test

Tencent · Nov 21, 2024 · f16caff · f16caff
1 parent f4cf6c2
commit f16caff
Show file tree

Hide file tree

Showing 4 changed files with 277 additions and 0 deletions.
diff --git a/tools/pnnx/tests/ncnn/test_torchaudio_InverseSpectrogram.py b/tools/pnnx/tests/ncnn/test_torchaudio_InverseSpectrogram.py
@@ -0,0 +1,73 @@
+# Tencent is pleased to support the open source community by making ncnn available.
+#
+# Copyright (C) 2024 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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 torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torchaudio
+
+class Model(nn.Module):
+    def __init__(self):
+        super(Model, self).__init__()
+
+        self.s0 = torchaudio.transforms.InverseSpectrogram(n_fft=64, window_fn=torch.hann_window, win_length=44, hop_length=16, pad=0, center=True, normalized='window')
+        self.s1 = torchaudio.transforms.InverseSpectrogram(n_fft=128, window_fn=torch.hann_window, win_length=128, hop_length=3, pad=0, center=True, onesided=True, normalized=False)
+        self.s2 = torchaudio.transforms.InverseSpectrogram(n_fft=512, window_fn=torch.hamming_window, win_length=256, hop_length=128, pad=0, center=True, onesided=True, normalized='frame_length')
+        self.s3 = torchaudio.transforms.InverseSpectrogram(n_fft=1024, window_fn=torch.hamming_window, win_length=512, hop_length=128, pad=0, center=True, onesided=True, normalized=False)
+
+    def forward(self, x, y, z, w):
+        x = torch.view_as_complex(x)
+        y = torch.view_as_complex(y)
+        z = torch.view_as_complex(z)
+        w = torch.view_as_complex(w)
+        out0 = self.s0(x)
+        out1 = self.s1(y)
+        out2 = self.s2(z)
+        out3 = self.s3(w)
+        return out0, out1, out2, out3
+
+def test():
+    net = Model()
+    net.eval()
+
+    torch.manual_seed(0)
+    x = torch.rand(33, 161, 2)
+    y = torch.rand(65, 77, 2)
+    z = torch.rand(257, 8, 2)
+    w = torch.rand(513, 4, 2)
+
+    a = net(x, y, z, w)
+
+    # export torchscript
+    mod = torch.jit.trace(net, (x, y, z, w))
+    mod.save("test_torchaudio_InverseSpectrogram.pt")
+
+    # torchscript to pnnx
+    import os
+    os.system("../../src/pnnx test_torchaudio_InverseSpectrogram.pt inputshape=[33,161,2],[65,77,2],[257,8,2],[513,4,2]")
+
+    # ncnn inference
+    import test_torchaudio_InverseSpectrogram_ncnn
+    b = test_torchaudio_InverseSpectrogram_ncnn.test_inference()
+
+    for a0, b0 in zip(a, b):
+        if not torch.allclose(a0, b0, 1e-4, 1e-4):
+            return False
+    return True
+
+if __name__ == "__main__":
+    if test():
+        exit(0)
+    else:
+        exit(1)
diff --git a/tools/pnnx/tests/ncnn/test_torchaudio_Spectrogram.py b/tools/pnnx/tests/ncnn/test_torchaudio_Spectrogram.py
@@ -0,0 +1,68 @@
+# Tencent is pleased to support the open source community by making ncnn available.
+#
+# Copyright (C) 2024 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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 torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torchaudio
+
+class Model(nn.Module):
+    def __init__(self):
+        super(Model, self).__init__()
+
+        self.s0 = torchaudio.transforms.Spectrogram(n_fft=64, window_fn=torch.hann_window, win_length=44, hop_length=16, pad=0, center=True, normalized='window', power=1)
+        self.s1 = torchaudio.transforms.Spectrogram(n_fft=128, window_fn=torch.hann_window, win_length=128, hop_length=3, pad=0, center=False, onesided=True, normalized=False, power=None)
+        self.s2 = torchaudio.transforms.Spectrogram(n_fft=512, window_fn=torch.hamming_window, win_length=256, hop_length=128, pad=0, center=True, pad_mode='constant', onesided=True, normalized='frame_length', power=2)
+        self.s3 = torchaudio.transforms.Spectrogram(n_fft=512, window_fn=torch.hamming_window, win_length=512, hop_length=128, pad=32, center=True, onesided=False, normalized=False, power=2)
+
+    def forward(self, x, y):
+        out0 = self.s0(x)
+        out1 = self.s1(x)
+        out2 = self.s2(y)
+        out3 = self.s3(y)
+        out1 = torch.view_as_real(out1)
+        return out0, out1, out2, out3
+
+def test():
+    net = Model()
+    net.eval()
+
+    torch.manual_seed(0)
+    x = torch.rand(2560)
+    y = torch.rand(1000)
+
+    a = net(x, y)
+
+    # export torchscript
+    mod = torch.jit.trace(net, (x, y))
+    mod.save("test_torchaudio_Spectrogram.pt")
+
+    # torchscript to pnnx
+    import os
+    os.system("../../src/pnnx test_torchaudio_Spectrogram.pt inputshape=[2560],[1000]")
+
+    # ncnn inference
+    import test_torchaudio_Spectrogram_ncnn
+    b = test_torchaudio_Spectrogram_ncnn.test_inference()
+
+    for a0, b0 in zip(a, b):
+        if not torch.allclose(a0, b0, 1e-4, 1e-4):
