1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/DKU.c	Mon Aug 27 02:14:35 2012 -0700
     1.3 @@ -0,0 +1,853 @@
     1.4 +/*
     1.5 + * Copyright 2010  OpenSourceCodeStewardshipFoundation
     1.6 + *
     1.7 + * Licensed under BSD
     1.8 + */
     1.9 +
    1.10 +#include <stdio.h>
    1.11 +#include <stdlib.h>
    1.12 +#include <malloc.h>
    1.13 +
    1.14 +#include "Queue_impl/PrivateQueue.h"
    1.15 +#include "Hash_impl/PrivateHash.h"
    1.16 +
    1.17 +#include "VSs.h"
    1.18 +#include "Measurement/VSs_Counter_Recording.h"
    1.19 +
    1.20 +//==========================================================================
    1.21 +
    1.22 +void
    1.23 +VSs__init();
    1.24 +
    1.25 +void
    1.26 +VSs__init_Helper();
    1.27 +//==========================================================================
    1.28 +
    1.29 +
    1.30 +
    1.31 +//===========================================================================
    1.32 +
    1.33 +
    1.34 +/*These are the library functions *called in the application*
    1.35 + * 
    1.36 + *There's a pattern for the outside sequential code to interact with the
    1.37 + * VMS_HW code.
    1.38 + *The VMS_HW system is inside a boundary..  every VSs system is in its
    1.39 + * own directory that contains the functions for each of the processor types.
    1.40 + * One of the processor types is the "seed" processor that starts the
    1.41 + * cascade of creating all the processors that do the work.
    1.42 + *So, in the directory is a file called "EntryPoint.c" that contains the
    1.43 + * function, named appropriately to the work performed, that the outside
    1.44 + * sequential code calls.  This function follows a pattern:
    1.45 + *1) it calls VSs__init()
    1.46 + *2) it creates the initial data for the seed processor, which is passed
    1.47 + *    in to the function
    1.48 + *3) it creates the seed VSs processor, with the data to start it with.
    1.49 + *4) it calls startVSsThenWaitUntilWorkDone
    1.50 + *5) it gets the returnValue from the transfer struc and returns that
    1.51 + *    from the function
    1.52 + *
    1.53 + *For now, a new VSs system has to be created via VSs__init every
    1.54 + * time an entry point function is called -- later, might add letting the
    1.55 + * VSs system be created once, and let all the entry points just reuse
    1.56 + * it -- want to be as simple as possible now, and see by using what makes
    1.57 + * sense for later..
    1.58 + */
    1.59 +
    1.60 +
    1.61 +
    1.62 +//===========================================================================
    1.63 +
    1.64 +/*This is the "border crossing" function -- the thing that crosses from the
    1.65 + * outside world, into the VMS_HW world.  It initializes and starts up the
    1.66 + * VMS system, then creates one processor from the specified function and
    1.67 + * puts it into the readyQ.  From that point, that one function is resp.
    1.68 + * for creating all the other processors, that then create others, and so
    1.69 + * forth.
    1.70 + *When all the processors, including the seed, have dissipated, then this
    1.71 + * function returns.  The results will have been written by side-effect via
    1.72 + * pointers read from, or written into initData.
    1.73 + *
    1.74 + *NOTE: no Threads should exist in the outside program that might touch
    1.75 + * any of the data reachable from initData passed in to here
    1.76 + */
    1.77 +void
    1.78 +VSs__create_seed_slave_and_do_work( TopLevelFnPtr fnPtr, void *initData )
    1.79 + { VSsSemEnv   *semEnv;
    1.80 +   SlaveVP     *seedSlv;
    1.81 +   VSsSemData  *semData;
    1.82 +   VSsTaskStub *threadTaskStub, *parentTaskStub;
    1.83 +
    1.84 +   VSs__init();      //normal multi-thd
    1.85 +   
    1.86 +   semEnv = _VMSMasterEnv->semanticEnv;
    1.87 +
    1.88 +      //VSs starts with one processor, which is put into initial environ,
    1.89 +      // and which then calls create() to create more, thereby expanding work
    1.90 +   seedSlv = VSs__create_slave_helper( fnPtr, initData,
    1.91 +                                     semEnv, semEnv->nextCoreToGetNewSlv++ );
    1.92 +   
    1.93 +      //seed slave is a thread slave, so make a thread's task stub for it
    1.94 +      // and then make another to stand for the seed's parent task.  Make
    1.95 +      // the parent be already ended, and have one child (the seed).  This
    1.96 +      // will make the dissipate handler do the right thing when the seed
    1.97 +      // is dissipated.
