Ignore:
File:
1 edited

Legend:

Unmodified
Added
Removed

  • libcfa/src/executor.cfa

    r8d462e5 ra51c0c0  
    44// buffer.
    55
     6#include <thread.hfa>
    67#include <containers/list.hfa>
    7 #include <thread.hfa>
    8 #include <stdio.h>
    98
    10 struct WRequest {                                       // client request, no return
    11     void (* action)( void );
    12     DLISTED_MGD_IMPL_IN(WRequest)
     9forall( dtype T | $dlistable(T, T) ) {
     10        monitor Buffer {                                                                        // unbounded buffer
     11                dlist( T, T ) queue;                                                    // unbounded list of work requests
     12                condition delay;
     13        }; // Buffer
     14
     15        void insert( Buffer(T) & mutex buf, T * elem ) with(buf) {
     16                dlist( T, T ) * qptr = &queue;                                  // workaround https://cforall.uwaterloo.ca/trac/ticket/166
     17                insert_last( *qptr, *elem );                                    // insert element into buffer
     18                signal( delay );                                                                // restart
     19        } // insert
     20
     21        T * remove( Buffer(T) & mutex buf ) with(buf) {
     22                dlist( T, T ) * qptr = &queue;                                  // workaround https://cforall.uwaterloo.ca/trac/ticket/166
     23                // if ( (*qptr)`is_empty ) wait( delay );                       // no request to process ? => wait
     24          if ( (*qptr)`is_empty ) return 0p;                            // no request to process ? => wait
     25                return &pop_first( *qptr );
     26        } // remove
     27} // forall
     28
     29struct WRequest {                                                                               // client request, no return
     30        void (* action)( void );
     31        DLISTED_MGD_IMPL_IN(WRequest)
    1332}; // WRequest
    1433DLISTED_MGD_IMPL_OUT(WRequest)
    … …  
    1938void doit( WRequest & req ) { req.action(); }
    2039
    21 monitor WRBuffer {                                      // unbounded buffer
    22     dlist( WRequest, WRequest ) queue;                  // unbounded list of work requests
    23     condition delay;
    24 }; // WRBuffer
    25 
    26 void insert( WRBuffer & mutex buf, WRequest * elem ) with(buf) {
    27     insert_last( queue, *elem );                        // insert element into buffer
    28     signal( delay );                                    // restart
    29 } // insert
    30 
    31 WRequest * remove( WRBuffer & mutex buf ) with(buf) {
    32     if ( queue`is_empty ) wait( delay );                // no request to process ? => wait
    33     return & pop_first( queue );
    34 } // remove
    35 
    36 // Each worker has its own work buffer to reduce contention between client and server. Hence, work requests arrive and
    37 // are distributed into buffers in a roughly round-robin order.
     40// Each worker has its own set (when requests buffers > workers) of work buffers to reduce contention between client
     41// and server, where work requests arrive and are distributed into buffers in a roughly round-robin order.
    3842
    3943thread Worker {
    40     WRBuffer * requests;
    41     unsigned int start, range;
     44        Buffer(WRequest) * requests;
     45        WRequest * request;
     46        unsigned int start, range;
    4247}; // Worker
    4348
    4449void main( Worker & w ) with(w) {
    45     for ( int i = 0;; i = (i + 1) % range ) {
    46         WRequest * request = remove( requests[i + start] );
    47       if ( ! request ) { yield(); continue; }
    48       if ( stop( *request ) ) break;
    49         doit( *request );
    50         delete( request );
    51     } // for
     50        for ( int i = 0;; i = (i + 1) % range ) {
     51                request = remove( requests[i + start] );
     52          if ( ! request ) { yield(); continue; }
     53          if ( stop( *request ) ) break;
     54                doit( *request );
     55                delete( request );
     56        } // for
    5257} // Worker::main
    5358
    54 void ?{}( Worker & worker, cluster * wc, WRBuffer * requests, unsigned int start, unsigned int range ) {
    55     (*get_thread(worker)){ *wc };                       // create on given cluster
    56     worker.[requests, start, range] = [requests, start, range];
     59void ?{}( Worker & worker, cluster * wc, Buffer(WRequest) * requests, unsigned int start, unsigned int range ) {
     60        ((thread &)worker){ *wc };
     61        worker.[requests, request, start, range] = [requests, 0p, start, range];
    5762} // ?{}
    5863
     64WRequest * current_request( Worker & worker ) { return worker.request; }
     65
    5966struct Executor {
    60     cluster * cluster;                                  // if workers execute on separate cluster
    61     processor ** processors;                            // array of virtual processors adding parallelism for workers
