seanhalle@0: /*
seanhalle@0:  * Copyright 2010  OpenSourceCodeStewardshipFoundation
seanhalle@0:  *
seanhalle@0:  * Licensed under BSD
seanhalle@0:  */
seanhalle@0: 
seanhalle@0: #include <stdio.h>
seanhalle@0: #include <stdlib.h>
seanhalle@0: #include <malloc.h>
seanhalle@0: 
seanhalle@0: #include "Queue_impl/PrivateQueue.h"
seanhalle@0: #include "Hash_impl/PrivateHash.h"
seanhalle@0: 
seanhalle@2: #include "VSs.h"
seanhalle@3: #include "Measurement/VSs_Counter_Recording.h"
seanhalle@0: 
seanhalle@0: //==========================================================================
seanhalle@0: 
seanhalle@0: void
seanhalle@2: VSs__init();
seanhalle@0: 
seanhalle@0: void
seanhalle@2: VSs__init_Helper();
seanhalle@0: //==========================================================================
seanhalle@0: 
seanhalle@0: 
seanhalle@0: 
seanhalle@0: //===========================================================================
seanhalle@0: 
seanhalle@0: 
seanhalle@0: /*These are the library functions *called in the application*
seanhalle@0:  * 
seanhalle@0:  *There's a pattern for the outside sequential code to interact with the
seanhalle@0:  * VMS_HW code.
seanhalle@2:  *The VMS_HW system is inside a boundary..  every VSs system is in its
seanhalle@0:  * own directory that contains the functions for each of the processor types.
seanhalle@0:  * One of the processor types is the "seed" processor that starts the
seanhalle@0:  * cascade of creating all the processors that do the work.
seanhalle@0:  *So, in the directory is a file called "EntryPoint.c" that contains the
seanhalle@0:  * function, named appropriately to the work performed, that the outside
seanhalle@0:  * sequential code calls.  This function follows a pattern:
seanhalle@2:  *1) it calls VSs__init()
seanhalle@0:  *2) it creates the initial data for the seed processor, which is passed
seanhalle@0:  *    in to the function
seanhalle@2:  *3) it creates the seed VSs processor, with the data to start it with.
seanhalle@2:  *4) it calls startVSsThenWaitUntilWorkDone
seanhalle@0:  *5) it gets the returnValue from the transfer struc and returns that
seanhalle@0:  *    from the function
seanhalle@0:  *
seanhalle@2:  *For now, a new VSs system has to be created via VSs__init every
seanhalle@0:  * time an entry point function is called -- later, might add letting the
seanhalle@2:  * VSs system be created once, and let all the entry points just reuse
seanhalle@0:  * it -- want to be as simple as possible now, and see by using what makes
seanhalle@0:  * sense for later..
seanhalle@0:  */
seanhalle@0: 
seanhalle@0: 
seanhalle@0: 
seanhalle@0: //===========================================================================
seanhalle@0: 
seanhalle@0: /*This is the "border crossing" function -- the thing that crosses from the
seanhalle@0:  * outside world, into the VMS_HW world.  It initializes and starts up the
seanhalle@0:  * VMS system, then creates one processor from the specified function and
seanhalle@0:  * puts it into the readyQ.  From that point, that one function is resp.
seanhalle@0:  * for creating all the other processors, that then create others, and so
seanhalle@0:  * forth.
seanhalle@0:  *When all the processors, including the seed, have dissipated, then this
seanhalle@0:  * function returns.  The results will have been written by side-effect via
seanhalle@0:  * pointers read from, or written into initData.
seanhalle@0:  *
seanhalle@0:  *NOTE: no Threads should exist in the outside program that might touch
seanhalle@0:  * any of the data reachable from initData passed in to here
seanhalle@0:  */
seanhalle@0: void
seanhalle@2: VSs__create_seed_slave_and_do_work( TopLevelFnPtr fnPtr, void *initData )
seanhalle@6:  { VSsSemEnv   *semEnv;
seanhalle@6:    SlaveVP     *seedSlv;
seanhalle@6:    VSsSemData  *semData;
seanhalle@8:    VSsTaskStub *threadTaskStub, *parentTaskStub;
seanhalle@0: 
seanhalle@2:    VSs__init();      //normal multi-thd
seanhalle@0:    
seanhalle@0:    semEnv = _VMSMasterEnv->semanticEnv;
seanhalle@0: 
seanhalle@2:       //VSs starts with one processor, which is put into initial environ,
seanhalle@0:       // and which then calls create() to create more, thereby expanding work
seanhalle@3:    seedSlv = VSs__create_slave_helper( fnPtr, initData,
seanhalle@6:                                      semEnv, semEnv->nextCoreToGetNewSlv++ );
seanhalle@3:    
seanhalle@8:       //seed slave is a thread slave, so make a thread's task stub for it
seanhalle@8:       // and then make another to stand for the seed's parent task.  Make
seanhalle@8:       // the parent be already ended, and have one child (the seed).  This
seanhalle@8:       // will make the dissipate handler do the right thing when the seed
seanhalle@8:       // is dissipated.
