Merge pull request #2418 from fortnern/select_io_tconv

Initial implementation of selection I/O with type conversion

src/H5Dchunk.c

@@ -2749,6 +2749,8 @@ H5D__chunk_read(H5D_io_info_t *io_info, H5D_dset_io_info_t *dset_info)
                     io_info->addrs[io_info->pieces_added]         = udata.chunk_block.offset;
                     io_info->element_sizes[io_info->pieces_added] = element_sizes[0];
                     io_info->rbufs[io_info->pieces_added]         = bufs[0];
+                    if (io_info->sel_pieces)
+                        io_info->sel_pieces[io_info->pieces_added] = chunk_info;
                     io_info->pieces_added++;
                 }
             } /* end if */
@@ -3139,6 +3141,8 @@ H5D__chunk_write(H5D_io_info_t *io_info, H5D_dset_io_info_t *dset_info)
                     io_info->addrs[io_info->pieces_added]         = udata.chunk_block.offset;
                     io_info->element_sizes[io_info->pieces_added] = element_sizes[0];
                     io_info->wbufs[io_info->pieces_added]         = bufs[0];
+                    if (io_info->sel_pieces)
+                        io_info->sel_pieces[io_info->pieces_added] = chunk_info;
                     io_info->pieces_added++;
                 }
             } /* end else */

src/H5Dcontig.c

@@ -757,18 +757,15 @@ H5D__contig_may_use_select_io(const H5D_io_info_t *io_info, const H5D_dset_io_in
 
     dataset = dset_info->dset;
 
-    /* Don't use selection I/O if it's globally disabled, if there is a type
-     * conversion, or if it's not a contiguous dataset, or if the sieve buffer
-     * exists (write) or is dirty (read) */
-    if (dset_info->io_ops.single_read != H5D__select_read ||
-        dset_info->layout_ops.readvv != H5D__contig_readvv ||
+    /* Don't use selection I/O if it's globally disabled, if it's not a contiguous dataset, or if the sieve
+     * buffer exists (write) or is dirty (read) */
+    if (dset_info->layout_ops.readvv != H5D__contig_readvv ||
         (op_type == H5D_IO_OP_READ && dataset->shared->cache.contig.sieve_dirty) ||
         (op_type == H5D_IO_OP_WRITE && dataset->shared->cache.contig.sieve_buf))
         ret_value = FALSE;
     else {
         hbool_t page_buf_enabled;
 
-        HDassert(dset_info->io_ops.single_write == H5D__select_write);
         HDassert(dset_info->layout_ops.writevv == H5D__contig_writevv);
 
         /* Check if the page buffer is enabled */
@@ -841,6 +838,8 @@ H5D__contig_read(H5D_io_info_t *io_info, H5D_dset_io_info_t *dinfo)
                 io_info->addrs[io_info->pieces_added]         = dinfo->store->contig.dset_addr;
                 io_info->element_sizes[io_info->pieces_added] = dinfo->type_info.src_type_size;
                 io_info->rbufs[io_info->pieces_added]         = dinfo->buf.vp;
+                if (io_info->sel_pieces)
+                    io_info->sel_pieces[io_info->pieces_added] = dinfo->layout_io_info.contig_piece_info;
                 io_info->pieces_added++;
             }
         }
@@ -911,6 +910,8 @@ H5D__contig_write(H5D_io_info_t *io_info, H5D_dset_io_info_t *dinfo)
                 io_info->addrs[io_info->pieces_added]         = dinfo->store->contig.dset_addr;
                 io_info->element_sizes[io_info->pieces_added] = dinfo->type_info.dst_type_size;
                 io_info->wbufs[io_info->pieces_added]         = dinfo->buf.cvp;
+                if (io_info->sel_pieces)
+                    io_info->sel_pieces[io_info->pieces_added] = dinfo->layout_io_info.contig_piece_info;
                 io_info->pieces_added++;
             }
         }

src/H5Dio.c

@@ -292,16 +292,16 @@ H5D__read(size_t count, H5D_dset_io_info_t *dset_info)
     if (io_op_init == 0)
         HGOTO_DONE(SUCCEED)
 
-    /* Perform second phase of type info initialization */
-    if (H5D__typeinfo_init_phase2(&io_info) < 0)
-        HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info (second phase)")
-
 #ifdef H5_HAVE_PARALLEL
     /* Adjust I/O info for any parallel I/O */
     if (H5D__ioinfo_adjust(&io_info) < 0)
         HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to adjust I/O info for parallel I/O")
 #endif /*H5_HAVE_PARALLEL*/
 
+    /* Perform second phase of type info initialization */
+    if (H5D__typeinfo_init_phase2(&io_info) < 0)
+        HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info (second phase)")
+
     /* If multi dataset I/O callback is not provided, perform read IO via
      * single-dset path with looping */
     if (io_info.md_io_ops.multi_read_md) {
@@ -339,6 +339,7 @@ H5D__read(size_t count, H5D_dset_io_info_t *dset_info)
     else {
         haddr_t prev_tag = HADDR_UNDEF;
 
+        /* Allocate selection I/O parameter arrays if necessary */
         if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info) && io_info.piece_count > 0) {
             if (NULL == (io_info.mem_spaces = H5MM_malloc(io_info.piece_count * sizeof(H5S_t *))))
                 HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
@@ -354,6 +355,11 @@ H5D__read(size_t count, H5D_dset_io_info_t *dset_info)
             if (NULL == (io_info.rbufs = H5MM_malloc(io_info.piece_count * sizeof(void *))))
                 HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
                             "memory allocation failed for read buffer list")
+            if (io_info.tconv_buf)
+                if (NULL ==
+                    (io_info.sel_pieces = H5MM_malloc(io_info.piece_count * sizeof(io_info.sel_pieces[0]))))
+                    HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                                "unable to allocate array of selected pieces")
         }
 
         /* Loop with serial & single-dset read IO path */
@@ -373,15 +379,24 @@ H5D__read(size_t count, H5D_dset_io_info_t *dset_info)
             H5AC_tag(prev_tag, NULL);
         }
 
