1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/SSR.c	Tue Mar 13 18:30:05 2012 -0700
     1.3 @@ -0,0 +1,785 @@
     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 "SSR.h"
    1.18 +#include "SSR_Counter_Recording.h"
    1.19 +
    1.20 +//==========================================================================
    1.21 +
    1.22 +void
    1.23 +SSR__init();
    1.24 +
    1.25 +void
    1.26 +SSR__init_Helper();
    1.27 +//==========================================================================
    1.28 +
    1.29 +
    1.30 +/*TODO: Q: dealing with library f()s and DKU vs WT vs FoR
    1.31 + * (still want to do FoR, with time-lines as syntax, could be super cool)
    1.32 + * A: thinking pin the coreCtlrs for all of BLIS -- let Master arbitrate
    1.33 + * among library, DKU, WT, FoR -- all the patterns in terms of virtual
    1.34 + * processors (or equivalently work-units), so Master picks which virt procr
    1.35 + * from which portions of app (DKU, WT, FoR) onto which sched slots
    1.36 + *Might even do hierarchy of masters -- group of sched slots for each core
    1.37 + * has its own master, that keeps generated work local
    1.38 + * single-reader-single-writer sync everywhere -- no atomic primitives
    1.39 + * Might have the different assigners talk to each other, to negotiate
    1.40 + * larger-grain sharing of resources, according to predicted critical
    1.41 + * path, and expansion of work
    1.42 + */
    1.43 +
    1.44 +
    1.45 +
    1.46 +//===========================================================================
    1.47 +
    1.48 +
    1.49 +/*These are the library functions *called in the application*
    1.50 + * 
    1.51 + *There's a pattern for the outside sequential code to interact with the
    1.52 + * VMS_HW code.
    1.53 + *The VMS_HW system is inside a boundary..  every SSR system is in its
    1.54 + * own directory that contains the functions for each of the processor types.
    1.55 + * One of the processor types is the "seed" processor that starts the
    1.56 + * cascade of creating all the processors that do the work.
    1.57 + *So, in the directory is a file called "EntryPoint.c" that contains the
    1.58 + * function, named appropriately to the work performed, that the outside
    1.59 + * sequential code calls.  This function follows a pattern:
    1.60 + *1) it calls SSR__init()
    1.61 + *2) it creates the initial data for the seed processor, which is passed
    1.62 + *    in to the function
    1.63 + *3) it creates the seed SSR processor, with the data to start it with.
    1.64 + *4) it calls startSSRThenWaitUntilWorkDone
    1.65 + *5) it gets the returnValue from the transfer struc and returns that
    1.66 + *    from the function
    1.67 + *
    1.68 + *For now, a new SSR system has to be created via SSR__init every
    1.69 + * time an entry point function is called -- later, might add letting the
    1.70 + * SSR system be created once, and let all the entry points just reuse
    1.71 + * it -- want to be as simple as possible now, and see by using what makes
    1.72 + * sense for later..
    1.73 + */
    1.74 +
    1.75 +
    1.76 +
    1.77 +//===========================================================================
    1.78 +
    1.79 +/*This is the "border crossing" function -- the thing that crosses from the
    1.80 + * outside world, into the VMS_HW world.  It initializes and starts up the
    1.81 + * VMS system, then creates one processor from the specified function and
    1.82 + * puts it into the readyQ.  From that point, that one function is resp.
    1.83 + * for creating all the other processors, that then create others, and so
    1.84 + * forth.
    1.85 + *When all the processors, including the seed, have dissipated, then this
    1.86 + * function returns.  The results will have been written by side-effect via
    1.87 + * pointers read from, or written into initData.