seanhalle@8:    threadTaskStub = create_thread_task_stub( initData );
seanhalle@8:    parentTaskStub = create_thread_task_stub( NULL );
seanhalle@8:    parentTaskStub->isEnded = TRUE;
seanhalle@8:    parentTaskStub->numLiveChildThreads = 1; //so dissipate works for seed
seanhalle@8:    threadTaskStub->parentTasksStub = parentTaskStub;
seanhalle@6:    
seanhalle@6:    semData = (VSsSemData *)seedSlv->semanticData;
seanhalle@8:       //seedVP is a thread, so has a permanent task
seanhalle@6:    semData->needsTaskAssigned = FALSE;
seanhalle@8:    semData->taskStub = threadTaskStub;
seanhalle@0: 
seanhalle@6:    resume_slaveVP( seedSlv, semEnv ); //returns right away, just queues Slv
seanhalle@0:    
seanhalle@0:    VMS_SS__start_the_work_then_wait_until_done();      //normal multi-thd
seanhalle@0: 
seanhalle@2:    VSs__cleanup_after_shutdown();
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: 
seanhalle@0: int32
seanhalle@2: VSs__giveMinWorkUnitCycles( float32 percentOverhead )
seanhalle@0:  {
seanhalle@0:    return MIN_WORK_UNIT_CYCLES;
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: int32
seanhalle@2: VSs__giveIdealNumWorkUnits()
seanhalle@0:  {
seanhalle@0:    return NUM_ANIM_SLOTS * NUM_CORES;
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: int32
seanhalle@2: VSs__give_number_of_cores_to_schedule_onto()
seanhalle@0:  {
seanhalle@0:    return NUM_CORES;
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: /*For now, use TSC -- later, make these two macros with assembly that first
seanhalle@0:  * saves jump point, and second jumps back several times to get reliable time
seanhalle@0:  */
seanhalle@0: void
seanhalle@2: VSs__start_primitive()
seanhalle@2:  { saveLowTimeStampCountInto( ((VSsSemEnv *)(_VMSMasterEnv->semanticEnv))->
seanhalle@0:                               primitiveStartTime );
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: /*Just quick and dirty for now -- make reliable later
seanhalle@0:  * will want this to jump back several times -- to be sure cache is warm
seanhalle@0:  * because don't want comm time included in calc-time measurement -- and
seanhalle@0:  * also to throw out any "weird" values due to OS interrupt or TSC rollover
seanhalle@0:  */
seanhalle@0: int32
seanhalle@2: VSs__end_primitive_and_give_cycles()
seanhalle@0:  { int32 endTime, startTime;
seanhalle@0:    //TODO: fix by repeating time-measurement
seanhalle@0:    saveLowTimeStampCountInto( endTime );
seanhalle@2:    startTime =((VSsSemEnv*)(_VMSMasterEnv->semanticEnv))->primitiveStartTime;
seanhalle@0:    return (endTime - startTime);
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: //===========================================================================
seanhalle@0: 
seanhalle@2: /*Initializes all the data-structures for a VSs system -- but doesn't
seanhalle@0:  * start it running yet!
seanhalle@0:  *
seanhalle@0:  *This runs in the main thread -- before VMS starts up
seanhalle@0:  * 
seanhalle@0:  *This sets up the semantic layer over the VMS system
seanhalle@0:  *
seanhalle@0:  *First, calls VMS_Setup, then creates own environment, making it ready
seanhalle@0:  * for creating the seed processor and then starting the work.
seanhalle@0:  */
seanhalle@0: void
seanhalle@2: VSs__init()
seanhalle@0:  {
seanhalle@0:    VMS_SS__init();
seanhalle@0:       //masterEnv, a global var, now is partially set up by init_VMS
seanhalle@0:       // after this, have VMS_int__malloc and VMS_int__free available
seanhalle@0: 
seanhalle@2:    VSs__init_Helper();
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: 
seanhalle@0: void idle_fn(void* data, SlaveVP *animatingSlv){
seanhalle@0:     while(1){
seanhalle@0:         VMS_int__suspend_slaveVP_and_send_req(animatingSlv);
seanhalle@0:     }
seanhalle@0: }
seanhalle@0: 
seanhalle@0: void
seanhalle@2: VSs__init_Helper()
seanhalle@2:  { VSsSemEnv       *semanticEnv;
seanhalle@6:    int32            i, coreNum, slotNum;
seanhalle@0:  
seanhalle@0:       //Hook up the semantic layer's plug-ins to the Master virt procr
seanhalle@2:    _VMSMasterEnv->requestHandler = &VSs__Request_Handler;
seanhalle@2:    _VMSMasterEnv->slaveAssigner  = &VSs__assign_slaveVP_to_slot;
seanhalle@0:    #ifdef HOLISTIC__TURN_ON_PERF_COUNTERS
seanhalle@2:    _VMSMasterEnv->counterHandler = &VSs__counter_handler;
seanhalle@0:    #endif
seanhalle@0: 
seanhalle@0:       //create the semantic layer's environment (all its data) and add to
seanhalle@0:       // the master environment
seanhalle@2:    semanticEnv = VMS_int__malloc( sizeof( VSsSemEnv ) );
seanhalle@0:    _VMSMasterEnv->semanticEnv = semanticEnv;
seanhalle@0:    
seanhalle@0:    #ifdef HOLISTIC__TURN_ON_PERF_COUNTERS
seanhalle@2:    VSs__init_counter_data_structs();
seanhalle@0:    #endif
seanhalle@3: 
seanhalle@0:    semanticEnv->shutdownInitiated = FALSE;
seanhalle@3:    semanticEnv->coreIsDone = VMS_int__malloc( NUM_CORES * sizeof( bool32 ) );
seanhalle@6:       //For each animation slot, there is an idle slave, and an initial
seanhalle@6:       // slave assigned as the current-task-slave.  Create them here.
