source: trunk/user/sort/sort.c @ 639

/*
 * sort.c - Parallel sort
 * 
 * Author     Cesar Fuguet Tortolero (2013)
 *            Alain Greiner (2019)
 *
 * Copyright (c) UPMC Sorbonne Universites
 *
 * This is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2.0 of the License.
 *
 * It is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with ALMOS-MKH; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 */

///////////////////////////////////////////////////////////////////////////////
// This multi-threaded application implement a multi-stage sort.
// It has been writen by Cesar Fuget Tortolero in 2013.
// It has been ported on ALMOS-MKH by Alain Greiner in 2019.
//
// There is one thread per physical cores. 
// Computation is organised as a binary tree: 
// - All threads execute in parallel a buble sort on a sub-array during the
//   the first stage of parallel sort,
// - The number of participating threads is divided by 2 at each next stage,
//   to make a merge sort, on two subsets of previous stage.
//
//       Number_of_stages = number of barriers = log2(Number_of_threads)
//
// The various stages are separated by synchronisation barriers, and the
// main thread uses the join syscall to check that all threads completed
// before printing the computation time (sequencial & parallel). 
// These results can be - optionnaly - registered in an instrumentation file.
//
// Constraints :
// - It supports up to 1024 cores: x_size, y_size, and ncores must be
//   power of 2 (max 16*16 clusters / max 4 cores per cluster)
// _ The array of values to be sorted (ARRAY_LENGTH) must be power of 2 
//   larger than the number of cores.
///////////////////////////////////////////////////////////////////////////////

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <pthread.h>
#include <almosmkh.h>
#include <hal_macros.h>

#define ARRAY_LENGTH        2048            // number of items
#define MAX_THREADS         1024            // 16 * 16 * 4

#define X_MAX               16              // max number of clusters in a row
#define Y_MAX               16              // max number of clusters in a column
#define CORES_MAX           4               // max number of cores in a cluster
#define CLUSTERS_MAX        X_MAX * Y_MAX

#define USE_DQT_BARRIER     1               // use DQT barrier if non zero
#define DISPLAY_ARRAY       0               // display items values before and after
#define DEBUG_MAIN          0               // trace main function
#define DEBUG_SORT          0               // trace sort function
#define CHECK_RESULT        0               // for debug
#define INSTRUMENTATION     1               // register computation times on file

///////////////////////////////////////////////////////////////////////////////////
//            Arguments for the sort() function 
///////////////////////////////////////////////////////////////////////////////////

typedef struct
{
    unsigned int        tid;                // continuous thread index
    unsigned int        threads;            // total number of threads
    pthread_barrier_t * parent_barrier;     // pointer on termination barrier
}
sort_args_t;

////////////////////////////////////////////////////////////////////////////////////
//            Sort specific global variables
////////////////////////////////////////////////////////////////////////////////////

int                 array0[ARRAY_LENGTH];    // values to sort
int                 array1[ARRAY_LENGTH];    

pthread_barrier_t   barrier;                 // synchronisation variables

/////////////////////////////////////////////////////////////////////////////////////
//             Global variables required by parallel_pthread_create()
/////////////////////////////////////////////////////////////////////////////////////

// 2D arrays of input arguments for the <sort> threads
// These arrays are initialised by the application main thread

sort_args_t       sort_args[CLUSTERS_MAX][CORES_MAX];  // sort function arguments 
sort_args_t     * sort_ptrs[CLUSTERS_MAX][CORES_MAX];  // pointers on arguments 

// 1D array of barriers to allow the <sort> threads to signal termination 
// this array is initialised by the pthread_parallel_create() function
  
pthread_barrier_t parent_barriers[CLUSTERS_MAX];       // termination barrier


////////////////////////////////////
static void bubbleSort( int * array,
                        unsigned int length,
                        unsigned int init_pos )
{
    unsigned int i;
    unsigned int j;
    int          aux;

    for(i = 0; i < length; i++)
    {
        for(j = init_pos; j < (init_pos + length - i - 1); j++)
        {
            if(array[j] > array[j + 1])
            {
                aux          = array[j + 1];
                array[j + 1] = array[j];
                array[j]     = aux;
            }
        }
    }
}  // end bubbleSort()