-        /* Make final multi dataset selection I/O call if we are using both
-         * features - in this case the multi_read callbacks did not perform the
-         * actual I/O */
-        H5_CHECK_OVERFLOW(io_info.pieces_added, size_t, uint32_t)
-        if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info))
-            if (H5F_shared_select_read(io_info.f_sh, H5FD_MEM_DRAW, (uint32_t)io_info.pieces_added,
-                                       io_info.mem_spaces, io_info.file_spaces, io_info.addrs,
-                                       io_info.element_sizes, io_info.rbufs) < 0)
-                HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "selection read failed")
+        /* Make final selection I/O call if the multi_read callbacks did not perform the actual I/O
+         * (if using selection I/O and either multi dataset or type conversion) */
+        if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info)) {
+            /* Check for type conversion */
+            if (io_info.tconv_buf) {
+                /* Type conversion pathway */
+                if (H5D__scatgath_read_select(&io_info) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "type conversion selection read failed")
+            }
+            else {
+                /* Call selection I/O directly */
+                H5_CHECK_OVERFLOW(io_info.pieces_added, size_t, uint32_t)
+                if (H5F_shared_select_read(io_info.f_sh, H5FD_MEM_DRAW, (uint32_t)io_info.pieces_added,
+                                           io_info.mem_spaces, io_info.file_spaces, io_info.addrs,
+                                           io_info.element_sizes, io_info.rbufs) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "selection read failed")
+            }
+        }
     }
 
 done:
@@ -663,16 +678,16 @@ H5D__write(size_t count, H5D_dset_io_info_t *dset_info)
     HDassert(type_info_init == count);
     HDassert(io_op_init == count);
 
-    /* Perform second phase of type info initialization */
-    if (H5D__typeinfo_init_phase2(&io_info) < 0)
-        HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info (second phase)")
-
 #ifdef H5_HAVE_PARALLEL
     /* Adjust I/O info for any parallel I/O */
     if (H5D__ioinfo_adjust(&io_info) < 0)
         HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to adjust I/O info for parallel I/O")
 #endif /*H5_HAVE_PARALLEL*/
 
+    /* Perform second phase of type info initialization */
+    if (H5D__typeinfo_init_phase2(&io_info) < 0)
+        HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info (second phase)")
+
     /* If multi dataset I/O callback is not provided, perform write IO via
      * single-dset path with looping */
     if (io_info.md_io_ops.multi_write_md) {
@@ -710,6 +725,7 @@ H5D__write(size_t count, H5D_dset_io_info_t *dset_info)
     else {
         haddr_t prev_tag = HADDR_UNDEF;
 
+        /* Allocate selection I/O parameter arrays if necessary */
         if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info) && io_info.piece_count > 0) {
             if (NULL == (io_info.mem_spaces = H5MM_malloc(io_info.piece_count * sizeof(H5S_t *))))
                 HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
@@ -724,7 +740,12 @@ H5D__write(size_t count, H5D_dset_io_info_t *dset_info)
                             "memory allocation failed for element size list")
             if (NULL == (io_info.wbufs = H5MM_malloc(io_info.piece_count * sizeof(const void *))))
                 HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
-                            "memory allocation failed for read buffer list")
+                            "memory allocation failed for write buffer list")
+            if (io_info.tconv_buf)
+                if (NULL ==
+                    (io_info.sel_pieces = H5MM_malloc(io_info.piece_count * sizeof(io_info.sel_pieces[0]))))
+                    HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                                "unable to allocate array of selected pieces")
         }
 
         /* loop with serial & single-dset write IO path */
@@ -742,15 +763,24 @@ H5D__write(size_t count, H5D_dset_io_info_t *dset_info)
             H5AC_tag(prev_tag, NULL);
         }
 
-        /* Make final multi dataset selection I/O call if we are using both
-         * features - in this case the multi_write callbacks did not perform the
-         * actual I/O */
-        H5_CHECK_OVERFLOW(io_info.pieces_added, size_t, uint32_t)
-        if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info))
-            if (H5F_shared_select_write(io_info.f_sh, H5FD_MEM_DRAW, (uint32_t)io_info.pieces_added,
-                                        io_info.mem_spaces, io_info.file_spaces, io_info.addrs,
-                                        io_info.element_sizes, io_info.wbufs) < 0)
-                HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "selection write failed")
+        /* Make final selection I/O call if the multi_write callbacks did not perform the actual I/O
+         * (if using selection I/O and either multi dataset or type conversion) */
+        if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info)) {
+            /* Check for type conversion */
+            if (io_info.tconv_buf) {
+                /* Type conversion pathway */
+                if (H5D__scatgath_write_select(&io_info) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "type conversion selection write failed")
+            }
+            else {
+                /* Call selection I/O directly */
+                H5_CHECK_OVERFLOW(io_info.pieces_added, size_t, uint32_t)
+                if (H5F_shared_select_write(io_info.f_sh, H5FD_MEM_DRAW, (uint32_t)io_info.pieces_added,
+                                            io_info.mem_spaces, io_info.file_spaces, io_info.addrs,
+                                            io_info.element_sizes, io_info.wbufs) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "selection write failed")
+            }
+        }
     }
 
 #ifdef OLD_WAY
@@ -998,9 +1028,9 @@ H5D__typeinfo_init(H5D_io_info_t *io_info, H5D_dset_io_info_t *dset_info, hid_t
         /* Check if the datatypes are compound subsets of one another */
         type_info->cmpd_subset = H5T_path_compound_subset(type_info->tpath);
 
-        /* Update io_info->max_type_size */
-        io_info->max_type_size =
-            MAX3(io_info->max_type_size, type_info->src_type_size, type_info->dst_type_size);
+        /* Update io_info->max_tconv_type_size */
+        io_info->max_tconv_type_size =
+            MAX3(io_info->max_tconv_type_size, type_info->src_type_size, type_info->dst_type_size);
 
         /* Check if we need a background buffer */
         if ((io_info->op_type == H5D_IO_OP_WRITE) && H5T_detect_class(dset->shared->type, H5T_VLEN, FALSE))
@@ -1026,7 +1056,8 @@ done:
  * Function:    H5D__typeinfo_init_phase2
  *
  * Purpose:     Finish initializing type info for all datasets after
- *              calculating the max type size across all datasets.
+ *              calculating the max type size across all datasets.  And
+ *              after H5D__ioinfo_adjust().
  *
  * Return:      Non-negative on success/Negative on failure
  *
@@ -1043,88 +1074,154 @@ H5D__typeinfo_init_phase2(H5D_io_info_t *io_info)
     HDassert(io_info);
 
     /* Check if we need to allocate a shared type conversion buffer */
-    if (io_info->max_type_size) {
-        void  *tconv_buf;    /* Temporary conversion buffer pointer */
-        void  *bkgr_buf;     /* Background conversion buffer pointer */
-        size_t max_temp_buf; /* Maximum temporary buffer size */
-        size_t target_size;  /* Desired buffer size	*/
-        size_t i;            /* Local index variable */
+    if (io_info->max_tconv_type_size) {
+        H5FD_mpio_xfer_t xfer_mode; /* Parallel transfer for this request */
+        size_t           i;         /* Local index variable */
 