    62     WRBuffer * requests;                                // list of work requests
    63     Worker ** workers;                                  // array of workers executing work requests
    64     unsigned int nprocessors, nworkers, nmailboxes;     // number of mailboxes/workers/processor tasks
    65     bool sepClus;                                       // use same or separate cluster for executor
     67        cluster * cluster;                                                                      // if workers execute on separate cluster
     68        processor ** processors;                                                        // array of virtual processors adding parallelism for workers
     69        Buffer(WRequest) * requests;                                            // list of work requests
     70        Worker ** workers;                                                                      // array of workers executing work requests
     71        unsigned int nprocessors, nworkers, nrqueues;           // number of processors/threads/request queues
     72        bool sepClus;                                                                           // use same or separate cluster for executor
     73        unsigned int next;                                                                      // demultiplexed across worker buffers
    6674}; // Executor
    6775
    68 static thread_local unsigned int next;                  // demultiplexed across worker buffers
    6976unsigned int tickets( Executor & ex ) with(ex) {
    70     //return uFetchAdd( next, 1 ) % nmailboxes;
    71     return next++ % nmailboxes;                         // no locking, interference randomizes
     77        //return uFetchAdd( next, 1 ) % nrqueues;
     78        return next++ % nrqueues;                                                       // no locking, interference randomizes
    7279} // tickets
    7380
    74 void ?{}( Executor & ex, unsigned int np, unsigned int nw, unsigned int nm, bool sc = false ) with(ex) {
    75     [nprocessors, nworkers, nmailboxes, sepClus] = [np, nw, nm, sc];
    76     assert( nmailboxes >= nworkers );
    77     cluster = sepClus ? new( "Executor" ) : active_cluster();
    78     processors = (processor **)anew( nprocessors );
    79     requests = (WRBuffer *)anew( nmailboxes );
    80     workers = (Worker **)anew( nworkers );
     81void ?{}( Executor & ex, unsigned int np, unsigned int nw, unsigned int nr, bool sc = false ) with(ex) {
     82        [nprocessors, nworkers, nrqueues, sepClus] = [np, nw, nr, sc];
     83        assert( nrqueues >= nworkers );
     84        cluster = sepClus ? new( "Executor" ) : active_cluster();
     85        processors = aalloc( nprocessors );
     86        requests = anew( nrqueues );
     87        workers = aalloc( nworkers );
    8188
    82     for ( i; nprocessors ) {
    83         processors[ i ] = new( *cluster );
    84     } // for
     89        for ( i; nprocessors ) {
     90                processors[i] = new( *cluster );
     91        } // for
    8592
    86     unsigned int reqPerWorker = nmailboxes / nworkers, extras = nmailboxes % nworkers;
    87     for ( unsigned int i = 0, step = 0; i < nworkers; i += 1, step += reqPerWorker + ( i < extras ? 1 : 0 ) ) {
    88         workers[ i ] = new( cluster, requests, step, reqPerWorker + ( i < extras ? 1 : 0 ) );
    89     } // for
     93        unsigned int reqPerWorker = nrqueues / nworkers, extras = nrqueues % nworkers;
     94//      for ( unsigned int i = 0, start = 0, range; i < nworkers; i += 1, start += range ) {
     95    for ( i; nworkers : start; 0u ~ @ ~ range : range; ) {
     96            range = reqPerWorker + ( i < extras ? 1 : 0 );
     97                workers[i] = new( cluster, requests, start, range );
     98        } // for
    9099} // ?{}
    91100
    92101void ?{}( Executor & ex, unsigned int nprocessors, unsigned int nworkers, bool sepClus = false ) {
    93     ex{ nprocessors, nworkers, nworkers, sepClus };
     102        ex{ nprocessors, nworkers, nworkers, sepClus };
    94103}
    95104void ?{}( Executor & ex, unsigned int nprocessors, bool sepClus = false ) {
    96     ex{ nprocessors, nprocessors, nprocessors, sepClus };
     105        ex{ nprocessors, nprocessors, nprocessors, sepClus };
    97106}
    98 void ?{}( Executor & ex ) {                             // special for current cluster
    99     ex{ 0, active_cluster()->nprocessors, false };
     107void ?{}( Executor & ex ) {                                                             // special for current cluster, no processors added
     108        ex{ 0, active_cluster()->nprocessors, false };
    100109}
    101110void ^?{}( Executor & ex ) with(ex) {
    102     // Add one sentinel per worker to stop them. Since in destructor, no new work should be queued.  Cannot combine next
    103     // two loops and only have a single sentinel because workers arrive in arbitrary order, so worker1 may take the
    104     // single sentinel while waiting for worker 0 to end.
