stats.c

体系机构仿真

}
#endif /* __GNUC__ */

/* register a float statistical variable */
struct stat_stat_t *
stat_reg_float(struct stat_sdb_t *sdb,	/* stat database */
	       char *name,		/* stat variable name */
	       char *desc,		/* stat variable description */
	       float *var,		/* stat variable */
	       float init_val,		/* stat variable initial value */
	       char *format)		/* optional variable output format */
{
  struct stat_stat_t *stat;

  stat = (struct stat_stat_t *)calloc(1, sizeof(struct stat_stat_t));
  if (!stat)
    fatal("out of virtual memory");

  stat->name = mystrdup(name);
  stat->desc = mystrdup(desc);
  stat->format = format ? format : "%12.4f";
  stat->sc = sc_float;
  stat->variant.for_float.var = var;
  stat->variant.for_float.init_val = init_val;

  /* link onto SDB chain */
  add_stat(sdb, stat);

  /* initialize stat */
  *var = init_val;

  return stat;
}

/* register a double statistical variable */
struct stat_stat_t *
stat_reg_double(struct stat_sdb_t *sdb,	/* stat database */
		char *name,		/* stat variable name */
		char *desc,		/* stat variable description */
		double *var,		/* stat variable */
		double init_val,	/* stat variable initial value */
		char *format)		/* optional variable output format */
{
  struct stat_stat_t *stat;

  stat = (struct stat_stat_t *)calloc(1, sizeof(struct stat_stat_t));
  if (!stat)
    fatal("out of virtual memory");

  stat->name = mystrdup(name);
  stat->desc = mystrdup(desc);
  stat->format = format ? format : "%12.4f";
  stat->sc = sc_double;
  stat->variant.for_double.var = var;
  stat->variant.for_double.init_val = init_val;

  /* link onto SDB chain */
  add_stat(sdb, stat);

  /* initialize stat */
  *var = init_val;

  return stat;
}

/* create an array distribution (w/ fixed size buckets) in stat database SDB,
   the array distribution has ARR_SZ buckets with BUCKET_SZ indicies in each
   bucked, PF specifies the distribution components to print for optional
   format FORMAT; the indicies may be optionally replaced with the strings from
   IMAP, or the entire distribution can be printed with the optional
   user-specified print function PRINT_FN */
struct stat_stat_t *
stat_reg_dist(struct stat_sdb_t *sdb,	/* stat database */
	      char *name,		/* stat variable name */
	      char *desc,		/* stat variable description */
	      unsigned int init_val,	/* dist initial value */
	      unsigned int arr_sz,	/* array size */
	      unsigned int bucket_sz,	/* array bucket size */
	      int pf,			/* print format, use PF_* defs */
	      char *format,		/* optional variable output format */
	      char **imap,		/* optional index -> string map */
	      print_fn_t print_fn)	/* optional user print function */
{
  unsigned int i;
  struct stat_stat_t *stat;
  unsigned int *arr;

  stat = (struct stat_stat_t *)calloc(1, sizeof(struct stat_stat_t));
  if (!stat)
    fatal("out of virtual memory");

  stat->name = mystrdup(name);
  stat->desc = mystrdup(desc);
  stat->format = format ? format : NULL;
  stat->sc = sc_dist;
  stat->variant.for_dist.init_val = init_val;
  stat->variant.for_dist.arr_sz = arr_sz;
  stat->variant.for_dist.bucket_sz = bucket_sz;
  stat->variant.for_dist.pf = pf;
  stat->variant.for_dist.imap = imap;
  stat->variant.for_dist.print_fn = print_fn;
  stat->variant.for_dist.overflows = 0;

  arr = (unsigned int *)calloc(arr_sz, sizeof(unsigned int));
  if (!arr)
    fatal("out of virtual memory");
  stat->variant.for_dist.arr = arr;

  /* link onto SDB chain */
  add_stat(sdb, stat);

  /* initialize stat */
  for (i=0; i < arr_sz; i++)
    arr[i] = init_val;

  return stat;
}

/* create a sparse array distribution in stat database SDB, while the sparse
   array consumes more memory per bucket than an array distribution, it can
   efficiently map any number of indicies from 0 to 2^32-1, PF specifies the
   distribution components to print for optional format FORMAT; the indicies
   may be optionally replaced with the strings from IMAP, or the entire
   distribution can be printed with the optional user-specified print function
   PRINT_FN */
struct stat_stat_t *
stat_reg_sdist(struct stat_sdb_t *sdb,	/* stat database */
	       char *name,		/* stat variable name */
	       char *desc,		/* stat variable description */
	       unsigned int init_val,	/* dist initial value */
	       int pf,			/* print format, use PF_* defs */
	       char *format,		/* optional variable output format */
	       print_fn_t print_fn)	/* optional user print function */
{
  struct stat_stat_t *stat;
  struct bucket_t **sarr;

  stat = (struct stat_stat_t *)calloc(1, sizeof(struct stat_stat_t));
  if (!stat)
    fatal("out of virtual memory");

  stat->name = mystrdup(name);
  stat->desc = mystrdup(desc);
  stat->format = format ? format : NULL;
  stat->sc = sc_sdist;
  stat->variant.for_sdist.init_val = init_val;
  stat->variant.for_sdist.pf = pf;
  stat->variant.for_sdist.print_fn = print_fn;