    1.88 + *
    1.89 + *NOTE: no Threads should exist in the outside program that might touch
    1.90 + * any of the data reachable from initData passed in to here
    1.91 + */
    1.92 +void
    1.93 +SSR__create_seed_procr_and_do_work( TopLevelFnPtr fnPtr, void *initData )
    1.94 + { SSRSemEnv *semEnv;
    1.95 +   SlaveVP *seedPr;
    1.96 +
    1.97 +   SSR__init();      //normal multi-thd
    1.98 +   
    1.99 +   semEnv = _VMSMasterEnv->semanticEnv;
   1.100 +
   1.101 +      //SSR starts with one processor, which is put into initial environ,
   1.102 +      // and which then calls create() to create more, thereby expanding work
   1.103 +   seedPr = SSR__create_procr_helper( fnPtr, initData,
   1.104 +                                      semEnv, semEnv->nextCoreToGetNewPr++ );
   1.105 +
   1.106 +   resume_slaveVP( seedPr, semEnv );
   1.107 +   
   1.108 +   VMS_SS__start_the_work_then_wait_until_done();      //normal multi-thd
   1.109 +
   1.110 +   SSR__cleanup_after_shutdown();
   1.111 + }
   1.112 +
   1.113 +
   1.114 +int32
   1.115 +SSR__giveMinWorkUnitCycles( float32 percentOverhead )
   1.116 + {
   1.117 +   return MIN_WORK_UNIT_CYCLES;
   1.118 + }
   1.119 +
   1.120 +int32
   1.121 +SSR__giveIdealNumWorkUnits()
   1.122 + {
   1.123 +   return NUM_SCHED_SLOTS * NUM_CORES;
   1.124 + }
   1.125 +
   1.126 +int32
   1.127 +SSR__give_number_of_cores_to_schedule_onto()
   1.128 + {
   1.129 +   return NUM_CORES;
   1.130 + }
   1.131 +
   1.132 +/*For now, use TSC -- later, make these two macros with assembly that first
   1.133 + * saves jump point, and second jumps back several times to get reliable time
   1.134 + */
   1.135 +void
   1.136 +SSR__start_primitive()
   1.137 + { saveLowTimeStampCountInto( ((SSRSemEnv *)(_VMSMasterEnv->semanticEnv))->
   1.138 +                              primitiveStartTime );
   1.139 + }
   1.140 +
   1.141 +/*Just quick and dirty for now -- make reliable later
   1.142 + * will want this to jump back several times -- to be sure cache is warm
   1.143 + * because don't want comm time included in calc-time measurement -- and
   1.144 + * also to throw out any "weird" values due to OS interrupt or TSC rollover
   1.145 + */
   1.146 +int32
   1.147 +SSR__end_primitive_and_give_cycles()
   1.148 + { int32 endTime, startTime;
   1.149 +   //TODO: fix by repeating time-measurement
   1.150 +   saveLowTimeStampCountInto( endTime );
   1.151 +   startTime =((SSRSemEnv*)(_VMSMasterEnv->semanticEnv))->primitiveStartTime;
   1.152 +   return (endTime - startTime);
   1.153 + }
   1.154 +
   1.155 +//===========================================================================
   1.156 +
   1.157 +/*Initializes all the data-structures for a SSR system -- but doesn't
   1.158 + * start it running yet!
   1.159 + *
   1.160 + *This runs in the main thread -- before VMS starts up
   1.161 + * 
   1.162 + *This sets up the semantic layer over the VMS system
   1.163 + *
   1.164 + *First, calls VMS_Setup, then creates own environment, making it ready
   1.165 + * for creating the seed processor and then starting the work.
   1.166 + */
   1.167 +void
   1.168 +SSR__init()
   1.169 + {
   1.170 +   VMS_SS__init();
   1.171 +      //masterEnv, a global var, now is partially set up by init_VMS
   1.172 +      // after this, have VMS_int__malloc and VMS_int__free available
   1.173 +
   1.174 +   SSR__init_Helper();
   1.175 + }
   1.176 +
   1.177 +
   1.178 +void idle_fn(void* data, SlaveVP *animatingSlv){
   1.179 +    while(1){
   1.180 +        VMS_int__suspend_slaveVP_and_send_req(animatingSlv);
   1.181 +    }
   1.182 +}
   1.183 +
   1.184 +void
   1.185 +SSR__init_Helper()
   1.186 + { SSRSemEnv       *semanticEnv;
   1.187 +   PrivQueueStruc **readyVPQs;
   1.188 +   int              coreIdx, i, j;
   1.189 + 
   1.190 +      //Hook up the semantic layer's plug-ins to the Master virt procr
   1.191 +   _VMSMasterEnv->requestHandler = &SSR__Request_Handler;
   1.192 +   _VMSMasterEnv->slaveAssigner  = &SSR__assign_slaveVP;
   1.193 +   #ifdef HOLISTIC__TURN_ON_PERF_COUNTERS
   1.194 +   _VMSMasterEnv->counterHandler = &SSR__counter_handler;
