[svn-r23081] ported revisions 22987 to 23080 from the trunk

14 files changed, 92 insertions, 292 deletions

diff --git a/src/H5Dmpio.c b/src/H5Dmpio.c
index c2d964e..900ad6a 100644
--- a/src/H5Dmpio.c
+++ b/src/H5Dmpio.c
@@ -105,10 +105,9 @@ static herr_t H5D__chunk_collective_io(H5D_io_info_t *io_info,
 static herr_t H5D__multi_chunk_collective_io(H5D_io_info_t *io_info,
     const H5D_type_info_t *type_info, H5D_chunk_map_t *fm,
     H5P_genplist_t *dx_plist);
-static herr_t H5D__multi_chunk_collective_io_no_opt(H5D_io_info_t *io_info,
-    const H5D_type_info_t *type_info, H5D_chunk_map_t *fm, H5P_genplist_t *dx_plist);
 static herr_t H5D__link_chunk_collective_io(H5D_io_info_t *io_info,
-    const H5D_type_info_t *type_info, H5D_chunk_map_t *fm, int sum_chunk);
+    const H5D_type_info_t *type_info, H5D_chunk_map_t *fm, int sum_chunk,
+    H5P_genplist_t *dx_plist);
 static herr_t H5D__inter_collective_io(H5D_io_info_t *io_info,
     const H5D_type_info_t *type_info, const H5S_t *file_space,
     const H5S_t *mem_space);
@@ -586,6 +585,12 @@ done:
  * Programmer:  Muqun Yang
  *              Monday, Feb. 13th, 2006
  *
+ * Modification:
+ *  - Refctore to remove multi-chunk-without-opimization feature and update for
+ *    multi-chunk-io accordingly
+ * Programmer: Jonathan Kim
+ * Date: 2012-10-10
+ *
  *-------------------------------------------------------------------------
  */
 static herr_t
@@ -594,8 +599,6 @@ H5D__chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *type_inf
 {
     H5P_genplist_t *dx_plist;           /* Pointer to DXPL */
     H5FD_mpio_chunk_opt_t chunk_opt_mode;
-    H5D_mpio_actual_chunk_opt_mode_t actual_chunk_opt_mode;
-    H5D_mpio_actual_io_mode_t actual_io_mode;
     int         io_option = H5D_MULTI_CHUNK_IO_MORE_OPT;
     int         sum_chunk = -1;
 #ifdef H5_HAVE_INSTRUMENTED_LIBRARY
@@ -617,10 +620,12 @@ H5D__chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *type_inf
 
     /* Check the optional property list on what to do with collective chunk IO. */
     chunk_opt_mode = (H5FD_mpio_chunk_opt_t)H5P_peek_unsigned(dx_plist, H5D_XFER_MPIO_CHUNK_OPT_HARD_NAME);
-    if(chunk_opt_mode == H5FD_MPIO_CHUNK_ONE_IO)
+    if(H5FD_MPIO_CHUNK_ONE_IO == chunk_opt_mode)
         io_option = H5D_ONE_LINK_CHUNK_IO;      /*no opt*/
-    else if(chunk_opt_mode == H5FD_MPIO_CHUNK_MULTI_IO)
-        io_option = H5D_MULTI_CHUNK_IO;         /*no opt */
+    /* direct request to multi-chunk-io */
+    else if(H5FD_MPIO_CHUNK_MULTI_IO == chunk_opt_mode)
+        io_option = H5D_MULTI_CHUNK_IO;         
+    /* via default path. branch by num threshold */
     else {
         unsigned one_link_chunk_io_threshold;   /* Threshhold to use single collective I/O for all chunks */
         int mpi_size;                   /* Number of processes in MPI job */
@@ -649,7 +654,7 @@ H5D__chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *type_inf
     /*** Test collective chunk user-input optimization APIs. ***/
     check_prop = H5Pexist(io_info->dxpl_id, H5D_XFER_COLL_CHUNK_LINK_HARD_NAME);
     if(check_prop > 0) {
-        if(io_option == H5D_ONE_LINK_CHUNK_IO) {
+        if(H5D_ONE_LINK_CHUNK_IO == io_option) {
             new_value = 0;
             if(H5Pset(io_info->dxpl_id, H5D_XFER_COLL_CHUNK_LINK_HARD_NAME, &new_value) < 0)
                 HGOTO_ERROR(H5E_IO, H5E_CANTSET, FAIL, "unable to set property value")
@@ -657,7 +662,7 @@ H5D__chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *type_inf
     } /* end if */
     check_prop = H5Pexist(io_info->dxpl_id, H5D_XFER_COLL_CHUNK_MULTI_HARD_NAME);
     if(check_prop > 0) {
-        if(io_option == H5D_MULTI_CHUNK_IO) {
+        if(H5D_MULTI_CHUNK_IO == io_option) {
             new_value = 0;
             if(H5Pset(io_info->dxpl_id, H5D_XFER_COLL_CHUNK_MULTI_HARD_NAME, &new_value) < 0)
                 HGOTO_ERROR(H5E_IO, H5E_CANTSET, FAIL, "unable to set property value")
@@ -665,7 +670,7 @@ H5D__chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *type_inf
     } /* end if */
     check_prop = H5Pexist(io_info->dxpl_id, H5D_XFER_COLL_CHUNK_LINK_NUM_TRUE_NAME);
     if(check_prop > 0) {
-        if(io_option == H5D_ONE_LINK_CHUNK_IO_MORE_OPT) {
+        if(H5D_ONE_LINK_CHUNK_IO_MORE_OPT == io_option) {
             new_value = 0;
             if(H5Pset(io_info->dxpl_id, H5D_XFER_COLL_CHUNK_LINK_NUM_TRUE_NAME, &new_value) < 0)
                 HGOTO_ERROR(H5E_IO, H5E_CANTSET, FAIL, "unable to set property value")
@@ -682,39 +687,16 @@ H5D__chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *type_inf
 #endif
 