seanhalle@6:    SlaveVP *idleSlv, *currTaskSlv;
seanhalle@6:    for( coreNum = 0; coreNum < NUM_CORES; coreNum++ )
seanhalle@6:     { semanticEnv->coreIsDone[coreNum] = FALSE; //use during shutdown
seanhalle@6:     
seanhalle@6:       for( slotNum = 0; slotNum < NUM_ANIM_SLOTS; ++slotNum )
seanhalle@6:        { idleSlv = VMS_int__create_slaveVP(&idle_fn,NULL);
seanhalle@6:          idleSlv->coreAnimatedBy                = coreNum;
seanhalle@9:          idleSlv->animSlotAssignedTo            =
seanhalle@9:                                _VMSMasterEnv->allAnimSlots[coreNum][slotNum];
seanhalle@6:          semanticEnv->idleSlv[coreNum][slotNum] = idleSlv;
seanhalle@6:          
seanhalle@6:          currTaskSlv = VMS_int__create_slaveVP( &idle_fn, NULL );
seanhalle@6:          currTaskSlv->coreAnimatedBy                 = coreNum;
seanhalle@9:          currTaskSlv->animSlotAssignedTo             = 
seanhalle@9:                                _VMSMasterEnv->allAnimSlots[coreNum][slotNum];
seanhalle@6:          semanticEnv->currTaskSlvs[coreNum][slotNum] = currTaskSlv;
seanhalle@0:        }
seanhalle@3:     }
seanhalle@0: 
seanhalle@6:       //create the ready queues, hash tables used for matching and so forth
seanhalle@6:    semanticEnv->slavesReadyToResumeQ = makeVMSQ();
seanhalle@9:    semanticEnv->freeExtraTaskSlvQ    = makeVMSQ();
seanhalle@6:    semanticEnv->taskReadyQ           = makeVMSQ();
seanhalle@0:    
seanhalle@3:    semanticEnv->argPtrHashTbl  = makeHashTable32( 16, &VMS_int__free );
seanhalle@4:    semanticEnv->commHashTbl    = makeHashTable32( 16, &VMS_int__free );
seanhalle@6:    
seanhalle@6:    semanticEnv->nextCoreToGetNewSlv = 0;
seanhalle@6:    
seanhalle@0: 
seanhalle@0:    //TODO: bug -- turn these arrays into dyn arrays to eliminate limit
seanhalle@0:    //semanticEnv->singletonHasBeenExecutedFlags = makeDynArrayInfo( );
seanhalle@0:    //semanticEnv->transactionStrucs = makeDynArrayInfo( );
seanhalle@0:    for( i = 0; i < NUM_STRUCS_IN_SEM_ENV; i++ )
seanhalle@0:     {
seanhalle@0:       semanticEnv->fnSingletons[i].endInstrAddr      = NULL;
seanhalle@0:       semanticEnv->fnSingletons[i].hasBeenStarted    = FALSE;
seanhalle@0:       semanticEnv->fnSingletons[i].hasFinished       = FALSE;
seanhalle@0:       semanticEnv->fnSingletons[i].waitQ             = makeVMSQ();
seanhalle@0:       semanticEnv->transactionStrucs[i].waitingVPQ   = makeVMSQ();
seanhalle@0:     }
seanhalle@6: 
seanhalle@8:    semanticEnv->numLiveExtraTaskSlvs   = 0; //must be last
seanhalle@8:    semanticEnv->numLiveThreadSlvs      = 1; //must be last, count the seed
seanhalle@6: 
seanhalle@6:    #ifdef HOLISTIC__TURN_ON_OBSERVE_UCC
seanhalle@6:    semanticEnv->unitList = makeListOfArrays(sizeof(Unit),128);
seanhalle@6:    semanticEnv->ctlDependenciesList = makeListOfArrays(sizeof(Dependency),128);
seanhalle@6:    semanticEnv->commDependenciesList = makeListOfArrays(sizeof(Dependency),128);
seanhalle@6:    semanticEnv->dynDependenciesList = makeListOfArrays(sizeof(Dependency),128);
seanhalle@6:    semanticEnv->ntonGroupsInfo = makePrivDynArrayOfSize((void***)&(semanticEnv->ntonGroups),8);
seanhalle@6:    
seanhalle@6:    semanticEnv->hwArcs = makeListOfArrays(sizeof(Dependency),128);
seanhalle@6:    memset(semanticEnv->last_in_slot,0,sizeof(NUM_CORES * NUM_ANIM_SLOTS * sizeof(Unit)));
seanhalle@6:    #endif
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: 
seanhalle@2: /*Frees any memory allocated by VSs__init() then calls VMS_int__shutdown
seanhalle@0:  */
seanhalle@0: void
seanhalle@2: VSs__cleanup_after_shutdown()
seanhalle@2:  { VSsSemEnv *semanticEnv;
seanhalle@0:    
seanhalle@0:    semanticEnv = _VMSMasterEnv->semanticEnv;
seanhalle@0: 
seanhalle@0:    #ifdef HOLISTIC__TURN_ON_OBSERVE_UCC
seanhalle@0:    //UCC
seanhalle@0:    FILE* output;
seanhalle@0:    int n;
seanhalle@0:    char filename[255];    
seanhalle@0:     for(n=0;n<255;n++)
seanhalle@0:     {
seanhalle@0:         sprintf(filename, "./counters/UCC.%d",n);
seanhalle@0:         output = fopen(filename,"r");
seanhalle@0:         if(output)
seanhalle@0:         {
seanhalle@0:             fclose(output);
seanhalle@0:         }else{
seanhalle@0:             break;
seanhalle@0:         }
seanhalle@0:     }
seanhalle@0:    if(n<255){
seanhalle@0:     printf("Saving UCC to File: %s ...\n", filename);
seanhalle@0:     output = fopen(filename,"w+");
seanhalle@0:     if(output!=NULL){
seanhalle@0:         set_dependency_file(output);
seanhalle@0:         //fprintf(output,"digraph Dependencies {\n");
seanhalle@0:         //set_dot_file(output);
seanhalle@0:         //FIXME:  first line still depends on counters being enabled, replace w/ unit struct!
