void
spread_i4 (gfc_array_i4 *ret, const gfc_array_i4 *source,
		 const index_type along, const index_type pncopies)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type rdelta = 0;
  index_type rrank;
  index_type rs;
  GFC_INTEGER_4 *rptr;
  GFC_INTEGER_4 * restrict dest;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type srank;
  const GFC_INTEGER_4 *sptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type n;
  index_type dim;
  index_type ncopies;

  srank = GFC_DESCRIPTOR_RANK(source);

  rrank = srank + 1;
  if (rrank > GFC_MAX_DIMENSIONS)
    runtime_error ("return rank too large in spread()");

  if (along > rrank)
      runtime_error ("dim outside of rank in spread()");

  ncopies = pncopies;

  if (ret->data == NULL)
    {
      /* The front end has signalled that we need to populate the
	 return array descriptor.  */
      ret->dtype = (source->dtype & ~GFC_DTYPE_RANK_MASK) | rrank;
      dim = 0;
      rs = 1;
      for (n = 0; n < rrank; n++)
	{
	  ret->dim[n].stride = rs;
	  ret->dim[n].lbound = 0;
	  if (n == along - 1)
	    {
	      ret->dim[n].ubound = ncopies - 1;
	      rdelta = rs;
	      rs *= ncopies;
	    }
	  else
	    {
	      count[dim] = 0;
	      extent[dim] = source->dim[dim].ubound + 1
		- source->dim[dim].lbound;
	      sstride[dim] = source->dim[dim].stride;
	      rstride[dim] = rs;

	      ret->dim[n].ubound = extent[dim]-1;
	      rs *= extent[dim];
	      dim++;
	    }
	}
      ret->offset = 0;
      if (rs > 0)
        ret->data = internal_malloc_size (rs * sizeof(GFC_INTEGER_4));
      else
	{
	  ret->data = internal_malloc_size (1);
	  return;
	}
    }
  else
    {
      int zero_sized;

      zero_sized = 0;

      dim = 0;
      if (GFC_DESCRIPTOR_RANK(ret) != rrank)
	runtime_error ("rank mismatch in spread()");

      if (unlikely (compile_options.bounds_check))
	{
	  for (n = 0; n < rrank; n++)
	    {
	      index_type ret_extent;

	      ret_extent = ret->dim[n].ubound + 1 - ret->dim[n].lbound;
	      if (n == along - 1)
		{
		  rdelta = ret->dim[n].stride;

		  if (ret_extent != ncopies)
		    runtime_error("Incorrect extent in return value of SPREAD"
				  " intrinsic in dimension %ld: is %ld,"
				  " should be %ld", (long int) n+1,
				  (long int) ret_extent, (long int) ncopies);
		}
	      else
		{
		  count[dim] = 0;
		  extent[dim] = source->dim[dim].ubound + 1
		    - source->dim[dim].lbound;
		  if (ret_extent != extent[dim])
		    runtime_error("Incorrect extent in return value of SPREAD"
				  " intrinsic in dimension %ld: is %ld,"
				  " should be %ld", (long int) n+1,
				  (long int) ret_extent,
				  (long int) extent[dim]);
		    
		  if (extent[dim] <= 0)
		    zero_sized = 1;
		  sstride[dim] = source->dim[dim].stride;
		  rstride[dim] = ret->dim[n].stride;
		  dim++;
		}
	    }
	}
      else
	{
	  for (n = 0; n < rrank; n++)
	    {
	      if (n == along - 1)
		{
		  rdelta = ret->dim[n].stride;
		}
	      else
		{
		  count[dim] = 0;
		  extent[dim] = source->dim[dim].ubound + 1
		    - source->dim[dim].lbound;
		  if (extent[dim] <= 0)
		    zero_sized = 1;
		  sstride[dim] = source->dim[dim].stride;
		  rstride[dim] = ret->dim[n].stride;
		  dim++;
		}
	    }
	}

      if (zero_sized)
	return;

      if (sstride[0] == 0)
	sstride[0] = 1;
    }
  sstride0 = sstride[0];
  rstride0 = rstride[0];
  rptr = ret->data;
  sptr = source->data;

  while (sptr)
    {
      /* Spread this element.  */
      dest = rptr;
      for (n = 0; n < ncopies; n++)
        {
	  *dest = *sptr;
          dest += rdelta;
        }
      /* Advance to the next element.  */
      sptr += sstride0;
      rptr += rstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          sptr -= sstride[n] * extent[n];
          rptr -= rstride[n] * extent[n];
          n++;
          if (n >= srank)
            {
              /* Break out of the loop.  */
              sptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              sptr += sstride[n];
              rptr += rstride[n];
            }
        }
    }
}
void
unpack1 (gfc_array_char *ret, const gfc_array_char *vector,
	 const gfc_array_l4 *mask, const gfc_array_char *field)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type rs;
  char *rptr;
  /* v.* indicates the vector array.  */
  index_type vstride0;
  char *vptr;
  /* f.* indicates the field array.  */
  index_type fstride[GFC_MAX_DIMENSIONS];
  index_type fstride0;
  const char *fptr;
  /* m.* indicates the mask array.  */
  index_type mstride[GFC_MAX_DIMENSIONS];
  index_type mstride0;
  const GFC_LOGICAL_4 *mptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type n;
  index_type dim;
  index_type size;
  index_type fsize;

  size = GFC_DESCRIPTOR_SIZE (ret);
  /* A field element size of 0 actually means this is a scalar.  */
  fsize = GFC_DESCRIPTOR_SIZE (field);
  if (ret->data == NULL)
    {
      /* The front end has signalled that we need to populate the
	 return array descriptor.  */
      dim = GFC_DESCRIPTOR_RANK (mask);
      rs = 1;
      for (n = 0; n < dim; n++)
	{
	  count[n] = 0;
	  ret->dim[n].stride = rs;
	  ret->dim[n].lbound = 0;
	  ret->dim[n].ubound = mask->dim[n].ubound - mask->dim[n].lbound;
	  extent[n] = ret->dim[n].ubound + 1;
	  rstride[n] = ret->dim[n].stride * size;
	  fstride[n] = field->dim[n].stride * fsize;
	  mstride[n] = mask->dim[n].stride;
	  rs *= extent[n];
	}
      ret->base = 0;
      ret->data = internal_malloc_size (rs * size);
    }
  else
    {
      dim = GFC_DESCRIPTOR_RANK (ret);
      for (n = 0; n < dim; n++)
	{
	  count[n] = 0;
	  extent[n] = ret->dim[n].ubound + 1 - ret->dim[n].lbound;
	  rstride[n] = ret->dim[n].stride * size;
	  fstride[n] = field->dim[n].stride * fsize;
	  mstride[n] = mask->dim[n].stride;
	}
      if (rstride[0] == 0)
	rstride[0] = size;
    }
  if (fstride[0] == 0)
    fstride[0] = fsize;
  if (mstride[0] == 0)
    mstride[0] = 1;

  vstride0 = vector->dim[0].stride * size;
  if (vstride0 == 0)
    vstride0 = size;
  rstride0 = rstride[0];
  fstride0 = fstride[0];
  mstride0 = mstride[0];
  rptr = ret->data;
  fptr = field->data;
  mptr = mask->data;
  vptr = vector->data;

  /* Use the same loop for both logical types. */
  if (GFC_DESCRIPTOR_SIZE (mask) != 4)
    {
      if (GFC_DESCRIPTOR_SIZE (mask) != 8)
        runtime_error ("Funny sized logical array");
      for (n = 0; n < dim; n++)
        mstride[n] <<= 1;
      mstride0 <<= 1;
      mptr = GFOR_POINTER_L8_TO_L4 (mptr);
    }

  while (rptr)
    {
      if (*mptr)
        {
          /* From vector.  */
          memcpy (rptr, vptr, size);
          vptr += vstride0;
        }
      else
        {
          /* From field.  */
          memcpy (rptr, fptr, size);
        }
      /* Advance to the next element.  */
      rptr += rstride0;
      fptr += fstride0;
      mptr += mstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          rptr -= rstride[n] * extent[n];
          fptr -= fstride[n] * extent[n];
          mptr -= mstride[n] * extent[n];
          n++;
          if (n >= dim)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              fptr += fstride[n];
              mptr += mstride[n];
            }
        }
    }
}
Пример #3
0
static void
eoshift0 (gfc_array_char * ret, const gfc_array_char * array,
	  int shift, const char * pbound, int which, index_type size,
	  char filler)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type roffset;
  char *rptr;
  char *dest;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type soffset;
  const char *sptr;
  const char *src;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type dim;
  index_type len;
  index_type n;

  /* The compiler cannot figure out that these are set, initialize
     them to avoid warnings.  */
  len = 0;
  soffset = 0;
  roffset = 0;

  if (ret->data == NULL)
    {
      int i;

      ret->data = internal_malloc_size (size * size0 ((array_t *)array));
      ret->offset = 0;
      ret->dtype = array->dtype;
      for (i = 0; i < GFC_DESCRIPTOR_RANK (array); i++)
        {
          ret->dim[i].lbound = 0;
          ret->dim[i].ubound = array->dim[i].ubound - array->dim[i].lbound;

          if (i == 0)
            ret->dim[i].stride = 1;
          else
            ret->dim[i].stride = (ret->dim[i-1].ubound + 1) * ret->dim[i-1].stride;
        }
    }

  which = which - 1;

  extent[0] = 1;
  count[0] = 0;
  sstride[0] = -1;
  rstride[0] = -1;
  n = 0;
  for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
    {
      if (dim == which)
        {
          roffset = ret->dim[dim].stride * size;
          if (roffset == 0)
            roffset = size;
          soffset = array->dim[dim].stride * size;
          if (soffset == 0)
            soffset = size;
          len = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
        }
      else
        {
          count[n] = 0;
          extent[n] = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
          rstride[n] = ret->dim[dim].stride * size;
          sstride[n] = array->dim[dim].stride * size;
          n++;
        }
    }
  if (sstride[0] == 0)
    sstride[0] = size;
  if (rstride[0] == 0)
    rstride[0] = size;

  dim = GFC_DESCRIPTOR_RANK (array);
  rstride0 = rstride[0];
  sstride0 = sstride[0];
  rptr = ret->data;
  sptr = array->data;

  if ((shift >= 0 ? shift : -shift) > len)
    {
      shift = len;
      len = 0;
    }
  else
    {
      if (shift > 0)
	len = len - shift;
      else
	len = len + shift;
    }

  while (rptr)
    {
      /* Do the shift for this dimension.  */
      if (shift > 0)
        {
          src = &sptr[shift * soffset];
          dest = rptr;
        }
      else
        {
          src = sptr;
          dest = &rptr[-shift * roffset];
        }
      for (n = 0; n < len; n++)
        {
          memcpy (dest, src, size);
          dest += roffset;
          src += soffset;
        }
      if (shift >= 0)
        {
          n = shift;
        }
      else
        {
          dest = rptr;
          n = -shift;
        }

      if (pbound)
	while (n--)
	  {
	    memcpy (dest, pbound, size);
	    dest += roffset;
	  }
      else
	while (n--)
	  {
	    memset (dest, filler, size);
	    dest += roffset;
	  }

      /* Advance to the next section.  */
      rptr += rstride0;
      sptr += sstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          rptr -= rstride[n] * extent[n];
          sptr -= sstride[n] * extent[n];
          n++;
          if (n >= dim - 1)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              sptr += sstride[n];
            }
        }
    }
}
void
unpack0_c16 (gfc_array_c16 *ret, const gfc_array_c16 *vector,
		 const gfc_array_l1 *mask, const GFC_COMPLEX_16 *fptr)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type rs;
  GFC_COMPLEX_16 * restrict rptr;
  /* v.* indicates the vector array.  */
  index_type vstride0;
  GFC_COMPLEX_16 *vptr;
  /* Value for field, this is constant.  */
  const GFC_COMPLEX_16 fval = *fptr;
  /* m.* indicates the mask array.  */
  index_type mstride[GFC_MAX_DIMENSIONS];
  index_type mstride0;
  const GFC_LOGICAL_1 *mptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type n;
  index_type dim;

  int empty;
  int mask_kind;

  empty = 0;

  mptr = mask->data;