///////////////////////////////////
static void merge( const int * src,               // source array
                   int       * dst,               // destination array
                   int         length,            // number of items in a subset
                   int         init_pos_src_a,    // index first item in src subset A
                   int         init_pos_src_b,    // index first item in src subset B
                   int         init_pos_dst )     // index first item in destination
{
    int i;
    int j;
    int k;

    i = 0;
    j = 0;
    k = init_pos_dst;

    while((i < length) || (j < length))
    {
        if((i < length) && (j < length))
        {
            if(src[init_pos_src_a + i] < src[init_pos_src_b + j])
            {
                dst[k++] = src[init_pos_src_a + i];
                i++;
            }
            else
            {
                dst[k++] = src[init_pos_src_b + j];
                j++;
            }
        }
        else if(i < length)
        {
            dst[k++] = src[init_pos_src_a + i];
            i++;
        }
        else
        {
            dst[k++] = src[init_pos_src_b + j];
            j++;
        }
    }
}  // end merge()

//////////////////////////////
void sort( sort_args_t * ptr )
{
    unsigned int        i;
    int               * src_array  = NULL;
    int               * dst_array  = NULL;

    // get arguments
    unsigned int        tid            = ptr->tid;
    unsigned int        threads        = ptr->threads;
    pthread_barrier_t * parent_barrier = ptr->parent_barrier;

    unsigned int        items      = ARRAY_LENGTH / threads;
    unsigned int        stages     = __builtin_ctz( threads ) + 1;

#if DEBUG_SORT
printf("\n[sort] start : ptr %x / tid %d / threads %d / barrier %x\n",
ptr, tid, threads, parent_barrier );
#endif

    bubbleSort( array0, items, items * tid );

#if DEBUG_SORT
printf("\n[sort] thread[%d] : stage 0 completed\n", tid );
#endif

    /////////////////////////////////
    pthread_barrier_wait( &barrier ); 

#if DEBUG_SORT
printf("\n[sort] thread[%d] exit barrier 0\n", tid );
#endif

    // the number of threads contributing to sort is divided by 2
    // and the number of items is multiplied by 2 at each next stage
    for ( i = 1 ; i < stages ; i++ )
    {
        if((i % 2) == 1)               // odd stage 
        {
            src_array = array0;
            dst_array = array1;
        }
        else                           // even stage
        {
            src_array = array1;
            dst_array = array0;
        }

        if( (tid & ((1<<i)-1)) == 0 )
        {

#if DEBUG_SORT
printf("\n[sort] thread[%d] : stage %d start\n", tid , i );
#endif
            merge( src_array, 
                   dst_array,
                   items << (i-1),
                   items * tid,
                   items * (tid + (1 << (i-1))),
                   items * tid );

#if DEBUG_SORT
printf("\n[sort] thread[%d] : stage %d completed\n", tid , i );
#endif
        }

        /////////////////////////////////
        pthread_barrier_wait( &barrier );

#if DEBUG_SORT
printf("\n[sort] thread[%d] exit barrier %d\n", tid , i );
#endif

    }  // en for stages 

    // sort thread signal completion to main thread
    pthread_barrier_wait( parent_barrier );

#if DEBUG_SORT
printf("\n[sort] thread[%d] exit\n", tid );
#endif

    // sort thread exit
    pthread_exit( NULL );

} // end sort()


/////////////////
void main( void )
{
    int                    error;
    unsigned int           x_size;             // number of rows
    unsigned int           y_size;             // number of columns
    unsigned int           ncores;             // number of cores per cluster
    unsigned int           total_threads;      // total number of threads
    unsigned int           x;                  // X coordinate for a sort thread
    unsigned int           y;                  // Y coordinate for a sort thread
    unsigned int           cxy;                // cluster identifier for a sort thead
    unsigned int           lid;                // core local index for a thread
    unsigned int           tid;                // sort thread continuous index
    pthread_barrierattr_t  barrier_attr;       // barrier attributes (used for DQT)
    unsigned int           n;                  // index in array to sort

    unsigned long long     start_cycle;
    unsigned long long     seq_end_cycle;
    unsigned long long     para_end_cycle;

    /////////////////////////
    get_cycle( &start_cycle );
 
    // compute number of threads (one thread per core)
    get_config( &x_size , &y_size , &ncores );
    total_threads = x_size * y_size * ncores;

    // compute covering DQT size an level
    unsigned int z = (x_size > y_size) ? x_size : y_size;
    unsigned int root_level = (z == 1) ? 0 : (z == 2) ? 1 : (z == 4) ? 2 : (z == 8) ? 3 : 4;

    // checks number of threads
    if ( (total_threads != 1)   && (total_threads != 2)   && (total_threads != 4)   && 
         (total_threads != 8)   && (total_threads != 16 ) && (total_threads != 32)  && 
         (total_threads != 64)  && (total_threads != 128) && (total_threads != 256) && 
         (total_threads != 512) && (total_threads != 1024) )
    {
        printf("\n[sort] ERROR : number of cores must be power of 2\n");
        exit( 0 );
    }

    // check array size
    if ( ARRAY_LENGTH % total_threads) 
    {
        printf("\n[sort] ERROR : array size must be multiple of number of threads\n");
        exit( 0 );
    }

    printf("\n[sort] main starts / %d threads / %d items / pid %x / cycle %d\n",
    total_threads, ARRAY_LENGTH, getpid(), (unsigned int)start_cycle );

    // initialize barrier
    if( USE_DQT_BARRIER )
    {
        barrier_attr.x_size   = x_size; 
        barrier_attr.y_size   = y_size;
        barrier_attr.nthreads = ncores;
        error = pthread_barrier_init( &barrier, &barrier_attr , total_threads );
    }
    else // use SIMPLE_BARRIER
    {
        error = pthread_barrier_init( &barrier, NULL , total_threads );
    }

    if( error )
    {
        printf("\n[sort] ERROR : cannot initialise barrier\n" );
        exit( 0 );
    }

#if DEBUG_MAIN
if( USE_DQT_BARRIER ) printf("\n[sort] main completes DQT barrier init\n");
else                  printf("\n[sort] main completes simple barrier init\n");
#endif

    // Array to sort initialization
    for ( n = 0 ; n < ARRAY_LENGTH ; n++ )
    {
        array0[n] = ARRAY_LENGTH - n - 1;
    }

#if DISPLAY_ARRAY
    printf("\n*** array before sort\n");
    for( n=0; n<ARRAY_LENGTH; n++) printf("array[%d] = %d\n", n , array0[n] );
#endif

#if DEBUG_MAIN
printf("\n[sort] main completes array init\n");
#endif

    // build array of arguments for the <sort> threads
    for (x = 0 ; x < x_size ; x++)
    {
        for (y = 0 ; y < y_size ; y++)
        {
            // compute cluster identifier
            cxy = HAL_CXY_FROM_XY( x , y );

            for ( lid = 0 ; lid < ncores ; lid++ )
            {
                // compute thread continuous index
                tid = (((x * y_size) + y) * ncores) + lid;

                // initialize 2D array of arguments
                sort_args[cxy][lid].tid            = tid;
                sort_args[cxy][lid].threads        = total_threads;
                sort_args[cxy][lid].parent_barrier = &parent_barriers[cxy];

                // initialize 2D array of pointers
                sort_ptrs[cxy][lid] = &sort_args[cxy][lid];
            }
        }
    }

    ///////////////////////////
    get_cycle( &seq_end_cycle );