seanhalle@0:         //forAllInDynArrayDo(_VMSMasterEnv->counter_history_array_info, &print_dot_node_info );
seanhalle@0:         forAllInListOfArraysDo(semanticEnv->unitList, &print_unit_to_file);
seanhalle@0:         forAllInListOfArraysDo( semanticEnv->commDependenciesList, &print_comm_dependency_to_file );
seanhalle@0:         forAllInListOfArraysDo( semanticEnv->ctlDependenciesList, &print_ctl_dependency_to_file );
seanhalle@0:         forAllInDynArrayDo(semanticEnv->ntonGroupsInfo,&print_nton_to_file);
seanhalle@0:         //fprintf(output,"}\n");
seanhalle@0:         fflush(output);
seanhalle@0: 
seanhalle@0:     } else
seanhalle@0:         printf("Opening UCC file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
seanhalle@0:    } else {
seanhalle@0:        printf("Could not open UCC file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
seanhalle@0:    }
seanhalle@0:    //Loop Graph
seanhalle@0:    for(n=0;n<255;n++)
seanhalle@0:     {
seanhalle@0:         sprintf(filename, "./counters/LoopGraph.%d",n);
seanhalle@0:         output = fopen(filename,"r");
seanhalle@0:         if(output)
seanhalle@0:         {
seanhalle@0:             fclose(output);
seanhalle@0:         }else{
seanhalle@0:             break;
seanhalle@0:         }
seanhalle@0:     }
seanhalle@0:    if(n<255){
seanhalle@0:     printf("Saving LoopGraph to File: %s ...\n", filename);
seanhalle@0:     output = fopen(filename,"w+");
seanhalle@0:     if(output!=NULL){
seanhalle@0:         set_dependency_file(output);
seanhalle@0:         //fprintf(output,"digraph Dependencies {\n");
seanhalle@0:         //set_dot_file(output);
seanhalle@0:         //FIXME:  first line still depends on counters being enabled, replace w/ unit struct!
seanhalle@0:         //forAllInDynArrayDo(_VMSMasterEnv->counter_history_array_info, &print_dot_node_info );
seanhalle@0:         forAllInListOfArraysDo( semanticEnv->unitList, &print_unit_to_file );
seanhalle@0:         forAllInListOfArraysDo( semanticEnv->commDependenciesList, &print_comm_dependency_to_file );
seanhalle@0:         forAllInListOfArraysDo( semanticEnv->ctlDependenciesList, &print_ctl_dependency_to_file );
seanhalle@0:         forAllInListOfArraysDo( semanticEnv->dynDependenciesList, &print_dyn_dependency_to_file );
seanhalle@0:         forAllInListOfArraysDo( semanticEnv->hwArcs, &print_hw_dependency_to_file );
seanhalle@0:         //fprintf(output,"}\n");
seanhalle@0:         fflush(output);
seanhalle@0: 
seanhalle@0:     } else
seanhalle@0:         printf("Opening LoopGraph file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
seanhalle@0:    } else {
seanhalle@0:        printf("Could not open LoopGraph file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
seanhalle@0:    }
seanhalle@0:    
seanhalle@0:    
seanhalle@0:    freeListOfArrays(semanticEnv->unitList);
seanhalle@0:    freeListOfArrays(semanticEnv->commDependenciesList);
seanhalle@0:    freeListOfArrays(semanticEnv->ctlDependenciesList);
seanhalle@0:    freeListOfArrays(semanticEnv->dynDependenciesList);
seanhalle@0:    
seanhalle@0:    #endif
seanhalle@0: #ifdef HOLISTIC__TURN_ON_PERF_COUNTERS    
seanhalle@0:     for(n=0;n<255;n++)
seanhalle@0:     {
seanhalle@0:         sprintf(filename, "./counters/Counters.%d.csv",n);
seanhalle@0:         output = fopen(filename,"r");
seanhalle@0:         if(output)
seanhalle@0:         {
seanhalle@0:             fclose(output);
seanhalle@0:         }else{
seanhalle@0:             break;
seanhalle@0:         }
seanhalle@0:     }
seanhalle@0:     if(n<255){
seanhalle@0:     printf("Saving Counter measurements to File: %s ...\n", filename);
seanhalle@0:     output = fopen(filename,"w+");
seanhalle@0:     if(output!=NULL){
seanhalle@0:         set_counter_file(output);
seanhalle@0:         int i;
seanhalle@0:         for(i=0;i<NUM_CORES;i++){
seanhalle@0:             forAllInListOfArraysDo( semanticEnv->counterList[i], &print_counter_events_to_file );
seanhalle@0:             fflush(output);
seanhalle@0:         }
seanhalle@0: 
seanhalle@0:     } else
seanhalle@0:         printf("Opening UCC file failed. Please check that folder \"counters\" exists in run directory and has write permission.\n");
seanhalle@0:    } else {
seanhalle@0:        printf("Could not open UCC file, please clean \"counters\" folder. (Must contain less than 255 files.)\n");
seanhalle@0:    }
seanhalle@0:     
seanhalle@0: #endif
seanhalle@0: /* It's all allocated inside VMS's big chunk -- that's about to be freed, so
seanhalle@0:  *  nothing to do here
seanhalle@0:    
seanhalle@0: 
seanhalle@0:    for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
seanhalle@0:     {
seanhalle@0:       VMS_int__free( semanticEnv->readyVPQs[coreIdx]->startOfData );
seanhalle@0:       VMS_int__free( semanticEnv->readyVPQs[coreIdx] );
seanhalle@0:     }
seanhalle@0:    VMS_int__free( semanticEnv->readyVPQs );
seanhalle@0:    
seanhalle@0:    freeHashTable( semanticEnv->commHashTbl );
seanhalle@0:    VMS_int__free( _VMSMasterEnv->semanticEnv );
seanhalle@0:  */
seanhalle@0:    VMS_SS__cleanup_at_end_of_shutdown();
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: 
seanhalle@0: //===========================================================================
seanhalle@0: 
seanhalle@2: SlaveVP *
seanhalle@7: VSs__create_thread( TopLevelFnPtr fnPtr,   void *initData,
seanhalle@7:                         SlaveVP *creatingThd )
seanhalle@2:  { VSsSemReq reqData;
seanhalle@0: 
seanhalle@0:       //the semantic request data is on the stack and disappears when this
seanhalle@0:       // call returns -- it's guaranteed to remain in the VP's stack for as
seanhalle@0:       // long as the VP is suspended.
