【Hackathon 5th No.96】add paddle unstack op

src/frontends/paddle/src/op/unstack.cpp

@@ -0,0 +1,36 @@
+// Copyright (C) 2018-2023 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include "default_opset.hpp"
+#include "openvino/frontend/paddle/node_context.hpp"
+
+namespace ov {
+namespace frontend {
+namespace paddle {
+namespace op {
+NamedOutputs unstack(const NodeContext& node) {
+    auto data = node.get_input("X");
+    auto input_shape = data.get_partial_shape();
+    PADDLE_OP_CHECK(node, input_shape.rank().is_static(), "rank of input data should be static");
+    auto dim = node.get_attribute<int32_t>("axis", 0);
+    if (dim < 0) {
+        dim = dim + static_cast<int32_t>(input_shape.rank().get_length());
+    }
+    auto axis = default_opset::Constant::create(element::i32, {}, {dim});
+    auto shape = input_shape.get_shape();
+    auto splits = std::make_shared<default_opset::Split>(data, axis, shape.at(dim));
+    auto split_outputs = splits->outputs();
+    NamedOutputs named_outputs;
+    auto out_names = node.get_output_names();
+    auto it = std::find(out_names.begin(), out_names.end(), "Y");
+    PADDLE_OP_CHECK(node, it != out_names.end(), "Expected output not found");
+    for (const auto& split_output : split_outputs) {
+        named_outputs[*it].push_back(std::make_shared<default_opset::Squeeze>(split_output, axis));
+    }
+    return named_outputs;
+}
+}  // namespace op
+}  // namespace paddle
+}  // namespace frontend
+}  // namespace ov

src/frontends/paddle/src/op_table.cpp

@@ -120,6 +120,7 @@ OP_CONVERTER(top_k_v2);
 OP_CONVERTER(transpose2);
 OP_CONVERTER(trilinear_interp_v2);
 OP_CONVERTER(unsqueeze);
+OP_CONVERTER(unstack);
 OP_CONVERTER(where);
 OP_CONVERTER(while_);
 OP_CONVERTER(write_to_array);
@@ -249,6 +250,7 @@ std::map<std::string, CreatorFunction> get_supported_ops() {
             {"transpose2", op::transpose2},
             {"trilinear_interp_v2", op::trilinear_interp_v2},
             {"unsqueeze2", op::unsqueeze},
+            {"unstack", op::unstack},
             {"where", op::where},
             {"while", op::while_},
             {"write_to_array", op::write_to_array},

src/frontends/paddle/tests/op_fuzzy.cpp

@@ -565,6 +565,11 @@ static const std::vector<std::string> models{
     std::string("trilinear_upsample_scales2/trilinear_upsample_scales2.pdmodel"),
     std::string("trilinear_upsample_true_0/trilinear_upsample_true_0.pdmodel"),
     std::string("unsqueeze"),
+    std::string("unstack_1"),
+    std::string("unstack_2"),
+    std::string("unstack_3"),
+    std::string("unstack_4"),
+    std::string("unstack_5"),
     std::string("where_1"),
     std::string("where_2"),
     std::string("where_3"),

src/frontends/paddle/tests/test_models/gen_scripts/generate_unstack.py

@@ -0,0 +1,46 @@
+# Copyright (C) 2018-2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+#
+# unstack paddle model generator
+#
+import paddle
+import numpy as np
+from save_model import saveModel
+import sys
+
+
+def unstack(name: str, x, axis):
+    paddle.enable_static()
+
+    with paddle.static.program_guard(paddle.static.Program(), paddle.static.Program()):
+        x_node = paddle.static.data(name="x", shape=x.shape, dtype=x.dtype)
+        out = paddle.unstack(x_node, axis) if axis is not None else paddle.unstack(x_node)
+        place = paddle.CPUPlace()
+        exe = paddle.static.Executor(place)
+        exe.run(paddle.static.default_startup_program())
+        outs = exe.run(feed={"x": x}, fetch_list=[out])
+        saveModel(name, exe, feedkeys=['x'], fetchlist=out, inputs=[x], outputs=outs, target_dir=sys.argv[1])
+
+    return outs
+
+
+def main():
+    dtype = np.float32
+    x = np.random.randn(2, 3, 4).astype(dtype)
+    unstack(name='unstack_1', x=x, axis=0)
+
+    dtype = np.int32
+    x = np.random.randn(2, 3, 4).astype(dtype)
+    unstack(name='unstack_2', x=x, axis=1)
+
+    dtype = np.int64
+    x = np.random.randn(3, 4).astype(dtype)
+    unstack(name='unstack_3', x=x, axis=-1)
+    unstack(name='unstack_4', x=x, axis=None)
+
+    x = np.random.randn(2, 1, 4).astype(dtype)
+    unstack(name='unstack_5', x=x, axis=0)
+
+if __name__ == "__main__":
+    main()