   1.195 +   #endif
   1.196 +
   1.197 +      //create the semantic layer's environment (all its data) and add to
   1.198 +      // the master environment
   1.199 +   semanticEnv = VMS_int__malloc( sizeof( SSRSemEnv ) );
   1.200 +   _VMSMasterEnv->semanticEnv = semanticEnv;
   1.201 +   
   1.202 +   #ifdef HOLISTIC__TURN_ON_PERF_COUNTERS
   1.203 +   SSR__init_counter_data_structs();
   1.204 +   #endif
   1.205 +   for(i=0;i<NUM_CORES;++i){
   1.206 +       for(j=0;j<NUM_SCHED_SLOTS;++j){
   1.207 +           semanticEnv->idlePr[i][j] = VMS_int__create_slaveVP(&idle_fn,NULL);
   1.208 +           semanticEnv->idlePr[i][j]->coreAnimatedBy = i;
   1.209 +       }
   1.210 +   }
   1.211 +
   1.212 +   #ifdef HOLISTIC__TURN_ON_OBSERVE_UCC
   1.213 +   semanticEnv->unitList = makeListOfArrays(sizeof(Unit),128);
   1.214 +   semanticEnv->ctlDependenciesList = makeListOfArrays(sizeof(Dependency),128);
   1.215 +   semanticEnv->commDependenciesList = makeListOfArrays(sizeof(Dependency),128);
   1.216 +   semanticEnv->dynDependenciesList = makeListOfArrays(sizeof(Dependency),128);
   1.217 +   semanticEnv->ntonGroupsInfo = makePrivDynArrayOfSize((void***)&(semanticEnv->ntonGroups),8);
   1.218 +   
   1.219 +   semanticEnv->hwArcs = makeListOfArrays(sizeof(Dependency),128);
   1.220 +   memset(semanticEnv->last_in_slot,0,sizeof(NUM_CORES * NUM_SCHED_SLOTS * sizeof(Unit)));
   1.221 +   #endif
   1.222 +
   1.223 +      //create the ready queue, hash tables used for pairing send to receive
   1.224 +      // and so forth
   1.225 +      //TODO: add hash tables for pairing sends with receives, and
   1.226 +      // initialize the data ownership system
   1.227 +   readyVPQs = VMS_int__malloc( NUM_CORES * sizeof(PrivQueueStruc *) );
   1.228 +
   1.229 +   for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
   1.230 +    {
   1.231 +      readyVPQs[ coreIdx ] = makeVMSQ();
   1.232 +    }
   1.233 +   
   1.234 +   semanticEnv->readyVPQs = readyVPQs;
   1.235 +   
   1.236 +   semanticEnv->nextCoreToGetNewPr = 0;
   1.237 +   semanticEnv->numSlaveVP = 0;
   1.238 +   
   1.239 +   semanticEnv->commHashTbl  = makeHashTable( 1<<16, &VMS_int__free );//start big
   1.240 +
   1.241 +   //TODO: bug -- turn these arrays into dyn arrays to eliminate limit
   1.242 +   //semanticEnv->singletonHasBeenExecutedFlags = makeDynArrayInfo( );
   1.243 +   //semanticEnv->transactionStrucs = makeDynArrayInfo( );
   1.244 +   for( i = 0; i < NUM_STRUCS_IN_SEM_ENV; i++ )
   1.245 +    {
   1.246 +      semanticEnv->fnSingletons[i].endInstrAddr      = NULL;
   1.247 +      semanticEnv->fnSingletons[i].hasBeenStarted    = FALSE;
   1.248 +      semanticEnv->fnSingletons[i].hasFinished       = FALSE;
   1.249 +      semanticEnv->fnSingletons[i].waitQ             = makeVMSQ();
   1.250 +      semanticEnv->transactionStrucs[i].waitingVPQ   = makeVMSQ();
   1.251 +    }
   1.252 + }
   1.253 +
   1.254 +
   1.255 +/*Frees any memory allocated by SSR__init() then calls VMS_int__shutdown
   1.256 + */
   1.257 +void
   1.258 +SSR__cleanup_after_shutdown()
   1.259 + { SSRSemEnv *semanticEnv;
   1.260 +   
   1.261 +   semanticEnv = _VMSMasterEnv->semanticEnv;
   1.262 +
   1.263 +   #ifdef HOLISTIC__TURN_ON_OBSERVE_UCC
   1.264 +   //UCC
   1.265 +   FILE* output;
   1.266 +   int n;
   1.267 +   char filename[255];    
   1.268 +    for(n=0;n<255;n++)
   1.269 +    {
   1.270 +        sprintf(filename, "./counters/UCC.%d",n);
   1.271 +        output = fopen(filename,"r");
   1.272 +        if(output)
   1.273 +        {
   1.274 +            fclose(output);
   1.275 +        }else{
   1.276 +            break;
   1.277 +        }
   1.278 +    }
   1.279 +   if(n<255){
   1.280 +    printf("Saving UCC to File: %s ...\n", filename);
   1.281 +    output = fopen(filename,"w+");
   1.282 +    if(output!=NULL){
   1.283 +        set_dependency_file(output);
   1.284 +        //fprintf(output,"digraph Dependencies {\n");
   1.285 +        //set_dot_file(output);
   1.286 +        //FIXME:  first line still depends on counters being enabled, replace w/ unit struct!
