# 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
from collections import OrderedDict

from paddle.distributed.auto_parallel.static.dist_attribute import (
    DistTensorSpec,
    TensorDistAttr,
)
from paddle.distributed.fleet import auto
from paddle.framework import core


class TestIndexSelectSPMDRule(unittest.TestCase):
    def setUp(self):
        x_shape = [4, 8, 16]
        index_shape = [3]
        process_mesh = auto.ProcessMesh(mesh=[[0, 1, 2], [3, 4, 5]])
        self.attrs = OrderedDict()
        self.attrs['axis'] = 1
        self.rule = core.get_phi_spmd_rule("index_select")

        x_dist_attr = TensorDistAttr()
        x_dist_attr.dims_mapping = [-1, -1, -1]
        x_dist_attr.process_mesh = process_mesh
        self.x_spec = DistTensorSpec(x_shape, x_dist_attr)

        index_dist_attr = TensorDistAttr()
        index_dist_attr.dims_mapping = [-1]
        index_dist_attr.process_mesh = process_mesh
        self.index_spec = DistTensorSpec(index_shape, index_dist_attr)

    def test_index_select_forward(self):
        # [-1, -1, -1], [0], axis=1 --> [-1, -1, -1], [0], [-1, 0, -1]
        self.x_spec.set_dims_mapping([-1, -1, -1])
        self.index_spec.set_dims_mapping([0])
        self.attrs["axis"] = 1
        result_dist_attrs = self.rule.infer_forward(
            self.x_spec,
            self.index_spec,
            self.attrs['axis'],
        )
        self.assertEqual(len(result_dist_attrs), 2)
        inferred_input_dist_attrs = result_dist_attrs[0]
        inferred_output_dist_attrs = result_dist_attrs[1]
        self.assertEqual(len(inferred_input_dist_attrs), 2)
        self.assertEqual(len(inferred_output_dist_attrs), 1)
        self.assertEqual(
            inferred_input_dist_attrs[0].dims_mapping, [-1, -1, -1]
        )
        self.assertEqual(inferred_input_dist_attrs[1].dims_mapping, [0])
        self.assertEqual(
            inferred_output_dist_attrs[0].dims_mapping, [-1, 0, -1]
        )

        # [-1, 0, -1], [0], axis=0 -> [-1, 0, -1], [-1], [-1, 0, -1]
        self.x_spec.set_dims_mapping([-1, 0, -1])
        self.index_spec.set_dims_mapping([0])
        self.attrs["axis"] = 0
        result_dist_attrs = self.rule.infer_forward(
            self.x_spec,
            self.index_spec,
            self.attrs['axis'],
        )
        self.assertEqual(len(result_dist_attrs), 2)
        inferred_input_dist_attrs = result_dist_attrs[0]
        inferred_output_dist_attrs = result_dist_attrs[1]
        self.assertEqual(len(inferred_input_dist_attrs), 2)
        self.assertEqual(len(inferred_output_dist_attrs), 1)
        self.assertEqual(inferred_input_dist_attrs[0].dims_mapping, [-1, 0, -1])
        self.assertEqual(inferred_input_dist_attrs[1].dims_mapping, [-1])
        self.assertEqual(
            inferred_output_dist_attrs[0].dims_mapping, [-1, 0, -1]
        )

        # [-1, 1, -1], [0], axis=0 -> [-1, 1, -1], [0], [0, 1, -1]
        self.x_spec.set_dims_mapping([-1, 1, -1])
        self.index_spec.set_dims_mapping([0])
        self.attrs["axis"] = 0
        result_dist_attrs = self.rule.infer_forward(
            self.x_spec,
            self.index_spec,
            self.attrs['axis'],
        )
        self.assertEqual(len(result_dist_attrs), 2)
        inferred_input_dist_attrs = result_dist_attrs[0]
        inferred_output_dist_attrs = result_dist_attrs[1]
        self.assertEqual(len(inferred_input_dist_attrs), 2)
        self.assertEqual(len(inferred_output_dist_attrs), 1)
        self.assertEqual(inferred_input_dist_attrs[0].dims_mapping, [-1, 1, -1])
        self.assertEqual(inferred_input_dist_attrs[1].dims_mapping, [0])
        self.assertEqual(inferred_output_dist_attrs[0].dims_mapping, [0, 1, -1])

