1.1 --- a/MasterLoop.c	Sat Jul 24 08:58:47 2010 -0700
     1.2 +++ b/MasterLoop.c	Mon Jul 26 15:25:53 2010 -0700
     1.3 @@ -6,7 +6,6 @@
     1.4  
     1.5  
     1.6  
     1.7 -#include <windows.h>
     1.8  #include <stdio.h>
     1.9  #include <malloc.h>
    1.10  #include <stddef.h>
    1.11 @@ -48,11 +47,11 @@
    1.12   */
    1.13  void masterLoop( void *initData, VirtProcr *masterPr )
    1.14   { 
    1.15 -   int slotIdx, numFilled, numInFirstChunk, filledSlotIdx;
    1.16 +   int             slotIdx, numFilled, filledSlotIdx, masterHasBeenQueued;
    1.17     VirtProcr      *schedVirtPr;
    1.18     SchedSlot      *currSlot, **schedSlots, **filledSlots;
    1.19     MasterEnv      *masterEnv;
    1.20 -   CASQueueStruc  *workQ;
    1.21 +   VMSQueueStruc  *workQ;
    1.22     void           *jmpPt, *stackPtrAddr, *framePtrAddr, *stillRunningAddr;
    1.23     void           *coreLoopFramePtr, *coreLoopStackPtr, *semanticEnv;
    1.24     
    1.25 @@ -65,7 +64,26 @@
    1.26        // of setup code..
    1.27     masterPr->nextInstrPt = &&masterLoopStartPt;
    1.28  
    1.29 - 
    1.30 +      //The second time MasterVP comes out of queue, the first animation of
    1.31 +      // it hasn't written the stackPtr and framePtr yet -- but the second
    1.32 +      // animation has already had its stackPtr and framePtr set to the old
    1.33 +      // value by the coreLoop.  Fix this by writing the correct stack and
    1.34 +      // frame pointers here, at which point they're correct in the first
    1.35 +      // animation of MasterVP.
    1.36 +      //TODO: remove writing stackPtr and framePtr at the bottom, for eff
    1.37 +   stackPtrAddr      = &(masterPr->stackPtr);
    1.38 +   framePtrAddr      = &(masterPr->framePtr);
    1.39 +
    1.40 +   asm volatile("movl %0,     %%eax;  \
    1.41 +                 movl %%esp, (%%eax); \
    1.42 +                 movl %1,     %%eax;  \
    1.43 +                 movl %%ebp, (%%eax); "
    1.44 +   /* outputs */ : "=g" (stackPtrAddr), "=g" (framePtrAddr)                 \
    1.45 +   /* inputs  */ :                                                          \
    1.46 +   /* clobber */ : "memory", "%eax", "%ebx"                                 \
    1.47 +                );
    1.48 +
    1.49 +
    1.50     masterLoopStartPt:
    1.51  
    1.52        //if another reference to same Master VirtProcr still going, busy-wait
    1.53 @@ -88,10 +106,11 @@
    1.54     semanticEnv      = masterEnv->semanticEnv;
    1.55  
    1.56        //prepare for scheduling
    1.57 -   masterEnv->numFilled = 0;
    1.58 +   numFilled = 0;
    1.59 +   masterHasBeenQueued = FALSE;
    1.60  
    1.61        //Poll each slot's Done flag -- slot 0 reserved for master, start at 1
    1.62 -   for( slotIdx = 1; slotIdx < NUM_SCHED_SLOTS; slotIdx++)
    1.63 +   for( slotIdx = 0; slotIdx < NUM_SCHED_SLOTS; slotIdx++)
    1.64      {
    1.65        currSlot = schedSlots[ slotIdx ];
    1.66  
    1.67 @@ -110,50 +129,63 @@
    1.68           
    1.69           if( schedVirtPr != NULL )
    1.70            { currSlot->procrAssignedToSlot = schedVirtPr;
    1.71 -            schedVirtPr->schedSlot = currSlot;
    1.72 +            schedVirtPr->schedSlot        = currSlot;