+            return False
+    return True
+
+if __name__ == "__main__":
+    if test():
+        exit(0)
+    else:
+        exit(1)
diff --git a/tools/pnnx/tests/test_torchaudio_InverseSpectrogram.py b/tools/pnnx/tests/test_torchaudio_InverseSpectrogram.py
@@ -0,0 +1,69 @@
+# Tencent is pleased to support the open source community by making ncnn available.
+#
+# Copyright (C) 2024 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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 torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torchaudio
+
+class Model(nn.Module):
+    def __init__(self):
+        super(Model, self).__init__()
+
+        self.s0 = torchaudio.transforms.InverseSpectrogram(n_fft=64, window_fn=torch.hann_window, win_length=44, hop_length=16, pad=0, center=True, normalized='window')
+        self.s1 = torchaudio.transforms.InverseSpectrogram(n_fft=128, window_fn=torch.hann_window, win_length=128, hop_length=3, pad=0, center=True, onesided=True, normalized=False)
+        self.s2 = torchaudio.transforms.InverseSpectrogram(n_fft=512, window_fn=torch.hamming_window, win_length=256, hop_length=128, pad=0, center=True, onesided=True, normalized='frame_length')
+        self.s3 = torchaudio.transforms.InverseSpectrogram(n_fft=512, window_fn=torch.hamming_window, win_length=512, hop_length=128, pad=0, center=True, onesided=False, normalized=False)
+
+    def forward(self, x, y, z, w):
+        out0 = self.s0(x)
+        out1 = self.s1(y)
+        out2 = self.s2(z)
+        out3 = self.s3(w)
+        return out0, out1, out2, out3
+
+def test():
+    net = Model()
+    net.eval()
+
+    torch.manual_seed(0)
+    x = torch.rand(3, 33, 161, dtype=torch.complex64)
+    y = torch.rand(1, 65, 77, dtype=torch.complex64)
+    z = torch.rand(257, 8, dtype=torch.complex64)
+    w = torch.rand(512, 4, dtype=torch.complex64)
+
+    a = net(x, y, z, w)
+
+    # export torchscript
+    mod = torch.jit.trace(net, (x, y, z, w))
+    mod.save("test_torchaudio_InverseSpectrogram.pt")
+
+    # torchscript to pnnx
+    import os
+    os.system("../src/pnnx test_torchaudio_InverseSpectrogram.pt inputshape=[3,33,161]c64,[1,65,77]c64,[257,8]c64,[512,4]c64")
+
+    # pnnx inference
+    import test_torchaudio_InverseSpectrogram_pnnx
+    b = test_torchaudio_InverseSpectrogram_pnnx.test_inference()
+
+    for a0, b0 in zip(a, b):
+        if not torch.allclose(a0, b0, 1e-4, 1e-4):
+            return False
+    return True
+
+if __name__ == "__main__":
+    if test():
+        exit(0)
+    else:
+        exit(1)
diff --git a/tools/pnnx/tests/test_torchaudio_Spectrogram.py b/tools/pnnx/tests/test_torchaudio_Spectrogram.py
@@ -0,0 +1,67 @@
+# Tencent is pleased to support the open source community by making ncnn available.
+#
+# Copyright (C) 2024 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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 torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torchaudio
+
+class Model(nn.Module):
+    def __init__(self):
+        super(Model, self).__init__()
+
+        self.s0 = torchaudio.transforms.Spectrogram(n_fft=64, window_fn=torch.hann_window, win_length=44, hop_length=16, pad=0, center=True, normalized='window', power=1)
+        self.s1 = torchaudio.transforms.Spectrogram(n_fft=128, window_fn=torch.hann_window, win_length=128, hop_length=3, pad=0, center=False, onesided=True, normalized=False, power=None)
+        self.s2 = torchaudio.transforms.Spectrogram(n_fft=512, window_fn=torch.hamming_window, win_length=256, hop_length=128, pad=0, center=True, pad_mode='constant', onesided=True, normalized='frame_length', power=2)
+        self.s3 = torchaudio.transforms.Spectrogram(n_fft=512, window_fn=torch.hamming_window, win_length=512, hop_length=128, pad=32, center=True, onesided=False, normalized=False, power=2)
+
+    def forward(self, x, y):
+        out0 = self.s0(x)
+        out1 = self.s1(x)
+        out2 = self.s2(y)
+        out3 = self.s3(y)
+        return out0, out1, out2, out3
+
+def test():
+    net = Model()
+    net.eval()
+
+    torch.manual_seed(0)
+    x = torch.rand(3, 2560)
+    y = torch.rand(1000)
+
+    a = net(x, y)
+
+    # export torchscript
+    mod = torch.jit.trace(net, (x, y))
+    mod.save("test_torchaudio_Spectrogram.pt")
+
+    # torchscript to pnnx
+    import os
+    os.system("../src/pnnx test_torchaudio_Spectrogram.pt inputshape=[3,2560],[1000]")
+
+    # pnnx inference
+    import test_torchaudio_Spectrogram_pnnx
+    b = test_torchaudio_Spectrogram_pnnx.test_inference()
+
+    for a0, b0 in zip(a, b):
+        if not torch.allclose(a0, b0, 1e-4, 1e-4):
+            return False
+    return True
+
+if __name__ == "__main__":
+    if test():
+        exit(0)
+    else:
+        exit(1)