hdf5/test/tgenprop.c

1626 lines
56 KiB
C
Raw Normal View History

/****************************************************************************
* NCSA HDF *
* Software Development Group *
* National Center for Supercomputing Applications *
* University of Illinois at Urbana-Champaign *
* 605 E. Springfield, Champaign IL 61820 *
* *
* For conditions of distribution and use, see the accompanying *
* hdf/COPYING file. *
* *
****************************************************************************/
/* $Id$ */
/***********************************************************
*
* Test program: tgenprop
*
* Test the Generic Property functionality
*
*************************************************************/
/* Define this macro to indicate that the testing APIs should be available */
#define H5P_TESTING
#include "testhdf5.h"
#include "hdf5.h"
#include "H5Dprivate.h" /* For Dataset creation property list names */
#include "H5Pprivate.h" /* For H5P testing functions */
#define FILENAME "tgenprop.h5"
/* Property definitions */
#define CLASS1_NAME "Class 1"
#define CLASS1_PATH "root/Class 1"
#define CLASS2_NAME "Class 2"
#define CLASS2_PATH "root/Class 1/Class 2"
/* Property definitions */
#define PROP1_NAME "Property 1"
int prop1_def=10; /* Property 1 default value */
#define PROP1_SIZE sizeof(prop1_def)
#define PROP1_DEF_VALUE (&prop1_def)
#define PROP2_NAME "Property 2"
float prop2_def=(float)3.14; /* Property 2 default value */
#define PROP2_SIZE sizeof(prop2_def)
#define PROP2_DEF_VALUE (&prop2_def)
#define PROP3_NAME "Property 3"
char prop3_def[10]="Ten chars"; /* Property 3 default value */
#define PROP3_SIZE sizeof(prop3_def)
#define PROP3_DEF_VALUE (&prop3_def)
#define PROP4_NAME "Property 4"
double prop4_def=1.41; /* Property 4 default value */
#define PROP4_SIZE sizeof(prop4_def)
#define PROP4_DEF_VALUE (&prop4_def)
/****************************************************************
**
** test_genprop_basic_class(): Test basic generic property list code.
** Tests creating new generic classes.
**
****************************************************************/
static void
test_genprop_basic_class(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t cid2; /* Generic Property class ID */
hid_t cid3; /* Generic Property class ID */
char *name; /* Name of class */
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Class Creation Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Check class name */
name = H5Pget_class_name(cid1);
CHECK_PTR(name, "H5Pget_class_name");
if(HDstrcmp(name,CLASS1_NAME)!=0) {
num_errs++;
printf("Class names don't match!, name=%s, CLASS1_NAME=%s\n",name,CLASS1_NAME);
} /* end if */
free(name);
/* Check class parent */
cid2 = H5Pget_class_parent(cid1);
CHECK_I(cid2, "H5Pget_class_parent");
/* Verify class parent correct */
ret = H5Pequal(cid2,H5P_NO_CLASS);
VERIFY(ret, 1, "H5Pequal");
/* Make certain false postives aren't being returned */
ret = H5Pequal(cid2,H5P_FILE_CREATE);
VERIFY(ret, 0, "H5Pequal");
/* Close parent class */
ret = H5Pclose_class(cid2);
CHECK_I(ret, "H5Pclose_class");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
/* Create another new generic class, derived from file creation class */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_FILE_CREATE,CLASS2_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Check class name */
name = H5Pget_class_name(cid1);
CHECK_PTR(name, "H5Pget_class_name");
if(HDstrcmp(name,CLASS2_NAME)!=0) {
num_errs++;
printf("Class names don't match!, name=%s, CLASS2_NAME=%s\n",name,CLASS2_NAME);
} /* end if */
free(name);
/* Check class parent */
cid2 = H5Pget_class_parent(cid1);
CHECK_I(cid2, "H5Pget_class_parent");
/* Verify class parent correct */
ret = H5Pequal(cid2,H5P_FILE_CREATE);
VERIFY(ret, 1, "H5Pequal");
/* Check class parent's parent */
cid3 = H5Pget_class_parent(cid2);
CHECK_I(cid3, "H5Pget_class_parent");
/* Verify class parent's parent correct */
ret = H5Pequal(cid3,H5P_NO_CLASS);
VERIFY(ret, 1, "H5Pequal");
/* Close parent class's parent */
ret = H5Pclose_class(cid3);
CHECK_I(ret, "H5Pclose_class");
/* Close parent class */
ret = H5Pclose_class(cid2);
CHECK_I(ret, "H5Pclose_class");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_basic_class() */
