[svn-r11312] Purpose:

Bug #281

Description:
A dataset created in serial mode with H5D_ALLOC_TIME_INCR allocation setting
was not extendible, either explicitly by H5Dextend or implicitly by writing
to unallocated chunks.  This was because parallel mode expects the allocation
mode be H5D_ALLOC_TIME_INCR only.

Solution:
Modified library to allocate more space when needed or directed if the
file is opened by parallel mode, independent of what the dataset allocation
mode is.

Platforms tested:
Heping pp.

5 files changed, 202 insertions, 3 deletions

diff --git a/testpar/Makefile.am b/testpar/Makefile.am
index 92985b8..7c4f567 100644
--- a/testpar/Makefile.am
+++ b/testpar/Makefile.am
@@ -31,7 +31,7 @@ check_PROGRAMS = $(TEST_PROG_PARA)
 check_SCRIPTS= $(TEST_SCRIPT)
 
 testphdf5_SOURCES=testphdf5.c t_dset.c t_file.c t_mdset.c t_ph5basic.c \
-        t_coll_chunk.c t_span_tree.c
+        t_coll_chunk.c t_span_tree.c t_chunk_alloc.c
 
 # The tests all depend on the hdf5 library and the test library
 LDADD = $(LIBHDF5) $(LIBH5TEST)
diff --git a/testpar/Makefile.in b/testpar/Makefile.in
index 0a7aea4..c886e66 100644
--- a/testpar/Makefile.in
+++ b/testpar/Makefile.in
@@ -78,7 +78,8 @@ t_mpi_LDADD = $(LDADD)
 t_mpi_DEPENDENCIES = $(am__DEPENDENCIES_1) $(am__DEPENDENCIES_2)
 am_testphdf5_OBJECTS = testphdf5.$(OBJEXT) t_dset.$(OBJEXT) \
 	t_file.$(OBJEXT) t_mdset.$(OBJEXT) t_ph5basic.$(OBJEXT) \
-	t_coll_chunk.$(OBJEXT) t_span_tree.$(OBJEXT)
+	t_coll_chunk.$(OBJEXT) t_span_tree.$(OBJEXT) \
+	t_chunk_alloc.$(OBJEXT)
 testphdf5_OBJECTS = $(am_testphdf5_OBJECTS)
 testphdf5_LDADD = $(LDADD)
 testphdf5_DEPENDENCIES = $(am__DEPENDENCIES_1) $(am__DEPENDENCIES_2)
@@ -321,7 +322,7 @@ TEST_PROG_PARA = t_mpi t_fphdf5 testphdf5
 TEST_SCRIPT_PARA = testph5.sh
 check_SCRIPTS = $(TEST_SCRIPT)
 testphdf5_SOURCES = testphdf5.c t_dset.c t_file.c t_mdset.c t_ph5basic.c \
-        t_coll_chunk.c t_span_tree.c
+        t_coll_chunk.c t_span_tree.c t_chunk_alloc.c
 
 
 # The tests all depend on the hdf5 library and the test library
@@ -403,6 +404,7 @@ mostlyclean-compile:
 distclean-compile:
 	-rm -f *.tab.c
 