-        /* Get info from API context */
-        if (H5CX_get_max_temp_buf(&max_temp_buf) < 0)
-            HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve max. temp. buf size")
-        if (H5CX_get_tconv_buf(&tconv_buf) < 0)
-            HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve temp. conversion buffer pointer")
-        if (H5CX_get_bkgr_buf(&bkgr_buf) < 0)
-            HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve background conversion buffer pointer")
+        /* Get the original state of parallel I/O transfer mode */
+        if (H5CX_get_io_xfer_mode(&xfer_mode) < 0)
+            HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode")
 
-        /* Set up datatype conversion/background buffers */
-        target_size = max_temp_buf;
+        /* Check if we're doing collective I/O */
+        if (xfer_mode == H5FD_MPIO_COLLECTIVE) {
+            size_t tconv_buf_size = 0; /* Size of shared type conversion buffer */
 
-        /* If the buffer is too small to hold even one element (in the dataset with the largest , try to make
-         * it bigger */
-        if (target_size < io_info->max_type_size) {
-            hbool_t default_buffer_info; /* Whether the buffer information are the defaults */
+            /* Only implemented for selection I/O */
+            HDassert(io_info->use_select_io);
 
-            /* Detect if we have all default settings for buffers */
-            default_buffer_info =
-                (hbool_t)((H5D_TEMP_BUF_SIZE == max_temp_buf) && (NULL == tconv_buf) && (NULL == bkgr_buf));
+            /* Collective I/O, conversion buffer must be large enough for entire I/O (for now).
+             * Stick with individual background buffers */
 
-            /* Check if we are using the default buffer info */
-            if (default_buffer_info)
-                /* OK to get bigger for library default settings */
-                target_size = io_info->max_type_size;
-            else
-                /* Don't get bigger than the application has requested */
-                HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "temporary buffer max size is too small")
-        } /* end if */
+            /* Calculate size of buffers */
+            for (i = 0; i < io_info->count; i++) {
+                H5D_type_info_t *type_info = &io_info->dsets_info[i].type_info;
 
-        /* Get a temporary buffer for type conversion unless the app has already
-         * supplied one through the xfer properties. Instead of allocating a
-         * buffer which is the exact size, we allocate the target size. This
-         * buffer is shared among all datasets in the operation, unlike for the
-         * background buffer, where each dataset gets its own. */
-        if (NULL == (io_info->tconv_buf = (uint8_t *)tconv_buf)) {
-            /* Allocate temporary buffer */
-            if (NULL == (io_info->tconv_buf = H5FL_BLK_MALLOC(type_conv, target_size)))
-                HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for type conversion")
-            io_info->tconv_buf_allocated = TRUE;
-        } /* end if */
+                /* Check if type conversion buffer is needed for this dataset */
+                if (!type_info->is_conv_noop || !type_info->is_xform_noop) {
+                    /* Calculate location and size of this dataset's portion of the global type
+                     * conversion buffer */
+                    tconv_buf_size += io_info->dsets_info[i].nelmts *
+                                      MAX(type_info->src_type_size, type_info->dst_type_size);
 
-        /* Don't use API provided background buffer if there's more than one dataset, since each
-         * dataset needs its own */
-        if (io_info->count > 1)
-            bkgr_buf = NULL;
+                    /* Allocate background buffer, if necessary */
+                    if (type_info->need_bkg) {
+                        if (NULL ==
+                            (type_info->bkg_buf = H5FL_BLK_CALLOC(type_conv, io_info->dsets_info[i].nelmts *
+                                                                                 type_info->dst_type_size)))
+                            HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL,
+                                        "memory allocation failed for type conversion")
+                        type_info->bkg_buf_allocated = TRUE;
 
-        /* Iterate over datasets */
-        for (i = 0; i < io_info->count; i++) {
-            H5D_type_info_t *type_info = &io_info->dsets_info[i].type_info;
+                        /* Check if we need to fill the background buffer with the destination contents */
+                        if (type_info->need_bkg == H5T_BKG_YES)
+                            io_info->must_fill_bkg = TRUE;
+                    }
+                }
+            }
 