  /* Use the same loop for all logical types, by using GFC_LOGICAL_1
     and using shifting to address size and endian issues.  */

  mask_kind = GFC_DESCRIPTOR_SIZE (mask);

  if (mask_kind == 1 || mask_kind == 2 || mask_kind == 4 || mask_kind == 8
#ifdef HAVE_GFC_LOGICAL_16
      || mask_kind == 16
#endif
      )
    {
      /*  Do not convert a NULL pointer as we use test for NULL below.  */
      if (mptr)
	mptr = GFOR_POINTER_TO_L1 (mptr, mask_kind);
    }
  else
    runtime_error ("Funny sized logical array");

  if (ret->data == NULL)
    {
      /* The front end has signalled that we need to populate the
	 return array descriptor.  */
      dim = GFC_DESCRIPTOR_RANK (mask);
      rs = 1;
      for (n = 0; n < dim; n++)
	{
	  count[n] = 0;
	  ret->dim[n].stride = rs;
	  ret->dim[n].lbound = 0;
	  ret->dim[n].ubound = mask->dim[n].ubound - mask->dim[n].lbound;
	  extent[n] = ret->dim[n].ubound + 1;
	  empty = empty || extent[n] <= 0;
	  rstride[n] = ret->dim[n].stride;
	  mstride[n] = mask->dim[n].stride * mask_kind;
	  rs *= extent[n];
	}
      ret->offset = 0;
      ret->data = internal_malloc_size (rs * sizeof (GFC_COMPLEX_16));
    }
  else
    {
      dim = GFC_DESCRIPTOR_RANK (ret);
      for (n = 0; n < dim; n++)
	{
	  count[n] = 0;
	  extent[n] = ret->dim[n].ubound + 1 - ret->dim[n].lbound;
	  empty = empty || extent[n] <= 0;
	  rstride[n] = ret->dim[n].stride;
	  mstride[n] = mask->dim[n].stride * mask_kind;
	}
      if (rstride[0] == 0)
	rstride[0] = 1;
    }

  if (empty)
    return;

  if (mstride[0] == 0)
    mstride[0] = 1;

  vstride0 = vector->dim[0].stride;
  if (vstride0 == 0)
    vstride0 = 1;
  rstride0 = rstride[0];
  mstride0 = mstride[0];
  rptr = ret->data;
  vptr = vector->data;

  while (rptr)
    {
      if (*mptr)
        {
	  /* From vector.  */
	  *rptr = *vptr;
	  vptr += vstride0;
        }
      else
        {
	  /* From field.  */
	  *rptr = fval;
        }
      /* Advance to the next element.  */
      rptr += rstride0;
      mptr += mstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          rptr -= rstride[n] * extent[n];
          mptr -= mstride[n] * extent[n];
          n++;
          if (n >= dim)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              mptr += mstride[n];
            }
        }
    }
}
Пример #5
0
static void
cshift0 (gfc_array_char * ret, const gfc_array_char * array,
	 ssize_t shift, int which, index_type size)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type roffset;
  char *rptr;

  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type soffset;
  const char *sptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type dim;
  index_type len;
  index_type n;
  int whichloop;

  if (which < 1 || which > GFC_DESCRIPTOR_RANK (array))
    runtime_error ("Argument 'DIM' is out of range in call to 'CSHIFT'");

  which = which - 1;

  extent[0] = 1;
  count[0] = 0;
  n = 0;

  /* The values assigned here must match the cases in the inner loop.  */
  whichloop = 0;
  switch (GFC_DESCRIPTOR_TYPE (array))
    {
    case GFC_DTYPE_LOGICAL:
    case GFC_DTYPE_INTEGER:
    case GFC_DTYPE_REAL:
      if (size == sizeof (int))
	whichloop = 1;
      else if (size == sizeof (long))
	whichloop = 2;
      else if (size == sizeof (double))
	whichloop = 3;
      else if (size == sizeof (long double))
	whichloop = 4;
      break;

    case GFC_DTYPE_COMPLEX:
      if (size == sizeof (_Complex float))
	whichloop = 5;
      else if (size == sizeof (_Complex double))
	whichloop = 6;
      break;

    default:
      break;
    }

  /* Initialized for avoiding compiler warnings.  */
  roffset = size;
  soffset = size;
  len = 0;

  if (ret->data == NULL)
    {
      int i;
      index_type arraysize = size0 ((array_t *)array);

      ret->offset = 0;
      ret->dtype = array->dtype;
      for (i = 0; i < GFC_DESCRIPTOR_RANK (array); i++)
        {
          ret->dim[i].lbound = 0;
          ret->dim[i].ubound = array->dim[i].ubound - array->dim[i].lbound;

          if (i == 0)
            ret->dim[i].stride = 1;
          else
            ret->dim[i].stride = (ret->dim[i-1].ubound + 1)
				 * ret->dim[i-1].stride;
        }

      if (arraysize > 0)
	ret->data = internal_malloc_size (size * arraysize);
      else
	{
	  ret->data = internal_malloc_size (1);
	  return;
	}
    }

  for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
    {
      if (dim == which)
        {
          roffset = ret->dim[dim].stride * size;
          if (roffset == 0)
            roffset = size;
          soffset = array->dim[dim].stride * size;
          if (soffset == 0)
            soffset = size;
          len = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
        }
      else
        {
          count[n] = 0;
          extent[n] = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
          rstride[n] = ret->dim[dim].stride * size;
          sstride[n] = array->dim[dim].stride * size;
          n++;
        }
    }
  if (sstride[0] == 0)
    sstride[0] = size;
  if (rstride[0] == 0)
    rstride[0] = size;

  dim = GFC_DESCRIPTOR_RANK (array);
  rstride0 = rstride[0];
  sstride0 = sstride[0];
  rptr = ret->data;
  sptr = array->data;

  shift = shift % (ssize_t)len;
  if (shift < 0)
    shift += len;

  while (rptr)
    {
      /* Do the shift for this dimension.  */

      /* If elements are contiguous, perform the operation
	 in two block moves.  */
      if (soffset == size && roffset == size)
	{
	  size_t len1 = shift * size;
	  size_t len2 = (len - shift) * size;
	  memcpy (rptr, sptr + len1, len2);
	  memcpy (rptr + len2, sptr, len1);
	}
      else
	{
	  /* Otherwise, we'll have to perform the copy one element at
	     a time.  We can speed this up a tad for common cases of 
	     fundamental types.  */
	  switch (whichloop)
	    {
	    case 0:
	      {
		char *dest = rptr;
		const char *src = &sptr[shift * soffset];

		for (n = 0; n < len - shift; n++)
		  {
		    memcpy (dest, src, size);
		    dest += roffset;
		    src += soffset;
		  }
		for (src = sptr, n = 0; n < shift; n++)
		  {
		    memcpy (dest, src, size);
		    dest += roffset;
		    src += soffset;
		  }
	      }
	      break;

	    case 1:
	      copy_loop_int (rptr, sptr, roffset, soffset, len, shift);
	      break;

	    case 2:
	      copy_loop_long (rptr, sptr, roffset, soffset, len, shift);
	      break;

	    case 3:
	      copy_loop_double (rptr, sptr, roffset, soffset, len, shift);
	      break;

	    case 4:
	      copy_loop_ldouble (rptr, sptr, roffset, soffset, len, shift);
	      break;

	    case 5:
	      copy_loop_cfloat (rptr, sptr, roffset, soffset, len, shift);
	      break;
	      
	    case 6:
	      copy_loop_cdouble (rptr, sptr, roffset, soffset, len, shift);
	      break;

	    default:
	      abort ();
	    }
	}

      /* Advance to the next section.  */
      rptr += rstride0;
      sptr += sstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          rptr -= rstride[n] * extent[n];
          sptr -= sstride[n] * extent[n];
          n++;
          if (n >= dim - 1)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              sptr += sstride[n];
            }
        }
    }
}
Пример #6
0
void
transpose_i4 (gfc_array_i4 * ret, gfc_array_i4 * source)
{
  /* r.* indicates the return array.  */
  index_type rxstride, rystride;
  GFC_INTEGER_4 *rptr;
  /* s.* indicates the source array.  */
  index_type sxstride, systride;
  const GFC_INTEGER_4 *sptr;

  index_type xcount, ycount;
  index_type x, y;

  assert (GFC_DESCRIPTOR_RANK (source) == 2);

  if (ret->data == NULL)
    {
      assert (GFC_DESCRIPTOR_RANK (ret) == 2);
      assert (ret->dtype == source->dtype);

      ret->dim[0].lbound = 0;
      ret->dim[0].ubound = source->dim[1].ubound - source->dim[1].lbound;
      ret->dim[0].stride = 1;

      ret->dim[1].lbound = 0;
      ret->dim[1].ubound = source->dim[0].ubound - source->dim[0].lbound;
      ret->dim[1].stride = ret->dim[0].ubound+1;

      ret->data = internal_malloc_size (sizeof (GFC_INTEGER_4) * size0 ((array_t *) ret));
      ret->offset = 0;
    }

  if (ret->dim[0].stride == 0)
    ret->dim[0].stride = 1;
  if (source->dim[0].stride == 0)
    source->dim[0].stride = 1;

  sxstride = source->dim[0].stride;
  systride = source->dim[1].stride;
  xcount = source->dim[0].ubound + 1 - source->dim[0].lbound;
  ycount = source->dim[1].ubound + 1 - source->dim[1].lbound;

  rxstride = ret->dim[0].stride;
  rystride = ret->dim[1].stride;

  rptr = ret->data;
  sptr = source->data;

  for (y=0; y < ycount; y++)
    {
      for (x=0; x < xcount; x++)
        {
          *rptr = *sptr;

          sptr += sxstride;
          rptr += rystride;
        }
        sptr += systride - (sxstride * xcount);
        rptr += rxstride - (rystride * xcount);
    }
}
void
maxval_r4 (gfc_array_r4 *retarray, gfc_array_r4 *array, index_type *pdim)
{
    index_type count[GFC_MAX_DIMENSIONS];
    index_type extent[GFC_MAX_DIMENSIONS];
    index_type sstride[GFC_MAX_DIMENSIONS];
    index_type dstride[GFC_MAX_DIMENSIONS];
    GFC_REAL_4 *base;
    GFC_REAL_4 *dest;
    index_type rank;
    index_type n;
    index_type len;
    index_type delta;
    index_type dim;

    /* Make dim zero based to avoid confusion.  */
    dim = (*pdim) - 1;
    rank = GFC_DESCRIPTOR_RANK (array) - 1;

    /* TODO:  It should be a front end job to correctly set the strides.  */

    if (array->dim[0].stride == 0)
        array->dim[0].stride = 1;

    len = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
    delta = array->dim[dim].stride;

    for (n = 0; n < dim; n++)
    {
        sstride[n] = array->dim[n].stride;
        extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
    }
    for (n = dim; n < rank; n++)
    {
        sstride[n] = array->dim[n + 1].stride;
        extent[n] =
            array->dim[n + 1].ubound + 1 - array->dim[n + 1].lbound;
    }

    if (retarray->data == NULL)
    {
        for (n = 0; n < rank; n++)
        {
            retarray->dim[n].lbound = 0;
            retarray->dim[n].ubound = extent[n]-1;
            if (n == 0)
                retarray->dim[n].stride = 1;
            else
                retarray->dim[n].stride = retarray->dim[n-1].stride * extent[n-1];
        }

        retarray->data
            = internal_malloc_size (sizeof (GFC_REAL_4)
                                    * retarray->dim[rank-1].stride
                                    * extent[rank-1]);
        retarray->base = 0;
        retarray->dtype = (array->dtype & ~GFC_DTYPE_RANK_MASK) | rank;
    }
    else
    {
        if (retarray->dim[0].stride == 0)
            retarray->dim[0].stride = 1;

        if (rank != GFC_DESCRIPTOR_RANK (retarray))
            runtime_error ("rank of return array incorrect");
    }

    for (n = 0; n < rank; n++)
    {
        count[n] = 0;
        dstride[n] = retarray->dim[n].stride;
        if (extent[n] <= 0)
            len = 0;
    }

    base = array->data;
    dest = retarray->data;

    while (base)
    {
        GFC_REAL_4 *src;
        GFC_REAL_4 result;
        src = base;
        {

            result = -GFC_REAL_4_HUGE;
            if (len <= 0)
                *dest = -GFC_REAL_4_HUGE;
            else
            {
                for (n = 0; n < len; n++, src += delta)
                {

                    if (*src > result)
                        result = *src;
                }
                *dest = result;
            }
        }
        /* Advance to the next element.  */
        count[0]++;
        base += sstride[0];
        dest += dstride[0];
        n = 0;
        while (count[n] == extent[n])
        {
            /* When we get to the end of a dimension, reset it and increment
               the next dimension.  */
            count[n] = 0;
            /* We could precalculate these products, but this is a less
               frequently used path so proabably not worth it.  */
            base -= sstride[n] * extent[n];
            dest -= dstride[n] * extent[n];
            n++;
            if (n == rank)
            {
                /* Break out of the look.  */
                base = NULL;
                break;
            }
            else
            {
                count[n]++;
                base += sstride[n];
                dest += dstride[n];
            }
        }
    }
}
Пример #8
0
void
eoshift3_4 (gfc_array_char *ret, gfc_array_char *array,
		       gfc_array_i4 *h, const gfc_array_char *bound,
		       GFC_INTEGER_4 *pwhich)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type roffset;
  char *rptr;
  char *dest;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type soffset;
  const char *sptr;
  const char *src;
  /* h.* indicates the shift array.  */
  index_type hstride[GFC_MAX_DIMENSIONS];
  index_type hstride0;
  const GFC_INTEGER_4 *hptr;
  /* b.* indicates the bound array.  */
  index_type bstride[GFC_MAX_DIMENSIONS];
  index_type bstride0;
  const char *bptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type dim;
  index_type size;
  index_type len;
  index_type n;
  int which;
  GFC_INTEGER_4 sh;
  GFC_INTEGER_4 delta;

  if (pwhich)
    which = *pwhich - 1;
  else
    which = 0;

  size = GFC_DESCRIPTOR_SIZE (ret);
  if (ret->data == NULL)
    {
      int i;

      ret->data = internal_malloc_size (size * size0 ((array_t *)array));
      ret->base = 0;
      ret->dtype = array->dtype;
      for (i = 0; i < GFC_DESCRIPTOR_RANK (array); i++)
        {
          ret->dim[i].lbound = 0;
          ret->dim[i].ubound = array->dim[i].ubound - array->dim[i].lbound;

          if (i == 0)
            ret->dim[i].stride = 1;
          else
            ret->dim[i].stride = (ret->dim[i-1].ubound + 1) * ret->dim[i-1].stride;
        }
    }


  extent[0] = 1;
  count[0] = 0;
  size = GFC_DESCRIPTOR_SIZE (array);
  n = 0;
  for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
    {
      if (dim == which)
        {
          roffset = ret->dim[dim].stride * size;
          if (roffset == 0)
            roffset = size;
          soffset = array->dim[dim].stride * size;
          if (soffset == 0)
            soffset = size;
          len = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
        }
      else
        {
          count[n] = 0;
          extent[n] = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
          rstride[n] = ret->dim[dim].stride * size;
          sstride[n] = array->dim[dim].stride * size;

          hstride[n] = h->dim[n].stride;
          if (bound)
            bstride[n] = bound->dim[n].stride * size;
          else
            bstride[n] = 0;
          n++;
        }
    }
  if (sstride[0] == 0)
    sstride[0] = size;
  if (rstride[0] == 0)
    rstride[0] = size;
  if (hstride[0] == 0)
    hstride[0] = 1;
  if (bound && bstride[0] == 0)
    bstride[0] = size;

  dim = GFC_DESCRIPTOR_RANK (array);
  rstride0 = rstride[0];
  sstride0 = sstride[0];
  hstride0 = hstride[0];
  bstride0 = bstride[0];
  rptr = ret->data;
  sptr = array->data;
  hptr = h->data;
  if (bound)
    bptr = bound->data;
  else
    bptr = zeros;

  while (rptr)
    {
      /* Do the shift for this dimension.  */
      sh = *hptr;
      if (( sh >= 0 ? sh : -sh ) > len)
	{
	  delta = len;
	  sh = len;
	}
      else
	delta = (sh >= 0) ? sh: -sh;

      if (sh > 0)
        {
          src = &sptr[delta * soffset];
          dest = rptr;
        }
      else
        {
          src = sptr;
          dest = &rptr[delta * roffset];
        }
      for (n = 0; n < len - delta; n++)
        {
          memcpy (dest, src, size);
          dest += roffset;
          src += soffset;
        }
      if (sh < 0)
        dest = rptr;
      n = delta;

      while (n--)
        {
          memcpy (dest, bptr, size);
          dest += roffset;
        }