/****************************************************************
**
** test_genprop_basic_class_prop(): Test basic generic property list code.
** Tests adding properties to generic classes.
**
****************************************************************/
static void
test_genprop_basic_class_prop(void)
{
hid_t cid1; /* Generic Property class ID */
size_t size; /* Size of property */
size_t nprops; /* Number of properties in class */
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Class Properties Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 0, "H5Pget_nprops");
/* Check the existance of the first property (should fail) */
ret = H5Pexist(cid1,PROP1_NAME);
VERIFY(ret, 0, "H5Pexist");
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Try to insert the first property again (should fail) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
VERIFY(ret, FAIL, "H5Pregister");
/* Check the existance of the first property */
ret = H5Pexist(cid1,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check the size of the first property */
ret = H5Pget_size(cid1,PROP1_NAME,&size);
CHECK_I(ret, "H5Pget_size");
VERIFY(size, PROP1_SIZE, "H5Pget_size");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 1, "H5Pget_nprops");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Try to insert the second property again (should fail) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
VERIFY(ret, FAIL, "H5Pregister");
/* Check the existance of the second property */
ret = H5Pexist(cid1,PROP2_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check the size of the second property */
ret = H5Pget_size(cid1,PROP2_NAME,&size);
CHECK_I(ret, "H5Pget_size");
VERIFY(size, PROP2_SIZE, "H5Pget_size");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 2, "H5Pget_nprops");
/* Insert third property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP3_NAME,PROP3_SIZE,PROP3_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Check the existance of the third property */
ret = H5Pexist(cid1,PROP3_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check the size of the third property */
ret = H5Pget_size(cid1,PROP3_NAME,&size);
CHECK_I(ret, "H5Pget_size");
VERIFY(size, PROP3_SIZE, "H5Pget_size");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 3, "H5Pget_nprops");
/* Unregister first property */
ret = H5Punregister(cid1,PROP1_NAME);
CHECK_I(ret, "H5Punregister");
/* Try to check the size of the first property (should fail) */
ret = H5Pget_size(cid1,PROP1_NAME,&size);
VERIFY(ret, FAIL, "H5Pget_size");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 2, "H5Pget_nprops");
/* Unregister second property */
ret = H5Punregister(cid1,PROP2_NAME);
CHECK_I(ret, "H5Punregister");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 1, "H5Pget_nprops");
/* Unregister third property */
ret = H5Punregister(cid1,PROP3_NAME);
CHECK_I(ret, "H5Punregister");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 0, "H5Pget_nprops");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_basic_class_prop() */
/****************************************************************
**
** test_genprop_iter1(): Property iterator for test_genprop_class_iter
**
****************************************************************/
static int
test_genprop_iter1(hid_t id, const char *name, void *iter_data)
{
struct { /* Struct for iterations */
int iter_count;
const char **names;
} *iter_struct=iter_data;
/* Shut compiler up */
id=id;
return(HDstrcmp(name,iter_struct->names[iter_struct->iter_count++]));
}
/****************************************************************
**
** test_genprop_class_iter(): Test basic generic property list code.
** Tests iterating over properties in a generic class.
**
****************************************************************/
static void
test_genprop_class_iter(void)
{
hid_t cid1; /* Generic Property class ID */
size_t nprops; /* Number of properties in class */
int idx; /* Index to start iteration at */
struct { /* Struct for iterations */
int iter_count;
const char **names;
} iter_struct;
const char *pnames[4]={ /* Names of properties for iterator */
PROP1_NAME,
PROP2_NAME,
PROP3_NAME,
PROP4_NAME};
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Class Property Iteration Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert third property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP3_NAME,PROP3_SIZE,PROP3_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert third property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP4_NAME,PROP4_SIZE,PROP4_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 4, "H5Pget_nprops");
/* Iterate over all properties in class */
iter_struct.iter_count=0;
iter_struct.names=pnames;
ret = H5Piterate(cid1,NULL,test_genprop_iter1,&iter_struct);
VERIFY(ret, 0, "H5Piterate");
/* Iterate over last three properties in class */
idx=iter_struct.iter_count=1;
ret = H5Piterate(cid1,&idx,test_genprop_iter1,&iter_struct);
VERIFY(ret, 0, "H5Piterate");
VERIFY(idx, (int)nprops, "H5Piterate");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_class_iter() */
/****************************************************************
**
** test_genprop_cls_cb1(): Property List callback for test_genprop_class_callback
**
****************************************************************/
static herr_t
test_genprop_cls_cb1(hid_t list_id, void *create_data)
{
struct { /* Struct for iterations */
int count;
hid_t id;
} *count_struct=create_data;
count_struct->count++;
count_struct->id=list_id;
return(SUCCEED);
}
static herr_t
test_genprop_cls_cb2(hid_t new_list_id, hid_t UNUSED old_list_id, void *create_data)
{
struct { /* Struct for iterations */
int count;
hid_t id;
} *count_struct=create_data;
count_struct->count++;
count_struct->id=new_list_id;
return(SUCCEED);
}
/****************************************************************
**
** test_genprop_class_callback(): Test basic generic property list code.
** Tests callbacks for property lists in a generic class.
**
****************************************************************/
static void
test_genprop_class_callback(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t lid1; /* Generic Property list ID */
hid_t lid2; /* Generic Property list ID */
size_t nprops; /* Number of properties in class */
struct { /* Struct for callbacks */
int count;
hid_t id;
} crt_cb_struct, cls_cb_struct;
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Class Callback Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,test_genprop_cls_cb1,&crt_cb_struct,NULL,NULL,test_genprop_cls_cb1,&cls_cb_struct);
CHECK_I(cid1, "H5Pcreate_class");
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert third property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP3_NAME,PROP3_SIZE,PROP3_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert fourth property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP4_NAME,PROP4_SIZE,PROP4_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 4, "H5Pget_nprops");
/* Initialize class callback structs */
crt_cb_struct.count=0;
crt_cb_struct.id=(-1);
cls_cb_struct.count=0;
cls_cb_struct.id=(-1);
/* Create a property list from the class */
lid1 = H5Pcreate(cid1);
CHECK_I(lid1, "H5Pcreate");
/* Verify that the creation callback occurred */
VERIFY(crt_cb_struct.count, 1, "H5Pcreate");
VERIFY(crt_cb_struct.id, lid1, "H5Pcreate");
/* Check the number of properties in list */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 4, "H5Pget_nprops");
/* Create another property list from the class */
lid2 = H5Pcreate(cid1);
CHECK_I(lid2, "H5Pcreate");
/* Verify that the creation callback occurred */
VERIFY(crt_cb_struct.count, 2, "H5Pcreate");
VERIFY(crt_cb_struct.id, lid2, "H5Pcreate");
/* Check the number of properties in list */
ret = H5Pget_nprops(lid2,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 4, "H5Pget_nprops");
/* Close first list */
ret = H5Pclose(lid1);
CHECK_I(ret, "H5Pclose");
/* Verify that the close callback occurred */
VERIFY(cls_cb_struct.count, 1, "H5Pclose");
VERIFY(cls_cb_struct.id, lid1, "H5Pclose");
/* Close second list */
ret = H5Pclose(lid2);
CHECK_I(ret, "H5Pclose");
/* Verify that the close callback occurred */
VERIFY(cls_cb_struct.count, 2, "H5Pclose");
VERIFY(cls_cb_struct.id, lid2, "H5Pclose");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_class_callback() */