-            /* Compute the number of elements that will fit into buffer */
-            type_info->request_nelmts = target_size / MAX(type_info->src_type_size, type_info->dst_type_size);
-            ;
-
-            /* Sanity check elements in temporary buffer */
-            if (type_info->request_nelmts == 0)
-                HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "temporary buffer max size is too small")
-
-            /* Allocate background buffer if necessary */
-            if (type_info->need_bkg && NULL == (type_info->bkg_buf = (uint8_t *)bkgr_buf)) {
-                size_t bkg_size; /* Desired background buffer size	*/
-
-                /* Compute the background buffer size */
-                /* (don't try to use buffers smaller than the default size) */
-                bkg_size = type_info->request_nelmts * type_info->dst_type_size;
-                if (bkg_size < max_temp_buf)
-                    bkg_size = max_temp_buf;
-
-                /* Allocate background buffer */
-                /* (Need calloc()-like call since memory needs to be initialized) */
-                if (NULL == (type_info->bkg_buf = H5FL_BLK_CALLOC(type_conv, bkg_size)))
+            /* Allocate global type conversion buffer (if any, could be none if datasets in this
+             * I/O have 0 elements selected) */
+            /* Allocating large buffers here will blow out all other type conversion buffers
+             * on the free list.  Should we change this to a regular malloc?  Would require
+             * keeping track of which version of free() to call. -NAF */
+            if (tconv_buf_size > 0) {
+                if (NULL == (io_info->tconv_buf = H5FL_BLK_MALLOC(type_conv, tconv_buf_size)))
                     HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL,
-                                "memory allocation failed for background conversion")
-                type_info->bkg_buf_allocated = TRUE;
+                                "memory allocation failed for type conversion")
+                io_info->tconv_buf_allocated = TRUE;
+            }
+        }
+        else {
+            /* No collective I/O, only need to make sure it's big enough for one element */
+            void  *tconv_buf;    /* Temporary conversion buffer pointer */
+            void  *bkgr_buf;     /* Background conversion buffer pointer */
+            size_t max_temp_buf; /* Maximum temporary buffer size */
+            size_t target_size;  /* Desired buffer size	*/
+
+            /* Disable selection I/O (for now) since the selection I/O type conversion code path
+             * assumes the tconv buf is large enough to hold all data in the I/O, and we don't want
+             * to impose this when there's no benefit */
+            io_info->use_select_io = FALSE;
+
+            /* Get info from API context */
+            if (H5CX_get_max_temp_buf(&max_temp_buf) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve max. temp. buf size")
+            if (H5CX_get_tconv_buf(&tconv_buf) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve temp. conversion buffer pointer")
+            if (H5CX_get_bkgr_buf(&bkgr_buf) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL,
+                            "can't retrieve background conversion buffer pointer")
+
+            /* Set up datatype conversion/background buffers */
+            target_size = max_temp_buf;
+
+            /* If the buffer is too small to hold even one element (in the dataset with the largest , try to
+             * make it bigger */
+            if (target_size < io_info->max_tconv_type_size) {
+                hbool_t default_buffer_info; /* Whether the buffer information are the defaults */
+
+                /* Detect if we have all default settings for buffers */
+                default_buffer_info = (hbool_t)((H5D_TEMP_BUF_SIZE == max_temp_buf) && (NULL == tconv_buf) &&
+                                                (NULL == bkgr_buf));
+
+                /* Check if we are using the default buffer info */
+                if (default_buffer_info)
+                    /* OK to get bigger for library default settings */
+                    target_size = io_info->max_tconv_type_size;
+                else
+                    /* Don't get bigger than the application has requested */
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "temporary buffer max size is too small")
             } /* end if */
+
+            /* Get a temporary buffer for type conversion unless the app has already
+             * supplied one through the xfer properties. Instead of allocating a
+             * buffer which is the exact size, we allocate the target size. This
+             * buffer is shared among all datasets in the operation, unlike for the
+             * background buffer, where each dataset gets its own. */
+            if (NULL == (io_info->tconv_buf = (uint8_t *)tconv_buf)) {
+                /* Allocate temporary buffer */
+                if (NULL == (io_info->tconv_buf = H5FL_BLK_MALLOC(type_conv, target_size)))
+                    HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL,
+                                "memory allocation failed for type conversion")
+                io_info->tconv_buf_allocated = TRUE;
+            } /* end if */
+
+            /* Don't use API provided background buffer if there's more than one dataset, since each
+             * dataset needs its own */
+            if (io_info->count > 1)
+                bkgr_buf = NULL;
+
+            /* Iterate over datasets */
+            for (i = 0; i < io_info->count; i++) {
+                H5D_type_info_t *type_info = &io_info->dsets_info[i].type_info;
+
+                /* Compute the number of elements that will fit into buffer */
+                type_info->request_nelmts =
+                    target_size / MAX(type_info->src_type_size, type_info->dst_type_size);
+                ;
+
+                /* Sanity check elements in temporary buffer */
+                if (type_info->request_nelmts == 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "temporary buffer max size is too small")
+
+                /* Allocate background buffer if necessary */
+                if (type_info->need_bkg && NULL == (type_info->bkg_buf = (uint8_t *)bkgr_buf)) {
+                    size_t bkg_size; /* Desired background buffer size	*/
+
+                    /* Compute the background buffer size */
+                    /* (don't try to use buffers smaller than the default size) */
+                    bkg_size = type_info->request_nelmts * type_info->dst_type_size;
+                    if (bkg_size < max_temp_buf)
+                        bkg_size = max_temp_buf;
+
+                    /* Allocate background buffer */
+                    /* (Need calloc()-like call since memory needs to be initialized) */
+                    if (NULL == (type_info->bkg_buf = H5FL_BLK_CALLOC(type_conv, bkg_size)))
+                        HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL,
+                                    "memory allocation failed for background conversion")
+                    type_info->bkg_buf_allocated = TRUE;
+                } /* end if */
+            }
         }
     }
 

src/H5Dmpio.c

@@ -626,13 +626,16 @@ H5D__mpio_opt_possible(H5D_io_info_t *io_info)
         if (!H5FD_mpi_opt_types_g)
             local_cause[0] |= H5D_MPIO_MPI_OPT_TYPES_ENV_VAR_DISABLED;
 
-        /* Don't allow collective operations if datatype conversions need to happen */
-        if (!type_info->is_conv_noop)
-            local_cause[0] |= H5D_MPIO_DATATYPE_CONVERSION;
+        /* Datatype conversions and transformations are allowed with selection I/O */
+        if (!io_info->use_select_io) {
+            /* Don't allow collective operations if datatype conversions need to happen */
+            if (!type_info->is_conv_noop)
+                local_cause[0] |= H5D_MPIO_DATATYPE_CONVERSION;
 
-        /* Don't allow collective operations if data transform operations should occur */
-        if (!type_info->is_xform_noop)
-            local_cause[0] |= H5D_MPIO_DATA_TRANSFORMS;
+            /* Don't allow collective operations if data transform operations should occur */
+            if (!type_info->is_xform_noop)
+                local_cause[0] |= H5D_MPIO_DATA_TRANSFORMS;
+        }
 
         /* Check whether these are both simple or scalar dataspaces */
         if (!((H5S_SIMPLE == H5S_GET_EXTENT_TYPE(mem_space) ||

src/H5Dpkg.h

@@ -70,7 +70,8 @@
 #define H5D_BT2_MERGE_PERC        40
 
 /* Macro to determine if the layout I/O callback should perform I/O */
-#define H5D_LAYOUT_CB_PERFORM_IO(IO_INFO) (!(IO_INFO)->use_select_io || (IO_INFO)->count == 1)
+#define H5D_LAYOUT_CB_PERFORM_IO(IO_INFO)                                                                    \
+    (!(IO_INFO)->use_select_io || ((IO_INFO)->count == 1 && !(IO_INFO)->tconv_buf))
 
 /****************************/
 /* Package Private Typedefs */
@@ -269,7 +270,10 @@ typedef struct H5D_io_info_t {
     hbool_t                 use_select_io;       /* Whether to use selection I/O */
     uint8_t                *tconv_buf;           /* Datatype conv buffer */
     hbool_t                 tconv_buf_allocated; /* Whether the type conversion buffer was allocated */
-    size_t                  max_type_size;       /* Largest of all source and destination type sizes */
+    size_t max_tconv_type_size; /* Largest of all source and destination type sizes involved in type
+                                   conversion */
+    hbool_t
+        must_fill_bkg; /* Whether any datasets need a background buffer filled with destination contents */
 } H5D_io_info_t;
 
 /* Created to pass both at once for callback func */
@@ -623,12 +627,14 @@ H5_DLL herr_t H5D__select_write(const H5D_io_info_t *io_info, const H5D_dset_io_
 H5_DLL herr_t H5D_select_io_mem(void *dst_buf, H5S_t *dst_space, const void *src_buf, H5S_t *src_space,
                                 size_t elmt_size, size_t nelmts);
 
-/* Functions that perform scatter-gather serial I/O operations */
+/* Functions that perform scatter-gather I/O operations */
 H5_DLL herr_t H5D__scatter_mem(const void *_tscat_buf, H5S_sel_iter_t *iter, size_t nelmts, void *_buf);
 H5_DLL size_t H5D__gather_mem(const void *_buf, H5S_sel_iter_t *iter, size_t nelmts,
                               void *_tgath_buf /*out*/);
 H5_DLL herr_t H5D__scatgath_read(const H5D_io_info_t *io_info, const H5D_dset_io_info_t *dset_info);
 H5_DLL herr_t H5D__scatgath_write(const H5D_io_info_t *io_info, const H5D_dset_io_info_t *dset_info);
+H5_DLL herr_t H5D__scatgath_read_select(H5D_io_info_t *io_info);
+H5_DLL herr_t H5D__scatgath_write_select(H5D_io_info_t *io_info);
 
