1.1 --- a/MasterLoop.c	Tue Aug 30 21:55:04 2011 -0700
     1.2 +++ b/MasterLoop.c	Sat Sep 03 20:41:51 2011 -0700
     1.3 @@ -49,9 +49,7 @@
     1.4   *So VMS__init just births the master virtual processor same way it births
     1.5   * all the others -- then does any extra setup needed and puts it into the
     1.6   * work queue.
     1.7 - *However means have to make masterEnv a global static volatile the same way
     1.8 - * did with readyToAnimateQ in core loop.  -- for performance, put the
     1.9 - * jump to the core loop directly in here, and have it directly jump back.
    1.10 + *However means have to make masterEnv a global static volatile.
    1.11   *
    1.12   *
    1.13   *Aug 18, 2010 -- Going to a separate MasterVP for each core, to see if this
    1.14 @@ -60,7 +58,7 @@
    1.15   *
    1.16   *So, this function is coupled to each of the MasterVPs, -- meaning this
    1.17   * function can't rely on a particular stack and frame -- each MasterVP that
    1.18 - * animates this function has a different one.
    1.19 + * animates this function has a different stack.
    1.20   *
    1.21   *At this point, the masterLoop does not write itself into the queue anymore,
    1.22   * instead, the coreLoop acquires the masterLock when it has nothing to
    1.23 @@ -96,24 +94,17 @@
    1.24        //So, setup values about stack ptr, jmp pt and all that
    1.25     //masterPr->nextInstrPt = &&masterLoopStartPt;
    1.26  
    1.27 -
    1.28 -      //Note, got rid of writing the stack and frame ptr up here, because
    1.29 -      // only one
    1.30 -      // core can ever animate a given MasterVP, so don't need to communicate
    1.31 -      // new frame and stack ptr to the MasterVP storage before a second
    1.32 -      // version of that MasterVP can get animated on a different core.
    1.33 -      //Also got rid of the busy-wait.
    1.34 -
    1.35 -   
    1.36 -   //masterLoopStartPt:
    1.37 -   //The animating materVP suspends at end of this loop, then later resumes and
    1.38 -   // comes back here 
    1.39 -   while(1){
    1.40 +      //Sept 2011
    1.41 +      //Old code jumped directly to this point, but doesn't work on x64
    1.42 +	  // So, just make this an endless loop, and do assembly function at end
    1.43 +	  // that saves its own return addr, then jumps to core_loop.
    1.44 +   while(1)
    1.45 +   {
    1.46         
    1.47     //============================= MEASUREMENT STUFF ========================
    1.48     #ifdef MEAS__TIME_MASTER
    1.49        //Total Master time includes one coreloop time -- just assume the core
    1.50 -      // loop time is same for Master as for AppVPs, even though it may be
    1.51 +      // loop time is same for Master as is for AppVPs, even though it may be
    1.52        // smaller due to higher predictability of the fixed jmp.
    1.53     saveLowTimeStampCountInto( masterPr->startMasterTSCLow );
    1.54     #endif
    1.55 @@ -122,7 +113,7 @@
    1.56     masterEnv        = (MasterEnv*)_VMSMasterEnv;
    1.57     
    1.58        //GCC may optimize so doesn't always re-define from frame-storage
    1.59 -   masterPr         = (VirtProcr*)volatileMasterPr; //on stack, to be sure after jmp
    1.60 +   masterPr         = (VirtProcr*)volatileMasterPr; //on stack, reload after jmp
    1.61     thisCoresIdx     = masterPr->coreAnimatedBy;
    1.62     readyToAnimateQ  = masterEnv->readyToAnimateQs[thisCoresIdx];
    1.63     schedSlots       = masterEnv->allSchedSlots[thisCoresIdx];
    1.64 @@ -132,15 +123,15 @@
    1.65     semanticEnv      = masterEnv->semanticEnv;
    1.66  
    1.67        //First, check for requests from other MasterVPs, and handle them
    1.68 -   if( masterEnv->requestsWaitingFor[thisCoresIdx] )
    1.69 -    { masterReqQ = masterEnv->masterReqQs[thisCoresIdx];
    1.70 -	  while( currReq = readVMSQ(masterReqQ) )
    1.71 -       { handleMasterReq( currReq, semanticEnv, masterPr );
    1.72 -       }	   
    1.73 -	}
    1.74 +   if( currReq = masterEnv->interMasterRequestsFor[thisCoresIdx] )
    1.75 +    { do
    1.76 +       { handleInterMasterReq( currReq, semanticEnv, masterPr );
    1.77 +       }
    1.78 +      while( currReq = currReq->nextReqst );
    1.79 +    }
    1.80        //Now, take care of the SlaveVPs
    1.81 -	  //Go through the slots -- if Slave there newly suspended, handle its request
    1.82 -	  // then, either way, ask assigner to fill each slot
    1.83 +      //Go through the slots -- if Slave there newly suspended, handle its request
    1.84 +      // then, either way, ask assigner to fill each slot
    1.85     numSlotsFilled = 0;
    1.86     for( slotIdx = 0; slotIdx < NUM_SCHED_SLOTS; slotIdx++)
    1.87      {
    1.88 @@ -208,14 +199,15 @@
    1.89   * master_loop, others are handed off to the plugin.
    1.90   */
    1.91  void inline
    1.92 -handleMasterReq( MasterReq *currReq, void *_semEnv, VirtProcr *masterPr )
    1.93 - {
    1.94 -   switch( currReq->reqType )
    1.95 -    { case interVMSReq:
    1.96 -            handleInterVMSReq(     (InterVMSCoreReq *)currReq, masterPr);
    1.97 +handleInterMasterReq( InterMasterReqst *currReq, void *_semEnv,
    1.98 +                                                    VirtProcr *masterPr )
    1.99 + { switch( currReq->reqType )
   1.100 +    { case destVMSCore:
   1.101 +            handleInterVMSCoreReq( (InterVMSCoreReqst *)currReq, masterPr);
   1.102           break;
   1.103 -      case interPluginReq:
   1.104 -	       (*interPluginReqHdlr)(  (InterPluginReq  *)currReq, _semEnv );
   1.105 +      case destPlugin:
   1.106 +	       (*interPluginReqHdlr)( ((InterPluginReqst  *)currReq)->pluginReq,
   1.107 +                                                                _semEnv );
   1.108           break;
   1.109        default:
   1.110           break;
   1.111 @@ -223,15 +215,16 @@
   1.112   }
   1.113  
   1.114  void inline
   1.115 -handleInterVMSReq( InterVMSCoreReq *currReq, VirtProcr *masterPr )
   1.116 +handleInterVMSReq( InterVMSCoreReqst *currReq, VirtProcr *masterPr )
   1.117   { 
   1.118     switch( currReq->reqType )
   1.119      {
   1.120 -      case transfer_free:        handleTransferFree( currReq, masterPr );
   1.121 +      case transfer_free_ptr:    handleTransferFree( currReq, masterPr );
   1.122           break;
   1.123      }
   1.124   }
   1.125   
   1.126 + 
   1.127  
   1.128  /*Work Stealing Alg -- racy one
   1.129   *This algorithm has a race condition -- the coreloops are accessing their