ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Add few OPAL_LIKELY/OPAL_UNLIKELY to the datatype engine.

Просмотр исходного кода 
This commit was SVN r15302.
Этот коммит содержится в:
George Bosilca 2007-07-07 04:31:06 +00:00
родитель d677db9b5f
Коммит 11ff1b2c20
2 изменённых файлов: 15 добавлений и 14 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

20
ompi/datatype/convertor.c
Просмотреть файл

@ -229,7 +229,7 @@ int32_t ompi_convertor_pack( ompi_convertor_t* pConv,
{ {
OMPI_CONVERTOR_SET_STATUS_BEFORE_PACK_UNPACK( pConv, iov, out_size, max_data ); OMPI_CONVERTOR_SET_STATUS_BEFORE_PACK_UNPACK( pConv, iov, out_size, max_data );
if( pConv->flags & CONVERTOR_NO_OP ) { if( OPAL_LIKELY(pConv->flags & CONVERTOR_NO_OP) ) {
/* We are doing conversion on a contiguous datatype on a homogeneous /* We are doing conversion on a contiguous datatype on a homogeneous
* environment. The convertor contain minimal informations, we only * environment. The convertor contain minimal informations, we only
* use the bConverted to manage the conversion. * use the bConverted to manage the conversion.
@ -245,7 +245,7 @@ int32_t ompi_convertor_pack( ompi_convertor_t* pConv,
if( iov[i].iov_len >= pending_length ) { if( iov[i].iov_len >= pending_length ) {
goto complete_contiguous_data_pack; goto complete_contiguous_data_pack;
} }
if( NULL == iov[i].iov_base ) if( OPAL_LIKELY(NULL == iov[i].iov_base) )
iov[i].iov_base = base_pointer; iov[i].iov_base = base_pointer;
else else
MEMCPY( iov[i].iov_base, base_pointer, iov[i].iov_len ); MEMCPY( iov[i].iov_base, base_pointer, iov[i].iov_len );
@ -257,7 +257,7 @@ int32_t ompi_convertor_pack( ompi_convertor_t* pConv,
return 0; return 0;
complete_contiguous_data_pack: complete_contiguous_data_pack:
iov[i].iov_len = pending_length; iov[i].iov_len = pending_length;
if( NULL == iov[i].iov_base ) if( OPAL_LIKELY(NULL == iov[i].iov_base) )
iov[i].iov_base = base_pointer; iov[i].iov_base = base_pointer;
else else
MEMCPY( iov[i].iov_base, base_pointer, iov[i].iov_len ); MEMCPY( iov[i].iov_base, base_pointer, iov[i].iov_len );
@ -276,7 +276,7 @@ int32_t ompi_convertor_unpack( ompi_convertor_t* pConv,
{ {
OMPI_CONVERTOR_SET_STATUS_BEFORE_PACK_UNPACK( pConv, iov, out_size, max_data ); OMPI_CONVERTOR_SET_STATUS_BEFORE_PACK_UNPACK( pConv, iov, out_size, max_data );
if( pConv->flags & CONVERTOR_NO_OP ) { if( OPAL_LIKELY(pConv->flags & CONVERTOR_NO_OP) ) {
/* We are doing conversion on a contiguous datatype on a homogeneous /* We are doing conversion on a contiguous datatype on a homogeneous
* environment. The convertor contain minimal informations, we only * environment. The convertor contain minimal informations, we only
* use the bConverted to manage the conversion. * use the bConverted to manage the conversion.
@ -340,7 +340,7 @@ int ompi_convertor_create_stack_with_pos_contig( ompi_convertor_t* pConvertor,
/* we save the current displacement starting from the begining /* we save the current displacement starting from the begining
* of this data. * of this data.
*/ */
if( 0 == count ) { if( OPAL_LIKELY(0 == count) ) {
pStack[1].type = pElems->elem.common.type; pStack[1].type = pElems->elem.common.type;
pStack[1].count = pElems->elem.count; pStack[1].count = pElems->elem.count;
pStack[1].disp = pElems->elem.disp; pStack[1].disp = pElems->elem.disp;
@ -403,7 +403,7 @@ int32_t ompi_convertor_set_position_nocheck( ompi_convertor_t* convertor,
rc = ompi_convertor_create_stack_at_begining( convertor, ompi_ddt_local_sizes ); rc = ompi_convertor_create_stack_at_begining( convertor, ompi_ddt_local_sizes );
if( 0 == (*position) ) return rc; if( 0 == (*position) ) return rc;
} }
if( convertor->flags & DT_FLAG_CONTIGUOUS ) { if( OPAL_LIKELY(convertor->flags & DT_FLAG_CONTIGUOUS) ) {
rc = ompi_convertor_create_stack_with_pos_contig( convertor, (*position), rc = ompi_convertor_create_stack_with_pos_contig( convertor, (*position),
ompi_ddt_local_sizes ); ompi_ddt_local_sizes );
} else { } else {
@ -443,13 +443,13 @@ int32_t ompi_convertor_set_position_nocheck( ompi_convertor_t* convertor,
* completed. With this flag set the pack and unpack functions \ * completed. With this flag set the pack and unpack functions \
* will not do anything. \ * will not do anything. \
*/ \ */ \
if( OPAL_UNLIKELY(0 == convertor->local_size) ) { \ if( OPAL_UNLIKELY((0 == count) || (0 == datatype->size)) ) { \
convertor->flags |= CONVERTOR_COMPLETED; \ convertor->flags |= CONVERTOR_COMPLETED; \
convertor->remote_size = 0; \ convertor->remote_size = 0; \
return OMPI_SUCCESS; \ return OMPI_SUCCESS; \
} \ } \
\ \
if( convertor->remoteArch == ompi_mpi_local_arch ) { \ if( OPAL_LIKELY(convertor->remoteArch == ompi_mpi_local_arch) ) { \
convertor->remote_size = convertor->local_size; \ convertor->remote_size = convertor->local_size; \
if( (convertor->flags & (CONVERTOR_WITH_CHECKSUM | DT_FLAG_NO_GAPS)) == DT_FLAG_NO_GAPS ) { \ if( (convertor->flags & (CONVERTOR_WITH_CHECKSUM | DT_FLAG_NO_GAPS)) == DT_FLAG_NO_GAPS ) { \
return OMPI_SUCCESS; \ return OMPI_SUCCESS; \
@ -599,7 +599,7 @@ int ompi_convertor_clone( const ompi_convertor_t* source,
destination->local_size = source->local_size; destination->local_size = source->local_size;
destination->remote_size = source->remote_size; destination->remote_size = source->remote_size;
/* create the stack */ /* create the stack */
if( source->stack_size > DT_STATIC_STACK_SIZE ) { if( OPAL_UNLIKELY(source->stack_size > DT_STATIC_STACK_SIZE) ) {
destination->pStack = (dt_stack_t*)malloc(sizeof(dt_stack_t) * source->stack_size ); destination->pStack = (dt_stack_t*)malloc(sizeof(dt_stack_t) * source->stack_size );
} else { } else {
destination->pStack = destination->static_stack; destination->pStack = destination->static_stack;
@ -607,7 +607,7 @@ int ompi_convertor_clone( const ompi_convertor_t* source,
destination->stack_size = source->stack_size; destination->stack_size = source->stack_size;
/* initialize the stack */ /* initialize the stack */
if( 0 == copy_stack ) { if( OPAL_LIKELY(0 == copy_stack) ) {
destination->bConverted = -1; destination->bConverted = -1;
destination->stack_pos = -1; destination->stack_pos = -1;
} else { } else {

