tehrengruber

# -*- coding: utf-8 -*-
from __future__ import annotations
from concurrent.futures import ThreadPoolExecutor
import multiprocessing
import threading
from typing import TYPE_CHECKING

import gt4py as gt
from gt4py.cartesian import gtscript, definitions

tehrengruber

__field_operator_fvm_advect(
  __sym_1,
  __sym_2,
  __sym_3,
  __sym_4,
  __sym_5,
  __sym_6,
  __sym_7,
  __sym_8,
  __sym_9,

tehrengruber

from __future__ import annotations
import numpy as np
from functional.iterator.embedded import MDIterator, np_as_located_field, lift, shift, deref, NeighborTableOffsetProvider
from fvm_advection_sphere.common import *

# v-e-v-e-v
#   c   c
num_vertices = 10
num_edges = num_vertices-1
num_cells = num_vertices-1

tehrengruber

import itertools
import ctypes

import numpy as np
import gt4py as gt
from gt4py.backend import REGISTRY

backend = "gt:cpu_kfirst" # (0, 1, 2)
#backend = "gt:cpu_ifirst" # (2, 0, 1)

tehrengruber

import ast
from copy import copy
import types
from types import LambdaType
from typing import List, Callable, Tuple, Any, Union, Literal
import inspect
import textwrap
import tempfile
import eve.datamodels as datamodels
import typing_inspect

tehrengruber

import ast
from copy import copy
import types
from types import LambdaType
from typing import List, Callable, Tuple, Any, Union, Literal
import inspect
import textwrap
import tempfile
import eve.datamodels as datamodels
import typing_inspect

tehrengruber

Benchmarks

Laptop:

CPU: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz (6 cores)

Memory: 16 GB DDR4 2667 MT/s

GPU: NVIDIA GeForce GTX 1650 Max-Q (4 GB, GDDR5)

tehrengruber

import numpy as np
import cupy as cp
import gt4py as gt

test = gt.storage.zeros(backend="gtcuda", dtype=np.float32,
                        shape=(14, 15, 16), default_origin=(0, 0, 0))
test[0:8, 0:8, 0:8] = 1
test.synchronize()

slic = test[4:12, 4:12, 4:12]

tehrengruber

def laplacian(grid: Grid, f: Field[I, J, K, Matrix[3, 3, dtype]]):
    def stencil((i, j, k), m):
        return l2_norm(m @ [1, 2, 3])
        
    #def stencil(i, j, k, val):
    #    if (i, j, k) in grid[(I, J, K), "interior"]:
    #      return -4*f[i, j, k]+f[i-1, j, k]+f[i+1, j, k]+f[i, j-1, k]+f[i, j+1, k]
    #    else:
            ...
    #return Field(f.domain, stencil)

tehrengruber

@stencil
def flux_x(phi):
    return flux(phi[I - 1], phi)

@stencil
def flux_y(phi):
    return flux(phi[J - 1], phi)

@stencil
def step(phi, fx, fy, dx, dy, dt):