
     hssize_t    select_nelmts;              /* Number of elements in selection */
+    hsize_t     bounds_start[H5S_MAX_RANK]; /* Selection bounds start */
+    hsize_t     bounds_end[H5S_MAX_RANK];   /* Selection bounds end */
+    int         rank;
+    hbool_t     bounds_init = FALSE;        /* Whether bounds_start, bounds_end, and rank are valid */
     size_t      i, j;                       /* Local index variables */
     herr_t      ret_value = SUCCEED;        /* Return value */
 
@@ -1519,12 +1523,44 @@ H5D__virtual_pre_io(H5D_io_info_t *io_info,
         /* Check for "printf" source dataset resolution */
         if(storage->list[i].parsed_source_file_name
                 || storage->list[i].parsed_source_dset_name) {
+            hbool_t partial_block;
+
+            HDassert(storage->list[i].unlim_dim_virtual >= 0);
+
+            /* Get selection bounds if necessary */
+            if(!bounds_init) {
+                /* Get rank of VDS */
+                if((rank = H5S_GET_EXTENT_NDIMS(io_info->dset->shared->space)) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "unable to get number of dimensions")
+
+                /* Get selection bounds */
+                if(H5S_SELECT_BOUNDS(file_space, bounds_start, bounds_end) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "unable to get selection bounds")
+
+                /* Adjust bounds_end to represent the extent just enclosing them
+                 * (add 1) */
+                for(j = 0; j < (size_t)rank; j++)
+                    bounds_end[j]++;
+
+                /* Bounds are now initialized */
+                bounds_init = TRUE;
+            } /* end if */
+
+            /* Get index of first block in virtual selection */
+            storage->list[i].sub_dset_io_start = (size_t)H5S_hyper_get_first_inc_block(storage->list[i].source_dset.virtual_select, bounds_start[storage->list[i].unlim_dim_virtual], NULL);
+
+            /* Get index of first block outside of virtual selection */
+            storage->list[i].sub_dset_io_end = (size_t)H5S_hyper_get_first_inc_block(storage->list[i].source_dset.virtual_select, bounds_end[storage->list[i].unlim_dim_virtual], &partial_block);
+            if(partial_block)
+                storage->list[i].sub_dset_io_end++;
+            if(storage->list[i].sub_dset_io_end > storage->list[i].sub_dset_nused)
+                storage->list[i].sub_dset_io_end = storage->list[i].sub_dset_nused;
+
             /* Iterate over sub-source dsets */
-            for(j = 0; j < storage->list[i].sub_dset_nused; j++) {
-                /* Check for no clipped selection.  This should be an assertion
-                 * once we have I/O scoping for printf working VDSINC */
-                if(!storage->list[i].sub_dset[j].clipped_virtual_select)
-                    break;
+            for(j = storage->list[i].sub_dset_io_start;
+                    j < storage->list[i].sub_dset_io_end; j++) {
+                /* There should always be a clipped virtual selection here */
+                HDassert(storage->list[i].sub_dset[j].clipped_virtual_select);
 
                 /* Project intersection of file space and mapping virtual space
                  * onto memory space */
@@ -1634,7 +1670,8 @@ H5D__virtual_post_io(H5O_storage_virtual_t *storage)
         if(storage->list[i].parsed_source_file_name
                 || storage->list[i].parsed_source_dset_name) {
             /* Iterate over sub-source dsets */
-            for(j = 0; j < storage->list[i].sub_dset_nused; j++)
+            for(j = storage->list[i].sub_dset_io_start;
+                    j < storage->list[i].sub_dset_io_end; j++)
                 /* Close projected memory space */
                 if(storage->list[i].sub_dset[j].projected_mem_space) {
                     if(H5S_close(storage->list[i].sub_dset[j].projected_mem_space) < 0)
@@ -1761,7 +1798,8 @@ H5D__virtual_read(H5D_io_info_t *io_info, const H5D_type_info_t *type_info,
         if(storage->list[i].parsed_source_file_name
                 || storage->list[i].parsed_source_dset_name) {
             /* Iterate over sub-source dsets */
-            for(j = 0; j < storage->list[i].sub_dset_nused; j++)
+            for(j = storage->list[i].sub_dset_io_start;
+                    j < storage->list[i].sub_dset_io_end; j++)
                 if(H5D__virtual_read_one(io_info, type_info, file_space, &storage->list[i].sub_dset[j]) < 0)
                     HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "unable to read source dataset")
         } /* end if */
@@ -1785,7 +1823,8 @@ H5D__virtual_read(H5D_io_info_t *io_info, const H5D_type_info_t *type_info,
             if(storage->list[i].parsed_source_file_name
                     || storage->list[i].parsed_source_dset_name) {
                 /* Iterate over sub-source dsets */
-                for(j = 0; j < storage->list[i].sub_dset_nused; j++)
+                for(j = storage->list[i].sub_dset_io_start;
+                        j < storage->list[i].sub_dset_io_end; j++)
                     if(storage->list[i].sub_dset[j].projected_mem_space)
                         if(H5S_select_subtract(fill_space, storage->list[i].sub_dset[j].projected_mem_space) < 0)
                             HGOTO_ERROR(H5E_DATASET, H5E_CANTCLIP, FAIL, "unable to clip fill selection")
@@ -1942,7 +1981,8 @@ H5D__virtual_write(H5D_io_info_t *io_info, const H5D_type_info_t *type_info,
         if(storage->list[i].parsed_source_file_name
                 || storage->list[i].parsed_source_dset_name) {
             /* Iterate over sub-source dsets */
-            for(j = 0; j < storage->list[i].sub_dset_nused; j++) {
+            for(j = storage->list[i].sub_dset_io_start;
+                    j < storage->list[i].sub_dset_io_end; j++) {
                 HDassert(0 && "Checking code coverage..."); //VDSINC
                 if(H5D__virtual_write_one(io_info, type_info, file_space, &storage->list[i].sub_dset[j]) < 0)
                     HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "unable to write to source dataset")
diff --git a/src/H5Oprivate.h b/src/H5Oprivate.h
index 604dd73..8462382 100644
--- a/src/H5Oprivate.h
+++ b/src/H5Oprivate.h
@@ -449,6 +449,8 @@ typedef struct H5O_storage_virtual_ent_t {
     H5O_storage_virtual_srcdset_t *sub_dset; /* Array of sub-source dataset info structs */
     size_t sub_dset_nalloc;             /* Number of slots allocated in sub_dset */
     size_t sub_dset_nused;              /* Number of slots "used" in sub_dset - essentially the farthest sub dataset in the extent */
+    size_t sub_dset_io_start;           /* First element in sub_dset involved in current I/O op.  Field has no meaning and may be uninitialized at all other times */
+    size_t sub_dset_io_end;             /* First element in sub_dset outside of current I/O op.  Field has no meaning and may be uninitialized at all other times */
     H5O_storage_virtual_name_seg_t *parsed_source_file_name; /* Parsed version of source_dset.file_name */
     size_t psfn_static_strlen;          /* Length of parsed_source_file_name without block number substitutions */
     size_t psfn_nsubs;                  /* Number of block number substitutions in parsed_source_file_name */
diff --git a/src/H5Shyper.c b/src/H5Shyper.c
index db81211..fb34981 100644
--- a/src/H5Shyper.c
+++ b/src/H5Shyper.c
@@ -10180,25 +10180,27 @@ H5S_hyper_get_first_inc_block(const H5S_t *space, hsize_t clip_size,
     HDassert(hslab);
     HDassert(hslab->unlim_dim >= 0);
     HDassert(hslab->opt_unlim_diminfo[hslab->unlim_dim].count == H5S_UNLIMITED);
-    HDassert(partial);
 