      /* Advance to the next section.  */
      rptr += rstride0;
      sptr += sstride0;
      hptr += hstride0;
      bptr += bstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          rptr -= rstride[n] * extent[n];
          sptr -= sstride[n] * extent[n];
	  hptr -= hstride[n] * extent[n];
          bptr -= bstride[n] * extent[n];
          n++;
          if (n >= dim - 1)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              sptr += sstride[n];
	      hptr += hstride[n];
              bptr += bstride[n];
            }
        }
    }
}
static void
reshape_internal (parray *ret, parray *source, shape_type *shape,
		  parray *pad, shape_type *order, index_type size)
{
  /* r.* indicates the return array.  */
  index_type rcount[GFC_MAX_DIMENSIONS];
  index_type rextent[GFC_MAX_DIMENSIONS];
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type rdim;
  index_type rsize;
  index_type rs;
  index_type rex;
  char * restrict rptr;
  /* s.* indicates the source array.  */
  index_type scount[GFC_MAX_DIMENSIONS];
  index_type sextent[GFC_MAX_DIMENSIONS];
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type sdim;
  index_type ssize;
  const char *sptr;
  /* p.* indicates the pad array.  */
  index_type pcount[GFC_MAX_DIMENSIONS];
  index_type pextent[GFC_MAX_DIMENSIONS];
  index_type pstride[GFC_MAX_DIMENSIONS];
  index_type pdim;
  index_type psize;
  const char *pptr;

  const char *src;
  int n;
  int dim;
  int sempty, pempty, shape_empty;
  index_type shape_data[GFC_MAX_DIMENSIONS];

  rdim = shape->dim[0].ubound - shape->dim[0].lbound + 1;
  if (rdim != GFC_DESCRIPTOR_RANK(ret))
    runtime_error("rank of return array incorrect in RESHAPE intrinsic");

  shape_empty = 0;

  for (n = 0; n < rdim; n++)
    {
      shape_data[n] = shape->data[n * shape->dim[0].stride];
      if (shape_data[n] <= 0)
	{
	  shape_data[n] = 0;
	  shape_empty = 1;
	}
    }

  if (ret->data == NULL)
    {
      rdim = shape->dim[0].ubound - shape->dim[0].lbound + 1;
      rs = 1;
      for (n = 0; n < rdim; n++)
	{
	  ret->dim[n].lbound = 0;
	  rex = shape_data[n];
	  ret->dim[n].ubound =  rex - 1;
	  ret->dim[n].stride = rs;
	  rs *= rex;
	}
      ret->offset = 0;
      ret->data = internal_malloc_size ( rs * size );
      ret->dtype = (source->dtype & ~GFC_DTYPE_RANK_MASK) | rdim;
    }

  if (shape_empty)
    return;

  if (pad)
    {
      pdim = GFC_DESCRIPTOR_RANK (pad);
      psize = 1;
      pempty = 0;
      for (n = 0; n < pdim; n++)
        {
          pcount[n] = 0;
          pstride[n] = pad->dim[n].stride;
          pextent[n] = pad->dim[n].ubound + 1 - pad->dim[n].lbound;
          if (pextent[n] <= 0)
	    {
	      pempty = 1;
              pextent[n] = 0;
	    }

          if (psize == pstride[n])
            psize *= pextent[n];
          else
            psize = 0;
        }
      pptr = pad->data;
    }
  else
    {
      pdim = 0;
      psize = 1;
      pempty = 1;
      pptr = NULL;
    }

  if (unlikely (compile_options.bounds_check))
    {
      index_type ret_extent, source_extent;

      rs = 1;
      for (n = 0; n < rdim; n++)
	{
	  rs *= shape_data[n];
	  ret_extent = ret->dim[n].ubound + 1 - ret->dim[n].lbound;
	  if (ret_extent != shape_data[n])
	    runtime_error("Incorrect extent in return value of RESHAPE"
			  " intrinsic in dimension %ld: is %ld,"
			  " should be %ld", (long int) n+1,
			  (long int) ret_extent, (long int) shape_data[n]);
	}

      source_extent = 1;
      sdim = GFC_DESCRIPTOR_RANK (source);
      for (n = 0; n < sdim; n++)
	{
	  index_type se;
	  se = source->dim[n].ubound + 1 - source->dim[0].lbound;
	  source_extent *= se > 0 ? se : 0;
	}

      if (rs > source_extent && (!pad || pempty))
	runtime_error("Incorrect size in SOURCE argument to RESHAPE"
		      " intrinsic: is %ld, should be %ld",
		      (long int) source_extent, (long int) rs);

      if (order)
	{
	  int seen[GFC_MAX_DIMENSIONS];
	  index_type v;

	  for (n = 0; n < rdim; n++)
	    seen[n] = 0;

	  for (n = 0; n < rdim; n++)
	    {
	      v = order->data[n * order->dim[0].stride] - 1;

	      if (v < 0 || v >= rdim)
		runtime_error("Value %ld out of range in ORDER argument"
			      " to RESHAPE intrinsic", (long int) v + 1);

	      if (seen[v] != 0)
		runtime_error("Duplicate value %ld in ORDER argument to"
			      " RESHAPE intrinsic", (long int) v + 1);
		
	      seen[v] = 1;
	    }
	}
    }

  rsize = 1;
  for (n = 0; n < rdim; n++)
    {
      if (order)
        dim = order->data[n * order->dim[0].stride] - 1;
      else
        dim = n;

      rcount[n] = 0;
      rstride[n] = ret->dim[dim].stride;
      rextent[n] = ret->dim[dim].ubound + 1 - ret->dim[dim].lbound;

      if (rextent[n] != shape_data[dim])
        runtime_error ("shape and target do not conform");

      if (rsize == rstride[n])
        rsize *= rextent[n];
      else
        rsize = 0;
      if (rextent[n] <= 0)
        return;
    }

  sdim = GFC_DESCRIPTOR_RANK (source);
  ssize = 1;
  sempty = 0;
  for (n = 0; n < sdim; n++)
    {
      scount[n] = 0;
      sstride[n] = source->dim[n].stride;
      sextent[n] = source->dim[n].ubound + 1 - source->dim[n].lbound;
      if (sextent[n] <= 0)
	{
	  sempty = 1;
	  sextent[n] = 0;
	}

      if (ssize == sstride[n])
        ssize *= sextent[n];
      else
        ssize = 0;
    }

  if (rsize != 0 && ssize != 0 && psize != 0)
    {
      rsize *= size;
      ssize *= size;
      psize *= size;
      reshape_packed (ret->data, rsize, source->data, ssize,
		      pad ? pad->data : NULL, psize);
      return;
    }
  rptr = ret->data;
  src = sptr = source->data;
  rstride0 = rstride[0] * size;
  sstride0 = sstride[0] * size;

  if (sempty && pempty)
    abort ();

  if (sempty)
    {
      /* Pretend we are using the pad array the first time around, too.  */
      src = pptr;
      sptr = pptr;
      sdim = pdim;
      for (dim = 0; dim < pdim; dim++)
	{
	  scount[dim] = pcount[dim];
	  sextent[dim] = pextent[dim];
	  sstride[dim] = pstride[dim];
	  sstride0 = pstride[0] * size;
	}
    }

  while (rptr)
    {
      /* Select between the source and pad arrays.  */
      memcpy(rptr, src, size);
      /* Advance to the next element.  */
      rptr += rstride0;
      src += sstride0;
      rcount[0]++;
      scount[0]++;

      /* Advance to the next destination element.  */
      n = 0;
      while (rcount[n] == rextent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          rcount[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          rptr -= rstride[n] * rextent[n] * size;
          n++;
          if (n == rdim)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              rcount[n]++;
              rptr += rstride[n] * size;
            }
	}

      /* Advance to the next source element.  */
      n = 0;
      while (scount[n] == sextent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          scount[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          src -= sstride[n] * sextent[n] * size;
          n++;
          if (n == sdim)
            {
              if (sptr && pad)
                {
                  /* Switch to the pad array.  */
                  sptr = NULL;
                  sdim = pdim;
                  for (dim = 0; dim < pdim; dim++)
                    {
                      scount[dim] = pcount[dim];
                      sextent[dim] = pextent[dim];
                      sstride[dim] = pstride[dim];
                      sstride0 = sstride[0] * size;
                    }
                }
              /* We now start again from the beginning of the pad array.  */
              src = pptr;
              break;
            }
          else
            {
              scount[n]++;
              src += sstride[n] * size;
            }
        }
    }
}
Пример #10
0
static void
cshift0 (gfc_array_char * ret, const gfc_array_char * array,
	 ssize_t shift, int which, index_type size)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type roffset;
  char *rptr;

  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type soffset;
  const char *sptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type dim;
  index_type len;
  index_type n;
  index_type arraysize;

  index_type type_size;

  if (which < 1 || which > GFC_DESCRIPTOR_RANK (array))
    runtime_error ("Argument 'DIM' is out of range in call to 'CSHIFT'");

  arraysize = size0 ((array_t *) array);

  if (ret->data == NULL)
    {
      int i;

      ret->offset = 0;
      ret->dtype = array->dtype;
      for (i = 0; i < GFC_DESCRIPTOR_RANK (array); i++)
        {
	  index_type ub, str;

          ub = GFC_DESCRIPTOR_EXTENT(array,i) - 1;

          if (i == 0)
            str = 1;
          else
            str = GFC_DESCRIPTOR_EXTENT(ret,i-1) *
	      GFC_DESCRIPTOR_STRIDE(ret,i-1);

	  GFC_DIMENSION_SET(ret->dim[i], 0, ub, str);
        }

      if (arraysize > 0)
	ret->data = internal_malloc_size (size * arraysize);
      else
	ret->data = internal_malloc_size (1);
    }
  else if (unlikely (compile_options.bounds_check))
    {
      bounds_equal_extents ((array_t *) ret, (array_t *) array,
				 "return value", "CSHIFT");
    }

  if (arraysize == 0)
    return;

  type_size = GFC_DTYPE_TYPE_SIZE (array);

  switch(type_size)
    {
    case GFC_DTYPE_LOGICAL_1:
    case GFC_DTYPE_INTEGER_1:
    case GFC_DTYPE_DERIVED_1:
      cshift0_i1 ((gfc_array_i1 *)ret, (gfc_array_i1 *) array, shift, which);
      return;

    case GFC_DTYPE_LOGICAL_2:
    case GFC_DTYPE_INTEGER_2:
      cshift0_i2 ((gfc_array_i2 *)ret, (gfc_array_i2 *) array, shift, which);
      return;

    case GFC_DTYPE_LOGICAL_4:
    case GFC_DTYPE_INTEGER_4:
      cshift0_i4 ((gfc_array_i4 *)ret, (gfc_array_i4 *) array, shift, which);
      return;

    case GFC_DTYPE_LOGICAL_8:
    case GFC_DTYPE_INTEGER_8:
      cshift0_i8 ((gfc_array_i8 *)ret, (gfc_array_i8 *) array, shift, which);
      return;

#ifdef HAVE_GFC_INTEGER_16
    case GFC_DTYPE_LOGICAL_16:
    case GFC_DTYPE_INTEGER_16:
      cshift0_i16 ((gfc_array_i16 *)ret, (gfc_array_i16 *) array, shift,
		   which);
      return;
#endif

    case GFC_DTYPE_REAL_4:
      cshift0_r4 ((gfc_array_r4 *)ret, (gfc_array_r4 *) array, shift, which);
      return;

    case GFC_DTYPE_REAL_8:
      cshift0_r8 ((gfc_array_r8 *)ret, (gfc_array_r8 *) array, shift, which);
      return;

