Add `pystencils.make_python_function` used for KernelFunction.compile

KernelFunction.compile = None is currently set by the create_kernel function of each respective backend as partial function of <backend>.make_python_function.

The code would be clearer with a unified make_python_function. KernelFunction.compile can then be implemented as a call to this function with the respective backend.

added 1 commit

• 24222c87 - Add `pystencils.make_python_function` used for KernelFunction.compile

Compare with previous version

added 5 commits

• 24222c87...9b9a4b54 - 4 commits from branch pycodegen:master
• 229e3950 - Add `pystencils.make_python_function` used for KernelFunction.compile

Compare with previous version

added 1 commit

• cef0089c - Add `pystencils.make_python_function` used for KernelFunction.compile

Compare with previous version

added 1 commit

• 0a9b03b8 - Fix llvm by manually setting backend in KernelFunction

Compare with previous version

added 1 commit

• 64c0fb22 - Fix llvm by manually setting backend in KernelFunction

Compare with previous version

added 1 commit

• 34509925 - Fix llvm by manually setting backend in KernelFunction

Compare with previous version

Maybe we want to add a llvm test that's not a Jupyter Notebook:

# -*- coding: utf-8 -*-
#
# Copyright © 2019 Stephan Seitz <stephan.seitz@fau.de>
#
# Distributed under terms of the GPLv3 license.

"""

"""
import llvmlite

from pystencils.session import *


def test_llvm():
    size = (60, 70)  # domain size

    u_arrays = [np.zeros(size), np.zeros(size), np.zeros(size)]

    u_fields = [ps.Field.create_from_numpy_array("u%s" % (name,), arr)
                for name, arr in zip(["0", "1", "2"], u_arrays)]

    for i, field in enumerate(u_fields):
        field.latex_name = "u^{(%d)}" % (i,)

    discretize = ps.fd.Discretization2ndOrder()

    def central2nd_time_derivative(fields):
        f_next, f_current, f_last = fields
        return (f_next[0, 0] - 2 * f_current[0, 0] + f_last[0, 0]) / discretize.dt**2

    rhs = ps.fd.diffusion(u_fields[1], 1)

    wave_eq = sp.Eq(central2nd_time_derivative(u_fields), discretize(rhs))

    wave_eq = sp.simplify(wave_eq)
    u_next_C = u_fields[-1][0, 0]
    update_rule = ps.Assignment(u_next_C, sp.solve(wave_eq, u_next_C)[0])
    update_rule

    update_rule = update_rule.subs({discretize.dx: 0.1, discretize.dt: 0.05})

    kernel = ps.create_kernel(update_rule, target='llvm')
    kernel.compile()

    def run_LLVM(timesteps=1):
        for t in range(timesteps):
            kernel(u0=u_arrays[0], u1=u_arrays[1], u2=u_arrays[2])
            u_arrays[0], u_arrays[1], u_arrays[2] = u_arrays[1], u_arrays[2], u_arrays[0]
        return u_arrays[2]
    print('Yeah. LLVM does something without crashing')


def main():
    test_llvm()


if __name__ == '__main__':
    main()

added 10 commits

• 34509925...305107fa - 8 commits from branch pycodegen:master
• a7e61dc4 - Add `pystencils.make_python_function` used for KernelFunction.compile
• c15b21cc - Fix llvm by manually setting backend in KernelFunction

Compare with previous version

I think this change resolved this pull request. Can I close it?

closed