   1.287 +        //forAllInDynArrayDo(_VMSMasterEnv->counter_history_array_info, &print_dot_node_info );
   1.288 +        forAllInListOfArraysDo(semanticEnv->unitList, &print_unit_to_file);
   1.289 +        forAllInListOfArraysDo( semanticEnv->commDependenciesList, &print_comm_dependency_to_file );
   1.290 +        forAllInListOfArraysDo( semanticEnv->ctlDependenciesList, &print_ctl_dependency_to_file );
   1.291 +        forAllInDynArrayDo(semanticEnv->ntonGroupsInfo,&print_nton_to_file);
   1.292 +        //fprintf(output,"}\n");
   1.293 +        fflush(output);
   1.294 +
   1.295 +    } else
   1.296 +        printf("Opening UCC file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
   1.297 +   } else {
   1.298 +       printf("Could not open UCC file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
   1.299 +   }
   1.300 +   //Loop Graph
   1.301 +   for(n=0;n<255;n++)
   1.302 +    {
   1.303 +        sprintf(filename, "./counters/LoopGraph.%d",n);
   1.304 +        output = fopen(filename,"r");
   1.305 +        if(output)
   1.306 +        {
   1.307 +            fclose(output);
   1.308 +        }else{
   1.309 +            break;
   1.310 +        }
   1.311 +    }
   1.312 +   if(n<255){
   1.313 +    printf("Saving LoopGraph to File: %s ...\n", filename);
   1.314 +    output = fopen(filename,"w+");
   1.315 +    if(output!=NULL){
   1.316 +        set_dependency_file(output);
   1.317 +        //fprintf(output,"digraph Dependencies {\n");
   1.318 +        //set_dot_file(output);
   1.319 +        //FIXME:  first line still depends on counters being enabled, replace w/ unit struct!
   1.320 +        //forAllInDynArrayDo(_VMSMasterEnv->counter_history_array_info, &print_dot_node_info );
   1.321 +        forAllInListOfArraysDo( semanticEnv->unitList, &print_unit_to_file );
   1.322 +        forAllInListOfArraysDo( semanticEnv->commDependenciesList, &print_comm_dependency_to_file );
   1.323 +        forAllInListOfArraysDo( semanticEnv->ctlDependenciesList, &print_ctl_dependency_to_file );
   1.324 +        forAllInListOfArraysDo( semanticEnv->dynDependenciesList, &print_dyn_dependency_to_file );
   1.325 +        forAllInListOfArraysDo( semanticEnv->hwArcs, &print_hw_dependency_to_file );
   1.326 +        //fprintf(output,"}\n");
   1.327 +        fflush(output);
   1.328 +
   1.329 +    } else
   1.330 +        printf("Opening LoopGraph file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
   1.331 +   } else {
   1.332 +       printf("Could not open LoopGraph file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
   1.333 +   }
   1.334 +   
   1.335 +   
   1.336 +   freeListOfArrays(semanticEnv->unitList);
   1.337 +   freeListOfArrays(semanticEnv->commDependenciesList);
   1.338 +   freeListOfArrays(semanticEnv->ctlDependenciesList);
   1.339 +   freeListOfArrays(semanticEnv->dynDependenciesList);
   1.340 +   
   1.341 +   #endif
   1.342 +#ifdef HOLISTIC__TURN_ON_PERF_COUNTERS    
   1.343 +    for(n=0;n<255;n++)
   1.344 +    {
   1.345 +        sprintf(filename, "./counters/Counters.%d.csv",n);
   1.346 +        output = fopen(filename,"r");
   1.347 +        if(output)
   1.348 +        {
   1.349 +            fclose(output);
   1.350 +        }else{
   1.351 +            break;
   1.352 +        }
   1.353 +    }
   1.354 +    if(n<255){
   1.355 +    printf("Saving Counter measurements to File: %s ...\n", filename);
   1.356 +    output = fopen(filename,"w+");
   1.357 +    if(output!=NULL){
   1.358 +        set_counter_file(output);
   1.359 +        int i;
   1.360 +        for(i=0;i<NUM_CORES;i++){
   1.361 +            forAllInListOfArraysDo( semanticEnv->counterList[i], &print_counter_events_to_file );
   1.362 +            fflush(output);
   1.363 +        }
   1.364 +
   1.365 +    } else
   1.366 +        printf("Opening UCC file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
   1.367 +   } else {
   1.368 +       printf("Could not open UCC file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
   1.369 +   }
   1.370 +    
   1.371 +#endif
   1.372 +/* It's all allocated inside VMS's big chunk -- that's about to be freed, so
   1.373 + *  nothing to do here
   1.374 +   
   1.375 +
   1.376 +   for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
   1.377 +    {
   1.378 +      VMS_int__free( semanticEnv->readyVPQs[coreIdx]->startOfData );
   1.379 +      VMS_int__free( semanticEnv->readyVPQs[coreIdx] );
   1.380 +    }
   1.381 +   VMS_int__free( semanticEnv->readyVPQs );
   1.382 +   
   1.383 +   freeHashTable( semanticEnv->commHashTbl );
   1.384 +   VMS_int__free( _VMSMasterEnv->semanticEnv );
   1.385 + */
   1.386 +   VMS_SS__cleanup_at_end_of_shutdown();
   1.387 + }
   1.388 +
   1.389 +
   1.390 +//===========================================================================
   1.391 +
   1.392 +/*
   1.393 + */
   1.394 +  SlaveVP *
   1.395 +SSR__create_procr_with( TopLevelFnPtr fnPtr,   void *initData,
   1.396 +                        SlaveVP *creatingPr )
   1.397 + { SSRSemReq reqData;
   1.398 +
   1.399 +      //the semantic request data is on the stack and disappears when this
   1.400 +      // call returns -- it's guaranteed to remain in the VP's stack for as
   1.401 +      // long as the VP is suspended.