/* FIXME: This here is a hack, which will have to be removed when
   the array descriptor is reworked.  Currently, we don't store the
   kind value for the type, but only the size.  Because on targets with
   __float128, we have sizeof(logn double) == sizeof(__float128),
   we cannot discriminate here and have to fall back to the generic
   handling (which is suboptimal).  */
#if !defined(GFC_REAL_16_IS_FLOAT128)
# ifdef HAVE_GFC_REAL_10
    case GFC_DTYPE_REAL_10:
      cshift0_r10 ((gfc_array_r10 *)ret, (gfc_array_r10 *) array, shift,
		   which);
      return;
# endif

# ifdef HAVE_GFC_REAL_16
    case GFC_DTYPE_REAL_16:
      cshift0_r16 ((gfc_array_r16 *)ret, (gfc_array_r16 *) array, shift,
		   which);
      return;
# endif
#endif

    case GFC_DTYPE_COMPLEX_4:
      cshift0_c4 ((gfc_array_c4 *)ret, (gfc_array_c4 *) array, shift, which);
      return;

    case GFC_DTYPE_COMPLEX_8:
      cshift0_c8 ((gfc_array_c8 *)ret, (gfc_array_c8 *) array, shift, which);
      return;

/* FIXME: This here is a hack, which will have to be removed when
   the array descriptor is reworked.  Currently, we don't store the
   kind value for the type, but only the size.  Because on targets with
   __float128, we have sizeof(logn double) == sizeof(__float128),
   we cannot discriminate here and have to fall back to the generic
   handling (which is suboptimal).  */
#if !defined(GFC_REAL_16_IS_FLOAT128)
# ifdef HAVE_GFC_COMPLEX_10
    case GFC_DTYPE_COMPLEX_10:
      cshift0_c10 ((gfc_array_c10 *)ret, (gfc_array_c10 *) array, shift,
		   which);
      return;
# endif

# ifdef HAVE_GFC_COMPLEX_16
    case GFC_DTYPE_COMPLEX_16:
      cshift0_c16 ((gfc_array_c16 *)ret, (gfc_array_c16 *) array, shift,
		   which);
      return;
# endif
#endif

    default:
      break;
    }

  switch (size)
    {
      /* Let's check the actual alignment of the data pointers.  If they
	 are suitably aligned, we can safely call the unpack functions.  */

    case sizeof (GFC_INTEGER_1):
      cshift0_i1 ((gfc_array_i1 *) ret, (gfc_array_i1 *) array, shift,
		  which);
      break;

    case sizeof (GFC_INTEGER_2):
      if (GFC_UNALIGNED_2(ret->data) || GFC_UNALIGNED_2(array->data))
	break;
      else
	{
	  cshift0_i2 ((gfc_array_i2 *) ret, (gfc_array_i2 *) array, shift,
		      which);
	  return;
	}

    case sizeof (GFC_INTEGER_4):
      if (GFC_UNALIGNED_4(ret->data) || GFC_UNALIGNED_4(array->data))
	break;
      else
	{
	  cshift0_i4 ((gfc_array_i4 *)ret, (gfc_array_i4 *) array, shift,
		      which);
	  return;
	}

    case sizeof (GFC_INTEGER_8):
      if (GFC_UNALIGNED_8(ret->data) || GFC_UNALIGNED_8(array->data))
	{
	  /* Let's try to use the complex routines.  First, a sanity
	     check that the sizes match; this should be optimized to
	     a no-op.  */
	  if (sizeof(GFC_INTEGER_8) != sizeof(GFC_COMPLEX_4))
	    break;

	  if (GFC_UNALIGNED_C4(ret->data) || GFC_UNALIGNED_C4(array->data))
	    break;

	  cshift0_c4 ((gfc_array_c4 *) ret, (gfc_array_c4 *) array, shift,
		      which);
	  return;
	}
      else
	{
	  cshift0_i8 ((gfc_array_i8 *)ret, (gfc_array_i8 *) array, shift,
		      which);
	  return;
	}

#ifdef HAVE_GFC_INTEGER_16
    case sizeof (GFC_INTEGER_16):
      if (GFC_UNALIGNED_16(ret->data) || GFC_UNALIGNED_16(array->data))
	{
	  /* Let's try to use the complex routines.  First, a sanity
	     check that the sizes match; this should be optimized to
	     a no-op.  */
	  if (sizeof(GFC_INTEGER_16) != sizeof(GFC_COMPLEX_8))
	    break;

	  if (GFC_UNALIGNED_C8(ret->data) || GFC_UNALIGNED_C8(array->data))
	    break;

	  cshift0_c8 ((gfc_array_c8 *) ret, (gfc_array_c8 *) array, shift,
		      which);
	  return;
	}
      else
	{
	  cshift0_i16 ((gfc_array_i16 *) ret, (gfc_array_i16 *) array,
		       shift, which);
	  return;
	}
#else
    case sizeof (GFC_COMPLEX_8):

      if (GFC_UNALIGNED_C8(ret->data) || GFC_UNALIGNED_C8(array->data))
	break;
      else
	{
	  cshift0_c8 ((gfc_array_c8 *) ret, (gfc_array_c8 *) array, shift,
		      which);
	  return;
	}
#endif

    default:
      break;
    }


  which = which - 1;
  sstride[0] = 0;
  rstride[0] = 0;

  extent[0] = 1;
  count[0] = 0;
  n = 0;
  /* Initialized for avoiding compiler warnings.  */
  roffset = size;
  soffset = size;
  len = 0;

  for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
    {
      if (dim == which)
        {
          roffset = GFC_DESCRIPTOR_STRIDE_BYTES(ret,dim);
          if (roffset == 0)
            roffset = size;
          soffset = GFC_DESCRIPTOR_STRIDE_BYTES(array,dim);
          if (soffset == 0)
            soffset = size;
          len = GFC_DESCRIPTOR_EXTENT(array,dim);
        }
      else
        {
          count[n] = 0;
          extent[n] = GFC_DESCRIPTOR_EXTENT(array,dim);
          rstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(ret,dim);
          sstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(array,dim);
          n++;
        }
    }
  if (sstride[0] == 0)
    sstride[0] = size;
  if (rstride[0] == 0)
    rstride[0] = size;

  dim = GFC_DESCRIPTOR_RANK (array);
  rstride0 = rstride[0];
  sstride0 = sstride[0];
  rptr = ret->data;
  sptr = array->data;

  shift = len == 0 ? 0 : shift % (ssize_t)len;
  if (shift < 0)
    shift += len;

  while (rptr)
    {
      /* Do the shift for this dimension.  */

      /* If elements are contiguous, perform the operation
	 in two block moves.  */
      if (soffset == size && roffset == size)
	{
	  size_t len1 = shift * size;
	  size_t len2 = (len - shift) * size;
	  memcpy (rptr, sptr + len1, len2);
	  memcpy (rptr + len2, sptr, len1);
	}
      else
	{
	  /* Otherwise, we'll have to perform the copy one element at
	     a time.  */
	  char *dest = rptr;
	  const char *src = &sptr[shift * soffset];

	  for (n = 0; n < len - shift; n++)
	    {
	      memcpy (dest, src, size);
	      dest += roffset;
	      src += soffset;
	    }
	  for (src = sptr, n = 0; n < shift; n++)
	    {
	      memcpy (dest, src, size);
	      dest += roffset;
	      src += soffset;
	    }
	}

      /* Advance to the next section.  */
      rptr += rstride0;
      sptr += sstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          rptr -= rstride[n] * extent[n];
          sptr -= sstride[n] * extent[n];
          n++;
          if (n >= dim - 1)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              sptr += sstride[n];
            }
        }
    }
}
Пример #11
0
void
reshape_8 (gfc_array_i8 * ret, gfc_array_i8 * source, shape_type * shape,
                      gfc_array_i8 * pad, shape_type * order)
{
  /* r.* indicates the return array.  */
  index_type rcount[GFC_MAX_DIMENSIONS];
  index_type rextent[GFC_MAX_DIMENSIONS];
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type rdim;
  index_type rsize;
  index_type rs;
  index_type rex;
  GFC_INTEGER_8 *rptr;
  /* s.* indicates the source array.  */
  index_type scount[GFC_MAX_DIMENSIONS];
  index_type sextent[GFC_MAX_DIMENSIONS];
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type sdim;
  index_type ssize;
  const GFC_INTEGER_8 *sptr;
  /* p.* indicates the pad array.  */
  index_type pcount[GFC_MAX_DIMENSIONS];
  index_type pextent[GFC_MAX_DIMENSIONS];
  index_type pstride[GFC_MAX_DIMENSIONS];
  index_type pdim;
  index_type psize;
  const GFC_INTEGER_8 *pptr;

  const GFC_INTEGER_8 *src;
  int n;
  int dim;

  if (source->dim[0].stride == 0)
    source->dim[0].stride = 1;
  if (shape->dim[0].stride == 0)
    shape->dim[0].stride = 1;
  if (pad && pad->dim[0].stride == 0)
    pad->dim[0].stride = 1;
  if (order && order->dim[0].stride == 0)
    order->dim[0].stride = 1;

  if (ret->data == NULL)
    {
      rdim = shape->dim[0].ubound - shape->dim[0].lbound + 1;
      rs = 1;
      for (n=0; n < rdim; n++)
	{
	  ret->dim[n].lbound = 0;
	  rex = shape->data[n * shape->dim[0].stride];
	  ret->dim[n].ubound =  rex - 1;
	  ret->dim[n].stride = rs;
	  rs *= rex;
	}
      ret->base = 0;
      ret->data = internal_malloc_size ( rs * sizeof (GFC_INTEGER_8));
      ret->dtype = (source->dtype & ~GFC_DTYPE_RANK_MASK) | rdim;
    }
  else
    {
      rdim = GFC_DESCRIPTOR_RANK (ret);
      if (ret->dim[0].stride == 0)
	ret->dim[0].stride = 1;
    }

  rsize = 1;
  for (n = 0; n < rdim; n++)
    {
      if (order)
        dim = order->data[n * order->dim[0].stride] - 1;
      else
        dim = n;

      rcount[n] = 0;
      rstride[n] = ret->dim[dim].stride;
      rextent[n] = ret->dim[dim].ubound + 1 - ret->dim[dim].lbound;

      if (rextent[n] != shape->data[dim * shape->dim[0].stride])
        runtime_error ("shape and target do not conform");

      if (rsize == rstride[n])
        rsize *= rextent[n];
      else
        rsize = 0;
      if (rextent[n] <= 0)
        return;
    }

  sdim = GFC_DESCRIPTOR_RANK (source);
  ssize = 1;
  for (n = 0; n < sdim; n++)
    {
      scount[n] = 0;
      sstride[n] = source->dim[n].stride;
      sextent[n] = source->dim[n].ubound + 1 - source->dim[n].lbound;
      if (sextent[n] <= 0)
        abort ();

      if (ssize == sstride[n])
        ssize *= sextent[n];
      else
        ssize = 0;
    }

  if (pad)
    {
      pdim = GFC_DESCRIPTOR_RANK (pad);
      psize = 1;
      for (n = 0; n < pdim; n++)
        {
          pcount[n] = 0;
          pstride[n] = pad->dim[n].stride;
          pextent[n] = pad->dim[n].ubound + 1 - pad->dim[n].lbound;
          if (pextent[n] <= 0)
            abort ();
          if (psize == pstride[n])
            psize *= pextent[n];
          else
            psize = 0;
        }
      pptr = pad->data;
    }
  else
    {
      pdim = 0;
      psize = 1;
      pptr = NULL;
    }

  if (rsize != 0 && ssize != 0 && psize != 0)
    {
      rsize *= sizeof (GFC_INTEGER_8);
      ssize *= sizeof (GFC_INTEGER_8);
      psize *= sizeof (GFC_INTEGER_8);
      reshape_packed ((char *)ret->data, rsize, (char *)source->data,
		      ssize, pad ? (char *)pad->data : NULL, psize);
      return;
    }
  rptr = ret->data;
  src = sptr = source->data;
  rstride0 = rstride[0];
  sstride0 = sstride[0];

  while (rptr)
    {
      /* Select between the source and pad arrays.  */
      *rptr = *src;
      /* Advance to the next element.  */
      rptr += rstride0;
      src += sstride0;
      rcount[0]++;
      scount[0]++;
      /* Advance to the next destination element.  */
      n = 0;
      while (rcount[n] == rextent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          rcount[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          rptr -= rstride[n] * rextent[n];
          n++;
          if (n == rdim)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              rcount[n]++;
              rptr += rstride[n];
            }
        }
      /* Advance to the next source element.  */
      n = 0;
      while (scount[n] == sextent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          scount[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          src -= sstride[n] * sextent[n];
          n++;
          if (n == sdim)
            {
              if (sptr && pad)
                {
                  /* Switch to the pad array.  */
                  sptr = NULL;
                  sdim = pdim;
                  for (dim = 0; dim < pdim; dim++)
                    {
                      scount[dim] = pcount[dim];
                      sextent[dim] = pextent[dim];
                      sstride[dim] = pstride[dim];
                      sstride0 = sstride[0];
                    }
                }
              /* We now start again from the beginning of the pad array.  */
              src = pptr;
              break;
            }
          else
            {
              scount[n]++;
              src += sstride[n];
            }
        }
    }
}
Пример #12
0
void
matmul_i8 (gfc_array_i8 * retarray, gfc_array_i8 * a, gfc_array_i8 * b)
{
  GFC_INTEGER_8 *abase;
  GFC_INTEGER_8 *bbase;
  GFC_INTEGER_8 *dest;

  index_type rxstride, rystride, axstride, aystride, bxstride, bystride;
  index_type x, y, n, count, xcount, ycount;

  assert (GFC_DESCRIPTOR_RANK (a) == 2
          || GFC_DESCRIPTOR_RANK (b) == 2);

