#  Copyright (c) Meta Platforms, Inc. and affiliates.
#
#  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 numpy as np
from aitemplate.frontend import nn
from aitemplate.testing import detect_target


class CNNBlockBase(nn.Module):
    """
    A CNN block is assumed to have input channels, output channels and a stride.
    The input and output of `forward()` method must be NHWC tensors.
    The method can perform arbitrary computation but must match the given
    channels and stride specification.
    Attribute:
        in_channels (int):
        out_channels (int):
        stride (int):
    """

    def __init__(self, in_channels, out_channels, stride):
        """
        The `__init__` method of any subclass should also contain these arguments.
        Args:
            in_channels (int):
            out_channels (int):
            stride (int):
        """
        super().__init__()
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.stride = stride


class BasicStem(CNNBlockBase):
    """
    The standard ResNet stem (layers before the first residual block),
    with a conv, relu and max_pool.
    """

    def __init__(self, in_channels=3, out_channels=64, norm="BN", activation="ReLU"):
        super().__init__(in_channels, out_channels, 4)
        conv_op = None
        if detect_target().name() == "cuda":
            if activation == "ReLU":
                conv_op = nn.Conv2dBiasReluFewChannels
            elif activation == "Hardswish":
                conv_op = nn.Conv2dBiasHardswishFewChannels
            else:
                raise NotImplementedError
        else:
            if activation == "ReLU":
                conv_op = nn.Conv2dBiasRelu
            elif activation == "Hardswish":
                conv_op = nn.Conv2dBiasHardswish
            else:
                raise NotImplementedError
        self.conv1 = conv_op(in_channels, out_channels, 7, 2, 7 // 2)
        self.pool = nn.MaxPool2d(3, 2, 1)

    def forward(self, x):
        x = self.conv1(x)
        x = self.pool(x)
        return x


class BasicBlock(CNNBlockBase):
    """
    The basic residual block for ResNet-18 and ResNet-34 defined in :paper:`ResNet`,
    with two 3x3 conv layers and a projection shortcut if needed.
    """

    def __init__(self, in_channels, out_channels, *, stride=1, norm="BN"):
        super().__init__(in_channels, out_channels, stride)

    def forward(self, x):
        raise NotImplementedError()


class BottleneckBlock(CNNBlockBase):
    """
    The standard bottleneck residual block used by ResNet-50, 101 and 152
    defined in :paper:`ResNet`.  It contains 3 conv layers with kernels
    1x1, 3x3, 1x1, and a projection shortcut if needed.
    """

    def __init__(
        self,
        in_channels,
        out_channels,
        *,
        bottleneck_channels,
        stride=1,
        num_groups=1,
        norm="BN",
        activation="ReLU",
        stride_in_1x1=False,
        dilation=1,
    ):
        """
        Args:
            bottleneck_channels (int): number of output channels for the 3x3
                "bottleneck" conv layers.
            num_groups (int): number of groups for the 3x3 conv layer.
            norm (str or callable): normalization for all conv layers.
                See :func:`layers.get_norm` for supported format.
            stride_in_1x1 (bool): when stride>1, whether to put stride in the
                first 1x1 convolution or the bottleneck 3x3 convolution.
            dilation (int): the dilation rate of the 3x3 conv layer.
        """
        super().__init__(in_channels, out_channels, stride)

        if in_channels != out_channels:
            self.downsample_0 = nn.Conv2dBias(in_channels, out_channels, 1, stride, 0)
        else:
            self.downsample_0 = None

        # The original MSRA ResNet models have stride in the first 1x1 conv
        # The subsequent fb.torch.resnet and Caffe2 ResNe[X]t implementations have
        # stride in the 3x3 conv
        stride_1x1, stride_3x3 = (stride, 1) if stride_in_1x1 else (1, stride)

        conv_op = None
        conv_op_add = None
        if activation == "ReLU":
            conv_op = nn.Conv2dBiasRelu
            conv_op_add = nn.Conv2dBiasAddRelu
        elif activation == "Hardswish":
            conv_op = nn.Conv2dBiasHardswish
            conv_op_add = nn.Conv2dBiasAddHardswish
        else:
            raise NotImplementedError

        self.conv1 = conv_op(in_channels, bottleneck_channels, 1, stride_1x1, 0)

        self.conv2 = conv_op(
            bottleneck_channels,
            bottleneck_channels,
            3,
            stride_3x3,
            1 * dilation,
            dilation,
        )

        self.conv3 = conv_op_add(bottleneck_channels, out_channels, 1, 1, 0)

        # for layer in [self.conv1, self.conv2, self.conv3, self.shortcut]:
        #     if layer is not None:  # shortcut can be None
        #         weight_init.c2_msra_fill(layer)

        # Zero-initialize the last normalization in each residual branch,
        # so that at the beginning, the residual branch starts with zeros,
        # and each residual block behaves like an identity.
        # See Sec 5.1 in "Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour":
        # "For BN layers, the learnable scaling coefficient γ is initialized
        # to be 1, except for each residual block's last BN
        # where γ is initialized to be 0."