seanhalle@0:    reqData.reqType            = 0; //know type because in a VMS create req
seanhalle@0:    reqData.fnPtr              = fnPtr;
seanhalle@0:    reqData.initData           = initData;
seanhalle@7:    reqData.callingSlv         = creatingThd;
seanhalle@0: 
seanhalle@7:    VMS_WL__send_create_slaveVP_req( &reqData, creatingThd );
seanhalle@0: 
seanhalle@7:    return creatingThd->dataRetFromReq;
seanhalle@0:  }
seanhalle@0: 
seanhalle@7: /*This is always the last thing done in the code animated by a thread.
seanhalle@7:  * Normally, this would be the last line of the thread's top level function.
seanhalle@7:  * But, if the thread exits from any point, it has to do so by calling
seanhalle@7:  * this.
seanhalle@7:  * 
seanhalle@7:  *This must update the count of active sub-tasks (sub-threads) of parents,
seanhalle@8:  * and the semantic data and task stub must stay.   
seanhalle@7:  */
seanhalle@2: void
seanhalle@7: VSs__end_thread( SlaveVP *thdToEnd )
seanhalle@8:  { VSsSemData *semData;
seanhalle@8:  
seanhalle@7:       //check whether all sub-tasks have ended.. if not, don't free the
seanhalle@7:       // semantic data nor task stub of this thread.
seanhalle@8:    semData = (VSsSemData *)thdToEnd->semanticData;
seanhalle@8:    if( semData->taskStub->numLiveChildTasks != 0 )
seanhalle@8:     { 
seanhalle@8:    fix_me();   
seanhalle@8:     }
seanhalle@7:    
seanhalle@7:       //Update the count of live sub-tasks in parent.  If parent was a
seanhalle@7:       // thread and has already ended, then if this was the last sub-task,
seanhalle@7:       // free the semantic data and task stub of the parent.
seanhalle@8:    
seanhalle@7:    VMS_WL__send_dissipate_req( thdToEnd );
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: 
seanhalle@0: //===========================================================================
seanhalle@0: 
seanhalle@0: 
seanhalle@4: //======================= task submit and end ==============================
seanhalle@4: /*
seanhalle@2:  */
seanhalle@4: void
seanhalle@2: VSs__submit_task( VSsTaskType *taskType, void *args, SlaveVP *animSlv)
seanhalle@2:  { VSsSemReq  reqData;
seanhalle@0: 
seanhalle@2:    reqData.reqType    = submit_task;
seanhalle@4:    
seanhalle@2:    reqData.taskType   = taskType;
seanhalle@2:    reqData.args       = args;
seanhalle@4:    reqData.callingSlv = animSlv;
seanhalle@2:   
seanhalle@4:    reqData.taskID     = NULL;
seanhalle@2:  
seanhalle@2:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@0:  }
seanhalle@0: 
seanhalle@4: inline int32 *
seanhalle@4: VSs__create_taskID_of_size( int32 numInts, SlaveVP *animSlv )
seanhalle@4:  { int32 *taskID;
seanhalle@4:    
seanhalle@4:    taskID    = VMS_WL__malloc( sizeof(int32) + numInts * sizeof(int32) );
seanhalle@4:    taskID[0] = numInts;
seanhalle@4:    return taskID;
seanhalle@4:  }
seanhalle@4: 
seanhalle@4: void
seanhalle@4: VSs__submit_task_with_ID( VSsTaskType *taskType, void *args, int32 *taskID, 
seanhalle@4:                           SlaveVP     *animSlv)
seanhalle@4:  { VSsSemReq  reqData;
seanhalle@4: 
seanhalle@4:    reqData.reqType    = submit_task;
seanhalle@4:    
seanhalle@4:    reqData.taskType   = taskType;
seanhalle@4:    reqData.args       = args;
seanhalle@4:    reqData.taskID     = taskID;
seanhalle@4:    reqData.callingSlv = animSlv;
seanhalle@4:  
seanhalle@4:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@4:  }
seanhalle@4: 
seanhalle@4: 
seanhalle@4: /*This call is the last to happen in every task.  It causes the slave to
seanhalle@2:  * suspend and get the next task out of the task-queue.  Notice there is no
seanhalle@2:  * assigner here.. only one slave, no slave ReadyQ, and so on..
seanhalle@2:  *Can either make the assigner take the next task out of the taskQ, or can
seanhalle@2:  * leave all as it is, and make task-end take the next task.
seanhalle@2:  *Note: this fits the case in the new VMS for no-context tasks, so will use
seanhalle@2:  * the built-in taskQ of new VMS, and should be local and much faster.
seanhalle@2:  * 
seanhalle@2:  *The task-stub is saved in the animSlv, so the request handler will get it
seanhalle@2:  * from there, along with the task-type which has arg types, and so on..
seanhalle@4:  * 
seanhalle@4:  * NOTE: if want, don't need to send the animating SlaveVP around.. 
seanhalle@4:  * instead, can make a single slave per core, and coreCtrlr looks up the
seanhalle@4:  * slave from having the core number.
seanhalle@4:  * 
seanhalle@4:  *But, to stay compatible with all the other VMS languages, leave it in..