/****************************************************************
**
** test_genprop_basic_list(): Test basic generic property list code.
** Tests creating new generic property lists.
**
****************************************************************/
static void
test_genprop_basic_list(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t cid2; /* Generic Property class ID */
hid_t lid1; /* Generic Property list ID */
size_t nprops; /* Number of properties */
size_t size; /* Size of property */
int prop1_value; /* Value for property #1 */
float prop2_value; /* Value for property #2 */
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Creation Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Add several properties (w/default values) */
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 2, "H5Pget_nprops");
/* Create a property list from the class */
lid1 = H5Pcreate(cid1);
CHECK_I(lid1, "H5Pcreate");
/* Get the list's class */
cid2 = H5Pget_class(lid1);
CHECK_I(cid2, "H5Pget_class");
/* Check that the list's class is correct */
ret = H5Pequal(cid1,cid2);
VERIFY(ret, 1, "H5Pequal");
/* Check correct "is a" class/list relationship */
ret = H5Pisa_class(lid1,cid1);
VERIFY(ret, 1, "H5Pisa_class");
/* Check "is a" class/list relationship another way */
ret = H5Pisa_class(lid1,cid2);
VERIFY(ret, 1, "H5Pisa_class");
/* Close class */
ret = H5Pclose_class(cid2);
CHECK_I(ret, "H5Pclose_class");
/* Check the number of properties in list */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 2, "H5Pget_nprops");
/* Check existence of properties */
ret = H5Pexist(lid1,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
ret = H5Pexist(lid1,PROP2_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check the sizes of the properties */
ret = H5Pget_size(lid1,PROP1_NAME,&size);
CHECK_I(ret, "H5Pget_size");
VERIFY(size, PROP1_SIZE, "H5Pget_size");
ret = H5Pget_size(lid1,PROP2_NAME,&size);
CHECK_I(ret, "H5Pget_size");
VERIFY(size, PROP2_SIZE, "H5Pget_size");
/* Check values of properties (set with default values) */
ret = H5Pget(lid1,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
ret = H5Pget(lid1,PROP2_NAME,&prop2_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop2_value, *PROP2_DEF_VALUE, "H5Pget");
/* Close list */
ret = H5Pclose(lid1);
CHECK_I(ret, "H5Pclose");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_basic_list() */
/****************************************************************
**
** test_genprop_basic_list_prop(): Test basic generic property list code.
** Tests creating new generic property lists and adding and
** removing properties from them.
**
****************************************************************/
static void
test_genprop_basic_list_prop(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t lid1; /* Generic Property list ID */
size_t nprops; /* Number of properties */
int prop1_value; /* Value for property #1 */
float prop2_value; /* Value for property #2 */
char prop3_value[10];/* Property #3 value */
double prop4_value; /* Property #4 value */
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Property Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Add several properties (several w/default values) */
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Create a property list from the class */
lid1 = H5Pcreate(cid1);
CHECK_I(lid1, "H5Pcreate");
/* Check the number of properties in list */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 2, "H5Pget_nprops");
/* Add temporary properties */
/* Insert first temporary property into class (with no callbacks) */
ret = H5Pinsert(lid1,PROP3_NAME,PROP3_SIZE,PROP3_DEF_VALUE,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pinsert");
/* Insert second temporary property into class (with no callbacks) */
ret = H5Pinsert(lid1,PROP4_NAME,PROP4_SIZE,PROP4_DEF_VALUE,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pinsert");
/* Check the number of properties in list */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 4, "H5Pget_nprops");
/* Check existence of all properties */
ret = H5Pexist(lid1,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
ret = H5Pexist(lid1,PROP2_NAME);
VERIFY(ret, 1, "H5Pexist");
ret = H5Pexist(lid1,PROP3_NAME);
VERIFY(ret, 1, "H5Pexist");
ret = H5Pexist(lid1,PROP4_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check values of permanent properties (set with default values) */
ret = H5Pget(lid1,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
ret = H5Pget(lid1,PROP2_NAME,&prop2_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop2_value, *PROP2_DEF_VALUE, "H5Pget");
/* Check values of temporary properties (set with regular values) */
ret = H5Pget(lid1,PROP3_NAME,&prop3_value);
CHECK_I(ret, "H5Pget");
if(memcmp(&prop3_value,PROP3_DEF_VALUE,PROP3_SIZE)!=0) {
num_errs++;
printf("Property #3 doesn't match!, line=%d\n",__LINE__);
} /* end if */
ret = H5Pget(lid1,PROP4_NAME,&prop4_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop4_value, *PROP4_DEF_VALUE, "H5Pget");
/* Delete permanent property */
ret = H5Premove(lid1,PROP2_NAME);
CHECK_I(ret, "H5Premove");
/* Check number of properties */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 3, "H5Pget_nprops");
/* Delete temporary property */
ret = H5Premove(lid1,PROP3_NAME);
CHECK_I(ret, "H5Premove");
/* Check number of properties */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 2, "H5Pget_nprops");
/* Check existence of remaining properties */
ret = H5Pexist(lid1,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
ret = H5Pexist(lid1,PROP4_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check values of permanent properties (set with default values) */
ret = H5Pget(lid1,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
/* Check values of temporary properties (set with regular values) */
ret = H5Pget(lid1,PROP4_NAME,&prop4_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop4_value, *PROP4_DEF_VALUE, "H5Pget");
/* Close list */
ret = H5Pclose(lid1);
CHECK_I(ret, "H5Pclose");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_basic_list_prop() */
/****************************************************************
**
** test_genprop_iter2(): Property iterator for test_genprop_list_iter
**
****************************************************************/
static int
test_genprop_iter2(hid_t id, const char *name, void *iter_data)
{
struct { /* Struct for iterations */
int iter_count;
const char **names;
} *iter_struct=iter_data;
/* Shut compiler up */
id=id;
return(HDstrcmp(name,iter_struct->names[iter_struct->iter_count++]));