    1.98 +   threadTaskStub = create_thread_task_stub( initData );
    1.99 +   parentTaskStub = create_thread_task_stub( NULL );
   1.100 +   parentTaskStub->isEnded = TRUE;
   1.101 +   parentTaskStub->numLiveChildThreads = 1; //so dissipate works for seed
   1.102 +   threadTaskStub->parentTaskStub = parentTaskStub;
   1.103 +   
   1.104 +   semData = (VSsSemData *)seedSlv->semanticData;
   1.105 +      //seedVP is a thread, so has a permanent task
   1.106 +   semData->needsTaskAssigned = FALSE;
   1.107 +   semData->taskStub = threadTaskStub;
   1.108 +   semData->slaveType = ThreadSlv;
   1.109 +
   1.110 +   resume_slaveVP( seedSlv, semEnv ); //returns right away, just queues Slv
   1.111 +   
   1.112 +   VMS_SS__start_the_work_then_wait_until_done();      //normal multi-thd
   1.113 +
   1.114 +   VSs__cleanup_after_shutdown();
   1.115 + }
   1.116 +
   1.117 +
   1.118 +int32
   1.119 +VSs__giveMinWorkUnitCycles( float32 percentOverhead )
   1.120 + {
   1.121 +   return MIN_WORK_UNIT_CYCLES;
   1.122 + }
   1.123 +
   1.124 +int32
   1.125 +VSs__giveIdealNumWorkUnits()
   1.126 + {
   1.127 +   return NUM_ANIM_SLOTS * NUM_CORES;
   1.128 + }
   1.129 +
   1.130 +int32
   1.131 +VSs__give_number_of_cores_to_schedule_onto()
   1.132 + {
   1.133 +   return NUM_CORES;
   1.134 + }
   1.135 +
   1.136 +/*For now, use TSC -- later, make these two macros with assembly that first
   1.137 + * saves jump point, and second jumps back several times to get reliable time
   1.138 + */
   1.139 +void
   1.140 +VSs__start_primitive()
   1.141 + { saveLowTimeStampCountInto( ((VSsSemEnv *)(_VMSMasterEnv->semanticEnv))->
   1.142 +                              primitiveStartTime );
   1.143 + }
   1.144 +
   1.145 +/*Just quick and dirty for now -- make reliable later
   1.146 + * will want this to jump back several times -- to be sure cache is warm
   1.147 + * because don't want comm time included in calc-time measurement -- and
   1.148 + * also to throw out any "weird" values due to OS interrupt or TSC rollover
   1.149 + */
   1.150 +int32
   1.151 +VSs__end_primitive_and_give_cycles()
   1.152 + { int32 endTime, startTime;
   1.153 +   //TODO: fix by repeating time-measurement
   1.154 +   saveLowTimeStampCountInto( endTime );
   1.155 +   startTime =((VSsSemEnv*)(_VMSMasterEnv->semanticEnv))->primitiveStartTime;
   1.156 +   return (endTime - startTime);
   1.157 + }
   1.158 +
   1.159 +//===========================================================================
   1.160 +
   1.161 +/*Initializes all the data-structures for a VSs system -- but doesn't
   1.162 + * start it running yet!
   1.163 + *
   1.164 + *This runs in the main thread -- before VMS starts up
   1.165 + * 
   1.166 + *This sets up the semantic layer over the VMS system
   1.167 + *
   1.168 + *First, calls VMS_Setup, then creates own environment, making it ready
   1.169 + * for creating the seed processor and then starting the work.
   1.170 + */
   1.171 +void
   1.172 +VSs__init()
   1.173 + {
   1.174 +   VMS_SS__init();
   1.175 +      //masterEnv, a global var, now is partially set up by init_VMS
   1.176 +      // after this, have VMS_int__malloc and VMS_int__free available
   1.177 +
   1.178 +   VSs__init_Helper();
   1.179 + }
   1.180 +
   1.181 +
   1.182 +void idle_fn(void* data, SlaveVP *animatingSlv){
   1.183 +    while(1){
   1.184 +        VMS_int__suspend_slaveVP_and_send_req(animatingSlv);
   1.185 +    }
   1.186 +}
   1.187 +
   1.188 +void
   1.189 +VSs__init_Helper()
   1.190 + { VSsSemEnv       *semanticEnv;
   1.191 +   int32            i, coreNum, slotNum;
   1.192 +   VSsSemData      *semData;
   1.193 + 
   1.194 +      //Hook up the semantic layer's plug-ins to the Master virt procr
   1.195 +   _VMSMasterEnv->requestHandler = &VSs__Request_Handler;
   1.196 +   _VMSMasterEnv->slaveAssigner  = &VSs__assign_slaveVP_to_slot;
   1.197 +
   1.198 +      //create the semantic layer's environment (all its data) and add to
   1.199 +      // the master environment
   1.200 +   semanticEnv = VMS_int__malloc( sizeof( VSsSemEnv ) );
   1.201 +   _VMSMasterEnv->semanticEnv = semanticEnv;
   1.202 +   
   1.203 +   #ifdef HOLISTIC__TURN_ON_PERF_COUNTERS
   1.204 +   _VMSMasterEnv->counterHandler = &VSs__counter_handler;
   1.205 +   VSs__init_counter_data_structs();
   1.206 +   #endif
   1.207 +
   1.208 +   semanticEnv->shutdownInitiated = FALSE;
   1.209 +   semanticEnv->coreIsDone = VMS_int__malloc( NUM_CORES * sizeof( bool32 ) );
   1.210 +      //For each animation slot, there is an idle slave, and an initial
   1.211 +      // slave assigned as the current-task-slave.  Create them here.
