Updating the Boundary Conditions of Model during run #3602
Answered
by
glwagner
RicksterB57
asked this question in
Helpdesk
-
|
Hello, I am attempting to simulate a storm event at high latitudes and using an array of ERA5 hourly sensible heat flux, latent heat flux, and wind at 10m. I am using these arrays to calculate wind drag and the surface heat flux to use as boundary conditions for my model. I have written functions that update the boundary conditions based on the time in the model. I have gotten this code to run successfully on cpu but when I adjust the code to run on the GPU, I run into some errors. This is the error that I recieve: and this is the code that produces the error: # Constants
const initial_rho = 1026.0; # kg m⁻³, average density at the surface of the world ocean
const rho_air = 1.225; # average density of air at sea level
const drag_coef = 2.5e-3; # drag coefficient
const heat_capacity = 3991.0;
# Convert constants to GPU arrays
ρ₀ = CUDA.fill(initial_rho);
ρₐ = CUDA.fill(rho_air);
cᴰ = CUDA.fill(drag_coef);
cᴾ = CUDA.fill(heat_capacity);
#Wind Drag
Qᵘ = -ρₐ ./ ρ₀ .* cᴰ .* hourly_mean_u10_gpu .* abs.(hourly_mean_u10_gpu);
#Surface Heat Flux
Qnet = hourly_mean_latent_gpu .+ hourly_mean_sensible_gpu;
Qᵀ = Qnet ./ (ρ₀ .* cᴾ);
function surface_heat_flux(x, y, t)
hour_index = Int.(floor.(t ./ 3600)) .+ 1
t_bnd = CUDA.@allowscalar Qᵀ[1, 1, hour_index]
return t_bnd
end
function wind_drag(x, y, t)
hour_index = Int.(floor.(t ./ 3600)) .+ 1
u_bnd = CUDA.@allowscalar Qᵘ[1, 1, hour_index]
return u_bnd
end
# Boundary Conditions
u_bcs = FieldBoundaryConditions(top = FluxBoundaryCondition(wind_drag));
T_bcs = FieldBoundaryConditions(top = FluxBoundaryCondition(surface_heat_flux));
# Define your model here
model = NonhydrostaticModel(; grid, buoyancy,
advection = UpwindBiasedFifthOrder(),
timestepper = :RungeKutta3,
tracers = (:T, :S),
coriolis = FPlane(rotation_rate=7.292115e-5, latitude=72),
closure = AnisotropicMinimumDissipation(),
boundary_conditions = (u=u_bcs, T=T_bcs));Any tips/help to diagnonse the error would be appreciated. |
Beta Was this translation helpful? Give feedback.
Answered by
glwagner
May 22, 2024
Replies: 1 comment 10 replies
-
|
Seems like a cool problem... ! Since your forcing is horizontally-homogeneous, try using julia> Q = Ref(0.0)
Base.RefValue{Float64}(0.0)
julia> Q[] = 1.0
1.0I believe I also suggest using parameters in your boundary condition functions to avoid global variables (otherwise you also need to declare Finally... I believe can actually be implemented now using |
Beta Was this translation helpful? Give feedback.
10 replies
-
|
Thanks for the tips I made some edits to the code that I have: But for some reason I keep on getting basically the same error as the original code: Is there a specific action/operation that I am doing within the boundary condition functions that is causing a problem with the GPU? |
Beta Was this translation helpful? Give feedback.
-
|
Can you edit your post to use triple backticks? It is using triple quotations. It's hard to read the code. Note that I edited your first post this way so you can see what I did. |
Beta Was this translation helpful? Give feedback.
-
|
I implemented something just like this here: https://github.com/glwagner/TropicalTurbulantics.jl/blob/main/tropical_turbulence_setup.jl Check out the function Note that |
Beta Was this translation helpful? Give feedback.
Answer selected by
glwagner
-
|
I edited the comment so the code should be more clear now. I will take a look at the setup that you sent. Thanks for the help. |
Beta Was this translation helpful? Give feedback.
-
|
Lines like this: Qᵘ = -p.ρₐ ./ p.ρ₀ .* p.cᴰ .* p.u10 .* abs.(p.u10);cannot be embedded inside a forcing function, because we cannot broadcast within a forcing function on the GPU. Basically for this to work, the momentum flux Another option is to use the pattern in my example which also avoids broadcasting. |
Beta Was this translation helpful? Give feedback.
-
|
Here's the documentation on |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for all the help. I got my model to start running. |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I implemented something just like this here: https://github.com/glwagner/TropicalTurbulantics.jl/blob/main/tropical_turbulence_setup.jl
Check out the function
interp_bc. that function is incorporated into a boundary condition here:https://github.com/glwagner/TropicalTurbulantics.jl/blob/8b5e882a3ea7b72d4cdc7cdbb9cde14f59d1d105/tropical_turbulence_setup.jl#L214
Note that
interp_bcapplies to interpolating any time-series at a point. There is alsointerp_forcing, which additionally depends on the vertical indexk(this applies to a horizontally uniform but vertically-varying forcing, such as lateral flux divergences derived from a GCM).