  /* allocate hash table */
  sarr = (struct bucket_t **)calloc(HTAB_SZ, sizeof(struct bucket_t *));
  if (!sarr)
    fatal("out of virtual memory");
  stat->variant.for_sdist.sarr = sarr;

  /* link onto SDB chain */
  add_stat(sdb, stat);

  return stat;
}

/* add NSAMPLES to array or sparse array distribution STAT */
void
stat_add_samples(struct stat_stat_t *stat,/* stat database */
		 unsigned int index,	/* distribution index of samples */
		 int nsamples)		/* number of samples to add to dist */
{
  switch (stat->sc)
    {
    case sc_dist:
      {
	unsigned int i;

	/* compute array index */
	i = index / stat->variant.for_dist.bucket_sz;

	/* check for overflow */
	if (i >= stat->variant.for_dist.arr_sz)
	  stat->variant.for_dist.overflows += nsamples;
	else
	  stat->variant.for_dist.arr[i] += nsamples;
      }
      break;
    case sc_sdist:
      {
	struct bucket_t *bucket;

	/* find bucket */
	for (bucket = stat->variant.for_sdist.sarr[HTAB_HASH(index)];
	     bucket != NULL;
	     bucket = bucket->next)
	  {
	    if (bucket->index == index)
	      break;
	  }
	if (!bucket)
	  {
	    /* add a new sample bucket */
	    bucket = (struct bucket_t *)calloc(1, sizeof(struct bucket_t));
	    if (!bucket)
	      fatal("out of virtual memory");
	    bucket->next = stat->variant.for_sdist.sarr[HTAB_HASH(index)];
	    stat->variant.for_sdist.sarr[HTAB_HASH(index)] = bucket;
	    bucket->index = index;
	    bucket->count = stat->variant.for_sdist.init_val;
	  }
	bucket->count += nsamples;
      }
      break;
    default:
      panic("stat variable is not an array distribution");
    }
}

/* add a single sample to array or sparse array distribution STAT */
void
stat_add_sample(struct stat_stat_t *stat,/* stat variable */
		unsigned int index)	/* index of sample */
{
  stat_add_samples(stat, index, 1);
}

/* register a double statistical formula, the formula is evaluated when the
   statistic is printed, the formula expression may reference any registered
   statistical variable and, in addition, the standard operators '(', ')', '+',
   '-', '*', and '/', and literal (i.e., C-format decimal, hexidecimal, and
   octal) constants are also supported; NOTE: all terms are immediately
   converted to double values and the result is a double value, see eval.h
   for more information on formulas */
struct stat_stat_t *
stat_reg_formula(struct stat_sdb_t *sdb,/* stat database */
		 char *name,		/* stat variable name */
		 char *desc,		/* stat variable description */
		 char *formula,		/* formula expression */
		 char *format)		/* optional variable output format */
{
  struct stat_stat_t *stat;

  stat = (struct stat_stat_t *)calloc(1, sizeof(struct stat_stat_t));
  if (!stat)
    fatal("out of virtual memory");

  stat->name = mystrdup(name);
  stat->desc = mystrdup(desc);
  stat->format = format ? format : "%12.4f";
  stat->sc = sc_formula;
  stat->variant.for_formula.formula = mystrdup(formula);

  /* link onto SDB chain */
  add_stat(sdb, stat);

  return stat;
}


/* compare two indicies in a sparse array hash table, used by qsort() */
static int
compare_fn(void *p1, void *p2)
{
  struct bucket_t **pb1 = p1, **pb2 = p2;

  /* compare indices */
  if ((*pb1)->index < (*pb2)->index)
    return -1;
  else if ((*pb1)->index > (*pb2)->index)
    return 1;
  else /* ((*pb1)->index == (*pb2)->index) */
    return 0;
}

/* print an array distribution */
static void
print_dist(struct stat_stat_t *stat,	/* stat variable */
	   FILE *fd)			/* output stream */
{
  unsigned int i, bcount, imax, imin;
  double btotal, bsum, bvar, bavg, bsqsum;
  int pf = stat->variant.for_dist.pf;