   1.212 +   SlaveVP *idleSlv, *slotTaskSlv;
   1.213 +   for( coreNum = 0; coreNum < NUM_CORES; coreNum++ )
   1.214 +    { semanticEnv->coreIsDone[coreNum] = FALSE; //use during shutdown
   1.215 +    
   1.216 +      for( slotNum = 0; slotNum < NUM_ANIM_SLOTS; ++slotNum )
   1.217 +       { idleSlv = VSs__create_slave_helper( &idle_fn, NULL, semanticEnv, 0);
   1.218 +         idleSlv->coreAnimatedBy                = coreNum;
   1.219 +         idleSlv->animSlotAssignedTo            =
   1.220 +                               _VMSMasterEnv->allAnimSlots[coreNum][slotNum];
   1.221 +         semanticEnv->idleSlv[coreNum][slotNum] = idleSlv;
   1.222 +         
   1.223 +         slotTaskSlv = VSs__create_slave_helper( &idle_fn, NULL, semanticEnv, 0);
   1.224 +         slotTaskSlv->coreAnimatedBy            = coreNum;
   1.225 +         slotTaskSlv->animSlotAssignedTo        = 
   1.226 +                               _VMSMasterEnv->allAnimSlots[coreNum][slotNum];
   1.227 +         
   1.228 +         semData                    = slotTaskSlv->semanticData;
   1.229 +         semData->needsTaskAssigned = TRUE;
   1.230 +         semData->slaveType         = SlotTaskSlv;
   1.231 +         semanticEnv->slotTaskSlvs[coreNum][slotNum] = slotTaskSlv;
   1.232 +       }
   1.233 +    }
   1.234 +
   1.235 +      //create the ready queues, hash tables used for matching and so forth
   1.236 +   semanticEnv->slavesReadyToResumeQ = makeVMSQ();
   1.237 +   semanticEnv->freeExtraTaskSlvQ    = makeVMSQ();
   1.238 +   semanticEnv->taskReadyQ           = makeVMSQ();
   1.239 +   
   1.240 +   semanticEnv->argPtrHashTbl  = makeHashTable32( 16, &VMS_int__free );
   1.241 +   semanticEnv->commHashTbl    = makeHashTable32( 16, &VMS_int__free );
   1.242 +   
   1.243 +   semanticEnv->nextCoreToGetNewSlv = 0;
   1.244 +   
   1.245 +
   1.246 +   //TODO: bug -- turn these arrays into dyn arrays to eliminate limit
   1.247 +   //semanticEnv->singletonHasBeenExecutedFlags = makeDynArrayInfo( );
   1.248 +   //semanticEnv->transactionStrucs = makeDynArrayInfo( );
   1.249 +   for( i = 0; i < NUM_STRUCS_IN_SEM_ENV; i++ )
   1.250 +    {
   1.251 +      semanticEnv->fnSingletons[i].endInstrAddr      = NULL;
   1.252 +      semanticEnv->fnSingletons[i].hasBeenStarted    = FALSE;
   1.253 +      semanticEnv->fnSingletons[i].hasFinished       = FALSE;
   1.254 +      semanticEnv->fnSingletons[i].waitQ             = makeVMSQ();
   1.255 +      semanticEnv->transactionStrucs[i].waitingVPQ   = makeVMSQ();
   1.256 +    }
   1.257 +
   1.258 +   semanticEnv->numLiveExtraTaskSlvs   = 0; //must be last
   1.259 +   semanticEnv->numLiveThreadSlvs      = 1; //must be last, counts the seed
   1.260 +
   1.261 +   #ifdef HOLISTIC__TURN_ON_OBSERVE_UCC
   1.262 +   semanticEnv->unitList = makeListOfArrays(sizeof(Unit),128);
   1.263 +   semanticEnv->ctlDependenciesList = makeListOfArrays(sizeof(Dependency),128);
   1.264 +   semanticEnv->commDependenciesList = makeListOfArrays(sizeof(Dependency),128);
   1.265 +   semanticEnv->dynDependenciesList = makeListOfArrays(sizeof(Dependency),128);
   1.266 +   semanticEnv->ntonGroupsInfo = makePrivDynArrayOfSize((void***)&(semanticEnv->ntonGroups),8);
   1.267 +   
   1.268 +   semanticEnv->hwArcs = makeListOfArrays(sizeof(Dependency),128);
   1.269 +   memset(semanticEnv->last_in_slot,0,sizeof(NUM_CORES * NUM_ANIM_SLOTS * sizeof(Unit)));
   1.270 +   #endif
   1.271 + }
   1.272 +
   1.273 +
   1.274 +/*Frees any memory allocated by VSs__init() then calls VMS_int__shutdown
   1.275 + */
   1.276 +void
   1.277 +VSs__cleanup_after_shutdown()
   1.278 + { VSsSemEnv *semanticEnv;
   1.279 +   
   1.280 +   semanticEnv = _VMSMasterEnv->semanticEnv;
   1.281 +
   1.282 +   #ifdef HOLISTIC__TURN_ON_OBSERVE_UCC
   1.283 +   //UCC
   1.284 +   FILE* output;
   1.285 +   int n;
   1.286 +   char filename[255];    
   1.287 +    for(n=0;n<255;n++)
   1.288 +    {
   1.289 +        sprintf(filename, "./counters/UCC.%d",n);
   1.290 +        output = fopen(filename,"r");
   1.291 +        if(output)
   1.292 +        {
   1.293 +            fclose(output);
   1.294 +        }else{
   1.295 +            break;
   1.296 +        }
   1.297 +    }
   1.298 +   if(n<255){
   1.299 +    printf("Saving UCC to File: %s ...\n", filename);
   1.300 +    output = fopen(filename,"w+");
   1.301 +    if(output!=NULL){
   1.302 +        set_dependency_file(output);
   1.303 +        //fprintf(output,"digraph Dependencies {\n");
   1.304 +        //set_dot_file(output);
   1.305 +        //FIXME:  first line still depends on counters being enabled, replace w/ unit struct!