     /* step 2:  Go ahead to do IO.*/
-    if(io_option == H5D_ONE_LINK_CHUNK_IO || io_option == H5D_ONE_LINK_CHUNK_IO_MORE_OPT) {
-        /* set the actual io mode properties to the correct values for link chunk io.
-         * Link chunk I/O does not break to independent, so we can set the actual_io mode
-         * as well as the optimisation mode. */
-        actual_chunk_opt_mode = H5D_MPIO_LINK_CHUNK;
-        actual_io_mode = H5D_MPIO_CHUNK_COLLECTIVE;
-
-        /* Set the actual chunk opt mode property. */
-        if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_CHUNK_OPT_MODE_NAME, &actual_chunk_opt_mode) < 0)
-            HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual chunk opt mode property")
-
-        if(H5D__link_chunk_collective_io(io_info, type_info, fm, sum_chunk) < 0)
+    if(H5D_ONE_LINK_CHUNK_IO == io_option || H5D_ONE_LINK_CHUNK_IO_MORE_OPT == io_option) {
+        if(H5D__link_chunk_collective_io(io_info, type_info, fm, sum_chunk, dx_plist) < 0)
             HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL, "couldn't finish linked chunk MPI-IO")
-
-        /* Set the actual io mode property. */
-        if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_IO_MODE_NAME, &actual_io_mode) < 0)
-            HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual io mode property")
     } /* end if */
-    else
-    if(io_option == H5D_MULTI_CHUNK_IO) {
-        /* Set the actual chunk opt mode property */
-        actual_chunk_opt_mode = H5D_MPIO_MULTI_CHUNK_NO_OPT;
-        if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_CHUNK_OPT_MODE_NAME, &actual_chunk_opt_mode) < 0)
-            HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual chunk opt mode property")
-
-        if(H5D__multi_chunk_collective_io_no_opt(io_info, type_info, fm, dx_plist) < 0)
-            HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL, "couldn't finish multiple chunk MPI-IO")
+    /* direct request to multi-chunk-io */
+    else if(H5D_MULTI_CHUNK_IO == io_option) {
+        if(H5D__multi_chunk_collective_io(io_info, type_info, fm, dx_plist) < 0)
+            HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL, "couldn't finish optimized multiple chunk MPI-IO")
     } /* end if */
-    else { /*multiple chunk IOs with opt */
-        actual_chunk_opt_mode = H5D_MPIO_MULTI_CHUNK;
-        if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_CHUNK_OPT_MODE_NAME, &actual_chunk_opt_mode) < 0)
-            HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual chunk opt mode property")
-
+    else { /* multiple chunk IO via threshold */
         if(H5D__multi_chunk_collective_io(io_info, type_info, fm, dx_plist) < 0)
             HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL, "couldn't finish optimized multiple chunk MPI-IO")
     } /* end else */
@@ -801,11 +783,16 @@ done:
  * Programmer:  Muqun Yang
  *              Monday, Feb. 13th, 2006
  *
+ * Modification:
+ *  - Set H5D_MPIO_ACTUAL_CHUNK_OPT_MODE_NAME and H5D_MPIO_ACTUAL_IO_MODE_NAME
+ *    dxpl in this.
+ * Programmer: Jonathan Kim
+ * Date: 2012-10-10
  *-------------------------------------------------------------------------
  */
 static herr_t
 H5D__link_chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *type_info,
-    H5D_chunk_map_t *fm, int sum_chunk)
+    H5D_chunk_map_t *fm, int sum_chunk, H5P_genplist_t *dx_plist)
 {
     H5D_chunk_addr_info_t *chunk_addr_info_array = NULL;
     MPI_Datatype chunk_final_mtype;         /* Final memory MPI datatype for all chunks with seletion */
@@ -824,10 +811,21 @@ H5D__link_chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *typ
     int                *chunk_mpi_file_counts = NULL;   /* Count of MPI file datatype for each chunk */
     int                *chunk_mpi_mem_counts = NULL;    /* Count of MPI memory datatype for each chunk */
     int                 mpi_code;           /* MPI return code */
+    H5D_mpio_actual_chunk_opt_mode_t actual_chunk_opt_mode = H5D_MPIO_LINK_CHUNK;
+    H5D_mpio_actual_io_mode_t actual_io_mode = H5D_MPIO_CHUNK_COLLECTIVE;
     herr_t              ret_value = SUCCEED;
 
