time_meas.h 1.94 KB
Newer Older
1 2 3
#ifdef OMP
#include <omp.h>
#endif
4 5 6
#include <unistd.h>
// global var to enable openair performance profiler 
extern int opp_enabled;
7 8 9 10 11 12 13 14 15 16 17 18

typedef struct {

  long long in;
  long long diff;
  long long max;
  int trials;
} time_stats_t;

static inline void start_meas(time_stats_t *ts) __attribute__((always_inline));
static inline void stop_meas(time_stats_t *ts) __attribute__((always_inline));

19 20
void print_meas(time_stats_t *ts, const char* name, time_stats_t * total_exec_time, time_stats_t * sf_exec_time);

21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38
#if defined(__i386__)
static inline unsigned long long rdtsc_oai(void) __attribute__((always_inline));
static inline unsigned long long rdtsc_oai(void) {
    unsigned long long int x;
    __asm__ volatile (".byte 0x0f, 0x31" : "=A" (x));
    return x;
}
#elif defined(__x86_64__)
static inline unsigned long long rdtsc_oai() __attribute__((always_inline));
static inline unsigned long long rdtsc_oai() { 
  unsigned long long a, d;
  __asm__ volatile ("rdtsc" : "=a" (a), "=d" (d));
  return (d<<32) | a;
}
#endif

static inline void start_meas(time_stats_t *ts) {

39
  if (opp_enabled) {
40

41 42 43 44 45
#ifdef OMP
    int tid;
    
    tid = omp_get_thread_num();
    if (tid==0)
46
#endif
47 48 49 50 51
      {
	ts->trials++;
	ts->in = rdtsc_oai();
      }
  }
52 53 54 55
}

static inline void stop_meas(time_stats_t *ts) {

56 57 58
  if (opp_enabled) {
    long long out = rdtsc_oai();
    
59
#ifdef OMP
60 61 62
    int tid;
    tid = omp_get_thread_num();
    if (tid==0)
63
#endif
64 65 66 67 68 69 70
      {
	ts->diff += (out-ts->in);
	if ((out-ts->in) > ts->max)
	  ts->max = out-ts->in;
	
      }
  }
71 72 73 74
}

static inline void reset_meas(time_stats_t *ts) {

75 76 77 78 79
  if (opp_enabled){
    ts->trials=0;
    ts->diff=0;
    ts->max=0;
  }
80
}
81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97

/*static inline double get_mean_meas_us(time_stats_t *ts, double cpu_freq_GHz) {

  return (double) ts->diff/ts->trials/cpu_freq_GHz/1000.0;

  }*/

static inline double get_cpu_freq_GHz(void) {

  time_stats_t ts;
  reset_meas(&ts);
  start_meas(&ts);
  sleep(1);
  stop_meas(&ts);
  return (double)ts.diff/1000000000;
}