seanhalle@0:  */
seanhalle@0: void
seanhalle@2: VSs__end_task( SlaveVP *animSlv )
seanhalle@2:  { VSsSemReq  reqData;
seanhalle@0: 
seanhalle@2:    reqData.reqType      = end_task;
seanhalle@2:    reqData.callingSlv   = animSlv;
seanhalle@2:    
seanhalle@2:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@0:  }
seanhalle@0: 
seanhalle@4: 
nengel@5: void
nengel@5: VSs__taskwait(SlaveVP *animSlv)
nengel@5: {
nengel@5:     VSsSemReq  reqData;
nengel@5: 
nengel@5:    reqData.reqType      = taskwait;
nengel@5:    reqData.callingSlv   = animSlv;
nengel@5:    
nengel@5:    VMS_WL__send_sem_request( &reqData, animSlv );
nengel@5: }
nengel@5: 
nengel@5: 
nengel@5: 
seanhalle@4: //==========================  send and receive ============================
seanhalle@4: //
seanhalle@4: 
seanhalle@4: inline int32 *
seanhalle@4: VSs__give_self_taskID( SlaveVP *animSlv )
seanhalle@4:  {
seanhalle@4:    return ((VSsSemData*)animSlv->semanticData)->taskStub->taskID;
seanhalle@4:  }
seanhalle@4: 
seanhalle@4: //================================ send ===================================
seanhalle@4: 
seanhalle@4: void
seanhalle@4: VSs__send_of_type_to( void *msg, const int32 type, int32 *receiverID,
seanhalle@4:                       SlaveVP *senderSlv )
seanhalle@4:  { VSsSemReq  reqData;
seanhalle@4: 
seanhalle@4:    reqData.reqType    = send_type_to;
seanhalle@4:    
seanhalle@4:    reqData.msg        = msg;
seanhalle@4:    reqData.msgType    = type;
seanhalle@4:    reqData.receiverID = receiverID;
seanhalle@4:    reqData.senderSlv  = senderSlv;
seanhalle@4:    
seanhalle@4:    reqData.nextReqInHashEntry = NULL;
seanhalle@4: 
seanhalle@4:    VMS_WL__send_sem_request( &reqData, senderSlv );
seanhalle@4: 
seanhalle@4:       //When come back from suspend, no longer own data reachable from msg
seanhalle@4:  }
seanhalle@4: 
seanhalle@4: void
seanhalle@4: VSs__send_from_to( void *msg, int32 *senderID, int32 *receiverID, SlaveVP *senderSlv )
seanhalle@4:  { VSsSemReq  reqData;
seanhalle@4: 
seanhalle@4:    reqData.reqType     = send_from_to;
seanhalle@4:    
seanhalle@4:    reqData.msg         = msg;
seanhalle@4:    reqData.senderID    = senderID;
seanhalle@4:    reqData.receiverID  = receiverID;
seanhalle@4:    reqData.senderSlv   = senderSlv;
seanhalle@4: 
seanhalle@4:    reqData.nextReqInHashEntry = NULL;
seanhalle@4: 
seanhalle@4:    VMS_WL__send_sem_request( &reqData, senderSlv );
seanhalle@4:  }
seanhalle@4: 
seanhalle@4: 
seanhalle@4: //================================ receive ================================
seanhalle@4: 
seanhalle@4: /*The "type" version of send and receive creates a many-to-one relationship.
seanhalle@4:  * The sender is anonymous, and many sends can stack up, waiting to be
seanhalle@4:  * received.  The same receiver can also have send from-to's
seanhalle@4:  * waiting for it, and those will be kept separate from the "type"
seanhalle@4:  * messages.
seanhalle@4:  */
seanhalle@4: void *
seanhalle@4: VSs__receive_type_to( const int32 type, int32* receiverID, SlaveVP *receiverSlv )
seanhalle@4:  {       DEBUG__printf1(dbgRqstHdlr,"WL: receive type to %d",receiverID[1] );
seanhalle@4:    VSsSemReq  reqData;
seanhalle@4: 
seanhalle@4:    reqData.reqType     = receive_type_to;
seanhalle@4:    
seanhalle@4:    reqData.msgType     = type;
seanhalle@4:    reqData.receiverID  = receiverID;
seanhalle@4:    reqData.receiverSlv = receiverSlv;
seanhalle@4:    
seanhalle@4:    reqData.nextReqInHashEntry = NULL;
seanhalle@4: 
seanhalle@4:    VMS_WL__send_sem_request( &reqData, receiverSlv );
seanhalle@4:    
seanhalle@4:    return receiverSlv->dataRetFromReq;
seanhalle@4:  }
seanhalle@4: 
seanhalle@4: 
seanhalle@4: 
seanhalle@4: /*Call this at the point a receiving task wants in-coming data.
seanhalle@4:  * Use this from-to form when know senderID -- it makes a direct channel
seanhalle@4:  * between sender and receiver.
seanhalle@4:  */
seanhalle@4: void *
seanhalle@4: VSs__receive_from_to( int32 *senderID, int32 *receiverID, SlaveVP *receiverSlv )
seanhalle@4:  { 
seanhalle@4:    VSsSemReq  reqData;
seanhalle@4: 
seanhalle@4:    reqData.reqType     = receive_from_to;
seanhalle@4: 
seanhalle@4:    reqData.senderID    = senderID;
seanhalle@4:    reqData.receiverID  = receiverID;
seanhalle@4:    reqData.receiverSlv = receiverSlv;
seanhalle@4: 
seanhalle@4:    reqData.nextReqInHashEntry = NULL;
seanhalle@4:       DEBUG__printf2(dbgRqstHdlr,"WL: receive from %d to: %d", reqData.senderID[1], reqData.receiverID[1]);
seanhalle@4:       
seanhalle@4:    VMS_WL__send_sem_request( &reqData, receiverSlv );
seanhalle@4: 
seanhalle@4:    return receiverSlv->dataRetFromReq;
seanhalle@4:  }
seanhalle@4: 
seanhalle@4: 
seanhalle@4: 
seanhalle@4: 
seanhalle@2: //==========================================================================
seanhalle@0: //
seanhalle@0: /*A function singleton is a function whose body executes exactly once, on a
seanhalle@0:  * single core, no matter how many times the fuction is called and no
seanhalle@0:  * matter how many cores or the timing of cores calling it.
