# HG changeset patch
# User Merten Sach <msach@mailbox.tu-berlin.de>
# Date 1325873378 -3600
# Node ID a1269b1549fc2cb4180809e8df40fb948e6c38e5
# Parent  1320dd56673ace11ad8766eeec9df4875dbdaed2
added preprocessor defines to disable perf_counters for vtune mesurement

diff -r 1320dd56673a -r a1269b1549fc src/Application/main.c
--- a/src/Application/main.c	Fri Dec 16 16:40:07 2011 +0100
+++ b/src/Application/main.c	Fri Jan 06 19:09:38 2012 +0100
@@ -19,6 +19,8 @@
 #undef DEBUG
 //#define DEBUG
 
+#define MEASURE_PERF
+
 #if !defined(unix) && !defined(__unix__)
 #ifdef __MACH__
 #define unix		1
@@ -104,8 +106,10 @@
    barr->counter++;
    if(barr->counter == barr->nthreads)
     { 
+#ifdef MEASURE_PERF
       read(cycles_counter_main_fd, &(barr->endBarrierCycles.cycles), \
                 sizeof(barr->endBarrierCycles.cycles));
+#endif
        
       barr->counter = 0;
       for(i=0; i < barr->nthreads; i++)
@@ -183,8 +187,9 @@
    uint64 numCycles;
    for(o=0; o < outer_iters; o++)
     {
-       
+#ifdef MEASURE_PERF
           saveCyclesAndInstrs(cpuid,startWorkload.cycles);
+#endif
        
       //workltask
       for(i=0; i < inner_iters; i++)
@@ -192,12 +197,14 @@
          workspace1 += (workspace1 + 32)/2;
          workspace2 += (workspace2 + 23.2)/1.4;
        }
-      
+  
+#ifdef MEASURE_PERF
           saveCyclesAndInstrs(cpuid,endWorkload.cycles);
           numCycles = endWorkload.cycles - startWorkload.cycles;
           //sanity check (400K is about 20K iters)
           if( numCycles < 400000 ) {totalWorkCycles += numCycles; numGoodTasks++;}
           else                     {totalBadCycles  += numCycles; }
+#endif
 
       //mutex access often causes switch to different Slave VP
       VPThread__mutex_lock(privateMutex, animatingPr);
@@ -266,9 +273,11 @@
    measurement_t *startExeCycles, *endExeCycles;
    startExeCycles = params->startExeCycles;
    
+#ifdef MEASURE_PERF
    int nread = read(cycles_counter_main_fd, &(startExeCycles->cycles),
                 sizeof(startExeCycles->cycles));
    if(nread<0) perror("Error reading cycles counter");
+#endif
    
    //create (which starts running) all threads
    for(i=0; i<num_threads; i++)
@@ -277,8 +286,10 @@
    //wait for all threads to finish
    barrier_wait(&barr, animatingPr);
   
+#ifdef MEASURE_PERF
    //endBarrierCycles read in barrier_wait()!  Merten, email me if want to chg
    params->endExeCycles->cycles = barr.endBarrierCycles.cycles;
+#endif
    
 
 /*
@@ -358,6 +369,7 @@
     }//for
    
    
+#ifdef MEASURE_PERF
    //setup performance counters
     hw_event = malloc(sizeof(struct perf_event_attr));
     memset(hw_event,0,sizeof(struct perf_event_attr));
@@ -419,6 +431,7 @@
       fprintf(stderr,"in main ");
       perror("Failed to open cycles counter");
     }
+#endif
    
    measurement_t startExeCycles, endExeCycles;
    BenchParams *benchParams;
@@ -435,6 +448,7 @@
    //This is the transition to the VMS runtime
    VPThread__create_seed_procr_and_do_work( &benchmark, benchParams );
    
+#ifdef MEASURE_PERF
    uint64_t totalWorkCyclesAcrossCores = 0, totalBadCyclesAcrossCores = 0;
    uint64_t totalSyncCyclesAcrossCores = 0, totalBadSyncCyclesAcrossCores = 0;
    for(i=0; i<num_threads; i++){ 
@@ -459,5 +473,8 @@
    printf("Overhead per sync: %f\n", (double)totalOverhead / (double)numSyncs );
    printf("ExeCycles/WorkCycles Ratio %f\n", 
           (double)totalExeCycles / (double)totalWorkCyclesAcrossCores);
+#else
+   printf("No measurement done!\n");
+#endif
    return 0;
  }