# HG changeset patch
# User Merten Sach <msach@mailbox.tu-berlin.de>
# Date 1323440892 -3600
# Node ID 662089f010bbe57e8771f393a53ffdcc1255dfea
# Parent  5d3b5e58456e325009300a03dd688333fb78d46e
Cycles measurement done with perf counter, correct command line parsing

diff -r 5d3b5e58456e -r 662089f010bb src/Application/main.c
--- a/src/Application/main.c	Wed Dec 07 06:17:46 2011 +0100
+++ b/src/Application/main.c	Fri Dec 09 15:28:12 2011 +0100
@@ -185,8 +185,8 @@
    for(o=0; o < outer_iters; o++)
     {
        
-//          saveCyclesAndInstrs(cpuid,startWorkload.cycles);
-          saveTSCLowHigh(startTask);
+          saveCyclesAndInstrs(cpuid,startWorkload.cycles);
+//          saveTSCLowHigh(startTask);
        
       //workload
       for(i=0; i < inner_iters; i++)
@@ -195,10 +195,10 @@
          workspace2 += (workspace2 + 23.2)/1.4;
        }
       
-          saveTSCLowHigh(endTask);
-          numCycles = endTask.longVal - startTask.longVal;
-//          saveCyclesAndInstrs(cpuid,endWorkload.cycles);
-//          numCycles = endWorkload.cycles - startWorkload.cycles;
+//          saveTSCLowHigh(endTask);
+//          numCycles = endTask.longVal - startTask.longVal;
+          saveCyclesAndInstrs(cpuid,endWorkload.cycles);
+          numCycles = endWorkload.cycles - startWorkload.cycles;
 
           //sanity check (400K is about 20K iters)
           if( numCycles < 400000 ) {totalWorkCycles += numCycles; numGoodTasks++;}
@@ -206,18 +206,22 @@
 
       //mutex access often causes switch to different Slave VP
       VPThread__mutex_lock(privateMutex, animatingPr);
+/*
           saveTSCLowHigh(endSync1);
           numCycles = endSync1.longVal - endTask.longVal;
           //sanity check (400K is about 20K iters)
           if( numCycles < 400000 ) {totalSyncCycles += numCycles; numGoodSyncs++;}
           else                     totalBadSyncCycles  += numCycles;
+*/
       
       VPThread__mutex_unlock(privateMutex, animatingPr);
+/*
           saveTSCLowHigh(endSync2);
           numCycles = endSync2.longVal - endSync1.longVal;
           //sanity check (400K is about 20K iters)
           if( numCycles < 400000 ) {totalSyncCycles += numCycles; numGoodSyncs++;}
           else                     totalBadSyncCycles  += numCycles;
+*/
 
     }
 
@@ -322,7 +326,6 @@
                   fprintf(stderr, "invalid number of threads specified: %d\n", num_threads);
                   return EXIT_FAILURE;
                 }
-               num_threads *= NUM_CORES;
             break;
             case 'o':
                if(!isdigit(argv[++i][0]))
@@ -451,13 +454,14 @@
 
    uint64_t totalExeCycles = endExeCycles.cycles - startExeCycles.cycles;
    totalExeCycles -= totalBadCyclesAcrossCores;
-   
+   uint64 totalOverhead = totalExeCycles - totalWorkCyclesAcrossCores;
+   int32  numSyncs = outer_iters * num_threads * 2;
+   printf("Total Execution Cycles: %lu\n", totalExeCycles);
    printf("Sum across threads of work cycles: %lu\n", totalWorkCyclesAcrossCores);
-   printf("Total Execution Cycles: %lu\n", totalExeCycles);
-   printf("Sum across threads of Sync cycles: %lu\n", totalSyncCyclesAcrossCores);
-   printf("Sum across threads of Bad Sync cycles: %lu\n", totalBadSyncCyclesAcrossCores);
+   printf("Sum across threads of bad work cycles: %lu\n", totalBadCyclesAcrossCores);
+//   printf("Sum across threads of Bad Sync cycles: %lu\n", totalBadSyncCyclesAcrossCores);
+   printf("Overhead per sync: %f\n", (double)totalOverhead / (double)numSyncs );
    printf("ExeCycles/WorkCycles Ratio %f\n", 
           (double)totalExeCycles / (double)totalWorkCyclesAcrossCores);
-
    return 0;
  }