seanhalle@0:  *
seanhalle@0:  *A data singleton is a ticket attached to data.  That ticket can be used
seanhalle@0:  * to get the data through the function exactly once, no matter how many
seanhalle@0:  * times the data is given to the function, and no matter the timing of
seanhalle@0:  * trying to get the data through from different cores.
seanhalle@0:  */
seanhalle@0: 
seanhalle@0: /*asm function declarations*/
seanhalle@2: void asm_save_ret_to_singleton(VSsSingleton *singletonPtrAddr);
seanhalle@2: void asm_write_ret_from_singleton(VSsSingleton *singletonPtrAddr);
seanhalle@0: 
seanhalle@0: /*Fn singleton uses ID as index into array of singleton structs held in the
seanhalle@0:  * semantic environment.
seanhalle@0:  */
seanhalle@0: void
seanhalle@3: VSs__start_fn_singleton( int32 singletonID,   SlaveVP *animSlv )
seanhalle@0:  {
seanhalle@2:    VSsSemReq  reqData;
seanhalle@0: 
seanhalle@0:       //
seanhalle@0:    reqData.reqType     = singleton_fn_start;
seanhalle@0:    reqData.singletonID = singletonID;
seanhalle@0: 
seanhalle@3:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@3:    if( animSlv->dataRetFromReq ) //will be 0 or addr of label in end singleton
seanhalle@0:     {
seanhalle@3:        VSsSemEnv *semEnv = VMS_int__give_sem_env_for( animSlv );
seanhalle@0:        asm_write_ret_from_singleton(&(semEnv->fnSingletons[ singletonID]));
seanhalle@0:     }
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: /*Data singleton hands addr of loc holding a pointer to a singleton struct.
seanhalle@0:  * The start_data_singleton makes the structure and puts its addr into the
seanhalle@0:  * location.
seanhalle@0:  */
seanhalle@0: void
seanhalle@3: VSs__start_data_singleton( VSsSingleton **singletonAddr,  SlaveVP *animSlv )
seanhalle@0:  {
seanhalle@2:    VSsSemReq  reqData;
seanhalle@0: 
seanhalle@0:    if( *singletonAddr && (*singletonAddr)->hasFinished )
seanhalle@0:        goto JmpToEndSingleton;
seanhalle@0:    
seanhalle@0:    reqData.reqType          = singleton_data_start;
seanhalle@0:    reqData.singletonPtrAddr = singletonAddr;
seanhalle@0: 
seanhalle@3:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@3:    if( animSlv->dataRetFromReq ) //either 0 or end singleton's return addr
seanhalle@0:     {    //Assembly code changes the return addr on the stack to the one
seanhalle@0:          // saved into the singleton by the end-singleton-fn
seanhalle@0:          //The return addr is at 0x4(%%ebp)
seanhalle@0:         JmpToEndSingleton:
seanhalle@0:           asm_write_ret_from_singleton(*singletonAddr);
seanhalle@0:     }
seanhalle@0:    //now, simply return
seanhalle@0:    //will exit either from the start singleton call or the end-singleton call
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: /*Uses ID as index into array of flags.  If flag already set, resumes from
seanhalle@0:  * end-label.  Else, sets flag and resumes normally.
seanhalle@0:  *
seanhalle@0:  *Note, this call cannot be inlined because the instr addr at the label
seanhalle@0:  * inside is shared by all invocations of a given singleton ID.
seanhalle@0:  */
seanhalle@0: void
seanhalle@3: VSs__end_fn_singleton( int32 singletonID, SlaveVP *animSlv )
seanhalle@0:  {
seanhalle@2:    VSsSemReq  reqData;
seanhalle@0: 
seanhalle@0:       //don't need this addr until after at least one singleton has reached
seanhalle@0:       // this function
seanhalle@3:    VSsSemEnv *semEnv = VMS_int__give_sem_env_for( animSlv );
seanhalle@0:    asm_write_ret_from_singleton(&(semEnv->fnSingletons[ singletonID]));
seanhalle@0: 
seanhalle@0:    reqData.reqType     = singleton_fn_end;
seanhalle@0:    reqData.singletonID = singletonID;
seanhalle@0: 
seanhalle@3:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@0: 
seanhalle@0: EndSingletonInstrAddr:
seanhalle@0:    return;
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: void
seanhalle@3: VSs__end_data_singleton(  VSsSingleton **singletonPtrAddr, SlaveVP *animSlv )
seanhalle@0:  {
seanhalle@2:    VSsSemReq  reqData;
seanhalle@0: 
seanhalle@0:       //don't need this addr until after singleton struct has reached
seanhalle@0:       // this function for first time
seanhalle@0:       //do assembly that saves the return addr of this fn call into the
seanhalle@0:       // data singleton -- that data-singleton can only be given to exactly
seanhalle@0:       // one instance in the code of this function.  However, can use this
seanhalle@0:       // function in different places for different data-singletons.
seanhalle@0: //   (*(singletonAddr))->endInstrAddr =  &&EndDataSingletonInstrAddr;
seanhalle@0: 
seanhalle@0: 
seanhalle@0:    asm_save_ret_to_singleton(*singletonPtrAddr);
seanhalle@0: 
seanhalle@0:    reqData.reqType          = singleton_data_end;
seanhalle@0:    reqData.singletonPtrAddr = singletonPtrAddr;
seanhalle@0: 
seanhalle@3:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: /*This executes the function in the masterVP, so it executes in isolation
seanhalle@0:  * from any other copies -- only one copy of the function can ever execute
seanhalle@0:  * at a time.
seanhalle@0:  *
seanhalle@0:  *It suspends to the master, and the request handler takes the function
seanhalle@0:  * pointer out of the request and calls it, then resumes the VP.