     FUNC_ENTER_STATIC
 
+    /* Set the actual-chunk-opt-mode property. */
+    if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_CHUNK_OPT_MODE_NAME, &actual_chunk_opt_mode) < 0)
+        HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual chunk opt mode property")
+
+    /* Set the actual-io-mode property.
+     * Link chunk I/O does not break to independent, so can set right away */
+    if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_IO_MODE_NAME, &actual_io_mode) < 0)
+        HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual io mode property")
+
     /* Get the sum # of chunks, if not already available */
     if(sum_chunk < 0) {
         if(H5D__mpio_get_sum_chunk(io_info, fm, &sum_chunk) < 0)
@@ -1075,6 +1073,12 @@ if(H5DEBUG(D))
  * Programmer:  Muqun Yang
  *              Monday, Feb. 13th, 2006
  *
+ * Modification:
+ *  - Set H5D_MPIO_ACTUAL_CHUNK_OPT_MODE_NAME dxpl in this to go along with
+ *    setting H5D_MPIO_ACTUAL_IO_MODE_NAME dxpl at the bottom.
+ * Programmer: Jonathan Kim
+ * Date: 2012-10-10
+ *
  *-------------------------------------------------------------------------
  */
 static herr_t
@@ -1096,11 +1100,16 @@ H5D__multi_chunk_collective_io(H5D_io_info_t *io_info, const H5D_type_info_t *ty
     int mpi_rank;
 #endif
     size_t              u;                    /* Local index variable */
+    H5D_mpio_actual_chunk_opt_mode_t actual_chunk_opt_mode = H5D_MPIO_MULTI_CHUNK;  /* actual chunk optimization mode */
     H5D_mpio_actual_io_mode_t actual_io_mode = H5D_MPIO_NO_COLLECTIVE; /* Local variable for tracking the I/O mode used. */
     herr_t              ret_value = SUCCEED;
 
     FUNC_ENTER_STATIC
 
+    /* Set the actual chunk opt mode property */
+    if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_CHUNK_OPT_MODE_NAME, &actual_chunk_opt_mode) < 0)
+        HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual chunk opt mode property")
+
 #ifdef H5Dmpio_DEBUG
     mpi_rank = H5F_mpi_get_rank(io_info->dset->oloc.file);
 #endif
@@ -1263,210 +1272,6 @@ done:
 } /* end H5D__multi_chunk_collective_io */
 
