mirror of
https://github.com/amd/blis.git
synced 2026-05-24 10:24:34 +00:00
259 lines
8.5 KiB
C
259 lines
8.5 KiB
C
/*
|
|
|
|
BLIS
|
|
An object-based framework for developing high-performance BLAS-like
|
|
libraries.
|
|
|
|
Copyright (C) 2014, The University of Texas
|
|
|
|
Redistribution and use in source and binary forms, with or without
|
|
modification, are permitted provided that the following conditions are
|
|
met:
|
|
- Redistributions of source code must retain the above copyright
|
|
notice, this list of conditions and the following disclaimer.
|
|
- Redistributions in binary form must reproduce the above copyright
|
|
notice, this list of conditions and the following disclaimer in the
|
|
documentation and/or other materials provided with the distribution.
|
|
- Neither the name of The University of Texas nor the names of its
|
|
contributors may be used to endorse or promote products derived
|
|
from this software without specific prior written permission.
|
|
|
|
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
#include "blis.h"
|
|
|
|
void bli_cleanup_communicator( thread_comm_t* communicator )
|
|
{
|
|
if( communicator == NULL ) return;
|
|
bli_destroy_lock( &communicator->barrier_lock );
|
|
}
|
|
|
|
void bli_setup_communicator( thread_comm_t* communicator, dim_t n_threads)
|
|
{
|
|
if( communicator == NULL ) return;
|
|
communicator->sent_object = NULL;
|
|
communicator->n_threads = n_threads;
|
|
communicator->barrier_sense = 0;
|
|
bli_init_lock( &communicator->barrier_lock );
|
|
communicator->barrier_threads_arrived = 0;
|
|
}
|
|
|
|
thread_comm_t* bli_create_communicator( dim_t n_threads )
|
|
{
|
|
thread_comm_t* comm = (thread_comm_t*) bli_malloc( sizeof(thread_comm_t) );
|
|
bli_setup_communicator( comm, n_threads );
|
|
return comm;
|
|
}
|
|
|
|
void* bli_broadcast_structure( thread_comm_t* communicator, dim_t id, void* to_send )
|
|
{
|
|
if( communicator == NULL || communicator->n_threads == 1 ) return to_send;
|
|
|
|
if( id == 0 ) communicator->sent_object = to_send;
|
|
|
|
bli_barrier( communicator, id );
|
|
void * object = communicator->sent_object;
|
|
bli_barrier( communicator, id );
|
|
|
|
return object;
|
|
}
|
|
|
|
void bli_init_lock( lock_t* lock )
|
|
{
|
|
omp_init_lock( lock );
|
|
}
|
|
void bli_destroy_lock( lock_t* lock )
|
|
{
|
|
omp_destroy_lock( lock );
|
|
}
|
|
void bli_set_lock( lock_t* lock )
|
|
{
|
|
omp_set_lock( lock );
|
|
}
|
|
void bli_unset_lock( lock_t* lock )
|
|
{
|
|
omp_unset_lock( lock );
|
|
}
|
|
|
|
//barrier routine taken from art of multicore programming or something
|
|
void bli_barrier( thread_comm_t* communicator, dim_t t_id )
|
|
{
|
|
if(communicator == NULL || communicator->n_threads == 1)
|
|
return;
|
|
bool_t my_sense = communicator->barrier_sense;
|
|
dim_t my_threads_arrived;
|
|
|
|
_Pragma("omp atomic capture")
|
|
my_threads_arrived = ++(communicator->barrier_threads_arrived);
|
|
|
|
/*
|
|
bli_set_lock(&communicator->barrier_lock);
|
|
my_threads_arrived = communicator->barrier_threads_arrived + 1;
|
|
communicator->barrier_threads_arrived = my_threads_arrived;
|
|
bli_unset_lock(&communicator->barrier_lock);
|
|
*/
|
|
|
|
if( my_threads_arrived == communicator->n_threads ) {
|
|
|
|
bli_set_lock(&communicator->barrier_lock);
|
|
communicator->barrier_threads_arrived = 0;
|
|
communicator->barrier_sense = !communicator->barrier_sense;
|
|
bli_unset_lock(&communicator->barrier_lock);
|
|
}
|
|
else {
|
|
volatile bool_t* listener = &communicator->barrier_sense;
|
|
while( *listener == my_sense ) {}
|
|
}
|
|
}
|
|
/*
|
|
//Recursively create thread communicators
|
|
void create_comms( dim_t* caucuses_at_level, dim_t n_levels, dim_t cur_level,
|
|
thread_comm_tree_t* parent, thread_comm_tree_t* leaves, dim_t global_id )
|
|
{
|
|
//Create a communicator
|
|
dim_t n_threads = 1;
|
|