   1.306 +        //forAllInDynArrayDo(_VMSMasterEnv->counter_history_array_info, &print_dot_node_info );
   1.307 +        forAllInListOfArraysDo(semanticEnv->unitList, &print_unit_to_file);
   1.308 +        forAllInListOfArraysDo( semanticEnv->commDependenciesList, &print_comm_dependency_to_file );
   1.309 +        forAllInListOfArraysDo( semanticEnv->ctlDependenciesList, &print_ctl_dependency_to_file );
   1.310 +        forAllInDynArrayDo(semanticEnv->ntonGroupsInfo,&print_nton_to_file);
   1.311 +        //fprintf(output,"}\n");
   1.312 +        fflush(output);
   1.313 +
   1.314 +    } else
   1.315 +        printf("Opening UCC file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
   1.316 +   } else {
   1.317 +       printf("Could not open UCC file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
   1.318 +   }
   1.319 +   //Loop Graph
   1.320 +   for(n=0;n<255;n++)
   1.321 +    {
   1.322 +        sprintf(filename, "./counters/LoopGraph.%d",n);
   1.323 +        output = fopen(filename,"r");
   1.324 +        if(output)
   1.325 +        {
   1.326 +            fclose(output);
   1.327 +        }else{
   1.328 +            break;
   1.329 +        }
   1.330 +    }
   1.331 +   if(n<255){
   1.332 +    printf("Saving LoopGraph to File: %s ...\n", filename);
   1.333 +    output = fopen(filename,"w+");
   1.334 +    if(output!=NULL){
   1.335 +        set_dependency_file(output);
   1.336 +        //fprintf(output,"digraph Dependencies {\n");
   1.337 +        //set_dot_file(output);
   1.338 +        //FIXME:  first line still depends on counters being enabled, replace w/ unit struct!
   1.339 +        //forAllInDynArrayDo(_VMSMasterEnv->counter_history_array_info, &print_dot_node_info );
   1.340 +        forAllInListOfArraysDo( semanticEnv->unitList, &print_unit_to_file );
   1.341 +        forAllInListOfArraysDo( semanticEnv->commDependenciesList, &print_comm_dependency_to_file );
   1.342 +        forAllInListOfArraysDo( semanticEnv->ctlDependenciesList, &print_ctl_dependency_to_file );
   1.343 +        forAllInListOfArraysDo( semanticEnv->dynDependenciesList, &print_dyn_dependency_to_file );
   1.344 +        forAllInListOfArraysDo( semanticEnv->hwArcs, &print_hw_dependency_to_file );
   1.345 +        //fprintf(output,"}\n");
   1.346 +        fflush(output);
   1.347 +
   1.348 +    } else
   1.349 +        printf("Opening LoopGraph file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
   1.350 +   } else {
   1.351 +       printf("Could not open LoopGraph file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
   1.352 +   }
   1.353 +   
   1.354 +   
   1.355 +   freeListOfArrays(semanticEnv->unitList);
   1.356 +   freeListOfArrays(semanticEnv->commDependenciesList);
   1.357 +   freeListOfArrays(semanticEnv->ctlDependenciesList);
   1.358 +   freeListOfArrays(semanticEnv->dynDependenciesList);
   1.359 +   
   1.360 +   #endif
   1.361 +#ifdef HOLISTIC__TURN_ON_PERF_COUNTERS    
   1.362 +    for(n=0;n<255;n++)
   1.363 +    {
   1.364 +        sprintf(filename, "./counters/Counters.%d.csv",n);
   1.365 +        output = fopen(filename,"r");
   1.366 +        if(output)
   1.367 +        {
   1.368 +            fclose(output);
   1.369 +        }else{
   1.370 +            break;
   1.371 +        }
   1.372 +    }
   1.373 +    if(n<255){
   1.374 +    printf("Saving Counter measurements to File: %s ...\n", filename);
   1.375 +    output = fopen(filename,"w+");
   1.376 +    if(output!=NULL){
   1.377 +        set_counter_file(output);
   1.378 +        int i;
   1.379 +        for(i=0;i<NUM_CORES;i++){
   1.380 +            forAllInListOfArraysDo( semanticEnv->counterList[i], &print_counter_events_to_file );
   1.381 +            fflush(output);
   1.382 +        }
   1.383 +
   1.384 +    } else
   1.385 +        printf("Opening UCC file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
   1.386 +   } else {
   1.387 +       printf("Could not open UCC file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
   1.388 +   }
   1.389 +    
   1.390 +#endif
   1.391 +/* It's all allocated inside VMS's big chunk -- that's about to be freed, so
   1.392 + *  nothing to do here
   1.393 +   
   1.394 +
   1.395 +   for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
   1.396 +    {
   1.397 +      VMS_int__free( semanticEnv->readyVPQs[coreIdx]->startOfData );
   1.398 +      VMS_int__free( semanticEnv->readyVPQs[coreIdx] );
   1.399 +    }
   1.400 +   VMS_int__free( semanticEnv->readyVPQs );
   1.401 +   
   1.402 +   freeHashTable( semanticEnv->commHashTbl );
   1.403 +   VMS_int__free( _VMSMasterEnv->semanticEnv );
   1.404 + */
   1.405 +   VMS_SS__cleanup_at_end_of_shutdown();
   1.406 + }
   1.407 +
   1.408 +
   1.409 +//===========================================================================
   1.410 +
   1.411 +SlaveVP *
   1.412 +VSs__create_thread( TopLevelFnPtr fnPtr,   void *initData,
   1.413 +                        SlaveVP *creatingThd )
   1.414 + { VSsSemReq reqData;
   1.415 +
   1.416 +      //the semantic request data is on the stack and disappears when this
   1.417 +      // call returns -- it's guaranteed to remain in the VP's stack for as
   1.418 +      // long as the VP is suspended.