        # [-1, 1, -1], [0], axis=1 -> [-1, -1, -1], [0], [-1, 0, -1]
        self.x_spec.set_dims_mapping([-1, 1, -1])
        self.index_spec.set_dims_mapping([0])
        self.attrs["axis"] = 1
        result_dist_attrs = self.rule.infer_forward(
            self.x_spec,
            self.index_spec,
            self.attrs['axis'],
        )
        self.assertEqual(len(result_dist_attrs), 2)
        inferred_input_dist_attrs = result_dist_attrs[0]
        inferred_output_dist_attrs = result_dist_attrs[1]
        self.assertEqual(len(inferred_input_dist_attrs), 2)
        self.assertEqual(len(inferred_output_dist_attrs), 1)
        self.assertEqual(
            inferred_input_dist_attrs[0].dims_mapping, [-1, -1, -1]
        )
        self.assertEqual(inferred_input_dist_attrs[1].dims_mapping, [0])
        self.assertEqual(
            inferred_output_dist_attrs[0].dims_mapping, [-1, 0, -1]
        )

    def test_index_select_backward(self):
        # [-1, -1, -1], [0], [-1, 0, -1], axis=1 --> [-1, -1, -1], [0], [-1, 0, -1], [-1, -1, -1](partial on axis=1 with 0)
        self.out_grad_spec = DistTensorSpec(self.x_spec)
        self.out_grad_spec.shape = [4, 3, 16]
        self.out_grad_spec.set_dims_mapping([-1, 0, -1])
        self.x_spec.set_dims_mapping([-1, -1, -1])
        self.index_spec.set_dims_mapping([0])
        self.attrs["axis"] = 1

        result_dist_attrs = self.rule.infer_backward(
            self.x_spec,
            self.index_spec,
            self.out_grad_spec,
            self.attrs['axis'],
        )
        self.assertEqual(len(result_dist_attrs), 2)
        inferred_input_dist_attrs = result_dist_attrs[0]
        inferred_output_dist_attrs = result_dist_attrs[1]
        self.assertEqual(len(inferred_input_dist_attrs), 3)
        self.assertEqual(len(inferred_output_dist_attrs), 1)

        self.assertEqual(
            inferred_input_dist_attrs[0].dims_mapping, [-1, -1, -1]
        )
        self.assertEqual(inferred_input_dist_attrs[1].dims_mapping, [0])
        self.assertEqual(inferred_input_dist_attrs[2].dims_mapping, [-1, 0, -1])
        self.assertEqual(
            inferred_output_dist_attrs[0].dims_mapping, [-1, -1, -1]
        )
        self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
        self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {0})

        # [1, -1, -1], [0], [1, 0, -1], axis=1 -> [1, -1, -1], [0], [1, 0, -1], [1, -1, -1](partial on axis=1 with mesh_dim 0)
        self.x_spec.set_dims_mapping([1, -1, -1])
        self.x_spec.shape = [3, 4, 2]
        self.index_spec.set_dims_mapping([0])
        self.out_grad_spec = DistTensorSpec(self.x_spec)
        self.out_grad_spec.shape = [3, 4, 2]
        self.out_grad_spec.set_dims_mapping([1, 0, -1])
        self.attrs["axis"] = 1

        result_dist_attrs = self.rule.infer_backward(
            self.x_spec,
            self.index_spec,
            self.out_grad_spec,
            self.attrs['axis'],
        )
        self.assertEqual(len(result_dist_attrs), 2)
        inferred_input_dist_attrs = result_dist_attrs[0]
        inferred_output_dist_attrs = result_dist_attrs[1]
        self.assertEqual(len(inferred_input_dist_attrs), 3)
        self.assertEqual(len(inferred_output_dist_attrs), 1)
        self.assertEqual(inferred_input_dist_attrs[0].dims_mapping, [1, -1, -1])
        self.assertEqual(inferred_input_dist_attrs[1].dims_mapping, [0])
        self.assertEqual(inferred_input_dist_attrs[2].dims_mapping, [1, 0, -1])
        self.assertEqual(
            inferred_output_dist_attrs[0].dims_mapping, [1, -1, -1]
        )
        self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
        self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {0})


if __name__ == "__main__":
    unittest.main()