 
-/*-------------------------------------------------------------------------
- * Function:    H5D__multi_chunk_collective_io_no_opt
- *
- * Purpose:     To do collective IO without any optimization per chunk base
- *              The internal independent IO inside HDF5 cannot handle
- *              non-contiguous(or with holes) storage efficiently.
- *              Under this case, the one independent IO call may consist of
- *              many small disk IOs. So we may use independent IO with derived datatype
- *              to replace the independent IO when we find this chunk is not good to
- *              do collective IO. However, according to our performance study,
- *              this approach may not overcome the overhead caused by MPI gather/scatter.
- *              So we decide to leave the original collective IO per chunk approach as
- *              an option for users. NO MPI gather/scatter calls are used.
- *              HDF5 will try to collective IO if possible.
- *              If users choose to use
- *              H5Pset_dxpl_mpio_chunk_opt(dxpl_id,H5FD_MPIO_OPT_MULTI_IO),
- *              this function will be called.
- *              The HDF5 library won't do any IO management but leave it to MPI-IO to figure
- *              out.
- *
- * Return:      Non-negative on success/Negative on failure
- *
- * Programmer:  Muqun Yang
- *              Monday, Feb. 13th, 2006
- *
- *-------------------------------------------------------------------------
- */
-static herr_t
-H5D__multi_chunk_collective_io_no_opt(H5D_io_info_t *io_info,
-    const H5D_type_info_t *type_info, H5D_chunk_map_t *fm, H5P_genplist_t *dx_plist)
-{
-    H5SL_node_t    *chunk_node;           /* Current node in chunk skip list */
-    H5D_io_info_t   ctg_io_info;          /* Contiguous I/O info object */
-    H5D_storage_t   ctg_store;            /* Chunk storage information as contiguous dataset */
-    H5D_io_info_t   cpt_io_info;          /* Compact I/O info object */
-    H5D_storage_t   cpt_store;            /* Chunk storage information as compact dataset */
-    hbool_t         cpt_dirty;            /* Temporary placeholder for compact storage "dirty" flag */
-    int             min_chunk = -1;       /* Minimum # of chunks all processes will operate on */
-    int             count_chunk;          /* How many chunks have we operated on? */
-    H5D_storage_t   store;                /* union of EFL and chunk pointer in file space */
-    H5D_mpio_actual_io_mode_t actual_io_mode = H5D_MPIO_NO_COLLECTIVE; /*Local variable for tracking the I/O modes used. */
-    herr_t          ret_value = SUCCEED;
-
-    FUNC_ENTER_STATIC
-
-#ifdef H5D_DEBUG
-if(H5DEBUG(D)) {
-    int mpi_rank;
-
-    mpi_rank = H5F_mpi_get_rank(io_info->dset->oloc.file);
-    HDfprintf(H5DEBUG(D), "Rank %d: coming to multi_chunk_collective_io_no_opt\n", mpi_rank);
-}
-#endif
-
-    /* Set up contiguous I/O info object */
-    HDmemcpy(&ctg_io_info, io_info, sizeof(ctg_io_info));
-    ctg_io_info.store = &ctg_store;
-    ctg_io_info.layout_ops = *H5D_LOPS_CONTIG;
-
-    /* Initialize temporary contiguous storage info */
-    ctg_store.contig.dset_size = (hsize_t)io_info->dset->shared->layout.u.chunk.size;
-
-    /* Set up compact I/O info object */
-    HDmemcpy(&cpt_io_info, io_info, sizeof(cpt_io_info));
-    cpt_io_info.store = &cpt_store;
-    cpt_io_info.layout_ops = *H5D_LOPS_COMPACT;
-
-    /* Initialize temporary compact storage info */
-    cpt_store.compact.dirty = &cpt_dirty;
-
-    /* Set dataset storage for I/O info */
-    io_info->store = &store;
-
-    /* Get the min. # of chunks */
-    if(H5D__mpio_get_min_chunk(io_info, fm, &min_chunk) < 0)
-        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get minimum number of chunk")
-    HDassert(min_chunk >= 0);
-
-    /* Get first node in chunk skip list */
-    chunk_node = H5SL_first(fm->sel_chunks);
-    count_chunk = 0;
-
-     /* Iterate through chunks to be operated on */
-    while(chunk_node) {
-        H5D_chunk_info_t *chunk_info;   /* chunk information */
-        H5D_chunk_ud_t udata;           /* B-tree pass-through */
-        hbool_t make_ind, make_coll;    /* Flags to indicate that the MPI mode should change */
-
-        /* Get the actual chunk information from the skip list node */
-        chunk_info = H5SL_item(chunk_node);
-
-        /* Pass in chunk's coordinates in a union. */
-        store.chunk.offset = chunk_info->coords;
-        store.chunk.index = chunk_info->index;
-
-        /* Reset flags for changing parallel I/O mode */
-        make_ind = make_coll = FALSE;
-
-        count_chunk++;
-
-        /* If the number of chunk is greater than minimum number of chunk,
-         *      Do independent read.
-         */
-        if(count_chunk > min_chunk)
-            /* Switch to independent I/O (permanently) */
-            make_ind = TRUE;
-
-        /* Retrieve the chunk's address */
-        if(H5D__chunk_lookup(io_info->dset, io_info->dxpl_id, chunk_info->coords,
-                chunk_info->index, &udata) < 0)
-            HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL, "couldn't get chunk info from skipped list")
-
-        /* Independent I/O */
-        if(make_ind) {
-            void *chunk;                    /* Pointer to the data chunk in cache */
-            H5D_io_info_t *chk_io_info;     /* Pointer to I/O info object for this chunk */
-            uint32_t accessed_bytes = 0;    /* Total accessed size in a chunk */
-            htri_t cacheable;               /* Whether the chunk is cacheable */
-
-            /* Switch to independent I/O */
-            if(H5D__ioinfo_xfer_mode(io_info, dx_plist, H5FD_MPIO_INDEPENDENT) < 0)
-                HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to independent I/O")
-
-            /* Update the local variable tracking the dxpl's actual io mode */
-            actual_io_mode = actual_io_mode | H5D_MPIO_CHUNK_INDEPENDENT;
-
-            /* Load the chunk into cache and lock it. */
-            if((cacheable = H5D__chunk_cacheable(io_info, udata.addr,
-                    io_info->op_type == H5D_IO_OP_WRITE)) < 0)
-                HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't tell if chunk is cacheable")
-            if(cacheable) {
-                hbool_t entire_chunk = TRUE;         /* Whether whole chunk is selected */
-
-                /* Compute # of bytes accessed in chunk */
-                accessed_bytes = chunk_info->chunk_points * type_info->src_type_size;
-
-                /* Determine if we will access all the data in the chunk */
-                if(((io_info->op_type == H5D_IO_OP_WRITE) && (accessed_bytes != ctg_store.contig.dset_size))
-                        || (io_info->op_type != H5D_IO_OP_WRITE))
-                    entire_chunk = FALSE;
-
-                /* Lock the chunk into the cache */
-                if(NULL == (chunk = H5D__chunk_lock(io_info, &udata, entire_chunk)))
-                    HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "unable to read raw data chunk")
-
-                /* Set up the storage buffer information for this chunk */
-                cpt_store.compact.buf = chunk;
-
-                /* Point I/O info at contiguous I/O info for this chunk */
-                chk_io_info = &cpt_io_info;
-            } /* end if */
-            else {
-                /* Set up the storage address information for this chunk */
-                ctg_store.contig.dset_addr = udata.addr;
-
-                /* No chunk cached */
-                chunk = NULL;
-
-                /* Point I/O info at temporary I/O info for this chunk */
-                chk_io_info = &ctg_io_info;
-            } /* end else */
-
-            if(io_info->op_type == H5D_IO_OP_WRITE) {
-                if((io_info->io_ops.single_write)(chk_io_info, type_info,
-                        (hsize_t)chunk_info->chunk_points, chunk_info->fspace, chunk_info->mspace) < 0)
-                    HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "optimized write failed")
-            } /* end if */
-            else {
-                if((io_info->io_ops.single_read)(chk_io_info, type_info,
-                        (hsize_t)chunk_info->chunk_points, chunk_info->fspace, chunk_info->mspace) < 0)
-                   HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "optimized read failed")
-            } /* end ese */
-
-            /* Release the cache lock on the chunk. */
-            if(chunk)
-                if(H5D__chunk_unlock(io_info, &udata, (io_info->op_type == H5D_IO_OP_WRITE), chunk, accessed_bytes) < 0)
-                    HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "unable to unlock raw data chunk")
-        } /* end if */
-        else { /*collective I/O */
-            /* Set up the storage address information for this chunk */
-            ctg_store.contig.dset_addr = udata.addr;
-
-            /* Update the local variable tracking the dxpl's actual io Mode. */
-            actual_io_mode = actual_io_mode | H5D_MPIO_CHUNK_COLLECTIVE;
-
-            if(H5D__inter_collective_io(&ctg_io_info, type_info, chunk_info->fspace, chunk_info->mspace) < 0)
-                HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO")
-        } /* end else */
-
-        if(make_coll)
-            if(H5D__ioinfo_xfer_mode(io_info, dx_plist, H5FD_MPIO_COLLECTIVE) < 0)
-                HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to independent I/O")
-
-        /* Get the next chunk node in the skip list */
-        chunk_node = H5SL_next(chunk_node);
-    } /* end while */
-
-    /* Write the local value of actual io mode to the DXPL. */
-    if(H5P_set(dx_plist, H5D_MPIO_ACTUAL_IO_MODE_NAME, &actual_io_mode) < 0)
-        HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "couldn't set actual io mode property")
-
-done:
-    FUNC_LEAVE_NOAPI(ret_value)
-} /* end H5D__multi_chunk_collective_io_no_opt */
 