}
/****************************************************************
**
** test_genprop_list_iter(): Test basic generic property list code.
** Tests iterating over generic property list properties.
**
****************************************************************/
static void
test_genprop_list_iter(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t lid1; /* Generic Property list ID */
size_t nprops; /* Number of properties */
int idx; /* Index to start iteration at */
struct { /* Struct for iterations */
int iter_count;
const char **names;
} iter_struct;
const char *pnames[4]={ /* Names of properties for iterator */
PROP3_NAME,
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
PROP4_NAME,
PROP1_NAME,
PROP2_NAME
};
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Generic Property List Iteration Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Add several properties (several w/default values) */
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Create a property list from the class */
lid1 = H5Pcreate(cid1);
CHECK_I(lid1, "H5Pcreate");
/* Check the number of properties in list */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 2, "H5Pget_nprops");
/* Add temporary properties */
/* Insert first temporary property into class (with no callbacks) */
ret = H5Pinsert(lid1,PROP3_NAME,PROP3_SIZE,PROP3_DEF_VALUE,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pinsert");
/* Insert second temporary property into class (with no callbacks) */
ret = H5Pinsert(lid1,PROP4_NAME,PROP4_SIZE,PROP4_DEF_VALUE,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pinsert");
/* Check the number of properties in list */
ret = H5Pget_nprops(lid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 4, "H5Pget_nprops");
/* Iterate over all properties in list */
iter_struct.iter_count=0;
iter_struct.names=pnames;
ret = H5Piterate(lid1,NULL,test_genprop_iter2,&iter_struct);
VERIFY(ret, 0, "H5Piterate");
/* Iterate over last three properties in list */
idx=iter_struct.iter_count=1;
ret = H5Piterate(lid1,&idx,test_genprop_iter2,&iter_struct);
VERIFY(ret, 0, "H5Piterate");
VERIFY(idx, (int)nprops, "H5Piterate");
/* Close list */
ret = H5Pclose(lid1);
CHECK_I(ret, "H5Pclose");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_list_iter() */
typedef struct {
/* Creation information */
int crt_count;
char *crt_name;
void *crt_value;
/* Set information */
int set_count;
hid_t set_plist_id;
char *set_name;
void *set_value;
/* Get information */
int get_count;
hid_t get_plist_id;
char *get_name;
void *get_value;
/* Delete information */
int del_count;
hid_t del_plist_id;
char *del_name;
void *del_value;
/* Copy information */
int cop_count;
char *cop_name;
void *cop_value;
/* Close information */
int cls_count;
char *cls_name;
void *cls_value;
} prop_cb_info;
/* Global variables for Callback information */
prop_cb_info prop1_cb_info; /* Callback statistics for property #1 */
prop_cb_info prop2_cb_info; /* Callback statistics for property #2 */
/****************************************************************
**
** test_genprop_prop_crt_cb1(): Property creation callback for test_genprop_list_callback
**
****************************************************************/
static herr_t
test_genprop_prop_crt_cb1(const char *name, size_t size, void *def_value)
{
/* Set the information from the creation call */
prop1_cb_info.crt_count++;
prop1_cb_info.crt_name=HDstrdup(name);
prop1_cb_info.crt_value=HDmalloc(size);
HDmemcpy(prop1_cb_info.crt_value,def_value,size);
return(SUCCEED);
}
/****************************************************************
**
** test_genprop_prop_set_cb1(): Property set callback for test_genprop_list_callback
**
****************************************************************/
static herr_t
test_genprop_prop_set_cb1(hid_t plist_id, const char *name, size_t size, void *value)
{
/* Set the information from the set call */
prop1_cb_info.set_count++;
prop1_cb_info.set_plist_id=plist_id;
if(prop1_cb_info.set_name==NULL)
prop1_cb_info.set_name=HDstrdup(name);
if(prop1_cb_info.set_value==NULL)
prop1_cb_info.set_value=HDmalloc(size);
HDmemcpy(prop1_cb_info.set_value,value,size);
return(SUCCEED);
}
/****************************************************************
**
** test_genprop_prop_get_cb1(): Property get callback for test_genprop_list_callback
**
****************************************************************/
static herr_t
test_genprop_prop_get_cb1(hid_t plist_id, const char *name, size_t size, void *value)
{
/* Set the information from the get call */
prop1_cb_info.get_count++;
prop1_cb_info.get_plist_id=plist_id;
if(prop1_cb_info.get_name==NULL)
prop1_cb_info.get_name=HDstrdup(name);
if(prop1_cb_info.get_value==NULL)
prop1_cb_info.get_value=HDmalloc(size);
HDmemcpy(prop1_cb_info.get_value,value,size);
return(SUCCEED);
}
/****************************************************************
**
** test_genprop_prop_cop_cb1(): Property copy callback for test_genprop_list_callback
**
****************************************************************/
static herr_t
test_genprop_prop_cop_cb1(const char *name, size_t size, void *value)
{
/* Set the information from the get call */
prop1_cb_info.cop_count++;
if(prop1_cb_info.cop_name==NULL)
prop1_cb_info.cop_name=HDstrdup(name);
if(prop1_cb_info.cop_value==NULL)
prop1_cb_info.cop_value=HDmalloc(size);
HDmemcpy(prop1_cb_info.cop_value,value,size);
return(SUCCEED);
}
/****************************************************************
**
** test_genprop_prop_cls_cb1(): Property close callback for test_genprop_list_callback
**
****************************************************************/
static herr_t
test_genprop_prop_cls_cb1(const char *name, size_t size, void *value)
{
/* Set the information from the close call */
prop1_cb_info.cls_count++;
if(prop1_cb_info.cls_name==NULL)
prop1_cb_info.cls_name=HDstrdup(name);
if(prop1_cb_info.cls_value==NULL)
prop1_cb_info.cls_value=HDmalloc(size);
HDmemcpy(prop1_cb_info.cls_value,value,size);
return(SUCCEED);
}
/****************************************************************
**
** test_genprop_prop_del_cb2(): Property delete callback for test_genprop_list_callback
**
****************************************************************/
static herr_t
test_genprop_prop_del_cb2(hid_t plist_id, const char *name, size_t size, void *value)
{
/* Set the information from the delete call */
prop2_cb_info.del_count++;
prop2_cb_info.del_plist_id=plist_id;
prop2_cb_info.del_name=HDstrdup(name);
prop2_cb_info.del_value=HDmalloc(size);
HDmemcpy(prop2_cb_info.del_value,value,size);
return(SUCCEED);