   1.419 +   reqData.reqType            = 0; //know type because in a VMS create req
   1.420 +   reqData.fnPtr              = fnPtr;
   1.421 +   reqData.initData           = initData;
   1.422 +   reqData.callingSlv         = creatingThd;
   1.423 +
   1.424 +   VMS_WL__send_create_slaveVP_req( &reqData, creatingThd );
   1.425 +
   1.426 +   return creatingThd->dataRetFromReq;
   1.427 + }
   1.428 +
   1.429 +/*This is always the last thing done in the code animated by a thread VP.
   1.430 + * Normally, this would be the last line of the thread's top level function.
   1.431 + * But, if the thread exits from any point, it has to do so by calling
   1.432 + * this.
   1.433 + *
   1.434 + *It simply sends a dissipate request, which handles all the state cleanup.
   1.435 + */
   1.436 +void
   1.437 +VSs__end_thread( SlaveVP *thdToEnd )
   1.438 + { VSsSemData *semData;
   1.439 +   
   1.440 +   VMS_WL__send_dissipate_req( thdToEnd );
   1.441 + }
   1.442 +
   1.443 +
   1.444 +
   1.445 +//===========================================================================
   1.446 +
   1.447 +
   1.448 +//======================= task submit and end ==============================
   1.449 +/*
   1.450 + */
   1.451 +void
   1.452 +VSs__submit_task( VSsTaskType *taskType, void *args, SlaveVP *animSlv)
   1.453 + { VSsSemReq  reqData;
   1.454 +
   1.455 +   reqData.reqType    = submit_task;
   1.456 +   
   1.457 +   reqData.taskType   = taskType;
   1.458 +   reqData.args       = args;
   1.459 +   reqData.callingSlv = animSlv;
   1.460 +  
   1.461 +   reqData.taskID     = NULL;
   1.462 + 
   1.463 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.464 + }
   1.465 +
   1.466 +inline int32 *
   1.467 +VSs__create_taskID_of_size( int32 numInts, SlaveVP *animSlv )
   1.468 + { int32 *taskID;
   1.469 +   
   1.470 +   taskID    = VMS_WL__malloc( sizeof(int32) + numInts * sizeof(int32) );
   1.471 +   taskID[0] = numInts;
   1.472 +   return taskID;
   1.473 + }
   1.474 +
   1.475 +void
   1.476 +VSs__submit_task_with_ID( VSsTaskType *taskType, void *args, int32 *taskID, 
   1.477 +                          SlaveVP     *animSlv)
   1.478 + { VSsSemReq  reqData;
   1.479 +
   1.480 +   reqData.reqType    = submit_task;
   1.481 +   
   1.482 +   reqData.taskType   = taskType;
   1.483 +   reqData.args       = args;
   1.484 +   reqData.taskID     = taskID;
   1.485 +   reqData.callingSlv = animSlv;
   1.486 + 
   1.487 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.488 + }
   1.489 +
   1.490 +
   1.491 +/*This call is the last to happen in every task.  It causes the slave to
   1.492 + * suspend and get the next task out of the task-queue.  Notice there is no
   1.493 + * assigner here.. only one slave, no slave ReadyQ, and so on..
   1.494 + *Can either make the assigner take the next task out of the taskQ, or can
   1.495 + * leave all as it is, and make task-end take the next task.
   1.496 + *Note: this fits the case in the new VMS for no-context tasks, so will use
   1.497 + * the built-in taskQ of new VMS, and should be local and much faster.
   1.498 + * 
   1.499 + *The task-stub is saved in the animSlv, so the request handler will get it
   1.500 + * from there, along with the task-type which has arg types, and so on..
   1.501 + * 
   1.502 + * NOTE: if want, don't need to send the animating SlaveVP around.. 
   1.503 + * instead, can make a single slave per core, and coreCtrlr looks up the
   1.504 + * slave from having the core number.
   1.505 + * 
   1.506 + *But, to stay compatible with all the other VMS languages, leave it in..