  /* count and sum entries */
  bcount = 0; btotal = 0.0; bvar = 0.0; bsqsum = 0.0;
  imax = 0; imin = UINT_MAX;
  for (i=0; i<stat->variant.for_dist.arr_sz; i++)
    {
      bcount++;
      btotal += stat->variant.for_dist.arr[i];
      /* on-line variance computation, tres cool, no!?! */
      bsqsum += ((double)stat->variant.for_dist.arr[i] *
		 (double)stat->variant.for_dist.arr[i]);
      bavg = btotal / MAX((double)bcount, 1.0);
      bvar = (bsqsum - ((double)bcount * bavg * bavg)) / 
	(double)(((bcount - 1) > 0) ? (bcount - 1) : 1);
    }

  /* print header */
  fprintf(fd, "\n");
  fprintf(fd, "%-22s # %s\n", stat->name, stat->desc);
  fprintf(fd, "%s.array_size = %u\n",
	  stat->name, stat->variant.for_dist.arr_sz);
  fprintf(fd, "%s.bucket_size = %u\n",
	  stat->name, stat->variant.for_dist.bucket_sz);

  fprintf(fd, "%s.count = %u\n", stat->name, bcount);
  fprintf(fd, "%s.total = %.0f\n", stat->name, btotal);
  if (bcount > 0)
    {
      fprintf(fd, "%s.imin = %u\n", stat->name, 0U);
      fprintf(fd, "%s.imax = %u\n", stat->name, bcount);
    }
  else
    {
      fprintf(fd, "%s.imin = %d\n", stat->name, -1);
      fprintf(fd, "%s.imax = %d\n", stat->name, -1);
    }
  fprintf(fd, "%s.average = %8.4f\n", stat->name, btotal/MAX(bcount, 1.0));
  fprintf(fd, "%s.std_dev = %8.4f\n", stat->name, sqrt(bvar));
  fprintf(fd, "%s.overflows = %u\n",
	  stat->name, stat->variant.for_dist.overflows);

  fprintf(fd, "# pdf == prob dist fn, cdf == cumulative dist fn\n");
  fprintf(fd, "# %14s ", "index");
  if (pf & PF_COUNT)
    fprintf(fd, "%10s ", "count");
  if (pf & PF_PDF)
    fprintf(fd, "%6s ", "pdf");
  if (pf & PF_CDF)
    fprintf(fd, "%6s ", "cdf");
  fprintf(fd, "\n");

  fprintf(fd, "%s.start_dist\n", stat->name);

  if (bcount > 0)
    {
      /* print the array */
      bsum = 0.0;
      for (i=0; i<bcount; i++)
	{
	  bsum += (double)stat->variant.for_dist.arr[i];
	  if (stat->variant.for_dist.print_fn)
	    {
	      stat->variant.for_dist.print_fn(stat,
					      i,
					      stat->variant.for_dist.arr[i],
					      bsum,
					      btotal);
	    }
	  else
	    {
	      if (stat->format == NULL)
		{
		  if (stat->variant.for_dist.imap)
		    fprintf(fd, "%-16s ", stat->variant.for_dist.imap[i]);
		  else
		    fprintf(fd, "%16u ",
			    i * stat->variant.for_dist.bucket_sz);
		  if (pf & PF_COUNT)
		    fprintf(fd, "%10u ", stat->variant.for_dist.arr[i]);
		  if (pf & PF_PDF)
		    fprintf(fd, "%6.2f ",
			    (double)stat->variant.for_dist.arr[i] /
			    MAX(btotal, 1.0) * 100.0);
		  if (pf & PF_CDF)
		    fprintf(fd, "%6.2f ", bsum/MAX(btotal, 1.0) * 100.0);
		}
	      else
		{
		  if (pf == (PF_COUNT|PF_PDF|PF_CDF))
		    {
		      if (stat->variant.for_dist.imap)
		        fprintf(fd, stat->format,
			        stat->variant.for_dist.imap[i],
			        stat->variant.for_dist.arr[i],
			        (double)stat->variant.for_dist.arr[i] /
			        MAX(btotal, 1.0) * 100.0,
			        bsum/MAX(btotal, 1.0) * 100.0);
		      else
		        fprintf(fd, stat->format,
			        i * stat->variant.for_dist.bucket_sz,
			        stat->variant.for_dist.arr[i],
			        (double)stat->variant.for_dist.arr[i] /
			        MAX(btotal, 1.0) * 100.0,
			        bsum/MAX(btotal, 1.0) * 100.0);
		    }
		  else
		    fatal("distribution format not yet implemented");