def test_io_3d():
if not ti.has_pytorch():
return
import torch
n = 16
@ti.kernel
def torch_kernel(t: ti.ext_arr(), o: ti.ext_arr()):
for i in range(n):
for j in range(n):
for k in range(n):
o[i, j, k] = t[i, j, k] * t[i, j, k]
class Sqr(torch.autograd.Function):
@staticmethod
def forward(ctx, inp):
outp = torch.zeros_like(inp)
torch_kernel(inp, outp)
return outp
sqr = Sqr.apply
X = torch.tensor(2 * np.ones((n, n, n), dtype=np.float32),
requires_grad=True)
val = sqr(X).sum()
assert val == 2 * 2 * n * n * n
def torch_test(func):
import taichi as ti
if ti.has_pytorch():
# OpenGL somehow crashes torch test without a reason, unforturnately
return ti.archs_excluding(ti.opengl)(func)
else:
return lambda: None
def test_io_simple():
if not ti.has_pytorch():
return
import torch
n = 32
x1 = ti.var(ti.f32, shape=(n, n))
t1 = torch.tensor(2 * np.ones((n, n), dtype=np.float32))
x2 = ti.Matrix(2, 3, ti.f32, shape=(n, n))
t2 = torch.tensor(2 * np.ones((n, n, 2, 3), dtype=np.float32))
x1.from_torch(t1)
for i in range(n):
for j in range(n):
assert x1[i, j] == 2
x2.from_torch(t2)
for i in range(n):
for j in range(n):
for k in range(2):
for l in range(3):
assert x2[i, j][k, l] == 2
t3 = x2.to_torch()
assert (t2 == t3).all()
def test_io_simple():
if not ti.has_pytorch():
return
import torch
mat = ti.Matrix(2, 6, dt=ti.f32, shape=(), needs_grad=True)
zeros = torch.zeros((2, 6))
zeros[1, 2] = 3
mat.from_torch(zeros + 1)
assert mat[None][1, 2] == 4
zeros = mat.to_torch()
assert zeros[1, 2] == 4
def test_io():
if not ti.has_pytorch():
return
import torch
n = 32
@ti.kernel
def torch_kernel(t: ti.ext_arr(), o: ti.ext_arr()):
for i in range(n):
o[i] = t[i] * t[i]
@ti.kernel
def torch_kernel_2(t_grad: ti.ext_arr(), t: ti.ext_arr(),
o_grad: ti.ext_arr()):
for i in range(n):
t_grad[i] = 2 * t[i] * o_grad[i]
class Sqr(torch.autograd.Function):
@staticmethod
def forward(ctx, inp):
outp = torch.zeros_like(inp)
ctx.save_for_backward(inp)
torch_kernel(inp, outp)
return outp
@staticmethod
def backward(ctx, outp_grad):
outp_grad = outp_grad.contiguous()
inp_grad = torch.zeros_like(outp_grad)
inp, = ctx.saved_tensors
torch_kernel_2(inp_grad, inp, outp_grad)
return inp_grad
#, device=torch.device('cuda:0')
sqr = Sqr.apply
X = torch.tensor(2 * np.ones((n, ), dtype=np.float32), requires_grad=True)
sqr(X).sum().backward()
ret = X.grad.cpu()
for i in range(n):
assert ret[i] == 4
def test_torch_ad():
if not ti.has_pytorch():
return
import torch
n = 32
x = ti.var(ti.f32, shape=n, needs_grad=True)
y = ti.var(ti.f32, shape=n, needs_grad=True)
@ti.kernel
def torch_kernel():
for i in range(n):
# Do whatever complex operations here
y[n - i - 1] = x[i] * x[i]
class Sqr(torch.autograd.Function):
@staticmethod
def forward(ctx, inp):
x.from_torch(inp)
torch_kernel()
outp = y.to_torch()
return outp
@staticmethod
def backward(ctx, outp_grad):
ti.clear_all_gradients()
y.grad.from_torch(outp_grad)
torch_kernel.grad()
inp_grad = x.grad.to_torch()
return inp_grad
sqr = Sqr.apply
for i in range(10):
X = torch.tensor(2 * np.ones((n, ), dtype=np.float32),
requires_grad=True)
sqr(X).sum().backward()
ret = X.grad.cpu().numpy()
for j in range(n):
assert ret[j] == 4
import taichi as ti
import numpy as np
if ti.has_pytorch():
import torch
@ti.torch_test
def test_io_devices():
n = 32
x = ti.var(dt=ti.i32, shape=n)
@ti.kernel
def load(y: ti.ext_arr()):
for i in x:
x[i] = y[i] + 10
@ti.kernel
def inc():
for i in x:
x[i] += i
@ti.kernel
def store(y: ti.ext_arr()):
for i in x:
y[i] = x[i] * 2
devices = ['cpu']
if torch.cuda.is_available():
devices.append('cuda:0')
for device in devices:
with pytest.raises(TypeError,
match=r"unsupported operand type\(s\) for '&'"):
bitwise_float()
# @ti.test(arch=ti.cpu)
# def test_ternary_op():
# @ti.kernel
# def select():
# a = 1.1
# b = 3
# c = 3.6
# d = b if a else c
#
# with pytest.raises(TypeError,
# match="for 'select': 'f32', 'i32' and 'f32'"):
# select()
@pytest.mark.skipif(not ti.has_pytorch(), reason='Pytorch not installed.')
@ti.test(arch=[ti.cpu, ti.opengl])
def test_subscript():
a = ti.ndarray(ti.i32, shape=(10, 10))
@ti.kernel
def any_array(x: ti.any_arr()):
b = x[3, 1.1]
with pytest.raises(TypeError, match="indices must be integers"):
any_array(a)
def torch_test(func):
import taichi as ti
if ti.has_pytorch():
return ti.all_archs(func)
else:
return lambda: None
def torch_test(_func):
if ti.has_pytorch():
# OpenGL somehow crashes torch test without a reason, unforturnately
return ti.test(exclude=[opengl])(_func)
return lambda: None