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/t_chunk_alloc.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/t_coll_chunk.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/t_dset.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/t_file.Po@am__quote@
diff --git a/testpar/t_chunk_alloc.c b/testpar/t_chunk_alloc.c
new file mode 100644
index 0000000..cd56c40
--- /dev/null
+++ b/testpar/t_chunk_alloc.c
@@ -0,0 +1,194 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ * Copyright by the Board of Trustees of the University of Illinois.         *
+ * All rights reserved.                                                      *
+ *                                                                           *
+ * This file is part of HDF5.  The full HDF5 copyright notice, including     *
+ * terms governing use, modification, and redistribution, is contained in    *
+ * the files COPYING and Copyright.html.  COPYING can be found at the root   *
+ * of the source code distribution tree; Copyright.html can be found at the  *
+ * root level of an installed copy of the electronic HDF5 document set and   *
+ * is linked from the top-level documents page.  It can also be found at     *
+ * http://hdf.ncsa.uiuc.edu/HDF5/doc/Copyright.html.  If you do not have     *
+ * access to either file, you may request a copy from hdfhelp@ncsa.uiuc.edu. *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+/* 
+ * This verifies if the storage space allocation methods are compatible between
+ * serial and parallel modes.
+ *
+ * Created by: Christian Chilan and Albert Cheng
+ * Date: 2005/07/05
+ */
+
+#include "testphdf5.h"
+
+#define DATASETNAME "ExtendibleArray" 
+#define ELEMS		 20000000
+
+/*
+ * This creates a dataset serially with 20000 chunks, each of 1000
+ * elements. It does not perform any writing on it. Another routine
+ * will open this in parallel for extension test.
+ */
+void
+create_chunked_dataset(const char *filename)
+{
+    hid_t       file;                          /* handles */
+    hid_t       dataspace, dataset;  
+    hid_t       cparms;                     
+    int 	mpi_size, mpi_rank;
+    hsize_t      dims[1]  = {20*1000000};     /* dataset dimensions
+                                                 at creation time */
+
+    hsize_t      maxdims[1] = {H5S_UNLIMITED};
+    herr_t       hrc;                             
+
+    /* Variables used in reading data back */
+    hsize_t      chunk_dims[1] ={1000};
+
+    /* set up MPI parameters */
+    MPI_Comm_size(MPI_COMM_WORLD,&mpi_size);
+    MPI_Comm_rank(MPI_COMM_WORLD,&mpi_rank);
+
+    /* Create the data space with unlimited dimensions. */
+    dataspace = H5Screate_simple (1, dims, maxdims); 
+    VRFY((dataspace >= 0), "");
+
+    /* Create a new file. If file exists its contents will be overwritten. */
+    file = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);
+    VRFY((file >= 0), "");
+
+    /* Modify dataset creation properties, i.e. enable chunking  */
+    cparms = H5Pcreate (H5P_DATASET_CREATE);
+    VRFY((cparms >= 0), "");
+
+    hrc = H5Pset_chunk ( cparms, 1, chunk_dims);
+    VRFY((hrc >= 0), "");
+
+    /* Create a new dataset within the file using cparms creation properties. */
+    dataset = H5Dcreate (file, DATASETNAME, H5T_NATIVE_INT, dataspace, cparms);
+    VRFY((dataset >= 0), "");
+
+    /* Close resources */
+    hrc = H5Dclose (dataset);
+    VRFY((hrc >= 0), "");
+
+    hrc = H5Fclose (file);
+    VRFY((hrc >= 0), "");
+}     
+
+/*
+ * This program extends the size of a small dataset to 20000 chunks (each of 
+ * 1000 elements) in parallel. The dataset was initially created in a serial
+ * environment. A correct extend operation should increase the file size