9
ompi/datatype/convertor.h
Просмотреть файл

@ -23,6 +23,7 @@
#include "ompi_config.h" #include "ompi_config.h"
#include "ompi/constants.h" #include "ompi/constants.h"
#include "ompi/datatype/datatype.h" #include "ompi/datatype/datatype.h"
#include "opal/prefetch.h"
#ifdef HAVE_SYS_TYPES_H #ifdef HAVE_SYS_TYPES_H
#include <sys/types.h> #include <sys/types.h>
#endif #endif
@ -143,7 +144,7 @@ OMPI_DECLSPEC ompi_convertor_t* ompi_convertor_create( int32_t remote_arch, int3
*/ */
static inline int ompi_convertor_cleanup( ompi_convertor_t* convertor ) static inline int ompi_convertor_cleanup( ompi_convertor_t* convertor )
{ {
if( convertor->stack_size > DT_STATIC_STACK_SIZE ) { if( OPAL_UNLIKELY(convertor->stack_size > DT_STATIC_STACK_SIZE) ) {
free( convertor->pStack ); free( convertor->pStack );
convertor->pStack = convertor->static_stack; convertor->pStack = convertor->static_stack;
convertor->stack_size = DT_STATIC_STACK_SIZE; convertor->stack_size = DT_STATIC_STACK_SIZE;
@ -256,13 +257,13 @@ ompi_convertor_set_position( ompi_convertor_t* convertor,
/* /*
* If the convertor is already at the correct position we are happy. * If the convertor is already at the correct position we are happy.
*/ */
if( (*position) == convertor->bConverted ) return OMPI_SUCCESS; if( OPAL_LIKELY((*position) == convertor->bConverted) ) return OMPI_SUCCESS;
/* /*
* Do not allow the convertor to go outside the data boundaries. This test include * Do not allow the convertor to go outside the data boundaries. This test include
* the check for datatype with size zero as well as for convertors with a count of zero. * the check for datatype with size zero as well as for convertors with a count of zero.
*/ */
if( convertor->local_size <= *position) { if( OPAL_UNLIKELY(convertor->local_size <= *position) ) {
convertor->flags |= CONVERTOR_COMPLETED; convertor->flags |= CONVERTOR_COMPLETED;
convertor->bConverted = convertor->local_size; convertor->bConverted = convertor->local_size;
*position = convertor->bConverted; *position = convertor->bConverted;
@ -291,7 +292,7 @@ ompi_convertor_personalize( ompi_convertor_t* convertor, uint32_t flags,
{ {
convertor->flags |= flags; convertor->flags |= flags;
if( NULL == position ) if( OPAL_UNLIKELY(NULL == position) )
return OMPI_SUCCESS; return OMPI_SUCCESS;
return ompi_convertor_set_position( convertor, position ); return ompi_convertor_set_position( convertor, position );
} }