User's Guide to NCAR CCM3.6 Search Page

3. CCM3.6 Internals

    3.3 Data Structures

        3.3.2 Grid-point Data Structures

3.3.2.1 Model Buffers

Figure 3.2.Layout of main model buffers, b1 and b2.
 

Listed within each buffer is the variable holding the storage location of its associated field. To the right of each variable is a description of the field including whether the field is single-level(s) or multi-level (m). Also shown are related parameters pflenb, pflena, pbflnb,  pbflna,  and plenalcl.

Buffer b1(pflenb)                            Buffer b2(pflena)
          variable                                                                                                variable

p  b  f  l  n  a	Nzt1	vorticity: half time-filtered (m)	p  l  e  n  a  l  c  l	p  b  f  l  n  a	Npsp1	sfc. pressure, time n(s)
	Ndt1	divergence: half time-filtered (m)			Nzp1	vorticity, time n (m)
	Nphis	surface geopotential (s)			Ndp1	divergence, time n (m)
	Nqminus  (PCNST)	q being advected in slt code (m)			Ndpthp1	horiz. temp. diffusion (m)
	Npsm2	surface pressure, time n-2 (s)			Ndpslp1	long. deriv. of lpns, time n (s)
	Num2	u-wind, time n-2 (m)			Ndpsmp1	lat. deriv. of lpns, time n (m)
	Nvm2	v-wind, time n-2 (m)			Nduhp1	horiz. u-momentum diff . (m)
	ntm2	temperature, time n-2 (m)			Hdvhp1	horiz. v-momentum diff. (m)
	nqm2  (PCNST)	constituents, H2O first (m)			ndpsp1	pressure gradient (s)
	npb1ht	PBL height (s)
	nqpert  (PCNST)	PBL moisture/constit. Perturb. (s)
	ntpert	PBL temperature perturb. (s)
	nfsns	surface absorbed solar flux (s)
	nqrs	shortwave rad. heating rate (m)
	nqrl	Longwave rad. heating rate (m)
p  b  f  l  n  a	npsm1	surface pressure, time n-1 (s)
	nzm1	vorticity, time n-1 (m)
	ndm1	divergence, time n-1 (m)
	not used	(m)
	not used	(s)
	not used	(s)

  These buffers hold only those fields which must be carried across discrete time levels, that are needed in more than one Gaussian latitude scan, or that need to be contiguous to other fields. One example of a field which must be carried from one time level to the next is the shortwave radiative heating rate (qrs), which must be passed forward in time between calls to the radiation routines.

 Surface pressure, vorticity, and divergence are in the buffer for contiguity purposes. Contiguity is required for optimal vectorization of the FFT package. This is important since the FFT routines used in CCM3.6 vectorize over the number of transforms being done. By putting surface pressure next to vorticity and divergence, an 18-level model produces a respectable vector length of 37 (i.e. 2*18 + 1) in the Fourier transform of these fields. If surface pressure were transformed separately, the vector length for this field would be 1, which would significantly degrade performance. Surface pressure is contained in memory (common /com3d/), and is copied to and from the model buffer for the sole purpose of avoiding the FFT performance penalty.

All buffer fields are accessed computationally as individual, appropriately dimensioned arrays. This approach yields readable code, and actually runs a bit faster than using direct buffer references throughout. It also enables usage of the array bounds checker (f90 -Rb on Cray machines). This functionality is obtained through Cray Fortran pointer variables, as illustrated in a section of subroutine spegrd below:  

 
real b2(plenalcl)    ! model buffer
pointer (pz,z)       ! declaration of pointer variable
real z(plond,plev)   ! part of model buffer
  .
  .
  .
pz = loc(b2(nzp1+1)) ! array z may now be accessed

 Search page

Questions on these pages can be sent to... erik@ucar.edu .