Skip to content

[WIP] Rayleigh Damping mixed precision fix #32

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
FlorianDeconinck wants to merge 5 commits into
base: develop
Choose a base branch
from
Draft

[WIP] Rayleigh Damping mixed precision fix #32

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
70ea601
Move RFF calculation to f64
FlorianDeconinck Nov 29, 2024
4c8403a
lint
FlorianDeconinck Nov 29, 2024
bad27a8
Use `ndsl.constants.SECONDS_PER_DAY`
FlorianDeconinck Dec 2, 2024
e4d65df
Move damping increment to own K stencils
FlorianDeconinck Mar 5, 2025
e077ab4
Docs
FlorianDeconinck Mar 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 0 additions & 6 deletions pyFV3/stencils/fv_dynamics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -483,12 +483,6 @@ def compute_preamble(self, state: DycoreState, is_root_rank: bool):
"Dynamical Core (fv_dynamics): compute total energy is not implemented"
)

if (not self.config.rf_fast) and self.config.tau != 0:
raise NotImplementedError(
"Dynamical Core (fv_dynamics): Rayleigh_Super,"
" called when rf_fast=False and tau !=0, is not implemented"
)

if self.config.adiabatic and self.config.kord_tm > 0:
raise NotImplementedError(
"Dynamical Core (fv_dynamics): Adiabatic with positive kord_tm"
Expand Down
25 changes: 19 additions & 6 deletions pyFV3/stencils/ray_fast.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@
FORWARD,
PARALLEL,
computation,
f64,
horizontal,
interval,
log,
Expand All @@ -18,7 +19,7 @@
from ndsl.dsl.typing import Float, FloatField, FloatFieldK


SDAY = 86400.0
SDAY = Float(86400.0)
Copy link

@bensonr bensonr Dec 2, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If seconds per day is used frequently in the the ndsl, dycore, and physics, would it be better to define it in the ndsl constants section to ensure uniformity?

Copy link
Collaborator Author

@FlorianDeconinck FlorianDeconinck Dec 2, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes we probably should be moving that to constant



# NOTE: The fortran version of this computes rf in the first timestep only. Then
Expand All @@ -36,7 +37,7 @@ def compute_rf_vals(pfull, bdt, rf_cutoff, tau0, ptop):
@gtscript.function
def compute_rff_vals(pfull, dt, rf_cutoff, tau0, ptop):
rffvals = compute_rf_vals(pfull, dt, rf_cutoff, tau0, ptop)
rffvals = 1.0 / (1.0 + rffvals)
rffvals = f64(1.0) / (f64(1.0) + rffvals)
Copy link

@bensonr bensonr Dec 2, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In an f32 run, aren't you only promoting the constants here and not rffvals?

Copy link
Collaborator Author

@FlorianDeconinck FlorianDeconinck Dec 2, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The upcaster will force in the rffvals to be f64. Alternatively you could, type hint rffvals e.g.

rffvals: f64 = ...

Copy link
Collaborator Author

@FlorianDeconinck FlorianDeconinck Dec 2, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The upcaster follows basic C up-casting rules

return rffvals


Expand Down Expand Up @@ -155,14 +156,26 @@ class RayleighDamping:
Fortran name: ray_fast.
"""

def __init__(self, stencil_factory: StencilFactory, rf_cutoff, tau, hydrostatic):
def __init__(
self,
stencil_factory: StencilFactory,
rf_cutoff: Float,
tau: Float,
hydrostatic: bool,
):
orchestrate(obj=self, config=stencil_factory.config.dace_config)
grid_indexing = stencil_factory.grid_indexing
self._rf_cutoff = rf_cutoff
self._rf_cutoff = Float(rf_cutoff)
origin, domain = grid_indexing.get_origin_domain(
[X_INTERFACE_DIM, Y_INTERFACE_DIM, Z_DIM]
)

if tau == 0:
raise NotImplementedError(
"Dynamical Core (fv_dynamics): RayleighDamping,"
" with tau <= 0, is not implemented"
)

ax_offsets = grid_indexing.axis_offsets(origin, domain)
local_axis_offsets = {}
for axis_offset_name, axis_offset_value in ax_offsets.items():
Expand All @@ -175,7 +188,7 @@ def __init__(self, stencil_factory: StencilFactory, rf_cutoff, tau, hydrostatic)
domain=domain,
externals={
"hydrostatic": hydrostatic,
"rf_cutoff": rf_cutoff,
"rf_cutoff": self._rf_cutoff,
"tau": tau,
**local_axis_offsets,
},
Expand All @@ -191,7 +204,7 @@ def __call__(
dt: Float,
ptop: Float,
):
rf_cutoff_nudge = self._rf_cutoff + min(100.0, 10.0 * ptop)
rf_cutoff_nudge = self._rf_cutoff + min(Float(100.0), Float(10.0) * ptop)

self._ray_fast_wind_compute(
u,
Expand Down
Loading