        # nn.init.constant_(self.conv3.norm.weight, 0)
        # TODO this somehow hurts performance when training GN models from scratch.
        # Add it as an option when we need to use this code to train a backbone.

    def forward(self, x):
        out = self.conv1(x)
        out = self.conv2(out)

        if self.downsample_0 is not None:
            downsample = self.downsample_0(x)
        else:
            downsample = x

        out = self.conv3(out, downsample)
        return out


class ResNet(nn.Module):
    """
    Implement :paper:`ResNet`.
    """

    def __init__(self, stem, stages, num_classes=None, out_features=None, freeze_at=0):
        """
        Args:
            stem (nn.Module): a stem module
            stages (list[list[CNNBlockBase]]): several (typically 4) stages,
                each contains multiple :class:`CNNBlockBase`.
            activation (str): activation function to use.
            num_classes (None or int): if None, will not perform classification.
                Otherwise, will create a linear layer.
            out_features (list[str]): name of the layers whose outputs should
                be returned in forward. Can be anything in "stem", "linear", or "res2" ...
                If None, will return the output of the last layer.
            freeze_at (int): The number of stages at the beginning to freeze.
                see :meth:`freeze` for detailed explanation.
        """
        super().__init__()
        self.stem = stem
        self.num_classes = num_classes

        current_stride = self.stem.stride
        self._out_feature_strides = {"stem": current_stride}
        self._out_feature_channels = {"stem": self.stem.out_channels}

        self.stage_names, self.stages = [], []

        if out_features is not None:
            # Avoid keeping unused layers in this module. They consume extra memory
            # and may cause allreduce to fail
            num_stages = max(
                [
                    {"layer1": 1, "layer2": 2, "layer3": 3, "layer4": 4}.get(f, 0)
                    for f in out_features
                ]
            )
            stages = stages[:num_stages]

        for i, blocks in enumerate(stages):
            assert len(blocks) > 0, len(blocks)
            for block in blocks:
                assert isinstance(block, CNNBlockBase), block

            name = "layer" + str(i + 1)
            stage = nn.Sequential(*blocks)

            self.add_module(name, stage)
            self.stage_names.append(name)
            self.stages.append(stage)

            self._out_feature_strides[name] = current_stride = int(
                current_stride * np.prod([k.stride for k in blocks])
            )
            self._out_feature_channels[name] = curr_channels = blocks[-1].out_channels

        self.stage_names = tuple(self.stage_names)  # Make it static for scripting

        if num_classes is not None:
            self.avgpool = nn.AvgPool2d(7, 1, 0)
            self.fc = nn.Linear(curr_channels, num_classes)

        if out_features is None:
            out_features = [name]
        self._out_features = out_features
        assert len(self._out_features)
        children = [x[0] for x in self.named_children()]
        for out_feature in self._out_features:
            assert out_feature in children, "Available children: {}".format(
                ", ".join(children)
            )
        self.reshape = nn.Reshape()

    def forward(self, x):
        """
        Args:
            x: Tensor of shape (N,C,H,W). H, W must be a multiple of ``self.size_divisibility``.
        Returns:
            dict[str->Tensor]: names and the corresponding features
        """
        # assert x.dim() == 4, f"ResNet takes an input of shape (N, C, H, W). Got {x.shape} instead!"
        outputs = {}
        x = self.stem(x)
        if "stem" in self._out_features:
            outputs["stem"] = x
        for name, stage in zip(self.stage_names, self.stages):
            x = stage(x)
            if name in self._out_features:
                outputs[name] = x
        if self.num_classes is not None:
            x = self.avgpool(x)
            x = self.fc(x)
            if x._rank() == 2:
                x = self.reshape(x, [x._size(0), 1, 1, x._size(1)])
            return x
        return outputs

    @staticmethod
    def make_stage(block_class, num_blocks, *, in_channels, out_channels, **kwargs):
        """
        Create a list of blocks of the same type that forms one ResNet stage.
        Args:
            block_class (type): a subclass of CNNBlockBase that's used to create all blocks in this
                stage. A module of this type must not change spatial resolution of inputs unless its
                stride != 1.
            num_blocks (int): number of blocks in this stage
            in_channels (int): input channels of the entire stage.
            out_channels (int): output channels of **every block** in the stage.
            kwargs: other arguments passed to the constructor of
                `block_class`. If the argument name is "xx_per_block", the
                argument is a list of values to be passed to each block in the
                stage. Otherwise, the same argument is passed to every block
                in the stage.
        Returns:
            list[CNNBlockBase]: a list of block module.
        Examples:
        ::
            stage = ResNet.make_stage(
                BottleneckBlock, 3, in_channels=16, out_channels=64,
                bottleneck_channels=16, num_groups=1,
                stride_per_block=[2, 1, 1],
                dilations_per_block=[1, 1, 2]
            )
        Usually, layers that produce the same feature map spatial size are defined as one
        "stage" (in :paper:`FPN`). Under such definition, ``stride_per_block[1:]`` should
        all be 1.
        """
        blocks = []
        for i in range(num_blocks):
            curr_kwargs = {}
            for k, v in kwargs.items():
                if k.endswith("_per_block"):
                    assert len(v) == num_blocks, (
                        f"Argument '{k}' of make_stage should have the "
                        f"same length as num_blocks={num_blocks}."
                    )
                    newk = k[: -len("_per_block")]
                    assert (
                        newk not in kwargs
                    ), f"Cannot call make_stage with both {k} and {newk}!"
                    curr_kwargs[newk] = v[i]
                else:
                    curr_kwargs[k] = v