for( dim_t i = cur_level; i < n_levels; i++)
|
|
n_threads *= caucuses_at_level[i];
|
|
|
|
|
|
thread_comm_t* comm = bli_create_communicator( n_threads );
|
|
thread_comm_tree_t* info;
|
|
if( cur_level == n_levels )
|
|
{
|
|
leaves[global_id].parent = parent;
|
|
leaves[global_id].comm = comm;
|
|
return;
|
|
}
|
|
else
|
|
{
|
|
info = (thread_comm_tree_t*)bli_malloc(sizeof(thread_comm_tree_t));
|
|
info->comm = comm;
|
|
info->parent = parent;
|
|
}
|
|
|
|
//Now create child communicators
|
|
dim_t caucuses = caucuses_at_level[cur_level];
|
|
for( dim_t i = 0; i < caucuses; i++)
|
|
create_comms( caucuses_at_level, n_levels, cur_level+1, info, leaves, global_id * caucuses + i);
|
|
}
|
|
*/
|
|
thrinfo_t* bli_create_thread_info( thread_comm_t* ocomm, dim_t ocomm_id, thread_comm_t* icomm, dim_t icomm_id,
|
|
dim_t n_way, dim_t work_id )
|
|
{
|
|
|
|
thrinfo_t* thr = (thrinfo_t*) bli_malloc( sizeof(thrinfo_t) );
|
|
bli_setup_thread_info( thr, ocomm, ocomm_id, icomm, icomm_id, n_way, work_id );
|
|
return thr;
|
|
}
|
|
|
|
void bli_setup_thread_info( thrinfo_t* thr, thread_comm_t* ocomm, dim_t ocomm_id, thread_comm_t* icomm, dim_t icomm_id,
|
|
dim_t n_way, dim_t work_id )
|
|
{
|
|
thr->ocomm = ocomm;
|
|
thr->ocomm_id = ocomm_id;
|
|
thr->icomm = icomm;
|
|
thr->icomm_id = icomm_id;
|
|
|
|
thr->n_way = n_way;
|
|
thr->work_id = work_id;
|
|
}
|
|
|
|
/*
|
|
thrinfo_t* bli_create_thread_info( dim_t* caucuses_at_level, dim_t n_levels )
|
|
{
|
|
//Calculate total number of threads
|
|
dim_t n_threads = 1;
|
|
for( dim_t i = 0; i < n_levels; i++)
|
|
n_threads *= caucuses_at_level[i];
|
|
|
|
//Create communicators
|
|
thread_comm_tree_t* comm_leaves = (thread_comm_tree_t*)bli_malloc( sizeof(thread_comm_tree_t) * n_threads);
|
|
create_comms( caucuses_at_level, n_levels, 0, NULL, comm_leaves, 0 );
|
|
thrinfo_t* info_paths = (thrinfo_t*)bli_malloc( sizeof(thrinfo_t) * n_threads );
|
|
|
|
//Now create paths upwards
|
|
for( dim_t i = 0; i < n_threads; i++ )
|
|
{
|
|
thread_comm_tree_t* comm_node = &comm_leaves[i];
|
|
|
|
//Setup thread info for the bottom-most level
|
|
thrinfo_t* bot = &BLIS_SINGLE_THREADED; //bli_create_thrinfo_t( comm_node->comm, 0, NULL, 1, 0 );
|
|
|
|
//Now build thread infos upwards
|
|
comm_node = comm_node->parent;
|
|
thrinfo_t* cur;
|
|
thrinfo_t* prev = bot;
|
|
for( dim_t j = 0; j < n_levels; j++ )
|
|
{
|
|
if( j == n_levels - 1 )
|
|
cur = &info_paths[i];
|
|
else
|
|
cur = (thrinfo_t*)bli_malloc(sizeof(thrinfo_t));
|
|
|
|
dim_t caucus_size = prev->ocomm->n_threads;
|
|
dim_t ocomm_id = i % comm_node->comm->n_threads;
|
|
dim_t caucus_id = ocomm_id / caucus_size;
|
|
|
|
bli_setup_thrinfo_t(cur, comm_node->comm, ocomm_id,
|
|
prev, caucuses_at_level[n_levels - j - 1], caucus_id );
|
|
|
|
prev = cur;
|
|
comm_node = comm_node->parent;
|
|
}
|
|
}
|
|
return info_paths;
|
|
}
|
|
*/
|
|
void bli_get_range( void* thr, dim_t size, dim_t block_factor, dim_t* start, dim_t* end )
|
|
{
|
|
thrinfo_t* thread = (thrinfo_t*) thr;
|
|
|
|
dim_t n_way = thread->n_way;
|
|
dim_t work_id = thread->work_id;
|
|
dim_t n_pt = size / n_way;
|
|
n_pt = (n_pt * n_way < size) ? n_pt + 1 : n_pt;
|
|
n_pt = (n_pt % block_factor == 0) ? n_pt : n_pt + block_factor - (n_pt % block_factor);
|
|
*start = work_id * n_pt;
|
|
*end = bli_min( *start + n_pt, size );
|
|
}
|
|
|
|
void bli_get_range_tri_weighted( void* thr, dim_t size, dim_t block_factor, bool_t forward, dim_t* start, dim_t* end)
|
|
{
|
|
}
|
|
|
|
void bli_level3_thread_decorator( dim_t n_threads,
|
|
level3_int_t func,
|
|
obj_t* alpha,
|
|
obj_t* a,
|
|
obj_t* b,
|
|
obj_t* beta,
|
|
obj_t* c,
|
|
void* cntl,
|
|
void** thread )
|
|
{
|
|
_Pragma( "omp parallel num_threads(n_threads)" )
|
|
{
|
|
dim_t omp_id = omp_get_thread_num();
|
|
|
|
func( alpha,
|
|
a,
|
|
b,
|
|
beta,
|
|
c,
|
|
cntl,
|
|
thread[omp_id] );
|
|
}
|
|
}
|