}
/****************************************************************
**
** test_genprop_list_callback(): Test basic generic property list code.
** Tests callbacks for properties in a generic property list.
**
****************************************************************/
static void
test_genprop_list_callback(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t lid1; /* Generic Property list ID */
hid_t lid2; /* 2nd Generic Property list ID */
size_t nprops; /* Number of properties in class */
int prop1_value; /* Value for property #1 */
int prop1_new_value=20; /* Property #1 new value */
float prop2_value; /* Value for property #2 */
char prop3_value[10];/* Property #3 value */
double prop4_value; /* Property #4 value */
struct { /* Struct for callbacks */
int count;
hid_t id;
} cop_cb_struct;
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Property Callback Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,test_genprop_cls_cb2,&cop_cb_struct,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Insert first property into class (with callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,test_genprop_prop_crt_cb1,test_genprop_prop_set_cb1,test_genprop_prop_get_cb1,NULL,test_genprop_prop_cop_cb1,test_genprop_prop_cls_cb1);
CHECK_I(ret, "H5Pregister");
/* Insert second property into class (with only delete callback) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,test_genprop_prop_del_cb2,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert third property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP3_NAME,PROP3_SIZE,PROP3_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert fourth property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP4_NAME,PROP4_SIZE,PROP4_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Check the number of properties in class */
ret = H5Pget_nprops(cid1,&nprops);
CHECK_I(ret, "H5Pget_nprops");
VERIFY(nprops, 4, "H5Pget_nprops");
/* Initialize class callback structs */
cop_cb_struct.count=0;
cop_cb_struct.id=(-1);
/* Initialize callback information for properties tracked */
HDmemset(&prop1_cb_info,0,sizeof(prop_cb_info));
HDmemset(&prop2_cb_info,0,sizeof(prop_cb_info));
/* Create a property list from the class */
lid1 = H5Pcreate(cid1);
CHECK_I(lid1, "H5Pcreate");
/* Verify creation callback information for properties tracked */
VERIFY(prop1_cb_info.crt_count, 1, "H5Pcreate");
if(HDstrcmp(prop1_cb_info.crt_name,PROP1_NAME)!=0) {
num_errs++;
printf("Property #1 name doesn't match!, line=%d\n",__LINE__);
} /* end if */
if(HDmemcmp(prop1_cb_info.crt_value,PROP1_DEF_VALUE,PROP1_SIZE)!=0) {
num_errs++;
printf("Property #1 value doesn't match!, line=%d\n",__LINE__);
} /* end if */
/* Check values of permanent properties (set with default values) */
ret = H5Pget(lid1,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
ret = H5Pget(lid1,PROP2_NAME,&prop2_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop2_value, *PROP2_DEF_VALUE, "H5Pget");
/* Check values of temporary properties (set with regular values) */
ret = H5Pget(lid1,PROP3_NAME,&prop3_value);
CHECK_I(ret, "H5Pget");
if(memcmp(&prop3_value,PROP3_DEF_VALUE,PROP3_SIZE)!=0) {
num_errs++;
printf("Property #3 doesn't match!, line=%d\n",__LINE__);
} /* end if */
ret = H5Pget(lid1,PROP4_NAME,&prop4_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop4_value, *PROP4_DEF_VALUE, "H5Pget");
/* Verify get callback information for properties tracked */
VERIFY(prop1_cb_info.get_count, 1, "H5Pget");
VERIFY(prop1_cb_info.get_plist_id, lid1, "H5Pget");
if(HDstrcmp(prop1_cb_info.get_name,PROP1_NAME)!=0) {
num_errs++;
printf("Property #1 name doesn't match!, line=%d\n",__LINE__);
} /* end if */
if(HDmemcmp(prop1_cb_info.get_value,PROP1_DEF_VALUE,PROP1_SIZE)!=0) {
num_errs++;
printf("Property #1 value doesn't match!, line=%d\n",__LINE__);
} /* end if */
/* Set value of property #1 to different value */
ret = H5Pset(lid1,PROP1_NAME,&prop1_new_value);
CHECK_I(ret, "H5Pset");
/* Verify set callback information for properties tracked */
VERIFY(prop1_cb_info.set_count, 1, "H5Pset");
VERIFY(prop1_cb_info.set_plist_id, lid1, "H5Pset");
if(HDstrcmp(prop1_cb_info.set_name,PROP1_NAME)!=0) {
num_errs++;
printf("Property #1 name doesn't match!, line=%d\n",__LINE__);
} /* end if */
if(HDmemcmp(prop1_cb_info.set_value,&prop1_new_value,PROP1_SIZE)!=0) {
num_errs++;
printf("Property #1 value doesn't match!, line=%d\n",__LINE__);
} /* end if */
/* Check new value of tracked properties */
ret = H5Pget(lid1,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, prop1_new_value, "H5Pget");
/* Verify get callback information again for properties tracked */
VERIFY(prop1_cb_info.get_count, 2, "H5Pget");
