"When creating the kernels we pass as target (which may be 'cpu' or 'gpu') the default target of the target handling. This enables to switch to a GPU simulation just by changing the parameter of the data handling.\n",
"When creating the kernels we pass as target (which may be 'Target.CPU' or 'Target.GPU') the default target of the target handling. This enables to switch to a GPU simulation just by changing the parameter of the data handling.\n",
"\n",
"The rest is similar to the previous tutorial."
]
...
...
@@ -457,4 +458,4 @@
},
"nbformat": 4,
"nbformat_minor": 2
}
}
\ No newline at end of file
%% Cell type:code id: tags:
``` python
from pystencils.session import *
import pystencils as ps
sp.init_printing()
frac = sp.Rational
```
%% Cell type:markdown id: tags:
# Tutorial 06: Phase-field simulation of dentritic solidification
This is the second tutorial on phase field methods with pystencils. Make sure to read the previous tutorial first.
In this tutorial we again implement a model described in **Programming Phase-Field Modelling** by S. Bulent Biner.
This time we implement the model from chapter 4.7 that describes dentritic growth. So get ready for some beautiful snowflake pictures.
We start again by adding all required arrays fields. This time we explicitly store the change of the phase variable φ in time, since the dynamics is calculated using an Allen-Cahn formulation where a term $\partial_t \phi$ occurs.
However, we just insert these parameters into the free energy formulation before doing the functional derivative, to make the dependence of $\epsilon$ on $\nabla \phi$ explicit.
When creating the kernels we pass as target (which may be 'cpu' or 'gpu') the default target of the target handling. This enables to switch to a GPU simulation just by changing the parameter of the data handling.
When creating the kernels we pass as target (which may be 'Target.CPU' or 'Target.GPU') the default target of the target handling. This enables to switch to a GPU simulation just by changing the parameter of the data handling.
In this tutorial we show how to use the finite difference module of *pystencils* to solve a 2D wave equations. The time derivative is discretized by a simple forward Euler method.
We begin by creating three *numpy* arrays for the current, the previous and the next timestep. Actually we will see later that two fields are enough, but let's keep it simple. From these *numpy* arrays we create *pystencils* fields to formulate our update rule.
To run simulation a suitable initial condition and boundary treatment is required. We chose an initial condition which is zero at the borders of the domain. The outermost layer is not changed by the update kernel, so we have an implicit homogenous Dirichlet boundary condition.
The next cell demonstrates how to run the same simulation with the LLVM backend. The only difference is that in the `create_kernel` function the `target` is set to llvm.
The run function has to be changed now slightly, since the data has to be transfered to the GPU first, then the kernel can be executed, and in the end the data has to be transfered back