    1.73 +            currSlot->needsProcrAssigned  = FALSE;
    1.74  
    1.75 -            filledSlots[ masterEnv->numFilled ] = currSlot;
    1.76 -            masterEnv->numFilled += 1;
    1.77 +            filledSlots[ numFilled ]      = currSlot;
    1.78 +            numFilled += 1;
    1.79  
    1.80 -            currSlot->needsProcrAssigned = FALSE;
    1.81 +            writeVMSQ( schedVirtPr, workQ );
    1.82 +            if( numFilled == masterEnv->numToPrecede )
    1.83 +             {
    1.84 +               writeVMSQ( masterEnv->masterVirtPr, workQ );
    1.85 +               masterHasBeenQueued = TRUE;
    1.86 +             }
    1.87 +
    1.88            }
    1.89         }
    1.90      }
    1.91  
    1.92 +   if( !masterHasBeenQueued )
    1.93 +    {
    1.94 +      writeVMSQ( masterEnv->masterVirtPr, workQ );
    1.95 +    }
    1.96 +
    1.97 +      //Adjust the number to precede, for next round -- assume rate of
    1.98 +      // finishing work is stable -- which is a bad assumption!  But, just
    1.99 +      // want something working for the moment, look at dynamic behavior
   1.100 +      // later
   1.101 +//TODO: look at dynamic behavior -- time-average numToPrecede or something
   1.102 +   if( numFilled < NUM_CORES - 1 )
   1.103 +    { 
   1.104 +      masterEnv->numToPrecede = 0;
   1.105 +    }
   1.106 +   else
   1.107 +    { masterEnv->numToPrecede = numFilled - NUM_CORES + 1;
   1.108 +    }
   1.109 +/*
   1.110        //put some scheduled slaves in, then Master continuation, then rest
   1.111        //Adjust position of master such that it maintains close to a fixed
   1.112        // ratio --> make NUM_CORES - 1  slots or fewer come after the master
   1.113 -   numFilled    = masterEnv->numFilled;
   1.114 -   
   1.115 -   int numPrecede = numFilled;
   1.116 -   int numFollow  = NUM_CORES - 1;
   1.117 -   
   1.118 -   if( numFilled < numFollow )
   1.119 -    { numFollow  = numFilled;
   1.120 -      numPrecede = 0;
   1.121 -    }
   1.122 -   else
   1.123 -    { numPrecede -= numFollow;
   1.124 -    }
   1.125 -
   1.126 + 
   1.127     for( filledSlotIdx = 0; filledSlotIdx < numPrecede; filledSlotIdx++)
   1.128      {
   1.129 -      writeCASQ( filledSlots[ filledSlotIdx ]->procrAssignedToSlot, workQ );
   1.130 +      writeVMSQ( filledSlots[ filledSlotIdx ]->procrAssignedToSlot, workQ );
   1.131      }
   1.132  
   1.133        //enqueue continuation of this loop
   1.134        // note that After this enqueue, continuation might sneak through
   1.135 -   writeCASQ( masterEnv->masterVirtPr, workQ );
   1.136 +   writeVMSQ( masterEnv->masterVirtPr, workQ );
   1.137  
   1.138     for( filledSlotIdx = numPrecede;
   1.139          filledSlotIdx < numFilled;
   1.140          filledSlotIdx++)
   1.141      {
   1.142 -      writeCASQ( filledSlots[ filledSlotIdx ]->procrAssignedToSlot, workQ );
   1.143 +      writeVMSQ( filledSlots[ filledSlotIdx ]->procrAssignedToSlot, workQ );
   1.144      }
   1.145  
   1.146     masterEnv->numFilled = 0;
   1.147 -   
   1.148 +*/
   1.149  
   1.150        //Save stack ptr and frame -- don't need to, take out later, but safe
   1.151        // Also, wait to set stillRunning to FALSE until just before jump, to