 
 /*-------------------------------------------------------------------------
@@ -1794,7 +1599,6 @@ H5D__obtain_mpio_mode(H5D_io_info_t* io_info, H5D_chunk_map_t *fm,
 {
     int               total_chunks;
     unsigned          percent_nproc_per_chunk, threshold_nproc_per_chunk;
-    H5FD_mpio_chunk_opt_t chunk_opt_mode;
     uint8_t*          io_mode_info = NULL;
     uint8_t*          recv_io_mode_info = NULL;
     uint8_t*          mergebuf = NULL;
@@ -1827,8 +1631,8 @@ H5D__obtain_mpio_mode(H5D_io_info_t* io_info, H5D_chunk_map_t *fm,
     /* Setup parameters */
     H5_ASSIGN_OVERFLOW(total_chunks, fm->layout->u.chunk.nchunks, hsize_t, int);
     percent_nproc_per_chunk = H5P_peek_unsigned(dx_plist, H5D_XFER_MPIO_CHUNK_OPT_RATIO_NAME);
-    chunk_opt_mode = (H5FD_mpio_chunk_opt_t)H5P_peek_unsigned(dx_plist, H5D_XFER_MPIO_CHUNK_OPT_HARD_NAME);
-    if((chunk_opt_mode == H5FD_MPIO_CHUNK_MULTI_IO) || (percent_nproc_per_chunk == 0)) {
+    /* if ratio is 0, perform collective io */
+    if(0 == percent_nproc_per_chunk) {
         if(H5D__chunk_addrmap(io_info, chunk_addr) < 0)
            HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address");
         for(ic = 0; ic < total_chunks; ic++)
diff --git a/src/H5Dprivate.h b/src/H5Dprivate.h
index 9adf6a5..1e2b91d 100644
--- a/src/H5Dprivate.h
+++ b/src/H5Dprivate.h
@@ -87,8 +87,6 @@
 #define H5D_XFER_COLL_CHUNK_LINK_NUM_FALSE_NAME "coll_chunk_link_false"
 #define H5D_XFER_COLL_CHUNK_MULTI_RATIO_COLL_NAME "coll_chunk_multi_coll"
 #define H5D_XFER_COLL_CHUNK_MULTI_RATIO_IND_NAME "coll_chunk_multi_ind"
-#define H5D_XFER_COLL_CHUNK_LINK_TO_MULTI "coll_chunk_link_mul"/* Internal transferring from link to multiple chunk */
-#define H5D_XFER_COLL_CHUNK_LINK_TO_MULTI_OPT "coll_chunk_link_mul_opt"/* Internal transferring from link opt to multiple chunk opt*/
 
 /* Definitions for all collective chunk instrumentation properties */
 #define H5D_XFER_COLL_CHUNK_SIZE        sizeof(unsigned)
diff --git a/src/H5FDcore.c b/src/H5FDcore.c
index 932e2af..07f4551 100644
--- a/src/H5FDcore.c
+++ b/src/H5FDcore.c
@@ -481,7 +481,7 @@ H5FD_core_open(const char *name, unsigned flags, hid_t fapl_id,
      * default value. But if the file access property list was zero then use
      * the default value instead.
      */
-    file->increment = (fa->increment>0) ?  fa->increment : H5FD_CORE_INCREMENT;
+    file->increment = (fa->increment > 0) ? fa->increment : H5FD_CORE_INCREMENT;
 
     /* If save data in backing store. */
     file->backing_store = fa->backing_store;
diff --git a/src/H5FDdirect.c b/src/H5FDdirect.c
index 4721a44..9715bc1 100644
--- a/src/H5FDdirect.c
+++ b/src/H5FDdirect.c
@@ -533,7 +533,8 @@ H5FD_direct_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxadd
     /* Get the driver specific information */
     if(NULL == (plist = H5P_object_verify(fapl_id,H5P_FILE_ACCESS)))
         HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, NULL, "not a file access property list")
-    fa = H5P_get_driver_info(plist);
+    if(NULL == (fa = (H5FD_direct_fapl_t *)H5P_get_driver_info(plist)))
+        HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, NULL, "bad VFL driver info")
 
     file->fd = fd;
     H5_ASSIGN_OVERFLOW(file->eof,sb.st_size,h5_stat_size_t,haddr_t);
@@ -562,9 +563,9 @@ H5FD_direct_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxadd
      * is to handle correctly the case that the file is in a different file system
      * than the one where the program is running.
      */
-    buf1 = (int*)HDmalloc(sizeof(int));
-    if (HDposix_memalign(&buf2, file->fa.mboundary, file->fa.fbsize) != 0)
-  HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, NULL, "HDposix_memalign failed")
+    buf1 = (int *)HDmalloc(sizeof(int));
+    if(HDposix_memalign(&buf2, file->fa.mboundary, file->fa.fbsize) != 0)
+        HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, NULL, "HDposix_memalign failed")
 
     if(o_flags & O_CREAT) {
         if(write(file->fd, (void*)buf1, sizeof(int))<0) {
diff --git a/src/H5FDfamily.c b/src/H5FDfamily.c
index f051269..dc535a2 100644
--- a/src/H5FDfamily.c
+++ b/src/H5FDfamily.c
@@ -675,8 +675,8 @@ H5FD_family_open(const char *name, unsigned flags, hid_t fapl_id,
 
         if(NULL == (plist = (H5P_genplist_t *)H5I_object(fapl_id)))
             HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, NULL, "not a file access property list")
-        fa = (H5FD_family_fapl_t *)H5P_get_driver_info(plist);
-        HDassert(fa);
+        if(NULL == (fa = (H5FD_family_fapl_t *)H5P_get_driver_info(plist)))
+            HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, NULL, "bad VFL driver info")
 
         /* Check for new family file size. It's used by h5repart only. */
         if(H5P_exist_plist(plist, H5F_ACS_FAMILY_NEWSIZE_NAME) > 0) {
diff --git a/src/H5FDlog.c b/src/H5FDlog.c
index 6a2425c..9f4abd3 100644
--- a/src/H5FDlog.c
+++ b/src/H5FDlog.c
@@ -529,8 +529,8 @@ H5FD_log_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr)
     /* Get the driver specific information */
     if(NULL == (plist = H5P_object_verify(fapl_id, H5P_FILE_ACCESS)))
         HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, NULL, "not a file access property list")
-    fa = (H5FD_log_fapl_t *)H5P_get_driver_info(plist);
-    HDassert(fa);
+    if(NULL == (fa = (H5FD_log_fapl_t *)H5P_get_driver_info(plist)))
+        HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, NULL, "bad VFL driver info")
 
 #ifdef H5_HAVE_GETTIMEOFDAY
     if(fa->flags & H5FD_LOG_TIME_OPEN)
@@ -644,9 +644,9 @@ H5FD_log_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr)
 
 #ifdef H5_HAVE_GETTIMEOFDAY
         if(file->fa.flags & H5FD_LOG_TIME_OPEN)
-            HDfprintf(file->logfp, "Open took: (%f s)\n", (double)open_timeval_diff.tv_sec + ((double)open_timeval_diff.tv_usec / (double)1000000.0));
+            HDfprintf(file->logfp, "Open took: (%f s)\n", (double)open_timeval_diff.tv_sec + ((double)open_timeval_diff.tv_usec / (double)1000000.0f));
         if(file->fa.flags & H5FD_LOG_TIME_STAT)
-            HDfprintf(file->logfp, "Stat took: (%f s)\n", (double)stat_timeval_diff.tv_sec + ((double)stat_timeval_diff.tv_usec / (double)1000000.0));
+            HDfprintf(file->logfp, "Stat took: (%f s)\n", (double)stat_timeval_diff.tv_sec + ((double)stat_timeval_diff.tv_usec / (double)1000000.0f));
 #endif /* H5_HAVE_GETTIMEOFDAY */
 
     } /* end if */
@@ -734,7 +734,7 @@ H5FD_log_close(H5FD_t *_file)
                  timeval_diff.tv_usec += 1000000;
                  timeval_diff.tv_sec--;
              } /* end if */
-            HDfprintf(file->logfp, "Close took: (%f s)\n", (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0));
+            HDfprintf(file->logfp, "Close took: (%f s)\n", (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0f));
         } /* end if */
 #endif /* H5_HAVE_GETTIMEOFDAY */
 
@@ -1195,7 +1195,7 @@ H5FD_log_read(H5FD_t *_file, H5FD_mem_t type, hid_t UNUSED dxpl_id, haddr_t addr
                     timeval_diff.tv_usec += 1000000;
                     timeval_diff.tv_sec--;
                 } /* end if */
-                time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0);
+                time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0f);
                 HDfprintf(file->logfp, " (%f s)\n", time_diff);
 
                 /* Add to total seek time */
@@ -1284,7 +1284,7 @@ H5FD_log_read(H5FD_t *_file, H5FD_mem_t type, hid_t UNUSED dxpl_id, haddr_t addr
                 timeval_diff.tv_usec += 1000000;
                 timeval_diff.tv_sec--;
             } /* end if */
-            time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0);
+            time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0f);
             HDfprintf(file->logfp, " (%f s)\n", time_diff);
 
             /* Add to total read time */
@@ -1401,7 +1401,7 @@ H5FD_log_write(H5FD_t *_file, H5FD_mem_t type, hid_t UNUSED dxpl_id, haddr_t add
                     timeval_diff.tv_usec += 1000000;
                     timeval_diff.tv_sec--;
                 } /* end if */
-                time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0);
+                time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0f);
                 HDfprintf(file->logfp, " (%f s)\n", time_diff);
 
                 /* Add to total seek time */
@@ -1487,7 +1487,7 @@ H5FD_log_write(H5FD_t *_file, H5FD_mem_t type, hid_t UNUSED dxpl_id, haddr_t add
                 timeval_diff.tv_usec += 1000000;
                 timeval_diff.tv_sec--;
             } /* end if */
-            time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0);
+            time_diff = (double)timeval_diff.tv_sec + ((double)timeval_diff.tv_usec / (double)1000000.0f);
             HDfprintf(file->logfp, " (%f s)\n", time_diff);
 
             /* Add to total write time */
diff --git a/src/H5FDmpio.c b/src/H5FDmpio.c
index cac101b..2db77c9 100644
--- a/src/H5FDmpio.c
+++ b/src/H5FDmpio.c
@@ -1013,8 +1013,8 @@ H5FD_mpio_open(const char *name, unsigned flags, hid_t fapl_id,
 	_fa.info = MPI_INFO_NULL; /*default*/
 	fa = &_fa;
     } else {
-	fa = (const H5FD_mpio_fapl_t *)H5P_get_driver_info(plist);
-	assert(fa);
+        if(NULL == (fa = (const H5FD_mpio_fapl_t *)H5P_get_driver_info(plist)))
+	    HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, NULL, "bad VFL driver info")
     }
 
     /* Duplicate communicator and Info object for use by this file. */
diff --git a/src/H5FDmpiposix.c b/src/H5FDmpiposix.c
index 0d0b839..092cd24 100644
--- a/src/H5FDmpiposix.c
+++ b/src/H5FDmpiposix.c
@@ -600,8 +600,8 @@ H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id,
         fa = &_fa;
     } /* end if */
     else {
-        fa = H5P_get_driver_info(plist);
-        HDassert(fa);
+        if(NULL == (fa = (const H5FD_mpiposix_fapl_t *)H5P_get_driver_info(plist)))
+            HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, NULL, "bad VFL driver info")
     } /* end else */
 
     /* Duplicate the communicator for use by this file. */
diff --git a/src/H5Pocpl.c b/src/H5Pocpl.c
index 6e20963..d8ddfee 100644
--- a/src/H5Pocpl.c
+++ b/src/H5Pocpl.c
@@ -1514,9 +1514,12 @@ H5P__ocrt_pipeline_dec(const void **_pp, void *_value)
         UINT64DECODE_VAR(*pp, enc_value, enc_size);
         filter.cd_nelmts = (size_t)enc_value;
 
-        if(filter.cd_nelmts)
+        if(filter.cd_nelmts) {
             if(NULL == (filter.cd_values = (unsigned *)H5MM_malloc(sizeof(unsigned) * filter.cd_nelmts)))
                 HGOTO_ERROR(H5E_PLIST, H5E_CANTALLOC, FAIL, "memory allocation failed for cd_values")
+        } /* end if */
+        else
+            filter.cd_values = NULL;
 
         /* decode values */
         for(v = 0; v < filter.cd_nelmts; v++)
diff --git a/src/H5Ppublic.h b/src/H5Ppublic.h
index 507bbb6..6b17e0b 100644
--- a/src/H5Ppublic.h
+++ b/src/H5Ppublic.h
@@ -131,8 +131,7 @@ typedef enum H5D_mpio_actual_chunk_opt_mode_t {
      */
     H5D_MPIO_NO_CHUNK_OPTIMIZATION = 0,
     H5D_MPIO_LINK_CHUNK,
-    H5D_MPIO_MULTI_CHUNK,
-    H5D_MPIO_MULTI_CHUNK_NO_OPT
+    H5D_MPIO_MULTI_CHUNK
 }  H5D_mpio_actual_chunk_opt_mode_t;
 