     111        // Add one sentinel per worker to stop them. Since in destructor, no new external work should be queued.  Cannot
     112        // combine next two loops and only have a single sentinel because workers arrive in arbitrary order, so worker1 may
     113        // take the single sentinel while waiting for worker 0 to end.
    105114
    106     WRequest sentinel[nworkers];
    107     unsigned int reqPerWorker = nmailboxes / nworkers;
    108     for ( unsigned int i = 0, step = 0; i < nworkers; i += 1, step += reqPerWorker ) {
    109         insert( requests[step], &sentinel[i] );         // force eventually termination
    110     } // for
    111     for ( i; nworkers ) {
    112         delete( workers[ i ] );
    113     } // for
    114     for ( i; nprocessors ) {
    115         delete( processors[ i ] );
    116     } // for
     115        WRequest sentinel[nworkers];
     116        unsigned int reqPerWorker = nrqueues / nworkers;
     117        for ( unsigned int i = 0, step = 0; i < nworkers; i += 1, step += reqPerWorker ) {
     118                insert( requests[step], &sentinel[i] );                 // force eventually termination
     119        } // for
     120        for ( i; nworkers ) {
     121                delete( workers[i] );
     122        } // for
     123        for ( i; nprocessors ) {
     124                delete( processors[i] );
     125        } // for
    117126
    118     delete( workers );
    119     delete( requests );
    120     delete( processors );
    121     if ( sepClus ) { delete( cluster ); }
     127        free( workers );
     128//      adelete( nrqueues, requests );
     129        for ( i; nrqueues ) ^?{}( requests[i] );                        // FIX ME: problem with resolver
     130        free( requests );
     131        free( processors );
     132        if ( sepClus ) { delete( cluster ); }
    122133} // ^?{}
    123134
    124135void send( Executor & ex, void (* action)( void ) ) {   // asynchronous call, no return value
    125     WRequest * node = new( action );
    126     insert( ex.requests[tickets( ex )], node );
     136        WRequest * node = new( action );
     137        insert( ex.requests[tickets( ex )], node );
    127138} // send
     139
    128140
    129141int counter = 0;
    130142
    131 void workie( void ) {
    132     __atomic_add_fetch( &counter, 1, __ATOMIC_SEQ_CST );
    133 //    fprintf( stderr, "workie\n" );
     143void work( void ) {
     144        __atomic_add_fetch( &counter, 1, __ATOMIC_SEQ_CST );
     145        // fprintf( stderr, "workie\n" );
    134146}
    135147
    136 int main() {
    137     {
    138         Executor exector;
    139         for ( i; 3000 ) {
    140             send( exector, workie );
    141             if ( i % 100 == 0 ) {
    142 //              fprintf( stderr, "%d\n", i );
    143                 yield();
    144             }
    145         } // for
    146     }
    147     printf( "%d\n", counter );
     148int main( int argc, char * argv[] ) {
     149        int times = 1_000_000;
     150        if ( argc == 2 ) times = atoi( argv[1] );
     151        processor p[7];
     152        {
     153                Executor exector;
     154                for ( i; times ) {
     155                        send( exector, work );
     156                        if ( i % 100 == 0 ) yield();
     157                } // for
     158        }
     159        printf( "%d\n", counter );
    148160}
    149161
    150162// Local Variables: //
     163// tab-width: 4" //
    151164// compile-command: "cfa executor.cfa" //
    152165// End: //
Note: See TracChangeset for help on using the changeset viewer.