Single Precision Post Processing Does not Work

[wilfonba]

Post process calls MPI_ABORT with no error messages when single precision is specified.

[sbryngelson]

single precision specified how? this is too vague.

[wilfonba]

Fair. Using 'precision' : 1 in the input file causes an abort. Although in hindsight this may be because the simulation was ran with 'precision' : 2 and I was just trying to shrink the silo files by running post_process with 'precision' : 1. If this is the issue we should look into an implementation that allows for this.

[sbryngelson]

I meant what options are you using for parallel_IO and format and so on...

[wilfonba]

'format' : 1,
'precision' : 1,
'prim_vars_wrt' : 'T'
'parallel_io' : 'T'

[wilfonba]

Compiling with GNU shows the real problem:

"Unsupported combination of precision and parallel IO. Please use precision == 2 when enabling parallel_io. Exiting ..."

It appears it's simply not been implemented yet. NVHPC doesn't show these error messages for checking the input files.

[sbryngelson]

This seems to signal a more significant problem. The GNU compilers are respecting our m_start_up.f90 checks (where this error comes from), but NVHPC compilers are not.

[sbryngelson]

MFC/src/post_process/m_start_up.f90

Lines 370 to 373 in b5fd492

	elseif ((precision /= 2) .and. (parallel_io .neqv. .false.)) then
	print '(A)', 'Unsupported combination of precision and parallel IO. '// &
	'Please use precision == 2 when enabling parallel_io. Exiting ...'
	call s_mpi_abort()

[wilfonba]

On a similar note, there appears to be an undocumented input option "coarsen_silo" that shows up in the post-processing code and appears to write silo_files using the precision variable. This option appears to allow you to average flow variables onto a coarser grid and simultaneously write to a single precision silo file.

MFC/src/post_process/m_data_output.f90

Lines 689 to 702 in b5fd492

	if (coarsen_silo) then
	if (grid_geometry == 3) then
	err = DBPUTQM(dbfile, 'rectilinear_grid', 16, &
	'x', 1, 'y', 1, 'z', 1, &
	coarse_y_cb, coarse_z_cb, coarse_x_cb, dims, 3, &
	precision, DB_COLLINEAR, &
	optlist, ierr)
	else
	err = DBPUTQM(dbfile, 'rectilinear_grid', 16, &
	'x', 1, 'y', 1, 'z', 1, &
	coarse_x_cb, coarse_y_cb, coarse_z_cb, dims, 3, &
	precision, DB_COLLINEAR, &
	optlist, ierr)
	end if

[sbryngelson]

yes this should be re-tested then documented.