 typedef enum H5D_mpio_actual_io_mode_t {
diff --git a/src/H5detect.c b/src/H5detect.c
index 87a4fd5..d6f6a3b 100644
--- a/src/H5detect.c
+++ b/src/H5detect.c
@@ -120,7 +120,7 @@ static void print_results(int nd, detected_t *d, int na, malign_t *m);
 static void iprint(detected_t *);
 static int byte_cmp(int, const void *, const void *);
 static int bit_cmp(int, int *, volatile void *, volatile void *);
-static void fix_order(int, int, int, int *, const char **);
+static void fix_order(int, int, int *, const char **);
 static int imp_bit(int, int *, volatile void *, volatile void *);
 static unsigned long find_bias(int, int, int *, void *);
 static void precision (detected_t*);
@@ -293,7 +293,7 @@ precision (detected_t *d)
 #define DETECT_F(TYPE,VAR,INFO) {					      \
    volatile TYPE _v1, _v2, _v3;						      \
    unsigned char _buf1[sizeof(TYPE)], _buf3[sizeof(TYPE)];		      \
-   int _i, _j, _first = (-1), _last = (-1);				      \
+   int _i, _j, _last = (-1);						      \
    char *_mesg;								      \
 									      \
    memset(&INFO, 0, sizeof(INFO));					      \
@@ -315,15 +315,11 @@ precision (detected_t *d)
       memcpy(_buf3, (const void *)&_v3, sizeof(TYPE));			      \
       _j = byte_cmp(sizeof(TYPE), &_buf3, &_buf1);			      \
       if(_j >= 0) {							      \
-	 if(0 == _i || INFO.perm[_i - 1] != _j) {			      \
-	    INFO.perm[_i] = _j;						      \
-	    _last = _i;							      \
-	    if(_first < 0)						      \
-                _first = _i;						      \
-	 }								      \
+          INFO.perm[_i] = _j;						      \
+          _last = _i;							      \
       }									      \
    }									      \
-   fix_order(sizeof(TYPE), _first, _last, INFO.perm, (const char**)&_mesg);   \
+   fix_order(sizeof(TYPE), _last, INFO.perm, (const char**)&_mesg);	      \
                                                                               \
    if(!strcmp(_mesg, "VAX"))                                                  \
       INFO.is_vax = TRUE;                                                     \
@@ -973,11 +969,11 @@ bit_cmp(int nbytes, int *perm, volatile void *_a, volatile void *_b)
  *-------------------------------------------------------------------------
  */
 static void
-fix_order(int n, int first, int last, int *perm, const char **mesg)
+fix_order(int n, int last, int *perm, const char **mesg)
 {
     int		i;
 
-    if (first + 1 < last) {
+    if (last > 1) {
 	/*
 	 * We have at least three points to consider.
 	 */
@@ -998,6 +994,9 @@ fix_order(int n, int first, int last, int *perm, const char **mesg)
 	} else {
 	    /*
 	     * Bi-endian machines like VAX.
+             * (NOTE: This is not an actual determination of the VAX-endianess.
+             *          It could have some other endianess and fall into this
+             *          case - JKM & QAK)
 	     */
 	    assert(0 == n % 2);
 	    if (mesg) *mesg = "VAX";
diff --git a/src/H5public.h b/src/H5public.h
index 88e514d..5d9e3ef 100644
--- a/src/H5public.h
+++ b/src/H5public.h
@@ -75,10 +75,10 @@ extern "C" {
 /* Version numbers */
 #define H5_VERS_MAJOR	1	/* For major interface/format changes  	     */
 #define H5_VERS_MINOR	9	/* For minor interface/format changes  	     */
-#define H5_VERS_RELEASE	132	/* For tweaks, bug-fixes, or development     */
+#define H5_VERS_RELEASE	136	/* For tweaks, bug-fixes, or development     */
 #define H5_VERS_SUBRELEASE ""	/* For pre-releases like snap0       */
 				/* Empty string for real releases.           */
-#define H5_VERS_INFO    "HDF5 library version: 1.9.132"      /* Full version string */
+#define H5_VERS_INFO    "HDF5 library version: 1.9.136"      /* Full version string */
 
 #define H5check()	H5check_version(H5_VERS_MAJOR,H5_VERS_MINOR,	      \
 				        H5_VERS_RELEASE)
diff --git a/src/H5trace.c b/src/H5trace.c
index 2dab8ec..92736d4 100644
--- a/src/H5trace.c
+++ b/src/H5trace.c
@@ -612,10 +612,6 @@ H5_trace(const double *returning, const char *func, const char *type, ...)
                                     fprintf(out, "H5D_MPIO_MULTI_CHUNK");
                                     break;
 
-                                case H5D_MPIO_MULTI_CHUNK_NO_OPT:
-                                    fprintf(out, "H5D_MPIO_MULTI_CHUNK_NO_OPT");
-                                    break;
-
                                 default:
                                     fprintf(out, "%ld", (long)chunk_opt_mode);
                                     break;
diff --git a/src/Makefile.in b/src/Makefile.in
index c673bcf..fc89be5 100644
--- a/src/Makefile.in
+++ b/src/Makefile.in
@@ -522,7 +522,7 @@ CHECK_CLEANFILES = *.chkexe *.chklog *.clog
 # Add libtool shared library version numbers to the HDF5 library
 # See libtool versioning documentation online.
 LT_VERS_INTERFACE = 6
-LT_VERS_REVISION = 122
+LT_VERS_REVISION = 126
 LT_VERS_AGE = 0
 H5detect_CFLAGS = -g $(AM_CFLAGS)