VERIFY(prop1_cb_info.get_plist_id, lid1, "H5Pget");
if(HDstrcmp(prop1_cb_info.get_name,PROP1_NAME)!=0) {
num_errs++;
printf("Property #1 name doesn't match!, line=%d\n",__LINE__);
} /* end if */
if(HDmemcmp(prop1_cb_info.get_value,&prop1_new_value,PROP1_SIZE)!=0) {
num_errs++;
printf("Property #1 value doesn't match!, line=%d\n",__LINE__);
} /* end if */
/* Delete property #2 */
ret = H5Premove(lid1,PROP2_NAME);
CHECK_I(ret, "H5Premove");
/* Verify delete callback information for properties tracked */
VERIFY(prop2_cb_info.del_count, 1, "H5Premove");
VERIFY(prop2_cb_info.del_plist_id, lid1, "H5Premove");
if(HDstrcmp(prop2_cb_info.del_name,PROP2_NAME)!=0) {
num_errs++;
printf("Property #2 name doesn't match!, line=%d\n",__LINE__);
} /* end if */
if(HDmemcmp(prop2_cb_info.del_value,PROP2_DEF_VALUE,PROP2_SIZE)!=0) {
num_errs++;
printf("Property #2 value doesn't match!, line=%d\n",__LINE__);
} /* end if */
/* Copy first list */
lid2 = H5Pcopy(lid1);
CHECK_I(lid2, "H5Pcopy");
/* Verify copy callback information for properties tracked */
VERIFY(prop1_cb_info.cop_count, 1, "H5Pcopy");
if(HDstrcmp(prop1_cb_info.cop_name,PROP1_NAME)!=0) {
num_errs++;
printf("Property #1 name doesn't match!, line=%d\n",__LINE__);
} /* end if */
if(HDmemcmp(prop1_cb_info.cop_value,&prop1_new_value,PROP1_SIZE)!=0) {
num_errs++;
printf("Property #1 value doesn't match!, line=%d\n",__LINE__);
} /* end if */
/* Verify that the class creation callback occurred */
VERIFY(cop_cb_struct.count, 1, "H5Pcopy");
VERIFY(cop_cb_struct.id, lid2, "H5Pcopy");
/* Close first list */
ret = H5Pclose(lid1);
CHECK_I(ret, "H5Pclose");
/* Verify close callback information for properties tracked */
VERIFY(prop1_cb_info.cls_count, 1, "H5Pclose");
if(HDstrcmp(prop1_cb_info.cls_name,PROP1_NAME)!=0) {
num_errs++;
printf("Property #1 name doesn't match!, line=%d\n",__LINE__);
} /* end if */
if(HDmemcmp(prop1_cb_info.cls_value,&prop1_new_value,PROP1_SIZE)!=0) {
num_errs++;
printf("Property #1 value doesn't match!, line=%d\n",__LINE__);
} /* end if */
/* Close second list */
ret = H5Pclose(lid2);
CHECK_I(ret, "H5Pclose");
/* Verify close callback information for properties tracked */
VERIFY(prop1_cb_info.cls_count, 2, "H5Pclose");
/* Free memory allocated for tracking properties */
HDfree(prop1_cb_info.crt_name);
HDfree(prop1_cb_info.crt_value);
HDfree(prop1_cb_info.get_name);
HDfree(prop1_cb_info.get_value);
HDfree(prop1_cb_info.set_name);
HDfree(prop1_cb_info.set_value);
HDfree(prop1_cb_info.cop_name);
HDfree(prop1_cb_info.cop_value);
HDfree(prop1_cb_info.cls_name);
HDfree(prop1_cb_info.cls_value);
HDfree(prop2_cb_info.del_name);
HDfree(prop2_cb_info.del_value);
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* end test_genprop_list_callback() */
/****************************************************************
**
** test_genprop_list_addprop(): Test adding properties to a
** standard HDF5 property list and verify that the library
** ignores the extra properties.
**
****************************************************************/
static void
test_genprop_list_addprop(void)
{
hid_t fid; /* File ID */
hid_t did; /* Dataset ID */
hid_t sid; /* Dataspace ID */
hid_t pid; /* Property List ID */
int prop1_value; /* Value for property #1 */
herr_t ret; /* Generic return value */
/* Create file */
fid = H5Fcreate(FILENAME, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);
CHECK(fid, FAIL, "H5Fcreate");
/* Create scalar dataspace for dataset */
sid = H5Screate(H5S_SCALAR);
CHECK(sid, FAIL, "H5Screate");
/* Create a dataset creation property list */
pid = H5Pcreate(H5P_DATASET_CREATE);
CHECK(pid, FAIL, "H5Pcreate");
/* Insert temporary property into class (with no callbacks) */
ret = H5Pinsert(pid,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pinsert");
/* Check existence of added property */
ret = H5Pexist(pid,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check values of property (set with default value) */
ret = H5Pget(pid,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
/* Create a dataset */
did=H5Dcreate(fid,"Dataset1",H5T_NATIVE_INT,sid,pid);
CHECK(did, FAIL, "H5Dcreate");
/* Check existence of added property (after using property list) */
ret = H5Pexist(pid,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check values of property (set with default value) (after using property list) */
ret = H5Pget(pid,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
/* Close property list */
ret = H5Pclose(pid);
CHECK(ret, FAIL, "H5Pclose");
/* Close disk dataspace */
ret = H5Sclose(sid);
CHECK(ret, FAIL, "H5Sclose");
/* Close Dataset */
ret = H5Dclose(did);
CHECK(ret, FAIL, "H5Dclose");
/* Close file */
ret = H5Fclose(fid);
CHECK(ret, FAIL, "H5Fclose");