seanhalle@0:  *Only very short functions should be called this way -- for longer-running
seanhalle@0:  * isolation, use transaction-start and transaction-end, which run the code
seanhalle@0:  * between as work-code.
seanhalle@0:  */
seanhalle@0: void
seanhalle@2: VSs__animate_short_fn_in_isolation( PtrToAtomicFn ptrToFnToExecInMaster,
seanhalle@3:                                     void *data, SlaveVP *animSlv )
seanhalle@0:  {
seanhalle@2:    VSsSemReq  reqData;
seanhalle@0: 
seanhalle@0:       //
seanhalle@0:    reqData.reqType          = atomic;
seanhalle@0:    reqData.fnToExecInMaster = ptrToFnToExecInMaster;
seanhalle@0:    reqData.dataForFn        = data;
seanhalle@0: 
seanhalle@3:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: 
seanhalle@0: /*This suspends to the master.
seanhalle@0:  *First, it looks at the VP's data, to see the highest transactionID that VP
seanhalle@0:  * already has entered.  If the current ID is not larger, it throws an
seanhalle@0:  * exception stating a bug in the code.  Otherwise it puts the current ID
seanhalle@0:  * there, and adds the ID to a linked list of IDs entered -- the list is
seanhalle@0:  * used to check that exits are properly ordered.
seanhalle@0:  *Next it is uses transactionID as index into an array of transaction
seanhalle@0:  * structures.
seanhalle@0:  *If the "VP_currently_executing" field is non-null, then put requesting VP
seanhalle@0:  * into queue in the struct.  (At some point a holder will request
seanhalle@0:  * end-transaction, which will take this VP from the queue and resume it.)
seanhalle@0:  *If NULL, then write requesting into the field and resume.
seanhalle@0:  */
seanhalle@0: void
seanhalle@3: VSs__start_transaction( int32 transactionID, SlaveVP *animSlv )
seanhalle@0:  {
seanhalle@2:    VSsSemReq  reqData;
seanhalle@0: 
seanhalle@0:       //
seanhalle@3:    reqData.callingSlv      = animSlv;
seanhalle@0:    reqData.reqType     = trans_start;
seanhalle@0:    reqData.transID     = transactionID;
seanhalle@0: 
seanhalle@3:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@0:  }
seanhalle@0: 
seanhalle@0: /*This suspends to the master, then uses transactionID as index into an
seanhalle@0:  * array of transaction structures.
seanhalle@0:  *It looks at VP_currently_executing to be sure it's same as requesting VP.
seanhalle@0:  * If different, throws an exception, stating there's a bug in the code.
seanhalle@0:  *Next it looks at the queue in the structure.
seanhalle@0:  *If it's empty, it sets VP_currently_executing field to NULL and resumes.
seanhalle@0:  *If something in, gets it, sets VP_currently_executing to that VP, then
seanhalle@0:  * resumes both.
seanhalle@0:  */
seanhalle@0: void
seanhalle@3: VSs__end_transaction( int32 transactionID, SlaveVP *animSlv )
seanhalle@0:  {
seanhalle@2:    VSsSemReq  reqData;
seanhalle@0: 
seanhalle@0:       //
seanhalle@3:    reqData.callingSlv      = animSlv;
seanhalle@0:    reqData.reqType     = trans_end;
seanhalle@0:    reqData.transID     = transactionID;
seanhalle@0: 
seanhalle@3:    VMS_WL__send_sem_request( &reqData, animSlv );
seanhalle@0:  }
seanhalle@7: 
seanhalle@7: //======================== Internal ==================================
seanhalle@7: /*
seanhalle@7:  */
seanhalle@7: SlaveVP *
seanhalle@7: VSs__create_slave_with( TopLevelFnPtr fnPtr,   void *initData,
seanhalle@7:                         SlaveVP *creatingSlv )
seanhalle@7:  { VSsSemReq reqData;
seanhalle@7: 
seanhalle@7:       //the semantic request data is on the stack and disappears when this
seanhalle@7:       // call returns -- it's guaranteed to remain in the VP's stack for as
seanhalle@7:       // long as the VP is suspended.
seanhalle@7:    reqData.reqType            = 0; //know type because in a VMS create req
seanhalle@7:    reqData.coreToAssignOnto = -1; //means round-robin assign
seanhalle@7:    reqData.fnPtr              = fnPtr;
seanhalle@7:    reqData.initData           = initData;
seanhalle@7:    reqData.callingSlv             = creatingSlv;
seanhalle@7: 
seanhalle@7:    VMS_WL__send_create_slaveVP_req( &reqData, creatingSlv );
seanhalle@7: 
seanhalle@7:    return creatingSlv->dataRetFromReq;
seanhalle@7:  }
seanhalle@7: 
seanhalle@7: SlaveVP *
seanhalle@7: VSs__create_slave_with_affinity( TopLevelFnPtr fnPtr, void *initData,
seanhalle@7:                         SlaveVP *creatingSlv,  int32  coreToAssignOnto )
seanhalle@7:  { VSsSemReq  reqData;
seanhalle@7: 
seanhalle@7:       //the semantic request data is on the stack and disappears when this
seanhalle@7:       // call returns -- it's guaranteed to remain in the VP's stack for as
seanhalle@7:       // long as the VP is suspended.
seanhalle@7:    reqData.reqType            = create_slave_w_aff; //not used, May 2012
seanhalle@7:    reqData.coreToAssignOnto   = coreToAssignOnto;
seanhalle@7:    reqData.fnPtr              = fnPtr;
seanhalle@7:    reqData.initData           = initData;
seanhalle@7:    reqData.callingSlv         = creatingSlv;
seanhalle@7: 
seanhalle@7:    VMS_WL__send_create_slaveVP_req( &reqData, creatingSlv );
seanhalle@7: 
seanhalle@7:    return creatingSlv->dataRetFromReq;
seanhalle@7:  }
seanhalle@7: