source: soft/giet_vm/applications/router/router.c @ 720

/////////////////////////////////////////////////////////////////////////////////////////////
// File   : router.c   
// Date   : november 2014
// author : Alain Greiner
/////////////////////////////////////////////////////////////////////////////////////////////
// This multi-threaded application illustrates "task-farm" parallelism.
// It is described as a TCG (Task and Communication Graph).
// It contains 2 + N threads : one "producer", one "consumer" and N "router"), 
// plus the "main" thread, that makes the global initialisation, launches the other
// threads, and exit.
// It contains 2 MWMR channels per cluster : "fifo_in" and "fifo_out", that are
// allocated by the main thread in the user heaps distributed in all clusters.
// - The "producer" task writes token in all the  "fifo_in".
// - The N "router" tasks read token from "fifo_in" and write them into "fifo_out".
// - The "consumer" task read token from "fifo_out" and displays instrumentation results.
// Token are indexed (by the producer) from 0 to NMAX-1.
// The router task contains a random delay emulating a variable processing time.
// For instrumentation, the "consumer_tab" array is filled by the "consumer" task.
// Each entry contain the arrival order to the consumer task.
/////////////////////////////////////////////////////////////////////////////////////////////


#include "stdio.h"
#include "mwmr_channel.h"
#include "mapping_info.h"
#include "hard_config.h"
#include "malloc.h"

#define VERBOSE  1
#define NMAX     32       // total number of token
#define DEPTH    32       // MWMR channels depth


// macro to use a shared TTY
#define printf(...);  { lock_acquire( &tty_lock ); \
                        giet_tty_printf(__VA_ARGS__);  \
                        lock_release( &tty_lock ); }

// lock protecting shared TTY
user_lock_t           tty_lock;

 
// arguments for the producer, consumer, and router functions
typedef struct
{
    mwmr_channel_t*  pin;       // pointer on the MWMR input fifo
    mwmr_channel_t*  pout;      // pointer on the MWMR output fifo
    unsigned int     x_size;    // number of clusters in a row
    unsigned int     y_size;    // number of clusters in a column
} args_t;


// arrays of pointers
mwmr_channel_t*  fifo_in[16][16];
mwmr_channel_t*  fifo_out[16][16];


/////////////////////////////////////////////
__attribute__ ((constructor)) void producer()
{
    unsigned int    x      = 0;             // destination cluster coordinate
    unsigned int    y      = 0;             // destination cluster coordinate
    unsigned int    token  = 0;             // token value

    // get plat-form parameters
    unsigned int x_size;                    // number of clusters in a row
    unsigned int y_size;                    // number of clusters in a column
    unsigned int nprocs;                    // unused
    giet_procs_number( &x_size , &y_size , &nprocs );

   // loop on the clusters
    while ( token < NMAX )
    {
        // try to write a token in fifo_in[x,y]
        if ( nb_mwmr_write( fifo_in[x][y] , &token , 1 ) == 1 )
        {
            if ( VERBOSE ) printf("[PRODUCER] token %d sent to cluster(%d,%d)"
                                  " at cycle %d\n", token , x , y , giet_proctime() );
            token++;
        }

        // compute next cluster coordinates
        x++;
        if ( x == x_size )
        {
            x = 0;
            y++;
            if ( y == y_size ) y = 0;
        }
    }

    giet_pthread_exit( "Producer task completed");

} // end producer()

/////////////////////////////////////////////
__attribute__ ((constructor)) void consumer()
{
    unsigned int    x      = 0;             // source cluster coordinate
    unsigned int    y      = 0;             // source cluster coordinate
    unsigned int    n      = 0;             // index of received token
    unsigned int    token;                  // token value
    unsigned int    consumer_tab[NMAX];     // received token array

    // get plat-form parameters
    unsigned int x_size;                    // number of clusters in a row
    unsigned int y_size;                    // number of clusters in a column
    unsigned int nprocs;                    // unused
    giet_procs_number( &x_size , &y_size , &nprocs );

    // loop on the clusters
    while ( n < NMAX )
    { 
        // try to read a token from fifo_out[x,y]
        if ( nb_mwmr_read( fifo_out[x][y] , &token , 1 ) == 1 )
        {
            consumer_tab[n] = token;
            n++;

            if ( VERBOSE ) printf("[CONSUMER] token %d received at cycle %d\n",
                                  token , giet_proctime() );
        }

        // compute next cluster coordinates
        x++;
        if ( x == x_size )
        {
            x = 0;
            y++;
            if ( y == y_size ) y = 0;
        }
    }

    // instrumentation display
    giet_tty_printf("\n[CONSUMER] displays instrumentation results\n");
    for( n = 0 ; n < NMAX ; n++ )
    {
        giet_tty_printf(" - arrival = %d / value = %d\n", n , consumer_tab[n] );
    }

    giet_pthread_exit( "Consumer task completed");

} // end consumer()

////////////////////////////////////////////
__attribute__ ((constructor)) void compute()
{
    unsigned int    token;           // token value
    unsigned int        count;           // tempo

    // get proc coordinates
    unsigned int  x;
    unsigned int  y;
    unsigned int  p;
    giet_proc_xyp( &x , &y , &p );

    // main loop
    while(1)
    {
        mwmr_read( fifo_in[x][y] , &token , 1 );
        for ( count = 0 ; count < (giet_rand() << 2) ; count++ ) asm volatile ( "nop" );
        mwmr_write( fifo_out[x][y] , &token , 1 );

        if ( VERBOSE ) printf("[COMPUTE] token %d handled at cycle %d on P[%d,%d,%d]\n", 
                              token , giet_proctime() , x , y , p );
    }
} // end compute()


/////////////////////////////////////////
__attribute__ ((constructor)) void main()
{
    // get plat-form parameters
    unsigned int x_size;                       // number of clusters in a row
    unsigned int y_size;                       // number of clusters in a column
    unsigned int nprocs;                       // number of processors per cluster
    giet_procs_number( &x_size , &y_size , &nprocs );

    // shared TTY allocation
    giet_tty_alloc( 1 );     
    lock_init( &tty_lock);

    // check plat-form parameters
    giet_pthread_assert( ((nprocs >= 3) && (nprocs <= 8)),
                         "[ROUTER ERROR] nprocs per cluster must be in [3...8]");

    giet_pthread_assert( ((x_size >= 1) && (x_size <= 16)),
                         "[ROUTER ERROR] x_size must be in [1...16]");

    giet_pthread_assert( ((y_size >= 1) && (y_size <= 16)),
                         "[ROUTER ERROR] y_size must be in [1...16]");

    // index for loops
    unsigned int x;
    unsigned int y;
    unsigned int n;

    // distributed heap initialisation, plus
    // MWMR channels and associated buffers allocation
    for ( x = 0 ; x < x_size ; x++ ) 
    {
        for ( y = 0 ; y < y_size ; y++ ) 
        {
            heap_init( x , y );

            // allocate MWMR channel descriptors in cluster[x][y]
                fifo_in[x][y]  = remote_malloc( sizeof( mwmr_channel_t ) , x , y );
                fifo_out[x][y] = remote_malloc( sizeof( mwmr_channel_t ) , x , y );

                // allocate data buffers in cluster[x][y]
                unsigned int* buf_in  = remote_malloc( 4*DEPTH , x , y );            
                unsigned int* buf_out = remote_malloc( 4*DEPTH , x , y );            

                // initialize MWMR channels
                mwmr_init( fifo_in[x][y]  , buf_in  , 1 , DEPTH );
                mwmr_init( fifo_out[x][y] , buf_out , 1 , DEPTH );
        }
    }

    printf("\n[ROUTER] main completes initialisation at cycle %d for %d cores\n", 
           giet_proctime(), (x_size * y_size * nprocs) );

    // thread index and function for pthread_create()
    pthread_t   trdid; 
    void*       function;

    // launch producer, consumer and router threads
    for ( x = 0 ; x < x_size ; x++ )
    {
        for ( y = 0 ; y < y_size ; y++ )
        {
            for ( n = 0 ; n < nprocs ; n++ )
            {
                if      ( (x==0) && (y==0) && (n==0) )  function = &producer;
                else if ( (x==0) && (y==0) && (n==1) )  function = &consumer;
                else                                    function = &compute;
                
                if ( giet_pthread_create( &trdid,
                                          NULL,                  // no attribute
                                          function,
                                          NULL ) )               // no argument
                {
                    printf("\n[ROUTER ERROR] launching thread on P[%d,%d,%d]\n", x, y, n );
                    giet_pthread_exit( NULL );
                }
            }
        }
    }

    giet_pthread_exit( "main completed" );
    
} // end main()