How do the modes affect the scheduling?

I scheduling my thorn to be run after the QLM thorn, because I need access to the spin components calculated by the QLM thorn..

If I schedule everything in my thorn in global-early, I get around the problems of not being able to output correctly..

but now, the thorn runs before the QLM thorn, even though the schedule tree indicates differently.. which means I don't have access to the variables from the QLM thorn, because they haven't been calculated at the time my thorn does its calculations..

how is this happening?

best wishes,

Vassili

On Sat, Sep 19, 2015 at 5:40 PM, Vassilios Mewes <vassilios.mewes@uv.es> wrote:
Sorry for the late reply, was (and am) still testing around...

Here is the schedule tree for my thorn:

      disk_analysis::Disk_Analysis_Setup_LoopCounter: [global] 
      while (Disk_Analysis::nradii_LoopCounter)
        GROUP DISK_ANALYSE: Disk_analysis main loop
          disk_analysis::Disk_Analysis_Decrement_LoopCounter: [global] 
          disk_analysis::Disk_Analysis_calculate_J_gridpoint_shell: [global] [loop-local] 
          disk_analysis::Disk_Analysis_reduce_J_total_shell: [global] 
          disk_analysis::Disk_Analysis_calculate_angles_shell: [global] 
      end while
      disk_analysis::Disk_Analysis_extraction_plane: [global] 
      GROUP DISK_GLOBAL_QUANTITIES: 
        disk_analysis::Disk_Analysis_set_global_quantities: [global] [loop-local] 
        disk_analysis::Disk_Analysis_reduce_global_quantities: [global] 
        disk_analysis::Disk_Analysis_set_global_specific_quantities: [global] [loop-local] 

The problem is the following:

I want to calculate some quantities in the function disk_analysis::Disk_Analysis_set_global_specific_quantities, these will be grid functions that I want to output (in 1D and 3D)

however, the calculation depends on disk_analysis::Disk_Analysis_extraction_plane, where I decide if I need to rotate the coordinate system in which I later want to calculate the grid functions... it basically calculates and stores a rotation matrix, as well as an integer switch that later tells us if we need to rotate..

now, when scheduling disk_analysis::Disk_Analysis_set_global_specific_quantities as global loop-local, the values obtained in disk_analysis::Disk_Analysis_extraction_plane (which i store as SCALARS and have assigned storage for) are the correct ones, however, I cannot output the grid functions calculated, which I guess is due to this response from Roland in the mailing list earlier this year:

On Tue, Jun 2, 2015 at 1:46 PM, Roland Haas <rhaas at aei.mpg.de> wrote:

> Hello all,
>
> > If you are scheduling everything in GLOBAL (which is the same as
> > GLOBAL-LATE in ANALYSIS) then your OUTPUT will be incorrect since OUTPUT
> > is happening in LEVEL mode thus refinement level 3 is output before the
> > GLOBAL routine (which runs along with level 0) is executed.
> Oha, obvious not the right statement. GLOBAL-LATE will happen last but
> this happens to be the highest refinement level. Still the effect is the
> same OUTPUT happens before GLOBAL-LATE for all but one refinement level
> (that refinement level being the finest one).
>
> This can be seen in Carpet's Evolve.cc file in the CallAnalysis routine.
>
> Yours,
> Roland

now, if i want correct output, i can use either local mode or global-early loop-local mode..

then however, the values of the rotation matrix and the integer switch of actually using it are not correct, because the routine seems to be called now not in the order indicated by the schedule tree, but rather before the values have been set...

so: when using global loop-local in disk_analysis::Disk_Analysis_set_global_specific_quantities, the grid functions are not output correctly (only in the innermost refinement level)

and when using local or global-early loop-local the function disk_analysis::Disk_Analysis_set_global_specific_quantities cannot access the data calculated by disk_analysis::Disk_Analysis_extraction_plane..

is there an obvious way around this? ie getting the correct scheduling for both being able to use the values in the function AND outputting the calculated grid functions correctly??

best wishes,

Vassili

On Fri, Sep 18, 2015 at 3:17 PM, Frank Loeffler <knarf@cct.lsu.edu> wrote:
On Fri, Sep 18, 2015 at 12:53:33PM +0200, Vassilios Mewes wrote:
> I have a scalar variable, which contains the elements of a rotation matrix
> that rotates the z-axis into a given vector...
>
> I want to calculate some quantities (that are grid functions) in
> coordinates that have been rotated by this matrix...
>
> I run into the following problem:
>
> when scheduling the calculation of the quantities in local mode, the
> function doesn't seem to have access to the rotation matrix elements...

You should have access to scalars in any mode. What happens when you try?

> when scheduling the calculation in global, loop-local mode, I can't output
> the data at the grid points (the output gives 1.96808407167164e+243 at each
> point for every refinement level apart from the innermost, where the values
> are output correctly)

It is difficult to see what is going on without the actual scheduling
statements, and in general more information. Could you please post at
least the declaration of both scalar and GFs from interface.ccl, and the
corresponding SCHEDULE entries in schedule.ccl? Also, I assume you do
have storage enabled for both? Also useful might be the output of the
schedule tree. You get that by default, but also when running with the
-S option (that aborts the run after the tree has been printed).

Frank