   1.402 +   reqData.reqType            = 0; //know type because in a VMS create req
   1.403 +   reqData.coreToAssignOnto = -1; //means round-robin assign
   1.404 +   reqData.fnPtr              = fnPtr;
   1.405 +   reqData.initData           = initData;
   1.406 +   reqData.sendPr             = creatingPr;
   1.407 +
   1.408 +   VMS_WL__send_create_slaveVP_req( &reqData, creatingPr );
   1.409 +
   1.410 +   return creatingPr->dataRetFromReq;
   1.411 + }
   1.412 +
   1.413 +  SlaveVP *
   1.414 +SSR__create_procr_with_affinity( TopLevelFnPtr fnPtr, void *initData,
   1.415 +                        SlaveVP *creatingPr,  int32  coreToAssignOnto )
   1.416 + { SSRSemReq  reqData;
   1.417 +
   1.418 +      //the semantic request data is on the stack and disappears when this
   1.419 +      // call returns -- it's guaranteed to remain in the VP's stack for as
   1.420 +      // long as the VP is suspended.
   1.421 +   reqData.reqType            = 0; //know type because in a VMS create req
   1.422 +   reqData.coreToAssignOnto = coreToAssignOnto;
   1.423 +   reqData.fnPtr              = fnPtr;
   1.424 +   reqData.initData           = initData;
   1.425 +   reqData.sendPr             = creatingPr;
   1.426 +
   1.427 +   VMS_WL__send_create_slaveVP_req( &reqData, creatingPr );
   1.428 +
   1.429 +   return creatingPr->dataRetFromReq;
   1.430 + }
   1.431 +
   1.432 +
   1.433 +  void
   1.434 +SSR__dissipate_procr( SlaveVP *procrToDissipate )
   1.435 + {
   1.436 +   VMS_WL__send_dissipate_req( procrToDissipate );
   1.437 + }
   1.438 +
   1.439 +
   1.440 +//===========================================================================
   1.441 +
   1.442 +void *
   1.443 +SSR__malloc_to( int32 sizeToMalloc, SlaveVP *owningPr )
   1.444 + { SSRSemReq reqData;
   1.445 +
   1.446 +   reqData.reqType      = malloc_req;
   1.447 +   reqData.sendPr       = owningPr;
   1.448 +   reqData.sizeToMalloc = sizeToMalloc;
   1.449 +
   1.450 +   VMS_WL__send_sem_request( &reqData, owningPr );
   1.451 +
   1.452 +   return owningPr->dataRetFromReq;
   1.453 + }
   1.454 +
   1.455 +
   1.456 +/*Sends request to Master, which does the work of freeing
   1.457 + */
   1.458 +void
   1.459 +SSR__free( void *ptrToFree, SlaveVP *owningPr )
   1.460 + { SSRSemReq reqData;
   1.461 +
   1.462 +   reqData.reqType      = free_req;
   1.463 +   reqData.sendPr       = owningPr;
   1.464 +   reqData.ptrToFree    = ptrToFree;
   1.465 +
   1.466 +   VMS_WL__send_sem_request( &reqData, owningPr );
   1.467 + }
   1.468 +
   1.469 +
   1.470 +void
   1.471 +SSR__transfer_ownership_of_from_to( void *data, SlaveVP *oldOwnerSlv,
   1.472 +                                                  SlaveVP *newOwnerPr )
   1.473 + {
   1.474 +   //TODO: put in the ownership system that automatically frees when no
   1.475 +   // owners of data left -- will need keeper for keeping data around when
   1.476 +   // future created processors might need it but don't exist yet
   1.477 + }
   1.478 +
   1.479 +
   1.480 +void
   1.481 +SSR__add_ownership_by_to( SlaveVP *newOwnerSlv, void *data )
   1.482 + {
   1.483 +
   1.484 + }
   1.485 +
   1.486 +
   1.487 +void
   1.488 +SSR__remove_ownership_by_from( SlaveVP *loserSlv, void *dataLosing )
   1.489 + {
   1.490 +
   1.491 + }
   1.492 +
   1.493 +
   1.494 +/*Causes the SSR system to remove internal ownership, so data won't be
   1.495 + * freed when SSR shuts down, and will persist in the external program.