/* C[xcount,ycount] = A[xcount, count] * B[count,ycount]

   Either A or B (but not both) can be rank 1:

   o One-dimensional argument A is implicitly treated as a row matrix
     dimensioned [1,count], so xcount=1.

   o One-dimensional argument B is implicitly treated as a column matrix
     dimensioned [count, 1], so ycount=1.
  */

  if (retarray->data == NULL)
    {
      if (GFC_DESCRIPTOR_RANK (a) == 1)
        {
          retarray->dim[0].lbound = 0;
          retarray->dim[0].ubound = b->dim[1].ubound - b->dim[1].lbound;
          retarray->dim[0].stride = 1;
        }
      else if (GFC_DESCRIPTOR_RANK (b) == 1)
        {
          retarray->dim[0].lbound = 0;
          retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
          retarray->dim[0].stride = 1;
        }
      else
        {
          retarray->dim[0].lbound = 0;
          retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
          retarray->dim[0].stride = 1;
          
          retarray->dim[1].lbound = 0;
          retarray->dim[1].ubound = b->dim[1].ubound - b->dim[1].lbound;
          retarray->dim[1].stride = retarray->dim[0].ubound+1;
        }
          
      retarray->data
	= internal_malloc_size (sizeof (GFC_INTEGER_8) * size0 (retarray));
      retarray->base = 0;
    }

  abase = a->data;
  bbase = b->data;
  dest = retarray->data;

  if (retarray->dim[0].stride == 0)
    retarray->dim[0].stride = 1;
  if (a->dim[0].stride == 0)
    a->dim[0].stride = 1;
  if (b->dim[0].stride == 0)
    b->dim[0].stride = 1;


  if (GFC_DESCRIPTOR_RANK (retarray) == 1)
    {
      /* One-dimensional result may be addressed in the code below
	 either as a row or a column matrix. We want both cases to
	 work. */
      rxstride = rystride = retarray->dim[0].stride;
    }
  else
    {
      rxstride = retarray->dim[0].stride;
      rystride = retarray->dim[1].stride;
    }


  if (GFC_DESCRIPTOR_RANK (a) == 1)
    {
      /* Treat it as a a row matrix A[1,count]. */
      axstride = a->dim[0].stride;
      aystride = 1;

      xcount = 1;
      count = a->dim[0].ubound + 1 - a->dim[0].lbound;
    }
  else
    {
      axstride = a->dim[0].stride;
      aystride = a->dim[1].stride;

      count = a->dim[1].ubound + 1 - a->dim[1].lbound;
      xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
    }

  assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);

  if (GFC_DESCRIPTOR_RANK (b) == 1)
    {
      /* Treat it as a column matrix B[count,1] */
      bxstride = b->dim[0].stride;

      /* bystride should never be used for 1-dimensional b.
	 in case it is we want it to cause a segfault, rather than
	 an incorrect result. */
      bystride = 0xDEADBEEF; 
      ycount = 1;
    }
  else
    {
      bxstride = b->dim[0].stride;
      bystride = b->dim[1].stride;
      ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
    }

  abase = a->data;
  bbase = b->data;
  dest = retarray->data;

  if (rxstride == 1 && axstride == 1 && bxstride == 1)
    {
      GFC_INTEGER_8 *bbase_y;
      GFC_INTEGER_8 *dest_y;
      GFC_INTEGER_8 *abase_n;
      GFC_INTEGER_8 bbase_yn;

      if (rystride == ycount)
	memset (dest, 0, (sizeof (GFC_INTEGER_8) * size0((array_t *) retarray)));
      else
	{
	  for (y = 0; y < ycount; y++)
	    for (x = 0; x < xcount; x++)
	      dest[x + y*rystride] = (GFC_INTEGER_8)0;
	}

      for (y = 0; y < ycount; y++)
	{
	  bbase_y = bbase + y*bystride;
	  dest_y = dest + y*rystride;
	  for (n = 0; n < count; n++)
	    {
	      abase_n = abase + n*aystride;
	      bbase_yn = bbase_y[n];
	      for (x = 0; x < xcount; x++)
		{
		  dest_y[x] += abase_n[x] * bbase_yn;
		}
	    }
	}
    }
  else
    {
      for (y = 0; y < ycount; y++)
	for (x = 0; x < xcount; x++)
	  dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;

      for (y = 0; y < ycount; y++)
	for (n = 0; n < count; n++)
	  for (x = 0; x < xcount; x++)
	    /* dest[x,y] += a[x,n] * b[n,y] */
	    dest[x*rxstride + y*rystride] += abase[x*axstride + n*aystride] * bbase[n*bxstride + y*bystride];
    }
}
static void
pack_s_internal (gfc_array_char *ret, const gfc_array_char *array,
		 const GFC_LOGICAL_4 *mask, const gfc_array_char *vector,
		 index_type size)
{
  /* r.* indicates the return array.  */
  index_type rstride0;
  char *rptr;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  const char *sptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type n;
  index_type dim;
  index_type ssize;
  index_type nelem;
  index_type total;

  dim = GFC_DESCRIPTOR_RANK (array);
  ssize = 1;
  for (n = 0; n < dim; n++)
    {
      count[n] = 0;
      extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
      if (extent[n] < 0)
	extent[n] = 0;

      sstride[n] = array->dim[n].stride * size;
      ssize *= extent[n];
    }
  if (sstride[0] == 0)
    sstride[0] = size;

  sstride0 = sstride[0];

  if (ssize != 0)
    sptr = array->data;
  else
    sptr = NULL;

  if (ret->data == NULL)
    {
      /* Allocate the memory for the result.  */

      if (vector != NULL)
	{
	  /* The return array will have as many elements as there are
	     in vector.  */
	  total = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
	  if (total <= 0)
	    {
	      total = 0;
	      vector = NULL;
	    }
	}
      else
	{
	  if (*mask)
	    {
	      /* The result array will have as many elements as the input
		 array.  */
	      total = extent[0];
	      for (n = 1; n < dim; n++)
		total *= extent[n];
	    }
	  else
	    /* The result array will be empty.  */
	    total = 0;
	}

      /* Setup the array descriptor.  */
      ret->dim[0].lbound = 0;
      ret->dim[0].ubound = total - 1;
      ret->dim[0].stride = 1;
      ret->offset = 0;

      if (total == 0)
	{
	  ret->data = internal_malloc_size (1);
	  return;
	}
      else
	ret->data = internal_malloc_size (size * total);
    }

  rstride0 = ret->dim[0].stride * size;
  if (rstride0 == 0)
    rstride0 = size;
  rptr = ret->data;

  /* The remaining possibilities are now:
       If MASK is .TRUE., we have to copy the source array into the
     result array. We then have to fill it up with elements from VECTOR.
       If MASK is .FALSE., we have to copy VECTOR into the result
     array. If VECTOR were not present we would have already returned.  */

  if (*mask && ssize != 0)
    {
      while (sptr)
	{
	  /* Add this element.  */
	  memcpy (rptr, sptr, size);
	  rptr += rstride0;

	  /* Advance to the next element.  */
	  sptr += sstride0;
	  count[0]++;
	  n = 0;
	  while (count[n] == extent[n])
	    {
	      /* When we get to the end of a dimension, reset it and
		 increment the next dimension.  */
	      count[n] = 0;
	      /* We could precalculate these products, but this is a
		 less frequently used path so probably not worth it.  */
	      sptr -= sstride[n] * extent[n];
	      n++;
	      if (n >= dim)
		{
		  /* Break out of the loop.  */
		  sptr = NULL;
		  break;
		}
	      else
		{
		  count[n]++;
		  sptr += sstride[n];
		}
	    }
	}
    }

  /* Add any remaining elements from VECTOR.  */
  if (vector)
    {
      n = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
      nelem = ((rptr - ret->data) / rstride0);
      if (n > nelem)
        {
          sstride0 = vector->dim[0].stride * size;
          if (sstride0 == 0)
            sstride0 = size;

          sptr = vector->data + sstride0 * nelem;
          n -= nelem;
          while (n--)
            {
              memcpy (rptr, sptr, size);
              rptr += rstride0;
              sptr += sstride0;
            }
        }
    }
}
Пример #14
0
void *
internal_pack (gfc_array_char * source)
{
  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type stride[GFC_MAX_DIMENSIONS];
  index_type stride0;
  index_type dim;
  index_type ssize;
  const char *src;
  char *dest;
  void *destptr;
  int n;
  int packed;
  index_type size;
  int type;

  if (source->dim[0].stride == 0)
    {
      source->dim[0].stride = 1;
      return source->data;
    }

  type = GFC_DESCRIPTOR_TYPE (source);
  size = GFC_DESCRIPTOR_SIZE (source);
  switch (type)
    {
    case GFC_DTYPE_INTEGER:
    case GFC_DTYPE_LOGICAL:
    case GFC_DTYPE_REAL:
      switch (size)
	{
	case 4:
	  return internal_pack_4 ((gfc_array_i4 *)source);
	  
	case 8:
	  return internal_pack_8 ((gfc_array_i8 *)source);
	}
      break;

    case GFC_DTYPE_COMPLEX:
      switch (size)
	{
	case 8:
	  return internal_pack_c4 ((gfc_array_c4 *)source);
	  
	case 16:
	  return internal_pack_c8 ((gfc_array_c8 *)source);
	}
      break;

    default:
      break;
    }

  dim = GFC_DESCRIPTOR_RANK (source);
  ssize = 1;
  packed = 1;
  for (n = 0; n < dim; n++)
    {
      count[n] = 0;
      stride[n] = source->dim[n].stride;
      extent[n] = source->dim[n].ubound + 1 - source->dim[n].lbound;
      if (extent[n] <= 0)
        {
          /* Do nothing.  */
          packed = 1;
          break;
        }

      if (ssize != stride[n])
        packed = 0;

      ssize *= extent[n];
    }

  if (packed)
    return source->data;

   /* Allocate storage for the destination.  */
  destptr = internal_malloc_size (ssize * size);
  dest = (char *)destptr;
  src = source->data;
  stride0 = stride[0] * size;

  while (src)
    {
      /* Copy the data.  */
      memcpy(dest, src, size);
      /* Advance to the next element.  */
      dest += size;
      src += stride0;
      count[0]++;
      /* Advance to the next source element.  */
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          src -= stride[n] * extent[n] * size;
          n++;
          if (n == dim)
            {
              src = NULL;
              break;
            }
          else
            {
              count[n]++;
              src += stride[n] * size;
            }
        }
    }
  return destptr;
}
void
pack_r16 (gfc_array_r16 *ret, const gfc_array_r16 *array,
	       const gfc_array_l1 *mask, const gfc_array_r16 *vector)
{
  /* r.* indicates the return array.  */
  index_type rstride0;
  GFC_REAL_16 * restrict rptr;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  const GFC_REAL_16 *sptr;
  /* m.* indicates the mask array.  */
  index_type mstride[GFC_MAX_DIMENSIONS];
  index_type mstride0;
  const GFC_LOGICAL_1 *mptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  int zero_sized;
  index_type n;
  index_type dim;
  index_type nelem;
  index_type total;
  int mask_kind;

  dim = GFC_DESCRIPTOR_RANK (array);

  mptr = mask->data;

  /* Use the same loop for all logical types, by using GFC_LOGICAL_1
     and using shifting to address size and endian issues.  */

  mask_kind = GFC_DESCRIPTOR_SIZE (mask);

  if (mask_kind == 1 || mask_kind == 2 || mask_kind == 4 || mask_kind == 8
#ifdef HAVE_GFC_LOGICAL_16
      || mask_kind == 16
#endif
      )
    {
      /*  Do not convert a NULL pointer as we use test for NULL below.  */
      if (mptr)
	mptr = GFOR_POINTER_TO_L1 (mptr, mask_kind);
    }
  else
    runtime_error ("Funny sized logical array");

  zero_sized = 0;
  for (n = 0; n < dim; n++)
    {
      count[n] = 0;
      extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
      if (extent[n] <= 0)
       zero_sized = 1;
      sstride[n] = array->dim[n].stride;
      mstride[n] = mask->dim[n].stride * mask_kind;
    }
  if (sstride[0] == 0)
    sstride[0] = 1;
  if (mstride[0] == 0)
    mstride[0] = mask_kind;

  if (zero_sized)
    sptr = NULL;
  else
    sptr = array->data;

  if (ret->data == NULL || compile_options.bounds_check)
    {
      /* Count the elements, either for allocating memory or
	 for bounds checking.  */

      if (vector != NULL)
	{
	  /* The return array will have as many
	     elements as there are in VECTOR.  */
	  total = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
	  if (total < 0)
	    {
	      total = 0;
	      vector = NULL;
	    }
	}
      else
	{
	  /* We have to count the true elements in MASK.  */

	  /* TODO: We could speed up pack easily in the case of only
	     few .TRUE. entries in MASK, by keeping track of where we
	     would be in the source array during the initial traversal
	     of MASK, and caching the pointers to those elements. Then,
	     supposed the number of elements is small enough, we would
	     only have to traverse the list, and copy those elements
	     into the result array. In the case of datatypes which fit
	     in one of the integer types we could also cache the
	     value instead of a pointer to it.
	     This approach might be bad from the point of view of
	     cache behavior in the case where our cache is not big
	     enough to hold all elements that have to be copied.  */

	  const GFC_LOGICAL_1 *m = mptr;

	  total = 0;
	  if (zero_sized)
	    m = NULL;

	  while (m)
	    {
	      /* Test this element.  */
	      if (*m)
		total++;