   1.507 + */
   1.508 +void
   1.509 +VSs__end_task( SlaveVP *animSlv )
   1.510 + { VSsSemReq  reqData;
   1.511 +
   1.512 +   reqData.reqType      = end_task;
   1.513 +   reqData.callingSlv   = animSlv;
   1.514 +   
   1.515 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.516 + }
   1.517 +
   1.518 +
   1.519 +void
   1.520 +VSs__taskwait(SlaveVP *animSlv)
   1.521 +{
   1.522 +    VSsSemReq  reqData;
   1.523 +
   1.524 +   reqData.reqType      = taskwait;
   1.525 +   reqData.callingSlv   = animSlv;
   1.526 +   
   1.527 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.528 +}
   1.529 +
   1.530 +
   1.531 +
   1.532 +//==========================  send and receive ============================
   1.533 +//
   1.534 +
   1.535 +inline int32 *
   1.536 +VSs__give_self_taskID( SlaveVP *animSlv )
   1.537 + {
   1.538 +   return ((VSsSemData*)animSlv->semanticData)->taskStub->taskID;
   1.539 + }
   1.540 +
   1.541 +//================================ send ===================================
   1.542 +
   1.543 +void
   1.544 +VSs__send_of_type_to( void *msg, const int32 type, int32 *receiverID,
   1.545 +                      SlaveVP *senderSlv )
   1.546 + { VSsSemReq  reqData;
   1.547 +
   1.548 +   reqData.reqType    = send_type_to;
   1.549 +   
   1.550 +   reqData.msg        = msg;
   1.551 +   reqData.msgType    = type;
   1.552 +   reqData.receiverID = receiverID;
   1.553 +   reqData.senderSlv  = senderSlv;
   1.554 +   
   1.555 +   reqData.nextReqInHashEntry = NULL;
   1.556 +
   1.557 +   VMS_WL__send_sem_request( &reqData, senderSlv );
   1.558 +
   1.559 +      //When come back from suspend, no longer own data reachable from msg
   1.560 + }
   1.561 +
   1.562 +void
   1.563 +VSs__send_from_to( void *msg, int32 *senderID, int32 *receiverID, SlaveVP *senderSlv )
   1.564 + { VSsSemReq  reqData;
   1.565 +
   1.566 +   reqData.reqType     = send_from_to;
   1.567 +   
   1.568 +   reqData.msg         = msg;
   1.569 +   reqData.senderID    = senderID;
   1.570 +   reqData.receiverID  = receiverID;
   1.571 +   reqData.senderSlv   = senderSlv;
   1.572 +
   1.573 +   reqData.nextReqInHashEntry = NULL;
   1.574 +
   1.575 +   VMS_WL__send_sem_request( &reqData, senderSlv );
   1.576 + }
   1.577 +
   1.578 +
   1.579 +//================================ receive ================================
   1.580 +
   1.581 +/*The "type" version of send and receive creates a many-to-one relationship.
   1.582 + * The sender is anonymous, and many sends can stack up, waiting to be
   1.583 + * received.  The same receiver can also have send from-to's
   1.584 + * waiting for it, and those will be kept separate from the "type"
   1.585 + * messages.
   1.586 + */
   1.587 +void *
   1.588 +VSs__receive_type_to( const int32 type, int32* receiverID, SlaveVP *receiverSlv )
   1.589 + {       DEBUG__printf1(dbgRqstHdlr,"WL: receive type to %d",receiverID[1] );
   1.590 +   VSsSemReq  reqData;
   1.591 +
   1.592 +   reqData.reqType     = receive_type_to;
   1.593 +   
   1.594 +   reqData.msgType     = type;
   1.595 +   reqData.receiverID  = receiverID;
   1.596 +   reqData.receiverSlv = receiverSlv;
   1.597 +   
   1.598 +   reqData.nextReqInHashEntry = NULL;
   1.599 +
   1.600 +   VMS_WL__send_sem_request( &reqData, receiverSlv );
   1.601 +   
   1.602 +   return receiverSlv->dataRetFromReq;
   1.603 + }
   1.604 +
   1.605 +
   1.606 +
   1.607 +/*Call this at the point a receiving task wants in-coming data.
   1.608 + * Use this from-to form when know senderID -- it makes a direct channel
   1.609 + * between sender and receiver.
   1.610 + */
   1.611 +void *
   1.612 +VSs__receive_from_to( int32 *senderID, int32 *receiverID, SlaveVP *receiverSlv )
   1.613 + { 
   1.614 +   VSsSemReq  reqData;
   1.615 +
   1.616 +   reqData.reqType     = receive_from_to;
   1.617 +
   1.618 +   reqData.senderID    = senderID;
   1.619 +   reqData.receiverID  = receiverID;
   1.620 +   reqData.receiverSlv = receiverSlv;
   1.621 +
   1.622 +   reqData.nextReqInHashEntry = NULL;
   1.623 +      DEBUG__printf2(dbgRqstHdlr,"WL: receive from %d to: %d", reqData.senderID[1], reqData.receiverID[1]);
   1.624 +      
   1.625 +   VMS_WL__send_sem_request( &reqData, receiverSlv );
   1.626 +
   1.627 +   return receiverSlv->dataRetFromReq;
   1.628 + }
   1.629 +
   1.630 +
   1.631 +
   1.632 +
   1.633 +//==========================================================================
   1.634 +//
   1.635 +/*A function singleton is a function whose body executes exactly once, on a
   1.636 + * single core, no matter how many times the fuction is called and no
   1.637 + * matter how many cores or the timing of cores calling it.