   1.496 + *
   1.497 + *Must be called from the processor that currently owns the data.
   1.498 + *
   1.499 + *IMPL: Transferring ownership touches two different virtual processor's
   1.500 + * state -- which means it has to be done carefully -- the VMS rules for
   1.501 + * semantic layers say that a work-unit is only allowed to touch the
   1.502 + * virtual processor it is part of, and that only a single work-unit per
   1.503 + * virtual processor be assigned to a slave at a time.  So, this has to
   1.504 + * modify the virtual processor that owns the work-unit that called this
   1.505 + * function, then create a request to have the other processor modified.
   1.506 + *However, in this case, the TO processor is the outside, and transfers
   1.507 + * are only allowed to be called by the giver-upper, so can mark caller of
   1.508 + * this function as no longer owner, and return -- done.
   1.509 + */
   1.510 +void
   1.511 +SSR__transfer_ownership_to_outside( void *data )
   1.512 + {
   1.513 +   //TODO: removeAllOwnersFrom( data );
   1.514 + }
   1.515 +
   1.516 +
   1.517 +//===========================================================================
   1.518 +
   1.519 +void
   1.520 +SSR__send_of_type_to( SlaveVP *sendPr, void *msg, const int type,
   1.521 +                        SlaveVP *receivePr)
   1.522 + { SSRSemReq  reqData;
   1.523 +
   1.524 +   reqData.receivePr = receivePr;
   1.525 +   reqData.sendPr    = sendPr;
   1.526 +   reqData.reqType   = send_type;
   1.527 +   reqData.msgType   = type;
   1.528 +   reqData.msg       = msg;
   1.529 +   reqData.nextReqInHashEntry = NULL;
   1.530 +
   1.531 +      //On ownership -- remove inside the send and let ownership sit in limbo
   1.532 +      // as a potential in an entry in the hash table, when this receive msg
   1.533 +      // gets paired to a send, the ownership gets added to the receivePr --
   1.534 +      // the next work-unit in the receivePr's trace will have ownership.
   1.535 +   VMS_WL__send_sem_request( &reqData, sendPr );
   1.536 +
   1.537 +      //When come back from suspend, no longer own data reachable from msg
   1.538 +      //TODO: release ownership here
   1.539 + }
   1.540 +
   1.541 +void
   1.542 +SSR__send_from_to( void *msg, SlaveVP *sendPr, SlaveVP *receivePr )
   1.543 + { SSRSemReq  reqData;
   1.544 +
   1.545 +      //hash on the receiver, 'cause always know it, but sometimes want to
   1.546 +      // receive from anonymous sender
   1.547 +
   1.548 +   reqData.receivePr = receivePr;
   1.549 +   reqData.sendPr    = sendPr;
   1.550 +   reqData.reqType   = send_from_to;
   1.551 +   reqData.msg       = msg;
   1.552 +   reqData.nextReqInHashEntry = NULL;
   1.553 +
   1.554 +   VMS_WL__send_sem_request( &reqData, sendPr );
   1.555 + }
   1.556 +
   1.557 +
   1.558 +//===========================================================================
   1.559 +
   1.560 +void *
   1.561 +SSR__receive_any_to( SlaveVP *receivePr )
   1.562 + {
   1.563 +
   1.564 + }
   1.565 +
   1.566 +void *
   1.567 +SSR__receive_type_to( const int type, SlaveVP *receivePr )
   1.568 + { 
   1.569 +   SSRSemReq  reqData;
   1.570 +
   1.571 +   reqData.receivePr = receivePr;
   1.572 +   reqData.reqType   = receive_type;
   1.573 +   reqData.msgType   = type;
   1.574 +   reqData.nextReqInHashEntry = NULL;
   1.575 +
   1.576 +   VMS_WL__send_sem_request( &reqData, receivePr );
   1.577 +   
   1.578 +   return receivePr->dataRetFromReq;
   1.579 + }
   1.580 +
   1.581 +
   1.582 +
   1.583 +/*Call this at point receiving virt pr wants in-coming data.
   1.584 + * 
   1.585 + *The reason receivePr must call this is that it modifies the receivPr
   1.586 + * loc structure directly -- and the VMS rules state a virtual processor
   1.587 + * loc structure can only be modified by itself.