	      /* Advance to the next element.  */
	      m += mstride[0];
	      count[0]++;
	      n = 0;
	      while (count[n] == extent[n])
		{
		  /* When we get to the end of a dimension, reset it
		     and increment the next dimension.  */
		  count[n] = 0;
		  /* We could precalculate this product, but this is a
		     less frequently used path so probably not worth
		     it.  */
		  m -= mstride[n] * extent[n];
		  n++;
		  if (n >= dim)
		    {
		      /* Break out of the loop.  */
		      m = NULL;
		      break;
		    }
		  else
		    {
		      count[n]++;
		      m += mstride[n];
		    }
		}
	    }
	}

      if (ret->data == NULL)
	{
	  /* Setup the array descriptor.  */
	  ret->dim[0].lbound = 0;
	  ret->dim[0].ubound = total - 1;
	  ret->dim[0].stride = 1;

	  ret->offset = 0;
	  if (total == 0)
	    {
	      /* In this case, nothing remains to be done.  */
	      ret->data = internal_malloc_size (1);
	      return;
	    }
	  else
	    ret->data = internal_malloc_size (sizeof (GFC_REAL_16) * total);
	}
      else 
	{
	  /* We come here because of range checking.  */
	  index_type ret_extent;

	  ret_extent = ret->dim[0].ubound + 1 - ret->dim[0].lbound;
	  if (total != ret_extent)
	    runtime_error ("Incorrect extent in return value of PACK intrinsic;"
			   " is %ld, should be %ld", (long int) total,
			   (long int) ret_extent);
	}
    }

  rstride0 = ret->dim[0].stride;
  if (rstride0 == 0)
    rstride0 = 1;
  sstride0 = sstride[0];
  mstride0 = mstride[0];
  rptr = ret->data;

  while (sptr && mptr)
    {
      /* Test this element.  */
      if (*mptr)
        {
          /* Add it.  */
	  *rptr = *sptr;
          rptr += rstride0;
        }
      /* Advance to the next element.  */
      sptr += sstride0;
      mptr += mstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          sptr -= sstride[n] * extent[n];
          mptr -= mstride[n] * extent[n];
          n++;
          if (n >= dim)
            {
              /* Break out of the loop.  */
              sptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              sptr += sstride[n];
              mptr += mstride[n];
            }
        }
    }

  /* Add any remaining elements from VECTOR.  */
  if (vector)
    {
      n = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
      nelem = ((rptr - ret->data) / rstride0);
      if (n > nelem)
        {
          sstride0 = vector->dim[0].stride;
          if (sstride0 == 0)
            sstride0 = 1;

          sptr = vector->data + sstride0 * nelem;
          n -= nelem;
          while (n--)
            {
	      *rptr = *sptr;
              rptr += rstride0;
              sptr += sstride0;
            }
        }
    }
}
void
matmul_l8 (gfc_array_l8 * retarray, gfc_array_l4 * a, gfc_array_l4 * b)
{
  GFC_INTEGER_4 *abase;
  GFC_INTEGER_4 *bbase;
  GFC_LOGICAL_8 *dest;
  index_type rxstride;
  index_type rystride;
  index_type xcount;
  index_type ycount;
  index_type xstride;
  index_type ystride;
  index_type x;
  index_type y;

  GFC_INTEGER_4 *pa;
  GFC_INTEGER_4 *pb;
  index_type astride;
  index_type bstride;
  index_type count;
  index_type n;

  assert (GFC_DESCRIPTOR_RANK (a) == 2
          || GFC_DESCRIPTOR_RANK (b) == 2);

  if (retarray->data == NULL)
    {
      if (GFC_DESCRIPTOR_RANK (a) == 1)
        {
          retarray->dim[0].lbound = 0;
          retarray->dim[0].ubound = b->dim[1].ubound - b->dim[1].lbound;
          retarray->dim[0].stride = 1;
        }
      else if (GFC_DESCRIPTOR_RANK (b) == 1)
        {
          retarray->dim[0].lbound = 0;
          retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
          retarray->dim[0].stride = 1;
        }
      else
        {
          retarray->dim[0].lbound = 0;
          retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
          retarray->dim[0].stride = 1;
          
          retarray->dim[1].lbound = 0;
          retarray->dim[1].ubound = b->dim[1].ubound - b->dim[1].lbound;
          retarray->dim[1].stride = retarray->dim[0].ubound+1;
        }
          
      retarray->data
	= internal_malloc_size (sizeof (GFC_LOGICAL_8) * size0 ((array_t *) retarray));
      retarray->base = 0;
    }

  abase = a->data;
  if (GFC_DESCRIPTOR_SIZE (a) != 4)
    {
      assert (GFC_DESCRIPTOR_SIZE (a) == 8);
      abase = GFOR_POINTER_L8_TO_L4 (abase);
    }
  bbase = b->data;
  if (GFC_DESCRIPTOR_SIZE (b) != 4)
    {
      assert (GFC_DESCRIPTOR_SIZE (b) == 8);
      bbase = GFOR_POINTER_L8_TO_L4 (bbase);
    }
  dest = retarray->data;

  if (retarray->dim[0].stride == 0)
    retarray->dim[0].stride = 1;
  if (a->dim[0].stride == 0)
    a->dim[0].stride = 1;
  if (b->dim[0].stride == 0)
    b->dim[0].stride = 1;


  if (GFC_DESCRIPTOR_RANK (retarray) == 1)
    {
      rxstride = retarray->dim[0].stride;
      rystride = rxstride;
    }
  else
    {
      rxstride = retarray->dim[0].stride;
      rystride = retarray->dim[1].stride;
    }

  /* If we have rank 1 parameters, zero the absent stride, and set the size to
     one.  */
  if (GFC_DESCRIPTOR_RANK (a) == 1)
    {
      astride = a->dim[0].stride;
      count = a->dim[0].ubound + 1 - a->dim[0].lbound;
      xstride = 0;
      rxstride = 0;
      xcount = 1;
    }
  else
    {
      astride = a->dim[1].stride;
      count = a->dim[1].ubound + 1 - a->dim[1].lbound;
      xstride = a->dim[0].stride;
      xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
    }
  if (GFC_DESCRIPTOR_RANK (b) == 1)
    {
      bstride = b->dim[0].stride;
      assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
      ystride = 0;
      rystride = 0;
      ycount = 1;
    }
  else
    {
      bstride = b->dim[0].stride;
      assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
      ystride = b->dim[1].stride;
      ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
    }

  for (y = 0; y < ycount; y++)
    {
      for (x = 0; x < xcount; x++)
        {
          /* Do the summation for this element.  For real and integer types
             this is the same as DOT_PRODUCT.  For complex types we use do
             a*b, not conjg(a)*b.  */
          pa = abase;
          pb = bbase;
          *dest = 0;

          for (n = 0; n < count; n++)
            {
              if (*pa && *pb)
                {
                  *dest = 1;
                  break;
                }
              pa += astride;
              pb += bstride;
            }

          dest += rxstride;
          abase += xstride;
        }
      abase -= xstride * xcount;
      bbase += ystride;
      dest += rystride - (rxstride * xcount);
    }
}
Пример #17
0
void
pack_c16 (gfc_array_c16 *ret, const gfc_array_c16 *array,
	       const gfc_array_l1 *mask, const gfc_array_c16 *vector)
{
  /* r.* indicates the return array.  */
  index_type rstride0;
  GFC_COMPLEX_16 * restrict rptr;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  const GFC_COMPLEX_16 *sptr;
  /* m.* indicates the mask array.  */
  index_type mstride[GFC_MAX_DIMENSIONS];
  index_type mstride0;
  const GFC_LOGICAL_1 *mptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  int zero_sized;
  index_type n;
  index_type dim;
  index_type nelem;
  index_type total;
  int mask_kind;

  dim = GFC_DESCRIPTOR_RANK (array);

  mptr = mask->data;

  /* Use the same loop for all logical types, by using GFC_LOGICAL_1
     and using shifting to address size and endian issues.  */

  mask_kind = GFC_DESCRIPTOR_SIZE (mask);

  if (mask_kind == 1 || mask_kind == 2 || mask_kind == 4 || mask_kind == 8
#ifdef HAVE_GFC_LOGICAL_16
      || mask_kind == 16
#endif
      )
    {
      /*  Do not convert a NULL pointer as we use test for NULL below.  */
      if (mptr)
	mptr = GFOR_POINTER_TO_L1 (mptr, mask_kind);
    }
  else
    runtime_error ("Funny sized logical array");

  zero_sized = 0;
  for (n = 0; n < dim; n++)
    {
      count[n] = 0;
      extent[n] = GFC_DESCRIPTOR_EXTENT(array,n);
      if (extent[n] <= 0)
       zero_sized = 1;
      sstride[n] = GFC_DESCRIPTOR_STRIDE(array,n);
      mstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(mask,n);
    }
  if (sstride[0] == 0)
    sstride[0] = 1;
  if (mstride[0] == 0)
    mstride[0] = mask_kind;

  if (zero_sized)
    sptr = NULL;
  else
    sptr = array->data;

  if (ret->data == NULL || unlikely (compile_options.bounds_check))
    {
      /* Count the elements, either for allocating memory or
	 for bounds checking.  */

      if (vector != NULL)
	{
	  /* The return array will have as many
	     elements as there are in VECTOR.  */
	  total = GFC_DESCRIPTOR_EXTENT(vector,0);
	  if (total < 0)
	    {
	      total = 0;
	      vector = NULL;
	    }
	}
      else
        {
      	  /* We have to count the true elements in MASK.  */
	  total = count_0 (mask);
        }

      if (ret->data == NULL)
	{
	  /* Setup the array descriptor.  */
	  GFC_DIMENSION_SET(ret->dim[0], 0, total-1, 1);

	  ret->offset = 0;

	  /* internal_malloc_size allocates a single byte for zero size.  */
	  ret->data = internal_malloc_size (sizeof (GFC_COMPLEX_16) * total);

	  if (total == 0)
	    return;
	}
      else 
	{
	  /* We come here because of range checking.  */
	  index_type ret_extent;

	  ret_extent = GFC_DESCRIPTOR_EXTENT(ret,0);
	  if (total != ret_extent)
	    runtime_error ("Incorrect extent in return value of PACK intrinsic;"
			   " is %ld, should be %ld", (long int) total,
			   (long int) ret_extent);
	}
    }

  rstride0 = GFC_DESCRIPTOR_STRIDE(ret,0);
  if (rstride0 == 0)
    rstride0 = 1;
  sstride0 = sstride[0];
  mstride0 = mstride[0];
  rptr = ret->data;

  while (sptr && mptr)
    {
      /* Test this element.  */
      if (*mptr)
        {
          /* Add it.  */
	  *rptr = *sptr;
          rptr += rstride0;
        }
      /* Advance to the next element.  */
      sptr += sstride0;
      mptr += mstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          sptr -= sstride[n] * extent[n];
          mptr -= mstride[n] * extent[n];
          n++;
          if (n >= dim)
            {
              /* Break out of the loop.  */
              sptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              sptr += sstride[n];
              mptr += mstride[n];
            }
        }
    }

  /* Add any remaining elements from VECTOR.  */
  if (vector)
    {
      n = GFC_DESCRIPTOR_EXTENT(vector,0);
      nelem = ((rptr - ret->data) / rstride0);
      if (n > nelem)
        {
          sstride0 = GFC_DESCRIPTOR_STRIDE(vector,0);
          if (sstride0 == 0)
            sstride0 = 1;

          sptr = vector->data + sstride0 * nelem;
          n -= nelem;
          while (n--)
            {
	      *rptr = *sptr;
              rptr += rstride0;
              sptr += sstride0;
            }
        }
    }
}
Пример #18
0
static void
cshift1 (gfc_array_char * ret, const gfc_array_char * array,
	 const gfc_array_i4 * h, const GFC_INTEGER_4 * pwhich, index_type size)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type roffset;
  char *rptr;
  char *dest;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type soffset;
  const char *sptr;
  const char *src;
  /* h.* indicates the  array.  */
  index_type hstride[GFC_MAX_DIMENSIONS];
  index_type hstride0;
  const GFC_INTEGER_4 *hptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type dim;
  index_type len;
  index_type n;
  int which;
  GFC_INTEGER_4 sh;

  if (pwhich)
    which = *pwhich - 1;
  else
    which = 0;

  if (which < 0 || (which + 1) > GFC_DESCRIPTOR_RANK (array))
    runtime_error ("Argument 'DIM' is out of range in call to 'CSHIFT'");

  if (ret->data == NULL)
    {
      int i;

      ret->data = internal_malloc_size (size * size0 ((array_t *)array));
      ret->offset = 0;
      ret->dtype = array->dtype;
      for (i = 0; i < GFC_DESCRIPTOR_RANK (array); i++)
        {
          ret->dim[i].lbound = 0;
          ret->dim[i].ubound = array->dim[i].ubound - array->dim[i].lbound;

          if (i == 0)
            ret->dim[i].stride = 1;
          else
            ret->dim[i].stride = (ret->dim[i-1].ubound + 1) * ret->dim[i-1].stride;
        }
    }

  extent[0] = 1;
  count[0] = 0;
  n = 0;