   1.638 + *
   1.639 + *A data singleton is a ticket attached to data.  That ticket can be used
   1.640 + * to get the data through the function exactly once, no matter how many
   1.641 + * times the data is given to the function, and no matter the timing of
   1.642 + * trying to get the data through from different cores.
   1.643 + */
   1.644 +
   1.645 +/*asm function declarations*/
   1.646 +void asm_save_ret_to_singleton(VSsSingleton *singletonPtrAddr);
   1.647 +void asm_write_ret_from_singleton(VSsSingleton *singletonPtrAddr);
   1.648 +
   1.649 +/*Fn singleton uses ID as index into array of singleton structs held in the
   1.650 + * semantic environment.
   1.651 + */
   1.652 +void
   1.653 +VSs__start_fn_singleton( int32 singletonID,   SlaveVP *animSlv )
   1.654 + {
   1.655 +   VSsSemReq  reqData;
   1.656 +
   1.657 +      //
   1.658 +   reqData.reqType     = singleton_fn_start;
   1.659 +   reqData.singletonID = singletonID;
   1.660 +
   1.661 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.662 +   if( animSlv->dataRetFromReq ) //will be 0 or addr of label in end singleton
   1.663 +    {
   1.664 +       VSsSemEnv *semEnv = VMS_int__give_sem_env_for( animSlv );
   1.665 +       asm_write_ret_from_singleton(&(semEnv->fnSingletons[ singletonID]));
   1.666 +    }
   1.667 + }
   1.668 +
   1.669 +/*Data singleton hands addr of loc holding a pointer to a singleton struct.
   1.670 + * The start_data_singleton makes the structure and puts its addr into the
   1.671 + * location.
   1.672 + */
   1.673 +void
   1.674 +VSs__start_data_singleton( VSsSingleton **singletonAddr,  SlaveVP *animSlv )
   1.675 + {
   1.676 +   VSsSemReq  reqData;
   1.677 +
   1.678 +   if( *singletonAddr && (*singletonAddr)->hasFinished )
   1.679 +       goto JmpToEndSingleton;
   1.680 +   
   1.681 +   reqData.reqType          = singleton_data_start;
   1.682 +   reqData.singletonPtrAddr = singletonAddr;
   1.683 +
   1.684 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.685 +   if( animSlv->dataRetFromReq ) //either 0 or end singleton's return addr
   1.686 +    {    //Assembly code changes the return addr on the stack to the one
   1.687 +         // saved into the singleton by the end-singleton-fn
   1.688 +         //The return addr is at 0x4(%%ebp)
   1.689 +        JmpToEndSingleton:
   1.690 +          asm_write_ret_from_singleton(*singletonAddr);
   1.691 +    }
   1.692 +   //now, simply return
   1.693 +   //will exit either from the start singleton call or the end-singleton call
   1.694 + }
   1.695 +
   1.696 +/*Uses ID as index into array of flags.  If flag already set, resumes from
   1.697 + * end-label.  Else, sets flag and resumes normally.
   1.698 + *
   1.699 + *Note, this call cannot be inlined because the instr addr at the label
   1.700 + * inside is shared by all invocations of a given singleton ID.
   1.701 + */
   1.702 +void
   1.703 +VSs__end_fn_singleton( int32 singletonID, SlaveVP *animSlv )
   1.704 + {
   1.705 +   VSsSemReq  reqData;
   1.706 +
   1.707 +      //don't need this addr until after at least one singleton has reached
   1.708 +      // this function
   1.709 +   VSsSemEnv *semEnv = VMS_int__give_sem_env_for( animSlv );
   1.710 +   asm_write_ret_from_singleton(&(semEnv->fnSingletons[ singletonID]));
   1.711 +
   1.712 +   reqData.reqType     = singleton_fn_end;
   1.713 +   reqData.singletonID = singletonID;
   1.714 +
   1.715 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.716 +
   1.717 +EndSingletonInstrAddr:
   1.718 +   return;
   1.719 + }
   1.720 +
   1.721 +void
   1.722 +VSs__end_data_singleton(  VSsSingleton **singletonPtrAddr, SlaveVP *animSlv )
   1.723 + {
   1.724 +   VSsSemReq  reqData;
   1.725 +
   1.726 +      //don't need this addr until after singleton struct has reached
   1.727 +      // this function for first time
   1.728 +      //do assembly that saves the return addr of this fn call into the
   1.729 +      // data singleton -- that data-singleton can only be given to exactly
   1.730 +      // one instance in the code of this function.  However, can use this
   1.731 +      // function in different places for different data-singletons.
   1.732 +//   (*(singletonAddr))->endInstrAddr =  &&EndDataSingletonInstrAddr;
   1.733 +
   1.734 +
   1.735 +   asm_save_ret_to_singleton(*singletonPtrAddr);
   1.736 +
   1.737 +   reqData.reqType          = singleton_data_end;
   1.738 +   reqData.singletonPtrAddr = singletonPtrAddr;
   1.739 +
   1.740 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.741 + }
   1.742 +
   1.743 +/*This executes the function in the masterVP, so it executes in isolation
   1.744 + * from any other copies -- only one copy of the function can ever execute
   1.745 + * at a time.
   1.746 + *
   1.747 + *It suspends to the master, and the request handler takes the function
   1.748 + * pointer out of the request and calls it, then resumes the VP.