   1.588 + */
   1.589 +void *
   1.590 +SSR__receive_from_to( SlaveVP *sendPr, SlaveVP *receivePr )
   1.591 + { SSRSemReq  reqData;
   1.592 +
   1.593 +      //hash on the receiver, 'cause always know it, but sometimes want to
   1.594 +      // receive from anonymous sender
   1.595 +
   1.596 +   reqData.receivePr = receivePr;
   1.597 +   reqData.sendPr    = sendPr;
   1.598 +   reqData.reqType   = receive_from_to;
   1.599 +   reqData.nextReqInHashEntry = NULL;
   1.600 +
   1.601 +   VMS_WL__send_sem_request( &reqData, receivePr );
   1.602 +
   1.603 +   return receivePr->dataRetFromReq;
   1.604 + }
   1.605 +
   1.606 +
   1.607 +//===========================================================================
   1.608 +//
   1.609 +/*A function singleton is a function whose body executes exactly once, on a
   1.610 + * single core, no matter how many times the fuction is called and no
   1.611 + * matter how many cores or the timing of cores calling it.
   1.612 + *
   1.613 + *A data singleton is a ticket attached to data.  That ticket can be used
   1.614 + * to get the data through the function exactly once, no matter how many
   1.615 + * times the data is given to the function, and no matter the timing of
   1.616 + * trying to get the data through from different cores.
   1.617 + */
   1.618 +
   1.619 +/*asm function declarations*/
   1.620 +void asm_save_ret_to_singleton(SSRSingleton *singletonPtrAddr);
   1.621 +void asm_write_ret_from_singleton(SSRSingleton *singletonPtrAddr);
   1.622 +
   1.623 +/*Fn singleton uses ID as index into array of singleton structs held in the
   1.624 + * semantic environment.
   1.625 + */
   1.626 +void
   1.627 +SSR__start_fn_singleton( int32 singletonID,   SlaveVP *animPr )
   1.628 + {
   1.629 +   SSRSemReq  reqData;
   1.630 +
   1.631 +      //
   1.632 +   reqData.reqType     = singleton_fn_start;
   1.633 +   reqData.singletonID = singletonID;
   1.634 +
   1.635 +   VMS_WL__send_sem_request( &reqData, animPr );
   1.636 +   if( animPr->dataRetFromReq ) //will be 0 or addr of label in end singleton
   1.637 +    {
   1.638 +       SSRSemEnv *semEnv = VMS_int__give_sem_env_for( animPr );
   1.639 +       asm_write_ret_from_singleton(&(semEnv->fnSingletons[ singletonID]));
   1.640 +    }
   1.641 + }
   1.642 +
   1.643 +/*Data singleton hands addr of loc holding a pointer to a singleton struct.
   1.644 + * The start_data_singleton makes the structure and puts its addr into the
   1.645 + * location.
   1.646 + */
   1.647 +void
   1.648 +SSR__start_data_singleton( SSRSingleton **singletonAddr,  SlaveVP *animPr )
   1.649 + {
   1.650 +   SSRSemReq  reqData;
   1.651 +
   1.652 +   if( *singletonAddr && (*singletonAddr)->hasFinished )
   1.653 +       goto JmpToEndSingleton;
   1.654 +   
   1.655 +   reqData.reqType          = singleton_data_start;
   1.656 +   reqData.singletonPtrAddr = singletonAddr;
   1.657 +
   1.658 +   VMS_WL__send_sem_request( &reqData, animPr );
   1.659 +   if( animPr->dataRetFromReq ) //either 0 or end singleton's return addr
   1.660 +    {    //Assembly code changes the return addr on the stack to the one
   1.661 +         // saved into the singleton by the end-singleton-fn
   1.662 +         //The return addr is at 0x4(%%ebp)
   1.663 +        JmpToEndSingleton:
   1.664 +          asm_write_ret_from_singleton(*singletonAddr);
   1.665 +    }
   1.666 +   //now, simply return
   1.667 +   //will exit either from the start singleton call or the end-singleton call
   1.668 + }
   1.669 +
   1.670 +/*Uses ID as index into array of flags.  If flag already set, resumes from
   1.671 + * end-label.  Else, sets flag and resumes normally.
   1.672 + *
   1.673 + *Note, this call cannot be inlined because the instr addr at the label
   1.674 + * inside is shared by all invocations of a given singleton ID.
   1.675 + */
   1.676 +void
   1.677 +SSR__end_fn_singleton( int32 singletonID, SlaveVP *animPr )
   1.678 + {
   1.679 +   SSRSemReq  reqData;
   1.680 +
   1.681 +      //don't need this addr until after at least one singleton has reached
   1.682 +      // this function
   1.683 +   SSRSemEnv *semEnv = VMS_int__give_sem_env_for( animPr );
   1.684 +   asm_write_ret_from_singleton(&(semEnv->fnSingletons[ singletonID]));
   1.685 +
   1.686 +   reqData.reqType     = singleton_fn_end;
   1.687 +   reqData.singletonID = singletonID;
   1.688 +
   1.689 +   VMS_WL__send_sem_request( &reqData, animPr );
   1.690 +
   1.691 +EndSingletonInstrAddr:
   1.692 +   return;
   1.693 + }
   1.694 +
   1.695 +void
   1.696 +SSR__end_data_singleton(  SSRSingleton **singletonPtrAddr, SlaveVP *animPr )
   1.697 + {
   1.698 +   SSRSemReq  reqData;
   1.699 +
   1.700 +      //don't need this addr until after singleton struct has reached
   1.701 +      // this function for first time
   1.702 +      //do assembly that saves the return addr of this fn call into the
   1.703 +      // data singleton -- that data-singleton can only be given to exactly
   1.704 +      // one instance in the code of this function.  However, can use this
   1.705 +      // function in different places for different data-singletons.