} /* end test_genprop_list_addprop() */
/****************************************************************
**
** test_genprop_class_addprop(): Test adding properties to a
** standard HDF5 property class and verify that the library
** ignores the extra properties and continues to recognize the
** derived class as a valid version of the derived-from class.
**
****************************************************************/
static void
test_genprop_class_addprop(void)
{
hid_t fid; /* File ID */
hid_t did; /* Dataset ID */
hid_t sid; /* Dataspace ID */
hid_t cid; /* Property Class ID */
hid_t pid; /* Property List ID */
int prop1_value; /* Value for property #1 */
herr_t ret; /* Generic return value */
/* Create file */
fid = H5Fcreate(FILENAME, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);
CHECK(fid, FAIL, "H5Fcreate");
/* Create scalar dataspace for dataset */
sid = H5Screate(H5S_SCALAR);
CHECK(sid, FAIL, "H5Screate");
/* Create a new class, dervied from the dataset creation property list class */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid = H5Pcreate_class(H5P_DATASET_CREATE,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid, "H5Pcreate_class");
/* Check existence of an original property */
ret = H5Pexist(cid,H5D_CRT_DATA_PIPELINE_NAME);
VERIFY(ret, 0, "H5Pexist");
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Create a derived dataset creation property list */
pid = H5Pcreate(cid);
CHECK(pid, FAIL, "H5Pcreate");
/* Check existence of an original property */
ret = H5Pexist(pid,H5D_CRT_DATA_PIPELINE_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check existence of added property */
ret = H5Pexist(pid,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check values of property (set with default value) */
ret = H5Pget(pid,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
/* Create a dataset */
did=H5Dcreate(fid,"Dataset1",H5T_NATIVE_INT,sid,pid);
CHECK(did, FAIL, "H5Dcreate");
/* Check existence of added property (after using property list) */
ret = H5Pexist(pid,PROP1_NAME);
VERIFY(ret, 1, "H5Pexist");
/* Check values of property (set with default value) (after using property list) */
ret = H5Pget(pid,PROP1_NAME,&prop1_value);
CHECK_I(ret, "H5Pget");
VERIFY(prop1_value, *PROP1_DEF_VALUE, "H5Pget");
/* Close property class */
ret = H5Pclose_class(cid);
CHECK(ret, FAIL, "H5Pclose");
/* Close property list */
ret = H5Pclose(pid);
CHECK(ret, FAIL, "H5Pclose");
/* Close disk dataspace */
ret = H5Sclose(sid);
CHECK(ret, FAIL, "H5Sclose");
/* Close Dataset */
ret = H5Dclose(did);
CHECK(ret, FAIL, "H5Dclose");
/* Close file */
ret = H5Fclose(fid);
CHECK(ret, FAIL, "H5Fclose");
} /* end test_genprop_class_addprop() */
/****************************************************************
**
** test_genprop_equal(): Test basic generic property list code.
** More tests for H5Pequal()
**
****************************************************************/
static void
test_genprop_equal(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t lid1; /* Generic Property list ID */
hid_t lid2; /* Generic Property list ID */
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Basic Generic Property List Equal Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Create a property list from the class */
lid1 = H5Pcreate(cid1);
CHECK_I(lid1, "H5Pcreate");
/* Copy the property list */
lid2 = H5Pcopy(lid1);
CHECK_I(lid2, "H5Pcopy");
/* Check that the lists are equal */
ret = H5Pequal(lid1,lid2);
VERIFY(ret, 1, "H5Pequal");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* ent test_genprop_equal() */
/****************************************************************
**
** test_genprop_path(): Test basic generic property list code.
** Tests for class paths
**
****************************************************************/
static void
test_genprop_path(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t cid2; /* Generic Property class ID */
hid_t cid3; /* Generic Property class ID */
char *path; /* Class path */
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Generic Property List Class Path Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Get full path for first class */
path=H5P_get_class_path_test(cid1);
CHECK_PTR(path, "H5P_get_class_path_test");
if(HDstrcmp(path,CLASS1_PATH)!=0) {
num_errs++;
printf("Class names don't match!, path=%s, CLASS1_PATH=%s\n",path,CLASS1_PATH);
} /* end if */
free(path);
/* Create another new generic class, derived from first class */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid2 = H5Pcreate_class(cid1,CLASS2_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid2, "H5Pcreate_class");
/* Insert second property into class (with no callbacks) */
ret = H5Pregister(cid2,PROP2_NAME,PROP2_SIZE,PROP2_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Get full path for second class */
path=H5P_get_class_path_test(cid2);
CHECK_PTR(path, "H5P_get_class_path_test");
if(HDstrcmp(path,CLASS2_PATH)!=0) {
num_errs++;
printf("Class names don't match!, path=%s, CLASS2_PATH=%s\n",path,CLASS2_PATH);
} /* end if */
/* Open a copy of the class with the path name */
cid3 = H5P_open_class_path_test(path);
CHECK_I(cid3, "H5Popen_class_path");
/* Check that the classes are equal */
ret = H5Pequal(cid2,cid3);
VERIFY(ret, 1, "H5Pequal");
/* Release the path string */
free(path);
/* Close class */
ret = H5Pclose_class(cid3);
CHECK_I(ret, "H5Pclose_class");
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
/* Close class */
ret = H5Pclose_class(cid2);
CHECK_I(ret, "H5Pclose_class");
} /* ent test_genprop_path() */
/****************************************************************
**
** test_genprop_refcount(): Test basic generic property list code.
** Tests for correct reference counting
**
****************************************************************/
static void
test_genprop_refcount(void)
{
hid_t cid1; /* Generic Property class ID */
hid_t lid1; /* Generic Property class ID */
char *name; /* Name of class */
herr_t ret; /* Generic return value */
/* Output message about test being performed */
MESSAGE(5, ("Testing Generic Property List Reference Count Functionality\n"));
/* Create a new generic class, derived from the root of the class hierarchy */
[svn-r6252] Purpose: Lots of performance improvements & a couple new internal API interfaces. Description: Performance Improvements: - Cached file offset & length sizes in shared file struct, to avoid constantly looking them up in the FCPL. - Generic property improvements: - Added "revision" number to generic property classes to speed up comparisons. - Changed method of storing properties from using a hash-table to the TBBT routines in the library. - Share the propery names between classes and the lists derived from them. - Removed redundant 'def_value' buffer from each property. - Switching code to use a "copy on write" strategy for properties in each list, where the properties in each list are shared with the properties in the class, until a property's value is changed in a list. - Fixed error in layout code which was allocating too many buffers. - Redefined public macros of the form (H5open()/H5check, <variable>) internally to only be (<variable>), avoiding innumerable useless calls to H5open() and H5check_version(). - Reuse already zeroed buffers in H5F_contig_fill instead of constantly re-zeroing them. - Don't write fill values if writing entire dataset. - Use