  /* Initialized for avoiding compiler warnings.  */
  roffset = size;
  soffset = size;
  len = 0;

  for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
    {
      if (dim == which)
        {
          roffset = ret->dim[dim].stride * size;
          if (roffset == 0)
            roffset = size;
          soffset = array->dim[dim].stride * size;
          if (soffset == 0)
            soffset = size;
          len = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
        }
      else
        {
          count[n] = 0;
          extent[n] = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
          rstride[n] = ret->dim[dim].stride * size;
          sstride[n] = array->dim[dim].stride * size;

          hstride[n] = h->dim[n].stride;
          n++;
        }
    }
  if (sstride[0] == 0)
    sstride[0] = size;
  if (rstride[0] == 0)
    rstride[0] = size;
  if (hstride[0] == 0)
    hstride[0] = 1;

  dim = GFC_DESCRIPTOR_RANK (array);
  rstride0 = rstride[0];
  sstride0 = sstride[0];
  hstride0 = hstride[0];
  rptr = ret->data;
  sptr = array->data;
  hptr = h->data;

  while (rptr)
    {
      /* Do the  for this dimension.  */
      sh = *hptr;
      sh = (div (sh, len)).rem;
      if (sh < 0)
        sh += len;

      src = &sptr[sh * soffset];
      dest = rptr;

      for (n = 0; n < len; n++)
        {
          memcpy (dest, src, size);
          dest += roffset;
          if (n == len - sh - 1)
            src = sptr;
          else
            src += soffset;
        }

      /* Advance to the next section.  */
      rptr += rstride0;
      sptr += sstride0;
      hptr += hstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          rptr -= rstride[n] * extent[n];
          sptr -= sstride[n] * extent[n];
	  hptr -= hstride[n] * extent[n];
          n++;
          if (n >= dim - 1)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              sptr += sstride[n];
	      hptr += hstride[n];
            }
        }
    }
}
void
mmaxval_r4 (gfc_array_r4 * retarray, gfc_array_r4 * array,
            index_type *pdim, gfc_array_l4 * mask)
{
    index_type count[GFC_MAX_DIMENSIONS];
    index_type extent[GFC_MAX_DIMENSIONS];
    index_type sstride[GFC_MAX_DIMENSIONS];
    index_type dstride[GFC_MAX_DIMENSIONS];
    index_type mstride[GFC_MAX_DIMENSIONS];
    GFC_REAL_4 *dest;
    GFC_REAL_4 *base;
    GFC_LOGICAL_4 *mbase;
    int rank;
    int dim;
    index_type n;
    index_type len;
    index_type delta;
    index_type mdelta;

    dim = (*pdim) - 1;
    rank = GFC_DESCRIPTOR_RANK (array) - 1;

    /* TODO:  It should be a front end job to correctly set the strides.  */

    if (array->dim[0].stride == 0)
        array->dim[0].stride = 1;

    if (mask->dim[0].stride == 0)
        mask->dim[0].stride = 1;

    len = array->dim[dim].ubound + 1 - array->dim[dim].lbound;
    if (len <= 0)
        return;
    delta = array->dim[dim].stride;
    mdelta = mask->dim[dim].stride;

    for (n = 0; n < dim; n++)
    {
        sstride[n] = array->dim[n].stride;
        mstride[n] = mask->dim[n].stride;
        extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
    }
    for (n = dim; n < rank; n++)
    {
        sstride[n] = array->dim[n + 1].stride;
        mstride[n] = mask->dim[n + 1].stride;
        extent[n] =
            array->dim[n + 1].ubound + 1 - array->dim[n + 1].lbound;
    }

    if (retarray->data == NULL)
    {
        for (n = 0; n < rank; n++)
        {
            retarray->dim[n].lbound = 0;
            retarray->dim[n].ubound = extent[n]-1;
            if (n == 0)
                retarray->dim[n].stride = 1;
            else
                retarray->dim[n].stride = retarray->dim[n-1].stride * extent[n-1];
        }

        retarray->data
            = internal_malloc_size (sizeof (GFC_REAL_4)
                                    * retarray->dim[rank-1].stride
                                    * extent[rank-1]);
        retarray->base = 0;
        retarray->dtype = (array->dtype & ~GFC_DTYPE_RANK_MASK) | rank;
    }
    else
    {
        if (retarray->dim[0].stride == 0)
            retarray->dim[0].stride = 1;

        if (rank != GFC_DESCRIPTOR_RANK (retarray))
            runtime_error ("rank of return array incorrect");
    }

    for (n = 0; n < rank; n++)
    {
        count[n] = 0;
        dstride[n] = retarray->dim[n].stride;
        if (extent[n] <= 0)
            return;
    }

    dest = retarray->data;
    base = array->data;
    mbase = mask->data;

    if (GFC_DESCRIPTOR_SIZE (mask) != 4)
    {
        /* This allows the same loop to be used for all logical types.  */
        assert (GFC_DESCRIPTOR_SIZE (mask) == 8);
        for (n = 0; n < rank; n++)
            mstride[n] <<= 1;
        mdelta <<= 1;
        mbase = (GFOR_POINTER_L8_TO_L4 (mbase));
    }

    while (base)
    {
        GFC_REAL_4 *src;
        GFC_LOGICAL_4 *msrc;
        GFC_REAL_4 result;
        src = base;
        msrc = mbase;
        {

            result = -GFC_REAL_4_HUGE;
            if (len <= 0)
                *dest = -GFC_REAL_4_HUGE;
            else
            {
                for (n = 0; n < len; n++, src += delta, msrc += mdelta)
                {

                    if (*msrc && *src > result)
                        result = *src;
                }
                *dest = result;
            }
        }
        /* Advance to the next element.  */
        count[0]++;
        base += sstride[0];
        mbase += mstride[0];
        dest += dstride[0];
        n = 0;
        while (count[n] == extent[n])
        {
            /* When we get to the end of a dimension, reset it and increment
               the next dimension.  */
            count[n] = 0;
            /* We could precalculate these products, but this is a less
               frequently used path so proabably not worth it.  */
            base -= sstride[n] * extent[n];
            mbase -= mstride[n] * extent[n];
            dest -= dstride[n] * extent[n];
            n++;
            if (n == rank)
            {
                /* Break out of the look.  */
                base = NULL;
                break;
            }
            else
            {
                count[n]++;
                base += sstride[n];
                mbase += mstride[n];
                dest += dstride[n];
            }
        }
    }
}
void
transpose (gfc_array_char *ret, gfc_array_char *source)
{
  /* r.* indicates the return array.  */
  index_type rxstride, rystride;
  char *rptr;
  /* s.* indicates the source array.  */
  index_type sxstride, systride;
  const char *sptr;

  index_type xcount, ycount;
  index_type x, y;
  index_type size;

  assert (GFC_DESCRIPTOR_RANK (source) == 2
          && GFC_DESCRIPTOR_RANK (ret) == 2);

  size = GFC_DESCRIPTOR_SIZE (source);

  if (ret->data == NULL)
    {
      assert (ret->dtype == source->dtype);

      ret->dim[0].lbound = 0;
      ret->dim[0].ubound = source->dim[1].ubound - source->dim[1].lbound;
      ret->dim[0].stride = 1;

      ret->dim[1].lbound = 0;
      ret->dim[1].ubound = source->dim[0].ubound - source->dim[0].lbound;
      ret->dim[1].stride = ret->dim[0].ubound+1;

      ret->data = internal_malloc_size (size * size0 ((array_t*)ret));
      ret->base = 0;
    }

  sxstride = source->dim[0].stride * size;
  if (sxstride == 0)
    sxstride = size;
  systride = source->dim[1].stride * size;
  xcount = source->dim[0].ubound + 1 - source->dim[0].lbound;
  ycount = source->dim[1].ubound + 1 - source->dim[1].lbound;

  rxstride = ret->dim[0].stride * size;
  if (rxstride == 0)
    rxstride = size;
  rystride = ret->dim[1].stride * size;

  rptr = ret->data;
  sptr = source->data;

  for (y = 0; y < ycount; y++)
    {
      for (x = 0; x < xcount; x++)
        {
          memcpy (rptr, sptr, size);

          sptr += sxstride;
          rptr += rystride;
        }
      sptr += systride - (sxstride * xcount);
      rptr += rxstride - (rystride * xcount);
    }
}
Пример #21
0
static void
unpack_internal (gfc_array_char *ret, const gfc_array_char *vector,
		 const gfc_array_l1 *mask, const gfc_array_char *field,
		 index_type size, index_type fsize)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type rs;
  char * restrict rptr;
  /* v.* indicates the vector array.  */
  index_type vstride0;
  char *vptr;
  /* f.* indicates the field array.  */
  index_type fstride[GFC_MAX_DIMENSIONS];
  index_type fstride0;
  const char *fptr;
  /* m.* indicates the mask array.  */
  index_type mstride[GFC_MAX_DIMENSIONS];
  index_type mstride0;
  const GFC_LOGICAL_1 *mptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type n;
  index_type dim;

  int empty;
  int mask_kind;

  empty = 0;

  mptr = mask->data;

  /* Use the same loop for all logical types, by using GFC_LOGICAL_1
     and using shifting to address size and endian issues.  */

  mask_kind = GFC_DESCRIPTOR_SIZE (mask);

  if (mask_kind == 1 || mask_kind == 2 || mask_kind == 4 || mask_kind == 8
#ifdef HAVE_GFC_LOGICAL_16
      || mask_kind == 16
#endif
      )
    {
      /*  Don't convert a NULL pointer as we use test for NULL below.  */
      if (mptr)
	mptr = GFOR_POINTER_TO_L1 (mptr, mask_kind);
    }
  else
    runtime_error ("Funny sized logical array");

  if (ret->data == NULL)
    {
      /* The front end has signalled that we need to populate the
	 return array descriptor.  */
      dim = GFC_DESCRIPTOR_RANK (mask);
      rs = 1;
      for (n = 0; n < dim; n++)
	{
	  count[n] = 0;
	  GFC_DIMENSION_SET(ret->dim[n], 0,
			    GFC_DESCRIPTOR_EXTENT(mask,n) - 1, rs);
	  extent[n] = GFC_DESCRIPTOR_EXTENT(ret,n);
	  empty = empty || extent[n] <= 0;
	  rstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(ret, n);
	  fstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(field, n);
	  mstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(mask, n);
	  rs *= extent[n];
	}
      ret->offset = 0;
      ret->data = internal_malloc_size (rs * size);
    }
  else
    {
      dim = GFC_DESCRIPTOR_RANK (ret);
      for (n = 0; n < dim; n++)
	{
	  count[n] = 0;
	  extent[n] = GFC_DESCRIPTOR_EXTENT(ret,n);
	  empty = empty || extent[n] <= 0;
	  rstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(ret, n);
	  fstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(field, n);
	  mstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(mask, n);
	}
      if (rstride[0] == 0)
	rstride[0] = size;
    }

  if (empty)
    return;

  if (fstride[0] == 0)
    fstride[0] = fsize;
  if (mstride[0] == 0)
    mstride[0] = 1;

  vstride0 = GFC_DESCRIPTOR_STRIDE_BYTES(vector,0);
  if (vstride0 == 0)
    vstride0 = size;
  rstride0 = rstride[0];
  fstride0 = fstride[0];
  mstride0 = mstride[0];
  rptr = ret->data;
  fptr = field->data;
  vptr = vector->data;

  while (rptr)
    {
      if (*mptr)
        {
          /* From vector.  */
          memcpy (rptr, vptr, size);
          vptr += vstride0;
        }
      else
        {
          /* From field.  */
          memcpy (rptr, fptr, size);
        }
      /* Advance to the next element.  */
      rptr += rstride0;
      fptr += fstride0;
      mptr += mstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          rptr -= rstride[n] * extent[n];
          fptr -= fstride[n] * extent[n];
          mptr -= mstride[n] * extent[n];
          n++;
          if (n >= dim)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              fptr += fstride[n];
              mptr += mstride[n];
            }
        }
    }
}
Пример #22
0
static void
eoshift2 (gfc_array_char *ret, const gfc_array_char *array,
	  int shift, const gfc_array_char *bound, int which,
	  const char *filler, index_type filler_len)
{
  /* r.* indicates the return array.  */
  index_type rstride[GFC_MAX_DIMENSIONS];
  index_type rstride0;
  index_type roffset;
  char * restrict rptr;
  char *dest;
  /* s.* indicates the source array.  */
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type sstride0;
  index_type soffset;
  const char *sptr;
  const char *src;
  /* b.* indicates the bound array.  */
  index_type bstride[GFC_MAX_DIMENSIONS];
  index_type bstride0;
  const char *bptr;

  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type dim;
  index_type len;
  index_type n;
  index_type arraysize;
  index_type size;

  /* The compiler cannot figure out that these are set, initialize
     them to avoid warnings.  */
  len = 0;
  soffset = 0;
  roffset = 0;

  size = GFC_DESCRIPTOR_SIZE (array);

  arraysize = size0 ((array_t *) array);

  if (ret->data == NULL)
    {
      int i;

      ret->offset = 0;
      ret->dtype = array->dtype;
      for (i = 0; i < GFC_DESCRIPTOR_RANK (array); i++)
        {
	  index_type ub, str;

          ub = GFC_DESCRIPTOR_EXTENT(array,i) - 1;

          if (i == 0)
	    str = 1;
          else
            str = GFC_DESCRIPTOR_EXTENT(ret,i-1)
	      * GFC_DESCRIPTOR_STRIDE(ret,i-1);

	  GFC_DIMENSION_SET(ret->dim[i], 0, ub, str);

          /* internal_malloc_size allocates a single byte for zero size.  */
	  ret->data = internal_malloc_size (size * arraysize);