     diminfo = &hslab->opt_unlim_diminfo[hslab->unlim_dim];
 
     /* Check for selection outside of clip_size */
     if(diminfo->start >= clip_size) {
         ret_value = 0;
-        partial = FALSE;
+        if(partial)
+            partial = FALSE;
     } /* end if */
     else {
         /* Calculate index of first incomplete block */
         ret_value = (clip_size - diminfo->start + diminfo->stride
                 - diminfo->block) / diminfo->stride;
 
-        /* Check for partial block */
-        if((diminfo->stride * ret_value) < (clip_size - diminfo->start))
-            *partial = TRUE;
-        else
-            *partial = FALSE;
+        if(partial) {
+            /* Check for partial block */
+            if((diminfo->stride * ret_value) < (clip_size - diminfo->start))
+                *partial = TRUE;
+            else
+                *partial = FALSE;
+        } /* end if */
     } /* end else */
 
     FUNC_LEAVE_NOAPI(ret_value)
diff --git a/test/vds.c b/test/vds.c
index 5173083..08df7ef 100644
--- a/test/vds.c
+++ b/test/vds.c
@@ -5515,10 +5515,6 @@ test_printf(unsigned config, hid_t fapl)
     if(mdims[1] != 20)
         TEST_ERROR
 
-    /* Close filespace */
-    if(H5Sclose(filespace) < 0)
-        TEST_ERROR
-
     /* Read data through virtual dataset */
     /* Reset rbuf */
     HDmemset(rbuf[0], 0, sizeof(rbuf));
@@ -5543,6 +5539,43 @@ test_printf(unsigned config, hid_t fapl)
                     TEST_ERROR
         } /* end for */
 
+    /* Now try with different selections */
+    count[0] = 10;
+    for(start[1] = (hsize_t)0; start[1] < (hsize_t)5; start[1]++)
+        for(count[1] = (hsize_t)1; count[1] < (hsize_t)11; count[1]++) {
+            /* Reset rbuf */
+            HDmemset(rbuf[0], 0, sizeof(rbuf));
+
+            /* Select hyperslab in memory space */
+            if(H5Sselect_hyperslab(memspace, H5S_SELECT_SET, start, NULL, count, NULL) < 0)
+                TEST_ERROR
+
+            /* Select hyperslab in file space */
+            if(H5Sselect_hyperslab(filespace, H5S_SELECT_SET, start, NULL, count, NULL) < 0)
+                TEST_ERROR
+
+            /* Read data */
+            if(H5Dread(vdset, H5T_NATIVE_INT, memspace, filespace, H5P_DEFAULT, rbuf[0]) < 0)
+                TEST_ERROR
+
+            /* Verify read data */
+            for(i = 0; i < (int)mdims[0]; i++)
+                for(j = 0; j < (int)mdims[1]; j++) {
+                    if((j < (int)start[1]) || (j >= (int)(start[1] + count[1]))) {
+                        if(rbuf[i][j] != 0)
+                            TEST_ERROR
+                    } /* end if */
+                    else
+                        if(rbuf[i][j] != erbuf[i][j])
+                            TEST_ERROR
+                } /* end for */
+        } /* end for */
+    start[1] = 0;
+
+    /* Close filespace */
+    if(H5Sclose(filespace) < 0)
+        TEST_ERROR
+
     /* Close */
     if(!(config & TEST_IO_CLOSE_SRC)) {
         if(H5Dclose(srcdset[0]) < 0)