与外部数组进行交互

External arrays refer to numpy.ndarray or torch.Tensor.

Taichi张量与外部数组之间的转换

使用 to_numpy/from_numpy/to_torch/from_torch：

import taichi as ti
import numpy as np
ti.init()
n = 4
m = 7
# Taichi tensors
val = ti.var(ti.i32, shape=(n, m))
vec = ti.Vector(3, dt=ti.i32, shape=(n, m))
mat = ti.Matrix(3, 4, dt=ti.i32, shape=(n, m))
# Scalar
arr = np.ones(shape=(n, m), dtype=np.int32)
val.from_numpy(arr)
arr = val.to_numpy()
# Vector
arr = np.ones(shape=(n, m, 3), dtype=np.int32)
vec.from_numpy(arr)
arr = np.ones(shape=(n, m, 3, 1), dtype=np.int32)
vec.from_numpy(arr)
arr = vec.to_numpy()
assert arr.shape == (n, m, 3)
arr = vec.to_numpy(keep_dims=True)
assert arr.shape == (n, m, 3, 1)
# Matrix
arr = np.ones(shape=(n, m, 3, 4), dtype=np.int32)
mat.from_numpy(arr)
arr = mat.to_numpy()
assert arr.shape == (n, m, 3, 4)

TODO: add API reference

Using external arrays as Taichi kernel parameters

外部数组参数的类型提示是 ti.ext_arr()。请参阅下面的示例。请注意，结构for循环不支持外部数组。

import taichi as ti
import numpy as np
ti.init()
n = 4
m = 7
val = ti.var(ti.i32, shape=(n, m))
@ti.kernel
def test_numpy(arr: ti.ext_arr()):
  for i in range(n):
    for j in range(m):
      arr[i, j] += i + j
a = np.empty(shape=(n, m), dtype=np.int32)
for i in range(n):
  for j in range(m):
    a[i, j] = i * j
test_numpy(a)
for i in range(n):
  for j in range(m):
    assert a[i, j] == i * j + i + j