388 lines
15 KiB
Python
388 lines
15 KiB
Python
# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
import unittest
|
|
|
|
import numpy as np
|
|
from op_test import (
|
|
convert_float_to_uint16,
|
|
convert_uint16_to_float,
|
|
get_device_place,
|
|
get_places,
|
|
is_custom_device,
|
|
)
|
|
|
|
import paddle
|
|
from paddle import base
|
|
from paddle.base import core
|
|
|
|
|
|
def np_sinc(x: np.ndarray):
|
|
tmp = np.sinc(x)
|
|
return np.where(~np.isnan(tmp), tmp, np.full_like(x, 1.0))
|
|
|
|
|
|
def np_sinc_gradient(x: np.ndarray):
|
|
x = np.pi * np.where(x == 0, 1.0e-20, x)
|
|
s = np.sin(x)
|
|
c = np.cos(x)
|
|
tmp = np.pi * (x * c - s) / x**2
|
|
return np.where(~np.isnan(tmp), tmp, np.full_like(x, 0.0))
|
|
|
|
|
|
class TestSincAPI(unittest.TestCase):
|
|
def setUp(self):
|
|
self.support_dtypes = [
|
|
'float32',
|
|
'float64',
|
|
]
|
|
self.place = get_places()
|
|
self.shapes = [[6], [16, 64]]
|
|
|
|
def test_dtype(self):
|
|
def run_dygraph(place):
|
|
paddle.disable_static(place)
|
|
for dtype in self.support_dtypes:
|
|
for shape in self.shapes:
|
|
x_data = np.random.rand(*shape).astype(dtype)
|
|
x = paddle.to_tensor(x_data)
|
|
x.stop_gradient = False
|
|
out = paddle.sinc(x)
|
|
out.backward()
|
|
x_grad = x.grad
|
|
out_expected = np_sinc(x_data)
|
|
np_grad_expected = np_sinc_gradient(x_data)
|
|
np.testing.assert_allclose(
|
|
out.numpy(), out_expected, rtol=1e-6, atol=1e-6
|
|
)
|
|
np.testing.assert_allclose(
|
|
x_grad.numpy(), np_grad_expected, rtol=1e-6, atol=0.02
|
|
)
|
|
|
|
def run_static(place):
|
|
paddle.enable_static()
|
|
for dtype in self.support_dtypes:
|
|
for shape in self.shapes:
|
|
x_data = np.random.rand(*shape).astype(dtype)
|
|
startup_program = paddle.static.Program()
|
|
main_program = paddle.static.Program()
|
|
exe = base.Executor(place)
|
|
with paddle.static.program_guard(
|
|
main_program, startup_program
|
|
):
|
|
x = paddle.static.data(
|
|
name='x', shape=shape, dtype=dtype
|
|
)
|
|
x.stop_gradient = False
|
|
res = paddle.sinc(x)
|
|
x_grad = paddle.static.gradients(res, x)
|
|
[static_result, static_grad_result] = exe.run(
|
|
feed={'x': x_data}, fetch_list=[res, x_grad]
|
|
)
|
|
out_expected = np_sinc(x_data)
|
|
np_grad_expected = np_sinc_gradient(x_data)
|
|
np.testing.assert_allclose(
|
|
static_result, out_expected, rtol=1e-6, atol=1e-6
|
|
)
|
|
np.testing.assert_allclose(
|
|
static_grad_result,
|
|
np_grad_expected,
|
|
rtol=1e-6,
|
|
atol=0.02,
|
|
)
|
|
|
|
for place in self.place:
|
|
run_dygraph(place)
|
|
run_static(place)
|
|
|
|
def test_zero(self):
|
|
def run_dygraph(place):
|
|
paddle.disable_static(place)
|
|
for dtype in self.support_dtypes:
|
|
for shape in self.shapes:
|
|
x_data = np.random.rand(*shape).astype(dtype)
|
|
mask = (
|
|
(np.random.rand(*shape) > 0.5)
|
|
.astype('int')
|
|
.astype(dtype)
|
|
)
|
|
x_data = x_data * mask
|
|
x = paddle.to_tensor(x_data)
|
|
x.stop_gradient = False
|
|
out = paddle.sinc(x)
|
|
out.backward()
|
|
x_grad = x.grad
|
|
out_expected = np_sinc(x_data)
|
|
np_grad_expected = np_sinc_gradient(x_data)
|
|
np.testing.assert_allclose(
|
|
out.numpy(), out_expected, rtol=1e-6, atol=1e-6
|
|
)
|
|
np.testing.assert_allclose(
|
|
x_grad.numpy(), np_grad_expected, rtol=1e-6, atol=0.02
|
|
)
|
|
|
|
for place in self.place:
|
|
run_dygraph(place)
|
|
|
|
def test_input_type_error(self):
|
|
with self.assertRaises(TypeError):
|
|
x = np.random.rand(6).astype('float32')
|
|
x = paddle.sinc(x)
|
|
|
|
def test_input_dype_error(self):
|
|
paddle.enable_static()
|
|
place = paddle.CPUPlace()
|
|
with self.assertRaises(TypeError):
|
|
x_data = np.random.rand(6).astype('int32')
|
|
