Coompare newbc and gym-arch #5
Conversation
| # with mpi support | ||
| h5py | ||
| mpi4py | ||
| gymnasium |
There was a problem hiding this comment.
these files dont have their versions pinned, so its possible they have been updated since this MR was originally created & have caused breaking changes
|
|
||
| [build-system] | ||
| requires = ["flit_core >=3.2,<4"] | ||
| build-backend = "flit_core.buildapi" |
There was a problem hiding this comment.
not sure why I decided to only now start packaging things up into a pyproject.toml but probably the right thing to do
| @@ -54,7 +54,7 @@ OBJ_FILES = alloc.o bcdf.o bcextr.o bcfree.o bc.o bcrecyc.o bcrelax.o bcshk.o bc | |||
| setup.o solver.o startmpi.o stats.o step.o target_reystress.o tbforce.o updateghost.o utility.o utyibm.o \ | |||
| visflx.o writedf.o writefield.o writefieldvtk.o writegridplot3d.o writerst.o \ | |||
| writestatbl.o writestatchann.o writestat.o writestatzbl.o write_wallpressure.o visflx_stag.o \ | |||
| write_probe_data.o bcblow.o dissipation.o | |||
| bcblow.o dissipation.o dealloc.o | |||
There was a problem hiding this comment.
basically just adds dealloc and removes write_probe_data files from the build.
deallocdeallocates all arrays if they exist. This is so we can start the solver again without needing to restart the python process. Otherwise, the fortran code would attempt to reallocate these arrays. This would be a problem because (1) not enough memory (2) they are already allocated so it would errorwrite_probe_datais legacy from when I first started working on streams. Since you can now (in theory) run streams from python, exporting data with python IO routines w/ h5py is much much simpler than doing it in fortran
| if (allocated(w_order)) deallocate(w_order) | ||
| if (allocated(wallpfield_gpu)) deallocate(wallpfield_gpu) | ||
| if (allocated(slicexy_gpu)) deallocate(slicexy_gpu) | ||
| if (allocated(vf_df_old)) deallocate(vf_df_old) |
There was a problem hiding this comment.
deallocate CPU-side arrays if they exist. deallocate_all is bound out to python
| if (allocated(fdm_grid_points)) deallocate(fdm_grid_points) | ||
| #ifdef USE_CUDA | ||
| if (allocated(fl_trans_gpu)) deallocate(fl_trans_gpu) | ||
| if (allocated(temperature_trans_gpu)) deallocate(temperature_trans_gpu) |
There was a problem hiding this comment.
deallocations for GPU arrays. Note that this section is in #ifdef USE_CUDA which evaluates to true for all practical lab applications (essentially, ensuring we are running w/ GPU accelerations)
| @@ -135,6 +135,8 @@ subroutine initurb | |||
| enddo | |||
| enddo | |||
| enddo | |||
| deallocate(ranf_av) | |||
|
|
|||
There was a problem hiding this comment.
I dont recall why I put this here, probably some constraint such that I couldnt put it in the main deallocate routine above
| call deallocate_all() | ||
| call setup() | ||
| call solver() | ||
| call finalize() ! Finalize computation |
There was a problem hiding this comment.
proof of concept, not meant for real use. This shows that after stopping the simulation (the first finalize) you can
- deallocate the arrays
- call
setupwhich reallocates them - run the solver
- stop again
and you will get the same data twice. Not sure if I ever proved it was the same data, but I do believe it ran without error.
| @@ -10,8 +10,6 @@ subroutine manage_solver | |||
| updatestat = .false. | |||
| savefield = .false. | |||
| saverst = .false. | |||
| saveprobe = .false. | |||
| savespanaverage = .false. | |||
There was a problem hiding this comment.
all this IO can now be done from python, removed
| if (save_probe_steps > 0) then | ||
| if (mod(icyc, save_probe_steps) == 0) saveprobe = .true. | ||
| end if | ||
|
|
There was a problem hiding this comment.
all this IO can now be done from python, removed
| if (saveprobe) then | ||
| call write_probe_data | ||
| endif | ||
|
|
There was a problem hiding this comment.
all this IO can now be done from python, removed
| implicit none | ||
|
|
||
| call deallocate_all() | ||
| end subroutine |
There was a problem hiding this comment.
bind the deallocate_all function to python under the name wrap_deallocate_all
| @@ -51,9 +51,6 @@ subroutine readinp | |||
| read (12,*) rand_start | |||
| read (12,*) | |||
| read (12,*) | |||
| read (12,*) save_probe_steps, save_span_average_steps | |||
| read (12,*) | |||
| read (12,*) | |||
There was a problem hiding this comment.
no longer need to read IO information from config files for these parameters since they are unused
| @@ -101,8 +101,6 @@ module mod_streams | |||
| !f2py integer :: slot_start_x_global, slot_end_x_global | |||
| integer :: slot_start_x_global, slot_end_x_global | |||
|
|
|||
| ! the number of solver steps between outputting probe information / span average information | |||
| integer :: save_probe_steps, save_span_average_steps | |||
There was a problem hiding this comment.
dead variables now, removed
| @@ -1,314 +0,0 @@ | |||
| module mod_probe | |||
| real(8), dimension(:,:,:), allocatable :: span_average | |||
| character(len=5) :: nxstr, nystr | |||
There was a problem hiding this comment.
this whole file was to export binary data, using hdf5 now so remove it
| deallocate(nyvec) | ||
| deallocate(retauvec) | ||
| deallocate(reydata) | ||
| deallocate(reydataycode) |
There was a problem hiding this comment.
dont recall why these are here, might be that they are not part of mod_streams and could not dealloc them in deallocate_all.f90
| #importlib.reload(streams) | ||
| #streams.wrap_startmpi() | ||
| setup_solver() | ||
|
|
There was a problem hiding this comment.
on the 5th timestep, stop the solver and set it up again
| from itertools import count | ||
| from typing import Deque | ||
|
|
||
| # pytorch imports |
There was a problem hiding this comment.
These are from the pytorch getting started tutorial, do not recall which one
| ObsType = Any | ||
| RenderFrame = Any | ||
|
|
||
| class StreamsSolver(Env): |
There was a problem hiding this comment.
framework for creating a gym environment with gymnasium. these are all placeholders
for illustration and discussion of the diffs, not intended to be merged