gettimeofday() system call instead of time() system when checking the modification time of a dataset. - Added reference counted string API and use it for tracking the names of objects opening in a file (for the ID->name code). - Removed redundant H5P_get() calls in B-tree routines. - Redefine H5T datatype macros internally to the library, to avoid calling H5check redundantly. - Keep dataspace information for dataset locally instead of reading from disk each time. Added new module to track open objects in a file, to allow this (which will be useful eventually for some FPH5 metadata caching issues). - Remove H5AC_find macro which was inlining metadata cache lookups, and call function instead. - Remove redundant memset() calls from H5G_namei() routine. - Remove redundant checking of object type when locating objects in metadata cache and rely on the address only. - Create default dataset object to use when default dataset creation property list is used to create datasets, bypassing querying for all the property list values. - Use default I/O vector size when performing raw data with the default dataset transfer property list, instead of querying for I/O vector size. - Remove H5P_DEFAULT internally to the library, replacing it with more specific default property list based on the type of property list needed. - Remove redundant memset() calls in object header message (H5O*) routines. - Remove redunant memset() calls in data I/O routines. - Split free-list allocation routines into malloc() and calloc()- like routines, instead of one combined routine. - Remove lots of indirection in H5O*() routines. - Simplify metadata cache entry comparison routine (used when flushing entire cache out). - Only enable metadata cache statistics when H5AC_DEBUG is turned on, instead of always tracking them. - Simplify address comparison macro (H5F_addr_eq). - Remove redundant metadata cache entry protections during dataset creation by protecting the object header once and making all the modifications necessary for the dataset creation before unprotecting it. - Reduce # of "number of element in extent" computations performed by computing and storing the value during dataspace creation. - Simplify checking for group location's file information, when file has not been involving in file-mounting operations. - Use binary encoding for modification time, instead of ASCII. - Hoist H5HL_peek calls (to get information in a local heap) out of loops in many group routine. - Use static variable for iterators of selections, instead of dynamically allocation them each time. - Lookup & insert new entries in one step, avoiding traversing group's B-tree twice. - Fixed memory leak in H5Gget_objname_idx() routine (tangential to performance improvements, but fixed along the way). - Use free-list for reference counted strings. - Don't bother copying object names into cached group entries, since they are re-created when an object is opened. The benchmark I used to measure these results created several thousand small (2K) datasets in a file and wrote out the data for them. This is Elena's "regular.c" benchmark. These changes resulted in approximately ~4.3x speedup of the development branch when compared to the previous code in the development branch and ~1.4x speedup compared to the release branch. Additionally, these changes reduce the total memory used (code and data) by the development branch by ~800KB, bringing the development branch back into the same ballpark as the release branch. I'll send out a more detailed description of the benchmark results as a followup note. New internal API routines: Added "reference counted strings" API for tracking strings that get used by multiple owners without duplicating the strings. Added "ternary search tree" API for text->object mappings. Platforms tested: Tested h5committest {arabica (fortran), eirene (fortran, C++) modi4 (parallel, fortran)} Other platforms/configurations tested? FreeBSD 4.7 (sleipnir) serial & parallel Solaris 2.6 (baldric) serial
2003-01-10 01:20:03 +08:00
cid1 = H5Pcreate_class(H5P_NO_CLASS,CLASS1_NAME,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(cid1, "H5Pcreate_class");
/* Insert first property into class (with no callbacks) */
ret = H5Pregister(cid1,PROP1_NAME,PROP1_SIZE,PROP1_DEF_VALUE,NULL,NULL,NULL,NULL,NULL,NULL);
CHECK_I(ret, "H5Pregister");
/* Create a new generic list, derived from the root of the class hierarchy */
lid1 = H5Pcreate(cid1);
CHECK_I(lid1, "H5Pcreate");
/* Check class name */
name = H5Pget_class_name(cid1);
CHECK_PTR(name, "H5Pget_class_name");
if(HDstrcmp(name,CLASS1_NAME)!=0) {
num_errs++;
printf("Class names don't match!, name=%s, CLASS1_NAME=%s\n",name,CLASS1_NAME);
} /* end if */
free(name);
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
/* Get the list's class */
cid1 = H5Pget_class(lid1);
CHECK_I(cid1, "H5Pget_class");
/* Check correct "is a" class/list relationship */
ret = H5Pisa_class(lid1,cid1);
VERIFY(ret, 1, "H5Pisa_class");
/* Check class name */
name = H5Pget_class_name(cid1);
CHECK_PTR(name, "H5Pget_class_name");
if(HDstrcmp(name,CLASS1_NAME)!=0) {
num_errs++;
printf("Class names don't match!, name=%s, CLASS1_NAME=%s\n",name,CLASS1_NAME);
} /* end if */
free(name);
/* Close list */
ret = H5Pclose(lid1);
CHECK_I(ret, "H5Pclose");
/* Check class name */
name = H5Pget_class_name(cid1);
CHECK_PTR(name, "H5Pget_class_name");
if(HDstrcmp(name,CLASS1_NAME)!=0) {
num_errs++;
printf("Class names don't match!, name=%s, CLASS1_NAME=%s\n",name,CLASS1_NAME);
} /* end if */
free(name);
/* Close class */
ret = H5Pclose_class(cid1);
CHECK_I(ret, "H5Pclose_class");
} /* ent test_genprop_refcount() */
/****************************************************************
**
** test_genprop(): Main generic property testing routine.
**
****************************************************************/
void
test_genprop(void)
{
/* Output message about test being performed */
MESSAGE(5, ("Testing Generic Properties\n"));
/* These tests use the same file... */
test_genprop_basic_class(); /* Test basic code for creating a generic class */
test_genprop_basic_class_prop(); /* Test basic code for adding properties to a generic class */
test_genprop_class_iter(); /* Test code for iterating over properties in a generic class */
test_genprop_class_callback(); /* Test code for property class callbacks */
test_genprop_basic_list(); /* Test basic code for creating a generic property list */
test_genprop_basic_list_prop(); /* Test basic code for adding properties to a generic property list */
test_genprop_list_iter(); /* Test basic code for iterating over properties in a generic property list */
test_genprop_list_callback(); /* Test code for property list callbacks */
test_genprop_list_addprop(); /* Test adding properties to HDF5 property list */
test_genprop_class_addprop(); /* Test adding properties to HDF5 property class */
test_genprop_equal(); /* Tests for more H5Pequal verification */
test_genprop_path(); /* Tests for class path verification */
test_genprop_refcount(); /* Tests for class reference counting */
} /* test_genprop() */
/*-------------------------------------------------------------------------
* Function: cleanup_genprop
*
* Purpose: Cleanup temporary test files
*
* Return: none
*
* Programmer: Quincey Koziol
* June 8, 1999
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
void
cleanup_genprop(void)
{
remove(FILENAME);
}