 /* Functions that operate on dataset's layout information */
 H5_DLL herr_t H5D__layout_set_io_ops(const H5D_t *dataset);

src/H5Dscatgath.c

@@ -726,6 +726,528 @@ done:
     FUNC_LEAVE_NOAPI(ret_value)
 } /* end H5D__scatgath_write() */
 
+/*-------------------------------------------------------------------------
+ * Function:	H5D__scatgath_read_select
+ *
+ * Purpose:	Perform scatter/gather read from a list of dataset pieces
+ *
+ * Return:	Non-negative on success/Negative on failure
+ *
+ *-------------------------------------------------------------------------
+ */
+herr_t
+H5D__scatgath_read_select(H5D_io_info_t *io_info)
+{
+    H5S_t             **tmp_mem_spaces = NULL;  /* Memory spaces to use for read from disk */
+    H5S_sel_iter_t     *mem_iter       = NULL;  /* Memory selection iteration info */
+    hbool_t             mem_iter_init  = FALSE; /* Memory selection iteration info has been initialized */
+    void              **tmp_bufs       = NULL;  /* Buffers to use for read from disk */
+    size_t              buf_bytes_used = 0; /* Number of bytes used so far in conversion/background buffer */
+    H5D_dset_io_info_t *last_bkg_buf_dset = NULL; /* Index of last dset that used a background buffer */
+    size_t              i;                        /* Local index variable */
+    herr_t              ret_value = SUCCEED;      /* Return value		*/
+
+    FUNC_ENTER_PACKAGE
+
+    /* Sanity check */
+    HDassert(io_info);
+    HDassert(io_info->count > 0);
+    HDassert(io_info->mem_spaces || io_info->pieces_added == 0);
+    HDassert(io_info->file_spaces || io_info->pieces_added == 0);
+    HDassert(io_info->addrs || io_info->pieces_added == 0);
+    HDassert(io_info->element_sizes || io_info->pieces_added == 0);
+    HDassert(io_info->rbufs || io_info->pieces_added == 0);
+
+    /* Allocate list of buffers (within the tconv buf) */
+    if (NULL == (tmp_bufs = H5MM_malloc(io_info->pieces_added * sizeof(void *))))
+        HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "memory allocation failed for temporary buffer list")
+
+    /* Allocate the iterator */
+    if (NULL == (mem_iter = H5FL_MALLOC(H5S_sel_iter_t)))
+        HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "can't allocate memory iterator")
+
+    /* Allocate list of block memory spaces */
+    /*!FIXME delay doing this until we find the first mem space that is non-contiguous or doesn't start at 0
+     */
+    if (NULL == (tmp_mem_spaces = H5MM_malloc(io_info->pieces_added * sizeof(H5S_t *))))
+        HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                    "memory allocation failed for temporary memory space list")
+
+    /* Build read operation to tconv buffer */
+    for (i = 0; i < io_info->pieces_added; i++) {
+        H5D_dset_io_info_t *dset_info = io_info->sel_pieces[i]->dset_info;
+
+        HDassert(io_info->sel_pieces[i]->piece_points > 0);
+
+        /* Check if this piece is involved in type conversion */
+        if (dset_info->type_info.is_xform_noop && dset_info->type_info.is_conv_noop) {
+            /* No type conversion, just copy the mem space and buffer */
+            tmp_mem_spaces[i] = io_info->mem_spaces[i];
+            tmp_bufs[i]       = io_info->rbufs[i];
+        }
+        else {
+            /* Create block memory space */
+            if (NULL ==
+                (tmp_mem_spaces[i] = H5S_create_simple(1, &io_info->sel_pieces[i]->piece_points, NULL))) {
+                HDmemset(&tmp_mem_spaces[i], 0, (io_info->pieces_added - i) * sizeof(tmp_mem_spaces[0]));
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTCREATE, FAIL, "unable to create simple memory dataspace")
+            }
+
+            /* Set buffer to point into type conversion buffer */
+            tmp_bufs[i] = io_info->tconv_buf + buf_bytes_used;
+            buf_bytes_used += io_info->sel_pieces[i]->piece_points *
+                              MAX(dset_info->type_info.src_type_size, dset_info->type_info.dst_type_size);
+        }
+    }
+
+    /* Read data from all pieces */
+    H5_CHECK_OVERFLOW(io_info->pieces_added, size_t, uint32_t)
+    if (H5F_shared_select_read(io_info->f_sh, H5FD_MEM_DRAW, (uint32_t)io_info->pieces_added, tmp_mem_spaces,
+                               io_info->file_spaces, io_info->addrs, io_info->element_sizes, tmp_bufs) < 0)
+        HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "selection read failed")
+
+    /* Perform type conversion and scatter data to memory buffers for datasets that need this */
+    for (i = 0; i < io_info->pieces_added; i++) {
+        H5D_dset_io_info_t *dset_info = io_info->sel_pieces[i]->dset_info;
+
+        HDassert(tmp_mem_spaces[i]);
+
+        /* Check if this piece is involved in type conversion */
+        if (tmp_mem_spaces[i] != io_info->mem_spaces[i]) {
+            H5_CHECK_OVERFLOW(io_info->sel_pieces[i]->piece_points, hsize_t, size_t);
+
+            /* Initialize memory iterator */
+            HDassert(!mem_iter_init);
+            if (H5S_select_iter_init(mem_iter, io_info->mem_spaces[i], dset_info->type_info.dst_type_size,
+                                     0) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL,
+                            "unable to initialize memory selection information")
+            mem_iter_init = TRUE; /* Memory selection iteration info has been initialized */
+
+            /* If the source and destination are compound types and subset of each other
+             * and no conversion is needed, copy the data directly into user's buffer and
+             * bypass the rest of steps.
+             */
+            if (dset_info->type_info.cmpd_subset &&
+                H5T_SUBSET_FALSE != dset_info->type_info.cmpd_subset->subset) {
+                if (H5D__compound_opt_read((size_t)io_info->sel_pieces[i]->piece_points, mem_iter,
+                                           &dset_info->type_info, tmp_bufs[i], io_info->rbufs[i] /*out*/) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "datatype conversion failed")
+            }
+            else {
+                void *tmp_bkg_buf = NULL;
+
+                /* Check for background buffer */
+                if (dset_info->type_info.need_bkg) {
+                    HDassert(dset_info->type_info.bkg_buf);
+
+                    /* Check if we moved to a new background buffer */
+                    if (dset_info != last_bkg_buf_dset) {
+                        /* Reset buf_bytes_used */
+                        buf_bytes_used    = 0;
+                        last_bkg_buf_dset = dset_info;
+                    }
+
+                    /* Calculate background buffer position */
+                    tmp_bkg_buf = dset_info->type_info.bkg_buf + buf_bytes_used;
+                    buf_bytes_used +=
+                        io_info->sel_pieces[i]->piece_points * dset_info->type_info.dst_type_size;
+
+                    /* Gather data from read buffer to background buffer if necessary */
+                    if (H5T_BKG_YES == dset_info->type_info.need_bkg) {