   1.706 +//   (*(singletonAddr))->endInstrAddr =  &&EndDataSingletonInstrAddr;
   1.707 +
   1.708 +
   1.709 +   asm_save_ret_to_singleton(*singletonPtrAddr);
   1.710 +
   1.711 +   reqData.reqType          = singleton_data_end;
   1.712 +   reqData.singletonPtrAddr = singletonPtrAddr;
   1.713 +
   1.714 +   VMS_WL__send_sem_request( &reqData, animPr );
   1.715 + }
   1.716 +
   1.717 +/*This executes the function in the masterVP, so it executes in isolation
   1.718 + * from any other copies -- only one copy of the function can ever execute
   1.719 + * at a time.
   1.720 + *
   1.721 + *It suspends to the master, and the request handler takes the function
   1.722 + * pointer out of the request and calls it, then resumes the VP.
   1.723 + *Only very short functions should be called this way -- for longer-running
   1.724 + * isolation, use transaction-start and transaction-end, which run the code
   1.725 + * between as work-code.
   1.726 + */
   1.727 +void
   1.728 +SSR__animate_short_fn_in_isolation( PtrToAtomicFn ptrToFnToExecInMaster,
   1.729 +                                    void *data, SlaveVP *animPr )
   1.730 + {
   1.731 +   SSRSemReq  reqData;
   1.732 +
   1.733 +      //
   1.734 +   reqData.reqType          = atomic;
   1.735 +   reqData.fnToExecInMaster = ptrToFnToExecInMaster;
   1.736 +   reqData.dataForFn        = data;
   1.737 +
   1.738 +   VMS_WL__send_sem_request( &reqData, animPr );
   1.739 + }
   1.740 +
   1.741 +
   1.742 +/*This suspends to the master.
   1.743 + *First, it looks at the VP's data, to see the highest transactionID that VP
   1.744 + * already has entered.  If the current ID is not larger, it throws an
   1.745 + * exception stating a bug in the code.  Otherwise it puts the current ID
   1.746 + * there, and adds the ID to a linked list of IDs entered -- the list is
   1.747 + * used to check that exits are properly ordered.
   1.748 + *Next it is uses transactionID as index into an array of transaction
   1.749 + * structures.
   1.750 + *If the "VP_currently_executing" field is non-null, then put requesting VP
   1.751 + * into queue in the struct.  (At some point a holder will request
   1.752 + * end-transaction, which will take this VP from the queue and resume it.)
   1.753 + *If NULL, then write requesting into the field and resume.
   1.754 + */
   1.755 +void
   1.756 +SSR__start_transaction( int32 transactionID, SlaveVP *animPr )
   1.757 + {
   1.758 +   SSRSemReq  reqData;
   1.759 +
   1.760 +      //
   1.761 +   reqData.sendPr      = animPr;
   1.762 +   reqData.reqType     = trans_start;
   1.763 +   reqData.transID     = transactionID;
   1.764 +
   1.765 +   VMS_WL__send_sem_request( &reqData, animPr );
   1.766 + }
   1.767 +
   1.768 +/*This suspends to the master, then uses transactionID as index into an
   1.769 + * array of transaction structures.
   1.770 + *It looks at VP_currently_executing to be sure it's same as requesting VP.
   1.771 + * If different, throws an exception, stating there's a bug in the code.
   1.772 + *Next it looks at the queue in the structure.
   1.773 + *If it's empty, it sets VP_currently_executing field to NULL and resumes.
   1.774 + *If something in, gets it, sets VP_currently_executing to that VP, then
   1.775 + * resumes both.
   1.776 + */
   1.777 +void
   1.778 +SSR__end_transaction( int32 transactionID, SlaveVP *animPr )
   1.779 + {
   1.780 +   SSRSemReq  reqData;
   1.781 +
   1.782 +      //
   1.783 +   reqData.sendPr      = animPr;
   1.784 +   reqData.reqType     = trans_end;
   1.785 +   reqData.transID     = transactionID;
   1.786 +
   1.787 +   VMS_WL__send_sem_request( &reqData, animPr );
   1.788 + }