Pr/performance tuning

ompi/mca/fcoll/base/fcoll_base_file_select.c

@@ -262,6 +262,9 @@ int mca_fcoll_base_query_table (struct mca_io_ompio_file_t *file, char *name)
             file->f_cc_size >= file->f_stripe_size) {
             return 1;
         }
+	if ( 2 >= (int)file->f_size ){
+	    return 1;
+	}
     }
     if (!strcmp (name, "dynamic")) {
         if ((int)file->f_cc_size < file->f_bytes_per_agg && 

ompi/mca/fcoll/individual/fcoll_individual_module.c

@@ -64,6 +64,9 @@ mca_fcoll_individual_component_file_query (mca_io_ompio_file_t *fh, int *priorit
         if (*priority < 50) {
             *priority = 50;
         }
+	if ( 2 >= fh->f_size ) {
+	    *priority = 100;
+	}
     }
 
     return &individual;

ompi/mca/io/ompio/io_ompio.c

@@ -102,7 +102,6 @@ static int mca_io_ompio_merge_groups(mca_io_ompio_file_t *fh,
                                      int num_merge_aggrs);
 
 
-
 int ompi_io_ompio_set_file_defaults (mca_io_ompio_file_t *fh)
 {
 
@@ -133,15 +132,16 @@ int ompi_io_ompio_set_file_defaults (mca_io_ompio_file_t *fh)
         fh->f_init_num_aggrs = -1;
         fh->f_init_aggr_list = NULL;
 
-	ompi_datatype_create_contiguous(1048576, 
+	ompi_datatype_create_contiguous(MCA_IO_DEFAULT_FILE_VIEW_SIZE,
 					&ompi_mpi_byte.dt,
 					&default_file_view);
 	ompi_datatype_commit (&default_file_view);
 
 	fh->f_etype = &ompi_mpi_byte.dt;
 	fh->f_filetype =  default_file_view;
-
-
+	ompi_datatype_duplicate ( &ompi_mpi_byte.dt, &fh->f_orig_filetype );
+
+
         /* Default file View */
         fh->f_iov_type = MPI_DATATYPE_NULL;
         fh->f_stripe_size = mca_io_ompio_bytes_per_agg;

ompi/mca/io/ompio/io_ompio.h

@@ -57,7 +57,9 @@ OMPI_DECLSPEC extern int mca_io_ompio_coll_timing_info;
 #define OMPIO_CONTIGUOUS_FVIEW       0x00000010
 #define OMPIO_AGGREGATOR_IS_SET      0x00000020
 #define OMPIO_SHAREDFP_IS_SET        0x00000040
+
 #define QUEUESIZE 2048
+#define MCA_IO_DEFAULT_FILE_VIEW_SIZE 4*1024*1024
 
 #define OMPIO_MIN(a, b) (((a) < (b)) ? (a) : (b))
 #define OMPIO_MAX(a, b) (((a) < (b)) ? (b) : (a))
@@ -320,6 +322,7 @@ struct mca_io_ompio_file_t {
     size_t            f_view_size;
     ompi_datatype_t  *f_etype;
     ompi_datatype_t  *f_filetype;
+    ompi_datatype_t  *f_orig_filetype; /* the fileview passed by the user to us */
     size_t            f_etype_size;
 
     /* contains IO requests that needs to be read/written */

ompi/mca/io/ompio/io_ompio_component.c

@@ -31,7 +31,7 @@
 #include "ompi/mca/io/io.h"
 #include "io_ompio.h"
 
-int mca_io_ompio_cycle_buffer_size = OMPIO_PREALLOC_MAX_BUF_SIZE;
+int mca_io_ompio_cycle_buffer_size = -1;
 int mca_io_ompio_bytes_per_agg = OMPIO_PREALLOC_MAX_BUF_SIZE;
 int mca_io_ompio_num_aggregators = -1;
 int mca_io_ompio_record_offset_info = 0;
@@ -162,10 +162,10 @@ static int register_component(void)
                                            MCA_BASE_VAR_SCOPE_READONLY,
                                            &mca_io_ompio_coll_timing_info);
 
-    mca_io_ompio_cycle_buffer_size = OMPIO_PREALLOC_MAX_BUF_SIZE;
+    mca_io_ompio_cycle_buffer_size = -1;
     (void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
                                            "cycle_buffer_size",
-                                           "Cycle buffer size of individual reads/writes",
+                                           "Data size issued by individual reads/writes per call",
                                            MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
                                            OPAL_INFO_LVL_9,
                                            MCA_BASE_VAR_SCOPE_READONLY,

ompi/mca/io/ompio/io_ompio_file_open.c

@@ -380,6 +380,18 @@ ompio_io_ompio_file_close (mca_io_ompio_file_t *ompio_fh)
         ompi_datatype_destroy (&ompio_fh->f_iov_type);
     }
 
