wehub-resource-sync
212 lines
8.4 KiB
Python
212 lines
8.4 KiB
Python
# Copyright (c) 2025 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
|
|
|
|
import paddle
|
|
|
|
|
|
class TestSinOutAndParamDecorator(unittest.TestCase):
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
self.x_np = np.random.rand(3, 4).astype(np.float32)
|
|
self.test_types = ["decorator", "out", "out_decorator"]
|
|
|
|
def do_test(self, test_type):
|
|
x = paddle.to_tensor(self.x_np, stop_gradient=False)
|
|
if test_type == 'raw':
|
|
result = paddle.sin(x)
|
|
result.mean().backward()
|
|
return result, x.grad
|
|
elif test_type == 'decorator':
|
|
result = paddle.sin(input=x)
|
|
result.mean().backward()
|
|
return result, x.grad
|
|
elif test_type == 'out':
|
|
out = paddle.empty_like(x)
|
|
out.stop_gradient = False
|
|
paddle.sin(x, out=out)
|
|
out.mean().backward()
|
|
return out, x.grad
|
|
elif test_type == 'out_decorator':
|
|
out = paddle.empty_like(x)
|
|
out.stop_gradient = False
|
|
paddle.sin(input=x, out=out)
|
|
out.mean().backward()
|
|
return out, x.grad
|
|
else:
|
|
raise ValueError(f"Unknown test type: {test_type}")
|
|
|
|
def test_all(self):
|
|
out_std, grad_std = self.do_test('raw')
|
|
for test_type in self.test_types:
|
|
out, grad = self.do_test(test_type)
|
|
np.testing.assert_allclose(out.numpy(), out_std.numpy(), rtol=1e-7)
|
|
np.testing.assert_allclose(
|
|
grad.numpy(), grad_std.numpy(), rtol=1e-7
|
|
)
|
|
|
|
|
|
class TestSinSleefVectorized(unittest.TestCase):
|
|
"""Test sin with shapes that exercise Sleef vectorized paths.
|
|
|
|
For AVX2:
|
|
- float32: VEC_SIZE = 8, so shapes >= 8 trigger vectorized path
|
|
- float64: VEC_SIZE = 4, so shapes >= 4 trigger vectorized path
|
|
|
|
Test both:
|
|
1. Shapes that are exact multiples of VEC_SIZE (only vectorized loop)
|
|
2. Shapes with remainder (vectorized loop + scalar tail)
|
|
|
|
Note: If MKL is available at runtime, the MKL VML path (mkl_sin) will be
|
|
triggered instead (see sleef_vectorized_math.h L611-612 for float,
|
|
L647-648 for double). Both paths produce correct results and are
|
|
tested through these tests.
|
|
"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_sin_float32_vectorized_exact(self):
|
|
"""Test float32 sin with shape that's exact multiple of 8.
|
|
Covers vsin_avx2_f32 main loop (lines 79-83).
|
|
"""
|
|
# Shape 16 = 8 * 2, exercises only vectorized loop
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(16,)).astype(np.float32)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-5, atol=1e-5
|
|
)
|
|
|
|
def test_sin_float32_vectorized_with_tail(self):
|
|
"""Test float32 sin with shape that has remainder when divided by 8.
|
|
Covers vsin_avx2_f32 both main loop (79-83) and scalar tail (86-88).
|
|
"""
|
|
# Shape 13 = 8 + 5, exercises both vectorized loop and scalar tail
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(13,)).astype(np.float32)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-5, atol=1e-5
|
|
)
|
|
|
|
def test_sin_float64_vectorized_exact(self):
|
|
"""Test float64 sin with shape that's exact multiple of 4.
|
|
Covers vsin_avx2_f64 main loop (lines 112-116).
|
|
"""
|
|
# Shape 12 = 4 * 3, exercises only vectorized loop
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(12,)).astype(np.float64)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-10, atol=1e-10
|
|
)
|
|
|
|
def test_sin_float64_vectorized_with_tail(self):
|
|
"""Test float64 sin with shape that has remainder when divided by 4.
|
|
Covers vsin_avx2_f64 both main loop (112-116) and scalar tail (118-120).
|
|
"""
|
|
# Shape 11 = 4 * 2 + 3, exercises both vectorized loop and scalar tail
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(11,)).astype(np.float64)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-10, atol=1e-10
|
|
)
|
|
|
|
def test_sin_float32_large_shape(self):
|
|
"""Test float32 sin with large shape for comprehensive coverage.
|
|
Tests MKL VML path (mkl_sin at sleef_vectorized_math.h L611-612)
|
|
if MKL is available, otherwise Sleef vectorized path.
|
|
"""
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(1024,)).astype(np.float32)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-5, atol=1e-5
|
|
)
|
|
|
|
def test_sin_float64_large_shape(self):
|
|
"""Test float64 sin with large shape for comprehensive coverage.
|
|
Tests MKL VML path (mkl_sin at sleef_vectorized_math.h L647-648)
|
|
if MKL is available, otherwise Sleef vectorized path.
|
|
"""
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(1024,)).astype(np.float64)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-10, atol=1e-10
|
|
)
|
|
|
|
def test_sin_float32_2d_shape(self):
|
|
"""Test float32 sin with 2D shape to verify flattened processing."""
|
|
# Shape (4, 5) = 20 elements, exercises vectorized path
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(4, 5)).astype(np.float32)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-5, atol=1e-5
|
|
)
|
|
|
|
def test_sin_float64_2d_shape(self):
|
|
"""Test float64 sin with 2D shape to verify flattened processing."""
|
|
# Shape (3, 5) = 15 elements, exercises vectorized path with tail
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(3, 5)).astype(np.float64)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-10, atol=1e-10
|
|
)
|
|
|
|
def test_sin_float32_small_shape_fallback(self):
|
|
"""Test float32 sin with small shape (numel < 8) to cover Eigen fallback path.
|
|
Covers VectorizedSinImpl fallback branch (lines 74-80 in activation_impl.h).
|
|
"""
|
|
# Shape 5 < 8, triggers Eigen fallback instead of SIMD
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(5,)).astype(np.float32)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-5, atol=1e-5
|
|
)
|
|
|
|
def test_sin_float64_small_shape_fallback(self):
|
|
"""Test float64 sin with small shape (numel < 8) to cover Eigen fallback path.
|
|
Covers VectorizedSinImpl fallback branch (lines 74-80 in activation_impl.h).
|
|
"""
|
|
# Shape 3 < 8, triggers Eigen fallback instead of SIMD
|
|
x_np = np.random.uniform(-np.pi, np.pi, size=(3,)).astype(np.float64)
|
|
x = paddle.to_tensor(x_np, place=paddle.CPUPlace())
|
|
result = paddle.sin(x)
|
|
expected = np.sin(x_np)
|
|
np.testing.assert_allclose(
|
|
result.numpy(), expected, rtol=1e-10, atol=1e-10
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|