+                        if ((size_t)io_info->sel_pieces[i]->piece_points !=
+                            H5D__gather_mem(io_info->rbufs[i], mem_iter,
+                                            (size_t)io_info->sel_pieces[i]->piece_points,
+                                            tmp_bkg_buf /*out*/))
+                            HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "mem gather failed")
+
+                        /* Reset selection iterator */
+                        HDassert(mem_iter_init);
+                        if (H5S_SELECT_ITER_RELEASE(mem_iter) < 0)
+                            HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "Can't release selection iterator")
+                        mem_iter_init = FALSE;
+
+                        if (H5S_select_iter_init(mem_iter, io_info->mem_spaces[i],
+                                                 dset_info->type_info.dst_type_size, 0) < 0)
+                            HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL,
+                                        "unable to initialize memory selection information")
+                        mem_iter_init = TRUE; /* Memory selection iteration info has been initialized */
+                    }
+                }
+
+                /*
+                 * Perform datatype conversion.
+                 */
+                if (H5T_convert(dset_info->type_info.tpath, dset_info->type_info.src_type_id,
+                                dset_info->type_info.dst_type_id,
+                                (size_t)io_info->sel_pieces[i]->piece_points, (size_t)0, (size_t)0,
+                                tmp_bufs[i], tmp_bkg_buf) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTCONVERT, FAIL, "datatype conversion failed")
+
+                /* Do the data transform after the conversion (since we're using type mem_type) */
+                if (!dset_info->type_info.is_xform_noop) {
+                    H5Z_data_xform_t *data_transform; /* Data transform info */
+
+                    /* Retrieve info from API context */
+                    if (H5CX_get_data_transform(&data_transform) < 0)
+                        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get data transform info")
+
+                    if (H5Z_xform_eval(data_transform, tmp_bufs[i],
+                                       (size_t)io_info->sel_pieces[i]->piece_points,
+                                       dset_info->type_info.mem_type) < 0)
+                        HGOTO_ERROR(H5E_DATASET, H5E_BADVALUE, FAIL, "Error performing data transform")
+                }
+
+                /* Scatter the data into memory */
+                if (H5D__scatter_mem(tmp_bufs[i], mem_iter, (size_t)io_info->sel_pieces[i]->piece_points,
+                                     io_info->rbufs[i] /*out*/) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "scatter failed")
+            }
+
+            /* Release selection iterator */
+            HDassert(mem_iter_init);
+            if (H5S_SELECT_ITER_RELEASE(mem_iter) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "Can't release selection iterator")
+            mem_iter_init = FALSE;
+        }
+    }
+
+done:
+    /* Release and free selection iterator */
+    if (mem_iter_init && H5S_SELECT_ITER_RELEASE(mem_iter) < 0)
+        HDONE_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "Can't release selection iterator")
+    if (mem_iter)
+        mem_iter = H5FL_FREE(H5S_sel_iter_t, mem_iter);
+
+    /* Free tmp_bufs */
+    H5MM_free(tmp_bufs);
+    tmp_bufs = NULL;
+
+    /* Clear and free tmp_mem_spaces */
+    if (tmp_mem_spaces) {
+        for (i = 0; i < io_info->pieces_added; i++)
+            if (tmp_mem_spaces[i] != io_info->mem_spaces[i] && tmp_mem_spaces[i] &&
+                H5S_close(tmp_mem_spaces[i]) < 0)
+                HDONE_ERROR(H5E_DATASET, H5E_CLOSEERROR, FAIL, "Can't close dataspace")
+        H5MM_free(tmp_mem_spaces);
+        tmp_mem_spaces = NULL;
+    }
+
+    FUNC_LEAVE_NOAPI(ret_value)
+} /* end H5D__scatgath_read_select() */
+
+/*-------------------------------------------------------------------------
+ * Function:	H5D__scatgath_write_select
+ *
+ * Purpose:	Perform scatter/gather write to a list of dataset pieces.
+ *
+ * Return:	Non-negative on success/Negative on failure
+ *
+ *-------------------------------------------------------------------------
+ */
+herr_t
+H5D__scatgath_write_select(H5D_io_info_t *io_info)
+{
+    H5S_t             **write_mem_spaces  = NULL;  /* Memory spaces to use for write to disk */
+    size_t              spaces_added      = 0;     /* Number of spaces added to write_mem_spaces */
+    H5S_sel_iter_t     *mem_iter          = NULL;  /* Memory selection iteration info */
+    hbool_t             mem_iter_init     = FALSE; /* Memory selection iteration info has been initialized */
+    const void        **write_bufs        = NULL;  /* Buffers to use for write to disk */
+    size_t              tconv_bytes_used  = 0;     /* Number of bytes used so far in conversion buffer */
+    size_t              bkg_bytes_used    = 0;     /* Number of bytes used so far in background buffer */
+    H5D_dset_io_info_t *last_bkg_buf_dset = NULL;  /* Index of last dset that used a background buffer */
+    H5S_t             **bkg_mem_spaces    = NULL;  /* Array of memory spaces for read to background buffer */
+    H5S_t             **bkg_file_spaces   = NULL;  /* Array of file spaces for read to background buffer */
+    haddr_t            *bkg_addrs         = NULL;  /* Array of file addresses for read to background buffer */
+    size_t             *bkg_element_sizes = NULL;  /* Array of element sizes for read to background buffer */
+    void              **bkg_bufs   = NULL; /* Array background buffers for read of existing file contents */
+    size_t              bkg_pieces = 0; /* Number of pieces that need to read the background data from disk */
+    size_t              i;              /* Local index variable */
+    herr_t              ret_value = SUCCEED; /* Return value		*/
+
+    FUNC_ENTER_PACKAGE
+
+    /* Sanity check */
+    HDassert(io_info);
+    HDassert(io_info->count > 0);
+    HDassert(io_info->mem_spaces || io_info->pieces_added == 0);
+    HDassert(io_info->file_spaces || io_info->pieces_added == 0);
+    HDassert(io_info->addrs || io_info->pieces_added == 0);
+    HDassert(io_info->element_sizes || io_info->pieces_added == 0);
+    HDassert(io_info->wbufs || io_info->pieces_added == 0);
+
+    /* Allocate list of buffers (within the tconv buf) */
+    if (NULL == (write_bufs = (const void **)H5MM_malloc(io_info->pieces_added * sizeof(const void *))))
+        HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "memory allocation failed for temporary buffer list")
+
+    /* Allocate the iterator */
+    if (NULL == (mem_iter = H5FL_MALLOC(H5S_sel_iter_t)))
+        HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "can't allocate memory iterator")
+
+    /* Allocate list of block memory spaces */