            blocks.append(
                block_class(
                    in_channels=in_channels, out_channels=out_channels, **curr_kwargs
                )
            )
            in_channels = out_channels
        return blocks

    @staticmethod
    def make_default_stages(depth, block_class=None, **kwargs):
        """
        Created list of ResNet stages from pre-defined depth (one of 18, 34, 50, 101, 152).
        If it doesn't create the ResNet variant you need, please use :meth:`make_stage`
        instead for fine-grained customization.
        Args:
            depth (int): depth of ResNet
            block_class (type): the CNN block class. Has to accept
                `bottleneck_channels` argument for depth > 50.
                By default it is BasicBlock or BottleneckBlock, based on the
                depth.
            kwargs:
                other arguments to pass to `make_stage`. Should not contain
                stride and channels, as they are predefined for each depth.
        Returns:
            list[list[CNNBlockBase]]: modules in all stages; see arguments of
                :class:`ResNet.__init__`.
        """
        num_blocks_per_stage = {
            18: [2, 2, 2, 2],
            34: [3, 4, 6, 3],
            50: [3, 4, 6, 3],
            101: [3, 4, 23, 3],
            152: [3, 8, 36, 3],
        }[depth]
        if block_class is None:
            block_class = BasicBlock if depth < 50 else BottleneckBlock
        if depth < 50:
            in_channels = [64, 64, 128, 256]
            out_channels = [64, 128, 256, 512]
        else:
            in_channels = [64, 256, 512, 1024]
            out_channels = [256, 512, 1024, 2048]
        ret = []
        for (n, s, i, o) in zip(
            num_blocks_per_stage, [1, 2, 2, 2], in_channels, out_channels
        ):
            if depth >= 50:
                kwargs["bottleneck_channels"] = o // 4
            ret.append(
                ResNet.make_stage(
                    block_class=block_class,
                    num_blocks=n,
                    stride_per_block=[s] + [1] * (n - 1),
                    in_channels=i,
                    out_channels=o,
                    **kwargs,
                )
            )
        return ret


def make_stage(*args, **kwargs):
    """
    Deprecated alias for backward compatibiltiy.
    """
    return ResNet.make_stage(*args, **kwargs)


def build_resnet_backbone(depth, activation):
    """
    Create a ResNet instance from config.
    Returns:
        ResNet: a :class:`ResNet` instance.
    """
    norm = "BN"
    activation = activation
    num_groups = 1
    stride_in_1x1 = False
    num_groups = 1
    width_per_group = 64
    bottleneck_channels = num_groups * width_per_group
    in_channels = 64
    out_channels = 256

    stem = BasicStem(in_channels=3, out_channels=64, norm=norm, activation=activation)

    num_blocks_per_stage = {
        18: [2, 2, 2, 2],
        34: [3, 4, 6, 3],
        50: [3, 4, 6, 3],
        101: [3, 4, 23, 3],
        152: [3, 8, 36, 3],
    }[depth]

    stages = []

    for idx, stage_idx in enumerate(range(2, 6)):
        # res5_dilation is used this way as a convention in R-FCN & Deformable Conv paper
        dilation = 1
        first_stride = 1 if idx == 0 or (stage_idx == 5 and dilation == 2) else 2
        stage_kargs = {
            "num_blocks": num_blocks_per_stage[idx],
            "stride_per_block": [first_stride] + [1] * (num_blocks_per_stage[idx] - 1),
            "in_channels": in_channels,
            "out_channels": out_channels,
            "norm": norm,
            "activation": activation,
        }
        # Use BasicBlock for R18 and R34.
        if depth in [18, 34]:
            stage_kargs["block_class"] = BasicBlock
        else:
            stage_kargs["bottleneck_channels"] = bottleneck_channels
            stage_kargs["stride_in_1x1"] = stride_in_1x1
            stage_kargs["dilation"] = dilation
            stage_kargs["num_groups"] = num_groups
            stage_kargs["block_class"] = BottleneckBlock
        blocks = ResNet.make_stage(**stage_kargs)
        in_channels = out_channels
        out_channels *= 2
        bottleneck_channels *= 2
        stages.append(blocks)

    return ResNet(stem, stages, num_classes=1000)