fix variable support issue for paddle top_k_v2

src/core/tests/frontend/paddle/op_fuzzy.cpp

@@ -378,6 +378,7 @@ static const std::vector<std::string> models{
     std::string("top_k_v2_test_3"),
     std::string("top_k_v2_test_4"),
     std::string("top_k_v2_test_5"),
+    std::string("top_k_v2_test_6"),
     std::string("trilinear_downsample_false_0"),
     std::string("trilinear_downsample_false_1"),
     std::string("trilinear_downsample_true_0"),

src/core/tests/frontend/paddle/test_models/gen_scripts/generate_top_k_v2.py

@@ -9,27 +9,32 @@
 data_type = 'float32'
 
 
-def top_k_v2(name: str, x, k: int, axis=None, largest=True, sorted=True):
+def top_k_v2(name: str, x, k: int, axis=None, largest=True, sorted=True, k_is_var=True):
 
     paddle.enable_static()
 
+    k = np.array([k], dtype='int32') if k_is_var else k
+
     with paddle.static.program_guard(paddle.static.Program(), paddle.static.Program()):
         node_x = paddle.static.data(name='x', shape=x.shape, dtype='float32')
+        input_k = paddle.static.data(name='k', shape=[1], dtype='int32') if k_is_var else k
         value, indices = paddle.topk(
-            node_x, k=k, axis=axis, largest=largest, sorted=sorted, name="top_k")
+            node_x, k=input_k, axis=axis, largest=largest, sorted=sorted, name="top_k")
         indices = paddle.cast(indices, np.float32)
 
         cpu = paddle.static.cpu_places(1)
         exe = paddle.static.Executor(cpu[0])
         # startup program will call initializer to initialize the parameters.
         exe.run(paddle.static.default_startup_program())
 
+        feed_list = {'x': x, 'k': k} if k_is_var else {'x': x}
         outs = exe.run(
-            feed={'x': x},
+            feed=feed_list,
             fetch_list=[value, indices])
 
-        saveModel(name, exe, feedkeys=['x'], fetchlist=[value, indices], inputs=[
-                  x], outputs=outs, target_dir=sys.argv[1])
+        feedkey_list = ['x', 'k'] if k_is_var else ['x']
+        input_list = [x, k] if k_is_var else [x]
+        saveModel(name, exe, feedkeys=feedkey_list, fetchlist=[value, indices], inputs=input_list, outputs=outs, target_dir=sys.argv[1])
 
     return outs[0]
 
@@ -43,6 +48,7 @@ def main():
     top_k_v2("top_k_v2_test_4", data, k=7,
              axis=None, largest=True, sorted=True)
     top_k_v2("top_k_v2_test_5", data, k=6, axis=2, largest=False, sorted=True)
+    top_k_v2("top_k_v2_test_6", data, k=6, axis=2, largest=False, sorted=True, k_is_var=False)
 
 
 if __name__ == "__main__":

src/frontends/paddle/src/op/top_k_v2.cpp

@@ -13,7 +13,8 @@ NamedOutputs top_k_v2(const NodeContext& node) {
     Output<Node> k_expected_node;
     if (node.has_input("K")) {
         auto k_variable = node.get_input("K");
-        k_expected_node = std::make_shared<default_opset::Convert>(k_variable, element::i32);
+        auto k_var_node = std::make_shared<default_opset::Convert>(k_variable, element::i32);
+        k_expected_node = std::make_shared<default_opset::Squeeze>(k_var_node);
     } else {
         const auto k_expected = node.get_attribute<int>("k", 1);
         k_expected_node = default_opset::Constant::create(element::i32, {}, {k_expected});