+    /*!FIXME delay doing this until we find the first mem space that is non-contiguous or doesn't start at 0
+     */
+    if (NULL == (write_mem_spaces = H5MM_malloc(io_info->pieces_added * sizeof(H5S_t *))))
+        HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                    "memory allocation failed for temporary memory space list")
+
+    /* Build operations to read data to background buffer and to write data */
+    for (i = 0; i < io_info->pieces_added; i++) {
+        H5D_dset_io_info_t *dset_info = io_info->sel_pieces[i]->dset_info;
+
+        HDassert(io_info->sel_pieces[i]->piece_points > 0);
+
+        /* Check if this piece is involved in type conversion */
+        if (dset_info->type_info.is_xform_noop && dset_info->type_info.is_conv_noop) {
+            /* No type conversion, just copy the mem space and buffer */
+            write_mem_spaces[i] = io_info->mem_spaces[i];
+            spaces_added++;
+            write_bufs[i] = io_info->wbufs[i];
+        }
+        else {
+            void *tmp_write_buf; /* To sidestep const warnings */
+            void *tmp_bkg_buf = NULL;
+
+            H5_CHECK_OVERFLOW(io_info->sel_pieces[i]->piece_points, hsize_t, size_t);
+
+            /* Initialize memory iterator */
+            HDassert(!mem_iter_init);
+            if (H5S_select_iter_init(mem_iter, io_info->mem_spaces[i], dset_info->type_info.src_type_size,
+                                     0) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL,
+                            "unable to initialize memory selection information")
+            mem_iter_init = TRUE; /* Memory selection iteration info has been initialized */
+
+            /* Create block memory space */
+            if (NULL ==
+                (write_mem_spaces[i] = H5S_create_simple(1, &io_info->sel_pieces[i]->piece_points, NULL)))
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTCREATE, FAIL, "unable to create simple memory dataspace")
+            spaces_added++;
+
+            /* Set buffer to point into type conversion buffer */
+            tmp_write_buf = io_info->tconv_buf + tconv_bytes_used;
+            write_bufs[i] = (const void *)tmp_write_buf;
+            tconv_bytes_used += io_info->sel_pieces[i]->piece_points *
+                                MAX(dset_info->type_info.src_type_size, dset_info->type_info.dst_type_size);
+
+            /* Gather data from application buffer into the datatype conversion buffer */
+            if ((size_t)io_info->sel_pieces[i]->piece_points !=
+                H5D__gather_mem(io_info->wbufs[i], mem_iter, (size_t)io_info->sel_pieces[i]->piece_points,
+                                tmp_write_buf /*out*/))
+                HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "mem gather failed")
+
+            /* If the source and destination are compound types and the destination is
+             * is a subset of the source and no conversion is needed, copy the data
+             * directly into user's buffer and bypass the rest of steps.  If the source
+             * is a subset of the destination, the optimization is done in conversion
+             * function H5T_conv_struct_opt to protect the background data.
+             */
+            if (dset_info->type_info.cmpd_subset &&
+                H5T_SUBSET_DST == dset_info->type_info.cmpd_subset->subset &&
+                dset_info->type_info.dst_type_size == dset_info->type_info.cmpd_subset->copy_size) {
+                if (H5D__compound_opt_write((size_t)io_info->sel_pieces[i]->piece_points,
+                                            &dset_info->type_info, tmp_write_buf) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "datatype conversion failed")
+
+                /* No background buffer necessary, prevent this element from being considered in the second
+                 * loop */
+                /* Add this to H5Tconv.c? -NAF */
+                dset_info->type_info.need_bkg = H5T_BKG_NO;
+            } /* end if */
+            else {
+                /* Check for background buffer */
+                if (dset_info->type_info.need_bkg) {
+                    HDassert(dset_info->type_info.bkg_buf);
+
+                    /* Check if we moved to a new background buffer */
+                    if (dset_info != last_bkg_buf_dset) {
+                        /* Reset bkg_bytes_used */
+                        bkg_bytes_used    = 0;
+                        last_bkg_buf_dset = dset_info;
+                    }
+
+                    /* Calculate background buffer position */
+                    tmp_bkg_buf = dset_info->type_info.bkg_buf + bkg_bytes_used;
+                    bkg_bytes_used +=
+                        io_info->sel_pieces[i]->piece_points * dset_info->type_info.dst_type_size;
+                }
+
+                /* Set up background buffer read operation if necessary */
+                if (H5T_BKG_YES == dset_info->type_info.need_bkg) {
+                    HDassert(io_info->must_fill_bkg);
+
+                    /* Allocate arrays of parameters for selection read to background buffer if necessary */
+                    if (!bkg_mem_spaces) {
+                        HDassert(!bkg_file_spaces && !bkg_addrs && !bkg_element_sizes && !bkg_bufs);
+                        if (NULL == (bkg_mem_spaces = H5MM_malloc(io_info->pieces_added * sizeof(H5S_t *))))
+                            HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                                        "memory allocation failed for memory space list")
+                        if (NULL == (bkg_file_spaces = H5MM_malloc(io_info->pieces_added * sizeof(H5S_t *))))
+                            HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                                        "memory allocation failed for file space list")
+                        if (NULL == (bkg_addrs = H5MM_malloc(io_info->pieces_added * sizeof(haddr_t))))
+                            HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                                        "memory allocation failed for piece address list")
+                        if (NULL == (bkg_element_sizes = H5MM_malloc(io_info->pieces_added * sizeof(size_t))))
+                            HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                                        "memory allocation failed for element size list")
+                        if (NULL == (bkg_bufs = H5MM_malloc(io_info->pieces_added * sizeof(const void *))))
+                            HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL,
+                                        "memory allocation failed for write buffer list")
+                    }
+
+                    /* Use same (block) memory space, file space, address, and element size as write operation
+                     */
+                    HDassert(bkg_mem_spaces && bkg_file_spaces && bkg_addrs && bkg_element_sizes && bkg_bufs);
+                    bkg_mem_spaces[bkg_pieces]    = write_mem_spaces[i];
+                    bkg_file_spaces[bkg_pieces]   = io_info->file_spaces[i];