#if DEBUG_MAIN
printf("\n[sort] main completes sequencial init at cycle %d\n",
(unsigned int)seq_end_cycle );
#endif

    // create and execute the working threads
    if( pthread_parallel_create( root_level,
                                 &sort,
                                 &sort_ptrs[0][0],
                                 &parent_barriers[0] ) )
    {
        printf("\n[sort] ERROR : cannot create threads\n");
        exit( 0 );
    }

    ////////////////////////////
    get_cycle( &para_end_cycle );

#if DEBUG_main
printf("\n[sort] main completes parallel sort at cycle %d\n", 
(unsigned int)para_end_cycle );
#endif

    // destroy barrier
    pthread_barrier_destroy( &barrier );

#if DISPLAY_ARRAY
    printf("\n*** array after merge %d\n", i );
    for( n=0; n<ARRAY_LENGTH; n++) printf("array[%d] = %d\n", n , dst_array[n] );
#endif

#if CHECK_RESULT
    int    success = 1;
    int *  res_array = ( (total_threads ==   2) ||
                         (total_threads ==   8) || 
                         (total_threads ==  32) || 
                         (total_threads == 128) || 
                         (total_threads == 512) ) ? array1 : array0;

    for( n=0 ; n<(ARRAY_LENGTH-2) ; n++ )
    {
        if ( res_array[n] > res_array[n+1] )
        {
            printf("\n[sort] array[%d] = %d > array[%d] = %d\n",
            n , res_array[n] , n+1 , res_array[n+1] );
            success = 0;
            break;
        }
    }

    if ( success ) printf("\n[sort] success\n");
    else           printf("\n[sort] failure\n");
#endif

#if INSTRUMENTATION
    char               name[64];
    char               path[128];
    unsigned long long instru_cycle;

    // build file name 
    if( USE_DQT_BARRIER )
    snprintf( name , 64 , "p_sort_dqt_%d_%d_%d", ARRAY_LENGTH, x_size * y_size, ncores );
    else
    snprintf( name , 64 , "p_sort_smp_%d_%d_%d", ARRAY_LENGTH, x_size * y_size, ncores );

    // build file pathname
    snprintf( path , 128 , "home/%s" , name );

    // compute results
    unsigned int sequencial = (unsigned int)(seq_end_cycle - start_cycle);
    unsigned int parallel   = (unsigned int)(para_end_cycle - seq_end_cycle);

    // display results on process terminal
    printf("\n----- %s -----\n"
           " - sequencial : %d cycles\n"
           " - parallel   : %d cycles\n", 
           name, sequencial, parallel );

    // open file
    get_cycle( &instru_cycle );
    FILE * stream = fopen( path , NULL );

    if( stream == NULL )
    {
        printf("\n[sort] ERROR : cannot open instrumentation file <%s>\n", path );
        exit(0);
    }

    printf("\n[sort] file <%s> open at cycle %d\n", path, (unsigned int)instru_cycle );

#if IDBG
idbg();
#endif

    // register results to file
    get_cycle( &instru_cycle );
    int ret = fprintf( stream , "\n----- %s -----\n"
                                " - sequencial : %d cycles\n"
                                " - parallel   : %d cycles\n", name, sequencial, parallel );
    if( ret < 0 )
    {
        printf("\n[sort] ERROR : cannot write to instrumentation file <%s>\n", path );
        exit(0);
    }

    printf("\n[sort] file <%s> written at cycle %d\n", path, (unsigned int)instru_cycle );

#if IDBG
idbg();
#endif

    // close instrumentation file
    get_cycle( &instru_cycle );
    ret = fclose( stream );

    if( ret )
    {
        printf("\n[sort] ERROR : cannot close instrumentation file <%s>\n", path );
        exit(0);
    }

    printf("\n[sort] file <%s> closed at cycle %d\n", path, (unsigned int)instru_cycle );

#endif

    exit( 0 );

}  // end main()

/* 
vim: tabstop=4 : shiftwidth=4 : expandtab
*/