        }
    }
  else if (unlikely (compile_options.bounds_check))
    {
      bounds_equal_extents ((array_t *) ret, (array_t *) array,
				 "return value", "EOSHIFT");
    }

  if (arraysize == 0)
    return;

  which = which - 1;

  extent[0] = 1;
  count[0] = 0;
  sstride[0] = -1;
  rstride[0] = -1;
  bstride[0] = -1;
  n = 0;
  for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
    {
      if (dim == which)
        {
          roffset = GFC_DESCRIPTOR_STRIDE_BYTES(ret,dim);
          if (roffset == 0)
            roffset = size;
          soffset = GFC_DESCRIPTOR_STRIDE_BYTES(array,dim);
          if (soffset == 0)
            soffset = size;
          len = GFC_DESCRIPTOR_EXTENT(array,dim);
        }
      else
        {
          count[n] = 0;
          extent[n] = GFC_DESCRIPTOR_EXTENT(array,dim);
          rstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(ret,dim);
          sstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(array,dim);
          if (bound)
            bstride[n] = GFC_DESCRIPTOR_STRIDE_BYTES(bound,n);
          else
            bstride[n] = 0;
          n++;
        }
    }
  if (sstride[0] == 0)
    sstride[0] = size;
  if (rstride[0] == 0)
    rstride[0] = size;
  if (bound && bstride[0] == 0)
    bstride[0] = size;

  dim = GFC_DESCRIPTOR_RANK (array);
  rstride0 = rstride[0];
  sstride0 = sstride[0];
  bstride0 = bstride[0];
  rptr = ret->data;
  sptr = array->data;

  if ((shift >= 0 ? shift : -shift ) > len)
    {
      shift = len;
      len = 0;
    }
  else
    {
      if (shift > 0)
	len = len - shift;
      else
	len = len + shift;
    }
  
  if (bound)
    bptr = bound->data;
  else
    bptr = NULL;

  while (rptr)
    {
      /* Do the shift for this dimension.  */
      if (shift > 0)
        {
          src = &sptr[shift * soffset];
          dest = rptr;
        }
      else
        {
          src = sptr;
          dest = &rptr[-shift * roffset];
        }
      for (n = 0; n < len; n++)
        {
          memcpy (dest, src, size);
          dest += roffset;
          src += soffset;
        }
      if (shift >= 0)
        {
          n = shift;
        }
      else
        {
          dest = rptr;
          n = -shift;
        }

      if (bptr)
	while (n--)
	  {
	    memcpy (dest, bptr, size);
	    dest += roffset;
	  }
      else
	while (n--)
	  {
	    index_type i;

	    if (filler_len == 1)
	      memset (dest, filler[0], size);
	    else
	      for (i = 0; i < size ; i += filler_len)
		memcpy (&dest[i], filler, filler_len);

	    dest += roffset;
	  }

      /* Advance to the next section.  */
      rptr += rstride0;
      sptr += sstride0;
      bptr += bstride0;
      count[0]++;
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          rptr -= rstride[n] * extent[n];
          sptr -= sstride[n] * extent[n];
          bptr -= bstride[n] * extent[n];
          n++;
          if (n >= dim - 1)
            {
              /* Break out of the loop.  */
              rptr = NULL;
              break;
            }
          else
            {
              count[n]++;
              rptr += rstride[n];
              sptr += sstride[n];
              bptr += bstride[n];
            }
        }
    }
}
Пример #23
0
GFC_REAL_16 *
internal_pack_r16 (gfc_array_r16 * source)
{
  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type stride[GFC_MAX_DIMENSIONS];
  index_type stride0;
  index_type dim;
  index_type ssize;
  const GFC_REAL_16 *src;
  GFC_REAL_16 * restrict dest;
  GFC_REAL_16 *destptr;
  int n;
  int packed;

  /* TODO: Investigate how we can figure out if this is a temporary
     since the stride=0 thing has been removed from the frontend.  */

  dim = GFC_DESCRIPTOR_RANK (source);
  ssize = 1;
  packed = 1;
  for (n = 0; n < dim; n++)
    {
      count[n] = 0;
      stride[n] = source->dim[n].stride;
      extent[n] = source->dim[n].ubound + 1 - source->dim[n].lbound;
      if (extent[n] <= 0)
        {
          /* Do nothing.  */
          packed = 1;
          break;
        }

      if (ssize != stride[n])
        packed = 0;

      ssize *= extent[n];
    }

  if (packed)
    return source->data;

  /* Allocate storage for the destination.  */
  destptr = (GFC_REAL_16 *)internal_malloc_size (ssize * sizeof (GFC_REAL_16));
  dest = destptr;
  src = source->data;
  stride0 = stride[0];


  while (src)
    {
      /* Copy the data.  */
      *(dest++) = *src;
      /* Advance to the next element.  */
      src += stride0;
      count[0]++;
      /* Advance to the next source element.  */
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so probably not worth it.  */
          src -= stride[n] * extent[n];
          n++;
          if (n == dim)
            {
              src = NULL;
              break;
            }
          else
            {
              count[n]++;
              src += stride[n];
            }
        }
    }
  return destptr;
}
void
mminloc0_4_i4 (gfc_array_i4 * retarray, gfc_array_i4 *array,
				  gfc_array_l4 * mask)
{
  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type mstride[GFC_MAX_DIMENSIONS];
  index_type dstride;
  GFC_INTEGER_4 *dest;
  GFC_INTEGER_4 *base;
  GFC_LOGICAL_4 *mbase;
  int rank;
  index_type n;

  rank = GFC_DESCRIPTOR_RANK (array);
  if (rank <= 0)
    runtime_error ("Rank of array needs to be > 0");

  if (retarray->data == NULL)
    {
      retarray->dim[0].lbound = 0;
      retarray->dim[0].ubound = rank-1;
      retarray->dim[0].stride = 1;
      retarray->dtype = (retarray->dtype & ~GFC_DTYPE_RANK_MASK) | 1;
      retarray->base = 0;
      retarray->data = internal_malloc_size (sizeof (GFC_INTEGER_4) * rank);
    }
  else
    {
      if (GFC_DESCRIPTOR_RANK (retarray) != 1)
	runtime_error ("rank of return array does not equal 1");

      if (retarray->dim[0].ubound + 1 - retarray->dim[0].lbound != rank)
        runtime_error ("dimension of return array incorrect");

      if (retarray->dim[0].stride == 0)
	retarray->dim[0].stride = 1;
    }

  /* TODO:  It should be a front end job to correctly set the strides.  */

  if (array->dim[0].stride == 0)
    array->dim[0].stride = 1;

  if (mask->dim[0].stride == 0)
    mask->dim[0].stride = 1;

  dstride = retarray->dim[0].stride;
  dest = retarray->data;
  for (n = 0; n < rank; n++)
    {
      sstride[n] = array->dim[n].stride;
      mstride[n] = mask->dim[n].stride;
      extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
      count[n] = 0;
      if (extent[n] <= 0)
	{
	  /* Set the return value.  */
	  for (n = 0; n < rank; n++)
	    dest[n * dstride] = 0;
	  return;
	}
    }

  base = array->data;
  mbase = mask->data;

  if (GFC_DESCRIPTOR_SIZE (mask) != 4)
    {
      /* This allows the same loop to be used for all logical types.  */
      assert (GFC_DESCRIPTOR_SIZE (mask) == 8);
      for (n = 0; n < rank; n++)
        mstride[n] <<= 1;
      mbase = (GFOR_POINTER_L8_TO_L4 (mbase));
    }


  /* Initialize the return value.  */
  for (n = 0; n < rank; n++)
    dest[n * dstride] = 1;
  {

  GFC_INTEGER_4 minval;

  minval = GFC_INTEGER_4_HUGE;

  while (base)
    {
      {
        /* Implementation start.  */

  if (*mbase && *base < minval)
    {
      minval = *base;
      for (n = 0; n < rank; n++)
        dest[n * dstride] = count[n] + 1;
    }
        /* Implementation end.  */
      }
      /* Advance to the next element.  */
      count[0]++;
      base += sstride[0];
      mbase += mstride[0];
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          base -= sstride[n] * extent[n];
          mbase -= mstride[n] * extent[n];
          n++;
          if (n == rank)
            {
              /* Break out of the loop.  */
              base = NULL;
              break;
            }
          else
            {
              count[n]++;
              base += sstride[n];
              mbase += mstride[n];
            }
        }
    }
  }
}
Пример #25
0
void
minloc0_8_r8 (gfc_array_i8 * retarray, gfc_array_r8 *array)
{
  index_type count[GFC_MAX_DIMENSIONS];
  index_type extent[GFC_MAX_DIMENSIONS];
  index_type sstride[GFC_MAX_DIMENSIONS];
  index_type dstride;
  GFC_REAL_8 *base;
  GFC_INTEGER_8 *dest;
  index_type rank;
  index_type n;

  rank = GFC_DESCRIPTOR_RANK (array);
  if (rank <= 0)
    runtime_error ("Rank of array needs to be > 0");

  if (retarray->data == NULL)
    {
      retarray->dim[0].lbound = 0;
      retarray->dim[0].ubound = rank-1;
      retarray->dim[0].stride = 1;
      retarray->dtype = (retarray->dtype & ~GFC_DTYPE_RANK_MASK) | 1;
      retarray->offset = 0;
      retarray->data = internal_malloc_size (sizeof (GFC_INTEGER_8) * rank);
    }
  else
    {
      if (GFC_DESCRIPTOR_RANK (retarray) != 1)
	runtime_error ("rank of return array does not equal 1");

      if (retarray->dim[0].ubound + 1 - retarray->dim[0].lbound != rank)
        runtime_error ("dimension of return array incorrect");

      if (retarray->dim[0].stride == 0)
	retarray->dim[0].stride = 1;
    }

  /* TODO:  It should be a front end job to correctly set the strides.  */

  if (array->dim[0].stride == 0)
    array->dim[0].stride = 1;

  dstride = retarray->dim[0].stride;
  dest = retarray->data;
  for (n = 0; n < rank; n++)
    {
      sstride[n] = array->dim[n].stride;
      extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
      count[n] = 0;
      if (extent[n] <= 0)
	{
	  /* Set the return value.  */
	  for (n = 0; n < rank; n++)
	    dest[n * dstride] = 0;
	  return;
	}
    }

  base = array->data;

  /* Initialize the return value.  */
  for (n = 0; n < rank; n++)
    dest[n * dstride] = 0;
  {

  GFC_REAL_8 minval;

  minval = GFC_REAL_8_HUGE;

  while (base)
    {
      {
        /* Implementation start.  */

  if (*base < minval || !dest[0])
    {
      minval = *base;
      for (n = 0; n < rank; n++)
        dest[n * dstride] = count[n] + 1;
    }
        /* Implementation end.  */
      }
      /* Advance to the next element.  */
      count[0]++;
      base += sstride[0];
      n = 0;
      while (count[n] == extent[n])
        {
          /* When we get to the end of a dimension, reset it and increment
             the next dimension.  */
          count[n] = 0;
          /* We could precalculate these products, but this is a less
             frequently used path so proabably not worth it.  */
          base -= sstride[n] * extent[n];
          n++;
          if (n == rank)
            {
              /* Break out of the loop.  */
              base = NULL;
              break;
            }
          else
            {
              count[n]++;
              base += sstride[n];
            }
        }
    }
  }
}
static void
transpose_internal (gfc_array_char *ret, gfc_array_char *source,
		    index_type size)
{
  /* r.* indicates the return array.  */
  index_type rxstride, rystride;
  char *rptr;
  /* s.* indicates the source array.  */
  index_type sxstride, systride;
  const char *sptr;

  index_type xcount, ycount;
  index_type x, y;

  assert (GFC_DESCRIPTOR_RANK (source) == 2
          && GFC_DESCRIPTOR_RANK (ret) == 2);

  if (ret->data == NULL)
    {
      assert (ret->dtype == source->dtype);

      ret->dim[0].lbound = 0;
      ret->dim[0].ubound = source->dim[1].ubound - source->dim[1].lbound;
      ret->dim[0].stride = 1;

      ret->dim[1].lbound = 0;
      ret->dim[1].ubound = source->dim[0].ubound - source->dim[0].lbound;
      ret->dim[1].stride = ret->dim[0].ubound+1;

      ret->data = internal_malloc_size (size * size0 ((array_t*)ret));
      ret->offset = 0;
    }
  else if (unlikely (compile_options.bounds_check))
    {
      index_type ret_extent, src_extent;

      ret_extent = ret->dim[0].ubound + 1 - ret->dim[0].lbound;
      src_extent = source->dim[1].ubound + 1 - source->dim[1].lbound;

      if (src_extent != ret_extent)
	runtime_error ("Incorrect extent in return value of TRANSPOSE"
		       " intrinsic in dimension 1: is %ld,"
		       " should be %ld", (long int) src_extent,
		       (long int) ret_extent);

      ret_extent = ret->dim[1].ubound + 1 - ret->dim[1].lbound;
      src_extent = source->dim[0].ubound + 1 - source->dim[0].lbound;

      if (src_extent != ret_extent)
	runtime_error ("Incorrect extent in return value of TRANSPOSE"
		       " intrinsic in dimension 2: is %ld,"
		       " should be %ld", (long int) src_extent,
		       (long int) ret_extent);

    }

  sxstride = source->dim[0].stride * size;
  systride = source->dim[1].stride * size;
  xcount = source->dim[0].ubound + 1 - source->dim[0].lbound;
  ycount = source->dim[1].ubound + 1 - source->dim[1].lbound;

  rxstride = ret->dim[0].stride * size;
  rystride = ret->dim[1].stride * size;

  rptr = ret->data;
  sptr = source->data;

  for (y = 0; y < ycount; y++)
    {
      for (x = 0; x < xcount; x++)
        {
          memcpy (rptr, sptr, size);

          sptr += sxstride;
          rptr += rystride;
        }
      sptr += systride - (sxstride * xcount);
      rptr += rxstride - (rystride * xcount);
    }
}