+                    bkg_addrs[bkg_pieces]         = io_info->addrs[i];
+                    bkg_element_sizes[bkg_pieces] = io_info->element_sizes[i];
+
+                    /* Use previously calculated background buffer position */
+                    bkg_bufs[bkg_pieces] = tmp_bkg_buf;
+
+                    /* Add piece */
+                    bkg_pieces++;
+                }
+                else {
+                    /* Perform type conversion here to avoid second loop if no dsets use the background buffer
+                     */
+                    /* Do the data transform before the type conversion (since
+                     * transforms must be done in the memory type). */
+                    if (!dset_info->type_info.is_xform_noop) {
+                        H5Z_data_xform_t *data_transform; /* Data transform info */
+
+                        /* Retrieve info from API context */
+                        if (H5CX_get_data_transform(&data_transform) < 0)
+                            HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get data transform info")
+
+                        if (H5Z_xform_eval(data_transform, tmp_write_buf,
+                                           (size_t)io_info->sel_pieces[i]->piece_points,
+                                           dset_info->type_info.mem_type) < 0)
+                            HGOTO_ERROR(H5E_DATASET, H5E_BADVALUE, FAIL, "Error performing data transform")
+                    }
+
+                    /*
+                     * Perform datatype conversion.
+                     */
+                    if (H5T_convert(dset_info->type_info.tpath, dset_info->type_info.src_type_id,
+                                    dset_info->type_info.dst_type_id,
+                                    (size_t)io_info->sel_pieces[i]->piece_points, (size_t)0, (size_t)0,
+                                    tmp_write_buf, tmp_bkg_buf) < 0)
+                        HGOTO_ERROR(H5E_DATASET, H5E_CANTCONVERT, FAIL, "datatype conversion failed")
+                }
+            }
+
+            /* Release selection iterator */
+            HDassert(mem_iter_init);
+            if (H5S_SELECT_ITER_RELEASE(mem_iter) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "Can't release selection iterator")
+            mem_iter_init = FALSE;
+        }
+    }
+
+    HDassert(spaces_added == io_info->pieces_added);
+
+    /* Gather data to background buffer if necessary */
+    if (io_info->must_fill_bkg) {
+        size_t j = 0; /* Index into array of background buffers */
+
+        /* Read data */
+        H5_CHECK_OVERFLOW(bkg_pieces, size_t, uint32_t)
+        if (H5F_shared_select_read(io_info->f_sh, H5FD_MEM_DRAW, (uint32_t)bkg_pieces, bkg_mem_spaces,
+                                   bkg_file_spaces, bkg_addrs, bkg_element_sizes, bkg_bufs) < 0)
+            HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "selection read to background buffer failed")
+
+        /* Perform type conversion on pieces with background buffers that were just read */
+        for (i = 0; i < io_info->pieces_added; i++) {
+            H5D_dset_io_info_t *dset_info = io_info->sel_pieces[i]->dset_info;
+
+            if (H5T_BKG_YES == dset_info->type_info.need_bkg) {
+                /* Non-const write_buf[i].  Use pointer math here to avoid const warnings.  When
+                 * there's a background buffer write_buf[i] always points inside the non-const tconv
+                 * buf so this is OK. */
+                void *tmp_write_buf =
+                    (void *)((uint8_t *)io_info->tconv_buf +
+                             ((const uint8_t *)write_bufs[i] - (const uint8_t *)io_info->tconv_buf));
+
+                /* Do the data transform before the type conversion (since
+                 * transforms must be done in the memory type). */
+                if (!dset_info->type_info.is_xform_noop) {
+                    H5Z_data_xform_t *data_transform; /* Data transform info */
+
+                    /* Retrieve info from API context */
+                    if (H5CX_get_data_transform(&data_transform) < 0)
+                        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get data transform info")
+
+                    if (H5Z_xform_eval(data_transform, tmp_write_buf,
+                                       (size_t)io_info->sel_pieces[i]->piece_points,
+                                       dset_info->type_info.mem_type) < 0)
+                        HGOTO_ERROR(H5E_DATASET, H5E_BADVALUE, FAIL, "Error performing data transform")
+                }
+
+                /*
+                 * Perform datatype conversion.
+                 */
+                HDassert(j < bkg_pieces);
+                if (H5T_convert(dset_info->type_info.tpath, dset_info->type_info.src_type_id,
+                                dset_info->type_info.dst_type_id,
+                                (size_t)io_info->sel_pieces[i]->piece_points, (size_t)0, (size_t)0,
+                                tmp_write_buf, bkg_bufs[j]) < 0)
+                    HGOTO_ERROR(H5E_DATASET, H5E_CANTCONVERT, FAIL, "datatype conversion failed")
+
+                /* Advance to next background buffer */
+                j++;
+            }
+        }
+
+        HDassert(j == bkg_pieces);
+    }
+
+    /* Write data to disk */
+    H5_CHECK_OVERFLOW(io_info->pieces_added, size_t, uint32_t)
+    if (H5F_shared_select_write(io_info->f_sh, H5FD_MEM_DRAW, (uint32_t)io_info->pieces_added,
+                                write_mem_spaces, io_info->file_spaces, io_info->addrs,
+                                io_info->element_sizes, write_bufs) < 0)
+        HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "selection write failed")
+
+done:
+    /* Release and free selection iterator */
+    if (mem_iter_init && H5S_SELECT_ITER_RELEASE(mem_iter) < 0)
+        HDONE_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "Can't release selection iterator")
+    if (mem_iter)
+        mem_iter = H5FL_FREE(H5S_sel_iter_t, mem_iter);
+
+    /* Free write_bufs */
+    H5MM_free(write_bufs);
+    write_bufs = NULL;
+
+    /* Clear and free write_mem_spaces */
+    if (write_mem_spaces) {
+        for (i = 0; i < spaces_added; i++) {
+            HDassert(write_mem_spaces[i]);
+            if (write_mem_spaces[i] != io_info->mem_spaces[i] && H5S_close(write_mem_spaces[i]) < 0)
+                HDONE_ERROR(H5E_DATASET, H5E_CLOSEERROR, FAIL, "Can't close dataspace")
+        }
+        H5MM_free(write_mem_spaces);
+        write_mem_spaces = NULL;
+    }
+
+    /* Free bakcground buffer parameter arrays */
+    H5MM_free(bkg_mem_spaces);
+    bkg_mem_spaces = NULL;
+    H5MM_free(bkg_file_spaces);
+    bkg_file_spaces = NULL;
+    H5MM_free(bkg_addrs);
+    bkg_addrs = NULL;
+    H5MM_free(bkg_element_sizes);
+    bkg_element_sizes = NULL;
+    H5MM_free(bkg_bufs);
+    bkg_bufs = NULL;
+
+    FUNC_LEAVE_NOAPI(ret_value)
+} /* end H5D__scatgath_write_select() */
+
 /*-------------------------------------------------------------------------
  * Function:	H5D__compound_opt_read
  *