+ * considerably.
+ */
+void
+extend_chunked_dataset(const char *filename)
+{
+    /* HDF5 gubbins */
+    hid_t    file_id, dataset;             /* HDF5 file identifier */
+    hid_t    access_plist;                 /* HDF5 ID for file access property list */
+    herr_t   hrc;                          /* HDF5 return code */
+
+    hsize_t  size[1];
+    int mpi_size, mpi_rank;
+
+    long int lb;
+
+    /* MPI Gubbins */
+    MPI_Comm     comm;
+    MPI_Info     info;
+    MPI_File	   thefile;
+    MPI_Offset   filesize;
+
+    int          mpierr, ret;
+
+
+printf("extend_chunked_dataset: filename=%s\n", filename);
+    /* Initialize MPI */
+    comm = MPI_COMM_WORLD;
+    info = MPI_INFO_NULL;
+
+    /* set up MPI parameters */
+    MPI_Comm_size(MPI_COMM_WORLD,&mpi_size);
+    MPI_Comm_rank(MPI_COMM_WORLD,&mpi_rank);
+
+    /* Set up HDF5 file creation, access and independent data transfer property lists */
+    access_plist  = H5Pcreate(H5P_FILE_ACCESS);
+    VRFY((access_plist >= 0), "");
+
+    hrc = H5Pset_fapl_mpio(access_plist, comm, info);
+    VRFY((hrc >= 0), "");
+
+    /* Open the file */
+    file_id = H5Fopen(filename, H5F_ACC_RDWR, access_plist);
+    VRFY((file_id >= 0), "");
+
+    /* Can close some plists */
+    hrc = H5Pclose(access_plist);
+    VRFY((hrc >= 0), "");
+
+    /* Open dataset*/
+    dataset   = H5Dopen(file_id, DATASETNAME);
+    VRFY((dataset >= 0), "");
+
+    size[0] = ELEMS;
+
+    /* Extend dataset*/
+    hrc = H5Dextend(dataset, size);
+    VRFY((hrc >= 0), "");
+
+    /* Close up */
+    hrc = H5Dclose(dataset);
+    VRFY((hrc >= 0), "");
+
+    hrc = H5Fclose(file_id);
+    VRFY((hrc >= 0), "");
+
+    mpierr = MPI_File_open(comm, filename, MPI_MODE_RDONLY, info, &thefile);
+    VRFY((mpierr == MPI_SUCCESS), "");
+
+    mpierr = MPI_File_get_size(thefile, &filesize);
+    VRFY((mpierr == MPI_SUCCESS), "");
+
+
+    /* Definition of lower bound */
+    lb = ELEMS*sizeof(H5T_NATIVE_INT);
+
+    /* Print size of file. */
+    printf("File size: %ld\n", (long int)filesize);
+
+    if (filesize>=lb){
+	printf("Test passed\n");
+	ret = 0;
+    } else {
+	printf("Test failed\n");
+	ret = 1;
+    }
+
+    mpierr = MPI_File_close(&thefile);
+    VRFY((mpierr == MPI_SUCCESS), "");
+
+    hrc = H5close();
+    VRFY((hrc >= 0), "");
+
+    /* Finalize */
+    mpierr = MPI_Finalize();
+    VRFY((mpierr == MPI_SUCCESS), "");
+}
+
+void
+test_chunk_alloc(void)
+{
+    const char *filename;
+  
+    filename = GetTestParameters();
+    if (VERBOSE_MED)
+	printf("Extend Chunked allocation test on file %s\n", filename);
+
+    /* create the datafile first */
+    create_chunked_dataset(filename);
+    /* reopen it in parallel and extend it. */
+    /*extend_chunked_dataset(filename); */
+}
diff --git a/testpar/testphdf5.c b/testpar/testphdf5.c
index 2f75e51..10ce989 100644
--- a/testpar/testphdf5.c
+++ b/testpar/testphdf5.c
@@ -381,6 +381,8 @@ int main(int argc, char **argv)
 	    "extendible dataset collective read", PARATESTFILE);
     AddTest("eidsetw2", extend_writeInd2, NULL,
 	    "extendible dataset independent write #2", PARATESTFILE);
+    AddTest("calloc", test_chunk_alloc, NULL,
+	    "parallel extend Chunked allocation on serial file", PARATESTFILE);
 
 #ifdef H5_HAVE_FILTER_DEFLATE
     AddTest("cmpdsetr", compress_readAll, NULL,
diff --git a/testpar/testphdf5.h b/testpar/testphdf5.h
index 14e315e..e085b62 100644
--- a/testpar/testphdf5.h
+++ b/testpar/testphdf5.h
@@ -187,6 +187,7 @@ void dataset_readInd(void);
 void dataset_readAll(void);
 void extend_readInd(void);
 void extend_readAll(void);
+void test_chunk_alloc(void);
 void compact_dataset(void);
 void null_dataset(void);
 void big_dataset(void);