Skip to content

Improve LOOP datatype exchanges performances for sizes above the eage… #6172

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
derbeyn wants to merge 1 commit into from
Closed

Improve LOOP datatype exchanges performances for sizes above the eage… #6172

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 32 additions & 1 deletion opal/datatype/opal_datatype_pack.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -286,6 +286,7 @@ opal_generic_simple_pack_function( opal_convertor_t* pConvertor,
unsigned char *conv_ptr, *iov_ptr;
size_t iov_len_local;
uint32_t iov_count;
int save_disp = 0;

DO_DEBUG( opal_output( 0, "opal_convertor_generic_simple_pack( %p:%p, {%p, %lu}, %d )\n",
(void*)pConvertor, (void*)pConvertor->pBaseBuf,
Expand Down Expand Up @@ -340,6 +341,34 @@ opal_generic_simple_pack_function( opal_convertor_t* pConvertor,
pConvertor->stack_pos--; /* go one position up on the stack */
pStack--;
pos_desc++; /* and move to the next element */
} else if (pStack->index != -1 &&
OPAL_DATATYPE_LOOP == description[pStack->index].loop.common.type &&
pStack->count > 0) {
/*
* if we are not at the end of the datatype
* AND we've got a LOOP to process in the stack
* AND the count of blocks remaining in the LOOP is greater
* than 0.
* ==> do what is needed to convert the remaining data into
* a LOOP element.
* Doing this, the remaining of the loop will be itself
* processed as a loop instead of being processed data block
* by data block.
*/
description = pConvertor->use_desc->desc;
pStack->disp += description[pStack->index].loop.extent;

pos_desc = pStack->index;
conv_ptr = pConvertor->pBaseBuf + pStack->disp;
count_desc = (uint32_t)pStack->count;

save_disp = (pStack)->disp;
pStack--;
save_disp -= (pStack)->disp;
pConvertor->stack_pos--;
pElem = &(description[pos_desc]);

goto loop_part;
} else {
pos_desc = pStack->index + 1; /* jump back to the begining of the loop */
if( pStack->index == -1 ) { /* If it's the datatype count loop */
Expand All @@ -355,6 +384,7 @@ opal_generic_simple_pack_function( opal_convertor_t* pConvertor,
(int)pStack->count, pConvertor->stack_pos, pos_desc,
count_desc, (long)pStack->disp, (unsigned long)iov_len_local ); );
}
loop_part:
if( OPAL_DATATYPE_LOOP == pElem->elem.common.type ) {
ptrdiff_t local_disp = (ptrdiff_t)conv_ptr;
if( pElem->loop.common.flags & OPAL_DATATYPE_FLAG_CONTIGUOUS ) {
Expand All @@ -368,9 +398,10 @@ opal_generic_simple_pack_function( opal_convertor_t* pConvertor,
}
local_disp = (ptrdiff_t)conv_ptr - local_disp;
PUSH_STACK( pStack, pConvertor->stack_pos, pos_desc, OPAL_DATATYPE_LOOP, count_desc,
pStack->disp + local_disp);
pStack->disp + local_disp + save_disp);
pos_desc++;
update_loop_description: /* update the current state */
save_disp = 0;
conv_ptr = pConvertor->pBaseBuf + pStack->disp;
UPDATE_INTERNAL_COUNTERS( description, pos_desc, pElem, count_desc );
DDT_DUMP_STACK( pConvertor->pStack, pConvertor->stack_pos, pElem, "advance loop" );
Expand Down
34 changes: 33 additions & 1 deletion opal/datatype/opal_datatype_unpack.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -273,6 +273,7 @@ opal_generic_simple_unpack_function( opal_convertor_t* pConvertor,
unsigned char *conv_ptr, *iov_ptr;
size_t iov_len_local;
uint32_t iov_count;
int save_disp = 0;

DO_DEBUG( opal_output( 0, "opal_convertor_generic_simple_unpack( %p, {%p, %lu}, %u )\n",
(void*)pConvertor, (void*)iov[0].iov_base, (unsigned long)iov[0].iov_len, *out_size ); );
Expand Down Expand Up @@ -363,6 +364,35 @@ opal_generic_simple_unpack_function( opal_convertor_t* pConvertor,
pConvertor->stack_pos--;
pStack--;
pos_desc++;
} else if (pStack->index != -1 &&
OPAL_DATATYPE_LOOP == description[pStack->index].loop.common.type &&
pStack->count > 0) {
/*
* if we are not at the end of the datatype
* AND we've got a LOOP to process in the stack
* AND the count of blocks remaining in the LOOP is greater
* than 0.
* ==> do what is needed to convert the remaining data into
* a LOOP element.
* Doing this, the remaining of the loop will be itself
* processed as a loop instead of being processed data block
* by data block.
*/
description = pConvertor->use_desc->desc;
pStack->disp += description[pStack->index].loop.extent;

pos_desc = pStack->index;
conv_ptr = pConvertor->pBaseBuf + pStack->disp;
count_desc = (uint32_t)pStack->count;

save_disp = (pStack)->disp;
pStack--;
save_disp -= (pStack)->disp;
pConvertor->stack_pos--;
pElem = &(description[pos_desc]);

goto loop_part;

} else {
pos_desc = pStack->index + 1;
if( pStack->index == -1 ) {
Expand All @@ -378,6 +408,7 @@ opal_generic_simple_unpack_function( opal_convertor_t* pConvertor,
(int)pStack->count, pConvertor->stack_pos, pos_desc,
(long)pStack->disp, (unsigned long)iov_len_local ); );
}
loop_part:
if( OPAL_DATATYPE_LOOP == pElem->elem.common.type ) {
ptrdiff_t local_disp = (ptrdiff_t)conv_ptr;
if( pElem->loop.common.flags & OPAL_DATATYPE_FLAG_CONTIGUOUS ) {
Expand All @@ -391,9 +422,10 @@ opal_generic_simple_unpack_function( opal_convertor_t* pConvertor,
}
local_disp = (ptrdiff_t)conv_ptr - local_disp;
PUSH_STACK( pStack, pConvertor->stack_pos, pos_desc, OPAL_DATATYPE_LOOP, count_desc,
pStack->disp + local_disp);
pStack->disp + local_disp + save_disp);
pos_desc++;
update_loop_description: /* update the current state */
save_disp = 0;
conv_ptr = pConvertor->pBaseBuf + pStack->disp;
UPDATE_INTERNAL_COUNTERS( description, pos_desc, pElem, count_desc );
DDT_DUMP_STACK( pConvertor->pStack, pConvertor->stack_pos, pElem, "advance loop" );
Expand Down