   1.749 + *Only very short functions should be called this way -- for longer-running
   1.750 + * isolation, use transaction-start and transaction-end, which run the code
   1.751 + * between as work-code.
   1.752 + */
   1.753 +void
   1.754 +VSs__animate_short_fn_in_isolation( PtrToAtomicFn ptrToFnToExecInMaster,
   1.755 +                                    void *data, SlaveVP *animSlv )
   1.756 + {
   1.757 +   VSsSemReq  reqData;
   1.758 +
   1.759 +      //
   1.760 +   reqData.reqType          = atomic;
   1.761 +   reqData.fnToExecInMaster = ptrToFnToExecInMaster;
   1.762 +   reqData.dataForFn        = data;
   1.763 +
   1.764 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.765 + }
   1.766 +
   1.767 +
   1.768 +/*This suspends to the master.
   1.769 + *First, it looks at the VP's data, to see the highest transactionID that VP
   1.770 + * already has entered.  If the current ID is not larger, it throws an
   1.771 + * exception stating a bug in the code.  Otherwise it puts the current ID
   1.772 + * there, and adds the ID to a linked list of IDs entered -- the list is
   1.773 + * used to check that exits are properly ordered.
   1.774 + *Next it is uses transactionID as index into an array of transaction
   1.775 + * structures.
   1.776 + *If the "VP_currently_executing" field is non-null, then put requesting VP
   1.777 + * into queue in the struct.  (At some point a holder will request
   1.778 + * end-transaction, which will take this VP from the queue and resume it.)
   1.779 + *If NULL, then write requesting into the field and resume.
   1.780 + */
   1.781 +void
   1.782 +VSs__start_transaction( int32 transactionID, SlaveVP *animSlv )
   1.783 + {
   1.784 +   VSsSemReq  reqData;
   1.785 +
   1.786 +      //
   1.787 +   reqData.callingSlv      = animSlv;
   1.788 +   reqData.reqType     = trans_start;
   1.789 +   reqData.transID     = transactionID;
   1.790 +
   1.791 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.792 + }
   1.793 +
   1.794 +/*This suspends to the master, then uses transactionID as index into an
   1.795 + * array of transaction structures.
   1.796 + *It looks at VP_currently_executing to be sure it's same as requesting VP.
   1.797 + * If different, throws an exception, stating there's a bug in the code.
   1.798 + *Next it looks at the queue in the structure.
   1.799 + *If it's empty, it sets VP_currently_executing field to NULL and resumes.
   1.800 + *If something in, gets it, sets VP_currently_executing to that VP, then
   1.801 + * resumes both.
   1.802 + */
   1.803 +void
   1.804 +VSs__end_transaction( int32 transactionID, SlaveVP *animSlv )
   1.805 + {
   1.806 +   VSsSemReq  reqData;
   1.807 +
   1.808 +      //
   1.809 +   reqData.callingSlv      = animSlv;
   1.810 +   reqData.reqType     = trans_end;
   1.811 +   reqData.transID     = transactionID;
   1.812 +
   1.813 +   VMS_WL__send_sem_request( &reqData, animSlv );
   1.814 + }
   1.815 +
   1.816 +//======================== Internal ==================================
   1.817 +/*
   1.818 + */
   1.819 +SlaveVP *
   1.820 +VSs__create_slave_with( TopLevelFnPtr fnPtr,   void *initData,
   1.821 +                        SlaveVP *creatingSlv )
   1.822 + { VSsSemReq reqData;
   1.823 +
   1.824 +      //the semantic request data is on the stack and disappears when this
   1.825 +      // call returns -- it's guaranteed to remain in the VP's stack for as
   1.826 +      // long as the VP is suspended.
   1.827 +   reqData.reqType            = 0; //know type because in a VMS create req
   1.828 +   reqData.coreToAssignOnto = -1; //means round-robin assign
   1.829 +   reqData.fnPtr              = fnPtr;
   1.830 +   reqData.initData           = initData;
   1.831 +   reqData.callingSlv             = creatingSlv;
   1.832 +
   1.833 +   VMS_WL__send_create_slaveVP_req( &reqData, creatingSlv );
   1.834 +
   1.835 +   return creatingSlv->dataRetFromReq;
   1.836 + }
   1.837 +
   1.838 +SlaveVP *
   1.839 +VSs__create_slave_with_affinity( TopLevelFnPtr fnPtr, void *initData,
   1.840 +                        SlaveVP *creatingSlv,  int32  coreToAssignOnto )
   1.841 + { VSsSemReq  reqData;
   1.842 +
   1.843 +      //the semantic request data is on the stack and disappears when this
   1.844 +      // call returns -- it's guaranteed to remain in the VP's stack for as
   1.845 +      // long as the VP is suspended.
   1.846 +   reqData.reqType            = create_slave_w_aff; //not used, May 2012
   1.847 +   reqData.coreToAssignOnto   = coreToAssignOnto;
   1.848 +   reqData.fnPtr              = fnPtr;
   1.849 +   reqData.initData           = initData;
   1.850 +   reqData.callingSlv         = creatingSlv;
   1.851 +
   1.852 +   VMS_WL__send_create_slaveVP_req( &reqData, creatingSlv );
   1.853 +
   1.854 +   return creatingSlv->dataRetFromReq;
   1.855 + }
   1.856 +