startup_program = paddle.static.Program()
|
|
main_program = paddle.static.Program()
|
|
exe = base.Executor(place)
|
|
with paddle.static.program_guard(main_program, startup_program):
|
|
x = paddle.static.data(name='x', shape=[6], dtype='int32')
|
|
res = paddle.sinc(x)
|
|
static_result = exe.run(feed={'x': x_data}, fetch_list=[res])[0]
|
|
|
|
with self.assertRaises(TypeError):
|
|
x_data = np.random.rand(6).astype('int64')
|
|
startup_program = paddle.static.Program()
|
|
main_program = paddle.static.Program()
|
|
exe = base.Executor(place)
|
|
with paddle.static.program_guard(main_program, startup_program):
|
|
x = paddle.static.data(name='x', shape=[6], dtype='int64')
|
|
res = paddle.sinc(x)
|
|
static_result = exe.run(feed={'x': x_data}, fetch_list=[res])[0]
|
|
|
|
|
|
class TestSincInplaceAPI(unittest.TestCase):
|
|
def setUp(self):
|
|
self.support_dtypes = [
|
|
'float32',
|
|
'float64',
|
|
]
|
|
self.place = get_places()
|
|
self.shapes = [[6], [16, 64]]
|
|
|
|
def test_inplace(self):
|
|
def run_dygraph(place):
|
|
paddle.disable_static(place)
|
|
for dtype in self.support_dtypes:
|
|
for shape in self.shapes:
|
|
x_data = np.random.rand(*shape).astype(dtype)
|
|
x = paddle.to_tensor(x_data)
|
|
paddle.sinc_(x)
|
|
out_expected = np_sinc(x_data)
|
|
np.testing.assert_allclose(
|
|
x.numpy(), out_expected, rtol=1e-6, atol=1e-6
|
|
)
|
|
|
|
for place in self.place:
|
|
run_dygraph(place)
|
|
|
|
def test_inplace_input_type_error(self):
|
|
with self.assertRaises(TypeError):
|
|
x = np.random.rand(6).astype('float32')
|
|
paddle.sinc_(x)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not core.is_float16_supported(get_device_place()),
|
|
"core is not compiled with CUDA and not support the float16",
|
|
)
|
|
class TestSincAPIFP16(unittest.TestCase):
|
|
def setUp(self):
|
|
self.shapes = [[6], [16, 64]]
|
|
self.dtype = 'float16'
|
|
self.place = get_device_place()
|
|
|
|
def test_dtype(self):
|
|
def run_static(place):
|
|
paddle.enable_static()
|
|
for shape in self.shapes:
|
|
x_data = np.random.rand(*shape).astype(self.dtype)
|
|
startup_program = paddle.static.Program()
|
|
main_program = paddle.static.Program()
|
|
exe = base.Executor(place)
|
|
with paddle.static.program_guard(main_program, startup_program):
|
|
x = paddle.static.data(
|
|
name='x', shape=shape, dtype=self.dtype
|
|
)
|
|
x.stop_gradient = False
|
|
res = paddle.sinc(x)
|
|
x_grad = paddle.static.gradients(res, x)
|
|
[static_result, static_grad_result] = exe.run(
|
|
feed={'x': x_data}, fetch_list=[res, x_grad]
|
|
)
|
|
out_expected = np_sinc(x_data)
|
|
np_grad_expected = np_sinc_gradient(x_data)
|
|
np.testing.assert_allclose(
|
|
static_result, out_expected, rtol=1e-6, atol=1e-6
|
|
)
|
|
np.testing.assert_allclose(
|
|
static_grad_result, np_grad_expected, rtol=0.1, atol=0.1
|
|
)
|
|
|
|
run_static(self.place)
|
|
|
|
def test_zero(self):
|
|
def run_static(place):
|
|
paddle.enable_static()
|
|
for shape in self.shapes:
|
|
x_data = np.random.rand(*shape).astype(self.dtype)
|
|
mask = (
|
|
(np.random.rand(*shape) > 0.5)
|
|
.astype('int')
|
|
.astype(self.dtype)
|
|
)
|
|
x_data = x_data * mask
|
|
startup_program = paddle.static.Program()
|
|
main_program = paddle.static.Program()
|
|
exe = base.Executor(place)
|
|
with paddle.static.program_guard(main_program, startup_program):
|
|
x = paddle.static.data(
|
|
name='x', shape=shape, dtype=self.dtype
|
|
)
|
|
x.stop_gradient = False
|
|
res = paddle.sinc(x)
|
|
x_grad = paddle.static.gradients(res, x)
|
|
[static_result, static_grad_result] = exe.run(
|
|
feed={'x': x_data}, fetch_list=[res, x_grad]
|
|
)
|
|
out_expected = np_sinc(x_data)
|
|
np_grad_expected = np_sinc_gradient(x_data)