+    if ( MPI_DATATYPE_NULL != ompio_fh->f_etype ) {
+	ompi_datatype_destroy (&ompio_fh->f_etype);
+    }
+    if ( MPI_DATATYPE_NULL != ompio_fh->f_filetype ){
+	ompi_datatype_destroy (&ompio_fh->f_filetype);
+    }
+
+    if ( MPI_DATATYPE_NULL != ompio_fh->f_orig_filetype ){
+	ompi_datatype_destroy (&ompio_fh->f_orig_filetype);
+    }
+
+
     if (MPI_COMM_NULL != ompio_fh->f_comm && (ompio_fh->f_flags & OMPIO_SHAREDFP_IS_SET) )  {
         ompi_comm_free (&ompio_fh->f_comm);
     }

ompi/mca/io/ompio/io_ompio_file_read.c

@@ -106,7 +106,12 @@ int ompio_io_ompio_file_read (mca_io_ompio_file_t *fh,
                                    &decoded_iov, 
                                    &iov_count);
 
-    bytes_per_cycle = mca_io_ompio_cycle_buffer_size;
+    if ( -1 == mca_io_ompio_cycle_buffer_size ) {
+	bytes_per_cycle = max_data;
+    }
+    else {
+	bytes_per_cycle = mca_io_ompio_cycle_buffer_size;
+    }
     cycles = ceil((float)max_data/bytes_per_cycle);
 
 #if 0

ompi/mca/io/ompio/io_ompio_file_set_view.c

@@ -9,7 +9,7 @@
  *                         University of Stuttgart.  All rights reserved.
  * Copyright (c) 2004-2005 The Regents of the University of California.
  *                         All rights reserved.
- * Copyright (c) 2008-2014 University of Houston. All rights reserved.
+ * Copyright (c) 2008-2015 University of Houston. All rights reserved.
  * Copyright (c) 2015      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *  $COPYRIGHT$
@@ -135,10 +135,16 @@ int mca_io_ompio_file_set_view (ompi_file_t *fp,
 {
     mca_io_ompio_data_t *data;
     mca_io_ompio_file_t *fh;
+    size_t ftype_size;
+    OPAL_PTRDIFF_TYPE ftype_extent, lb;
 
     data = (mca_io_ompio_data_t *) fp->f_io_selected_data;
     fh = &data->ompio_fh;
-
+
+    ompi_datatype_destroy (&fh->f_etype);
+    ompi_datatype_destroy (&fh->f_filetype);
+    ompi_datatype_destroy (&fh->f_orig_filetype);
+
     if (NULL != fh->f_decoded_iov) {
         free (fh->f_decoded_iov);
         fh->f_decoded_iov = NULL;
@@ -154,14 +160,35 @@ int mca_io_ompio_file_set_view (ompi_file_t *fp,
 
     fh->f_flags |= OMPIO_FILE_VIEW_IS_SET;
     fh->f_datarep = strdup (datarep);
-
-    mca_io_ompio_set_view_internal (fh,
-				    disp,
-				    etype,
-				    filetype,
-				    datarep,
-				    info);
+    ompi_datatype_duplicate (filetype, &fh->f_orig_filetype );
 
+    opal_datatype_get_extent(&filetype->super, &lb, &ftype_extent);
+    opal_datatype_type_size (&filetype->super, &ftype_size);
+
+    if ( etype == filetype && 
+	 ompi_datatype_is_predefined (filetype ) &&
+	 ftype_extent == (OPAL_PTRDIFF_TYPE)ftype_size ){
+	ompi_datatype_t *newfiletype;
+	ompi_datatype_create_contiguous(MCA_IO_DEFAULT_FILE_VIEW_SIZE,
+					&ompi_mpi_byte.dt,
+					&newfiletype);
+	ompi_datatype_commit (&newfiletype);
+	mca_io_ompio_set_view_internal (fh,
+					disp,
+					etype,
+					newfiletype,
+					datarep,
+					info);
+	ompi_datatype_destroy ( &newfiletype );
+    } 
+    else {
+	mca_io_ompio_set_view_internal (fh,
+					disp,
+					etype,
+					filetype,
+					datarep,
+					info);
+    }
 
     if (OMPI_SUCCESS != mca_fcoll_base_file_select (&data->ompio_fh,
                                                     NULL)) {
@@ -186,7 +213,7 @@ int mca_io_ompio_file_get_view (struct ompi_file_t *fp,
 
     *disp = fh->f_disp;
     ompi_datatype_duplicate (fh->f_etype, etype);
-    ompi_datatype_duplicate (fh->f_filetype, filetype);
+    ompi_datatype_duplicate (fh->f_orig_filetype, filetype);
     strcpy (datarep, fh->f_datarep);
 
     return OMPI_SUCCESS;

ompi/mca/io/ompio/io_ompio_file_write.c

@@ -103,7 +103,12 @@ int ompio_io_ompio_file_write (mca_io_ompio_file_t *fh,
                                    &decoded_iov, 
                                    &iov_count);
 
-    bytes_per_cycle = mca_io_ompio_cycle_buffer_size;
+    if ( -1 == mca_io_ompio_cycle_buffer_size ) {
+	bytes_per_cycle = max_data;
+    }
+    else {
+	bytes_per_cycle = mca_io_ompio_cycle_buffer_size;
+    }
     cycles = ceil((float)max_data/bytes_per_cycle);
 
 #if 0