[Torch] Add support for max_pool1d

python/tvm/relay/frontend/pytorch.py

@@ -213,12 +213,33 @@ def _impl(inputs, input_types):
         pool_size = _infer_shape(inputs[1])
         strides = _infer_shape(inputs[2])
         padding = _infer_shape(inputs[3])
-
+        dilation = _infer_shape(inputs[4])
         ceil_mode = int(inputs[5])
 
+        if dilation != (1, 1):
+            msg = "MaxPool2d with dilation %s is not implemented" % (str(dilation), )
+            raise NotImplementedError(msg)
+
         return _op.nn.max_pool2d(data, pool_size, strides, padding, "NCHW", ceil_mode)
     return _impl
 
+def _maxpool_1d():
+    def _impl(inputs, input_types):
+        data = inputs[0]
+
+        pool_size = _infer_shape(inputs[1])
+        strides = _infer_shape(inputs[2])
+        padding = _infer_shape(inputs[3])
+        dilation = _infer_shape(inputs[4])
+        ceil_mode = int(inputs[5])
+
+        if dilation != (1,):
+            msg = "MaxPool1d with dilation %s is not implemented" % (str(dilation), )
+            raise NotImplementedError(msg)
+
+        return _op.nn.max_pool1d(data, pool_size, strides, padding, "NCW", ceil_mode)
+    return _impl
+
 def _hardtanh():
     def _impl(inputs, input_types):
         a = inputs[0]
@@ -863,6 +884,7 @@ def _wrap_const(c):
     "aten::adaptive_max_pool2d"             : _adaptive_max_pool_2d(),
     "aten::max_pool2d"                      : _maxpool_2d(),
     "aten::max_pool2d_with_indices"         : _maxpool_2d(),
+    "aten::max_pool1d"                      : _maxpool_1d(),
     "aten::hardtanh"                        : _hardtanh(),
     "aten::hardtanh_"                       : _hardtanh(),
     "aten::_convolution"                    : _convolution(),

src/relay/op/nn/pooling.cc

@@ -987,6 +987,10 @@ Array<te::Tensor> Pool1DCompute(const Attrs& attrs,
       << " or 4-D input (e.g. NCWc on for vector instructions)"
       << " or 5-D input (e.g. NCWnc for tensor accelerators)";
 
+  if (param->padding.size() == 1) {
+    padding.push_back(padding[0]);
+  }
+
   if (mode == topi::nn::kAvgPool) {
     bool count_include_pad = reinterpret_cast<const AvgPool1DAttrs*>(param)->count_include_pad;
     return Array<te::Tensor>{

tests/python/frontend/pytorch/test_forward.py

@@ -351,7 +351,7 @@ def forward(self, *args):
     verify_model(AdaptiveAvgPool2D1().float().eval(), input_data=input_data)
     verify_model(AdaptiveAvgPool2D2().float().eval(), input_data=input_data)
 
-def test_forward_maxpool():
+def test_forward_maxpool2d():
     torch.set_grad_enabled(False)
     input_shape = [1, 3, 10, 10]
 
@@ -363,9 +363,35 @@ class MaxPool2D2(Module):
         def forward(self, *args):
             return torch.nn.MaxPool2d(kernel_size=[10, 10])(args[0])
 
+    class MaxPool2D3(Module):
+        def forward(self, *args):
+            return torch.nn.MaxPool2d(kernel_size=[4, 4], padding=2, stride=2)(args[0])
+
     input_data = torch.rand(input_shape).float()
     verify_model(MaxPool2D1().float().eval(), input_data=input_data)
     verify_model(MaxPool2D2().float().eval(), input_data=input_data)
+    verify_model(MaxPool2D3().float().eval(), input_data=input_data)
+
+def test_forward_maxpool1d():
+    torch.set_grad_enabled(False)
+    input_shape = [1, 3, 10]
+
+    class MaxPool1D1(Module):
+        def forward(self, *args):
+            return torch.nn.MaxPool1d(kernel_size=1)(args[0])
+
+    class MaxPool1D2(Module):
+        def forward(self, *args):
+            return torch.nn.MaxPool1d(kernel_size=10)(args[0])
+
+    class MaxPool1D3(Module):
+        def forward(self, *args):
+            return torch.nn.MaxPool1d(kernel_size=4, padding=2, stride=2)(args[0])
+
+    input_data = torch.rand(input_shape).float()
+    verify_model(MaxPool1D1().float().eval(), input_data=input_data)
+    verify_model(MaxPool1D2().float().eval(), input_data=input_data)
+    verify_model(MaxPool1D3().float().eval(), input_data=input_data)
 
 def test_forward_avgpool():
     torch.set_grad_enabled(False)
@@ -1028,7 +1054,8 @@ def forward(self, xs):
     test_forward_concatenate()
     test_forward_relu()
     test_forward_adaptiveavgpool()
-    test_forward_maxpool()
+    test_forward_maxpool2d()
+    test_forward_maxpool1d()
     test_forward_hardtanh()
     test_forward_conv()
     test_forward_threshold()