|
|
np.testing.assert_allclose(
|
|
static_result, out_expected, rtol=1e-6, atol=1e-6
|
|
)
|
|
np.testing.assert_allclose(
|
|
static_grad_result, np_grad_expected, rtol=0.1, atol=0.1
|
|
)
|
|
|
|
run_static(self.place)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not core.is_bfloat16_supported(get_device_place()),
|
|
"core is not compiled with CUDA and not support the bfloat16",
|
|
)
|
|
class TestSincAPIBF16(unittest.TestCase):
|
|
def setUp(self):
|
|
self.shapes = [[6], [16, 64]]
|
|
self.dtype = 'uint16'
|
|
self.place = get_device_place()
|
|
|
|
def test_dtype(self):
|
|
def run(place):
|
|
paddle.enable_static()
|
|
for shape in self.shapes:
|
|
x_data_np = np.random.rand(*shape).astype('float32')
|
|
x_data = convert_float_to_uint16(x_data_np)
|
|
startup_program = paddle.static.Program()
|
|
main_program = paddle.static.Program()
|
|
exe = base.Executor(place)
|
|
with paddle.static.program_guard(main_program, startup_program):
|
|
x = paddle.static.data(
|
|
name='x', shape=shape, dtype=self.dtype
|
|
)
|
|
x.stop_gradient = False
|
|
res = paddle.sinc(x)
|
|
x_grad = paddle.static.gradients(res, x)
|
|
[static_result, static_grad_result] = exe.run(
|
|
feed={'x': x_data}, fetch_list=[res, x_grad]
|
|
)
|
|
out_expected = np_sinc(x_data_np)
|
|
np_grad_expected = np_sinc_gradient(x_data_np)
|
|
result = convert_uint16_to_float(static_result)
|
|
grad_result = convert_uint16_to_float(static_grad_result)
|
|
np.testing.assert_allclose(
|
|
result, out_expected, rtol=1e-3, atol=1e-2
|
|
)
|
|
np.testing.assert_allclose(
|
|
grad_result, np_grad_expected, atol=0.2
|
|
)
|
|
|
|
run(self.place)
|
|
|
|
def test_zero(self):
|
|
def run(place):
|
|
paddle.enable_static()
|
|
for shape in self.shapes:
|
|
x_data_np = np.random.rand(*shape).astype('float32')
|
|
mask = (
|
|
(np.random.rand(*shape) > 0.5)
|
|
.astype('int')
|
|
.astype('float32')
|
|
)
|
|
x_data_np = x_data_np * mask
|
|
x_data = convert_float_to_uint16(x_data_np)
|
|
startup_program = paddle.static.Program()
|
|
main_program = paddle.static.Program()
|
|
exe = base.Executor(place)
|
|
with paddle.static.program_guard(main_program, startup_program):
|
|
x = paddle.static.data(
|
|
name='x', shape=shape, dtype=self.dtype
|
|
)
|
|
x.stop_gradient = False
|
|
res = paddle.sinc(x)
|
|
x_grad = paddle.static.gradients(res, x)
|
|
[static_result, static_grad_result] = exe.run(
|
|
feed={'x': x_data}, fetch_list=[res, x_grad]
|
|
)
|
|
out_expected = np_sinc(x_data_np)
|
|
np_grad_expected = np_sinc_gradient(x_data_np)
|
|
result = convert_uint16_to_float(static_result)
|
|
grad_result = convert_uint16_to_float(static_grad_result)
|
|
np.testing.assert_allclose(
|
|
result, out_expected, rtol=1e-3, atol=1e-2
|
|
)
|
|
np.testing.assert_allclose(
|
|
grad_result, np_grad_expected, atol=0.2
|
|
)
|
|
|
|
run(self.place)
|
|
|
|
|
|
class TestSincAPI_ZeroSize(unittest.TestCase):
|
|
def setUp(self):
|
|
self.support_dtypes = [
|
|
'float32',
|
|
'float64',
|
|
]
|
|
self.place = get_places()
|
|
self.shapes = [[0], [16, 0]]
|
|
|
|
def test_dygraph(self):
|
|
def run_dygraph(place):
|
|
paddle.disable_static(place)
|
|
for dtype in self.support_dtypes:
|
|
for shape in self.shapes:
|
|
x_data = np.random.rand(*shape).astype(dtype)
|
|
x = paddle.to_tensor(x_data)
|
|
x.stop_gradient = False
|
|
out = paddle.sinc(x)
|
|
out_expected = np_sinc(x_data)
|
|
np.testing.assert_allclose(
|
|
out.numpy(), out_expected, rtol=1e-6, atol=1e-6
|
|
)
|
|
loss = paddle.sum(out)
|
|
loss.backward()
|
|
np.testing.assert_allclose(x.grad.shape, x.shape)
|
|
|
|
for place in self.place:
|
|
run_dygraph(place)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|