import math
import torch
import torch.nn as nn
from torch.nn import functional as F
import numpy as np
from einops import rearrange
from typing import Optional, Any

from .distributions import DiagonalGaussianDistribution
from .config import Config, AttnMode


def nonlinearity(x):
    # swish
    return x * torch.sigmoid(x)


def Normalize(in_channels, num_groups=32):
    return torch.nn.GroupNorm(
        num_groups=num_groups, num_channels=in_channels, eps=1e-6, affine=True
    )


class Upsample(nn.Module):
    def __init__(self, in_channels, with_conv):
        super().__init__()
        self.with_conv = with_conv
        if self.with_conv:
            self.conv = torch.nn.Conv2d(
                in_channels, in_channels, kernel_size=3, stride=1, padding=1
            )

    def forward(self, x):
        x = torch.nn.functional.interpolate(x, scale_factor=2, mode="bilinear", align_corners=False, antialias=True)
        if self.with_conv:
            x = self.conv(x)
        return x


class Downsample(nn.Module):
    def __init__(self, in_channels, with_conv):
        super().__init__()
        self.with_conv = with_conv
        if self.with_conv:
            # no asymmetric padding in torch conv, must do it ourselves
            self.conv = torch.nn.Conv2d(
                in_channels, in_channels, kernel_size=3, stride=2, padding=0
            )

    def forward(self, x):
        if self.with_conv:
            pad = (0, 1, 0, 1)
            x = torch.nn.functional.pad(x, pad, mode="constant", value=0)
            x = self.conv(x)
        else:
            x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2)
        return x


class ResnetBlock(nn.Module):
    def __init__(
        self,
        *,
        in_channels,
        out_channels=None,
        conv_shortcut=False,
        dropout,
        temb_channels=512,
    ):
        super().__init__()
        self.in_channels = in_channels
        out_channels = in_channels if out_channels is None else out_channels
        self.out_channels = out_channels
        self.use_conv_shortcut = conv_shortcut

        self.norm1 = Normalize(in_channels)
        self.conv1 = torch.nn.Conv2d(
            in_channels, out_channels, kernel_size=3, stride=1, padding=1
        )
        if temb_channels > 0:
            self.temb_proj = torch.nn.Linear(temb_channels, out_channels)
        self.norm2 = Normalize(out_channels)
        self.dropout = torch.nn.Dropout(dropout)
        self.conv2 = torch.nn.Conv2d(
            out_channels, out_channels, kernel_size=3, stride=1, padding=1
        )
        if self.in_channels != self.out_channels:
            if self.use_conv_shortcut:
                self.conv_shortcut = torch.nn.Conv2d(
                    in_channels, out_channels, kernel_size=3, stride=1, padding=1
                )
            else:
                self.nin_shortcut = torch.nn.Conv2d(
                    in_channels, out_channels, kernel_size=1, stride=1, padding=0
                )

    def forward(self, x, temb):
        h = x
        h = self.norm1(h)
        h = nonlinearity(h)
        h = self.conv1(h)

        if temb is not None:
            h = h + self.temb_proj(nonlinearity(temb))[:, :, None, None]

        h = self.norm2(h)
        h = nonlinearity(h)
        h = self.dropout(h)
        h = self.conv2(h)

        if self.in_channels != self.out_channels:
            if self.use_conv_shortcut:
                x = self.conv_shortcut(x)
            else:
                x = self.nin_shortcut(x)

        return x + h


class AttnBlock(nn.Module):
    def __init__(self, in_channels):
        super().__init__()
        print(f"building AttnBlock (vanilla) with {in_channels} in_channels")

        self.in_channels = in_channels

        self.norm = Normalize(in_channels)
        self.q = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.k = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.v = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.proj_out = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )

    def forward(self, x):
        h_ = x
        h_ = self.norm(h_)
        q = self.q(h_)
        k = self.k(h_)
        v = self.v(h_)

        # compute attention
        b, c, h, w = q.shape
        q = q.reshape(b, c, h * w)
        q = q.permute(0, 2, 1)  # b,hw,c
        k = k.reshape(b, c, h * w)  # b,c,hw
        w_ = torch.bmm(q, k)  # b,hw,hw    w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
        w_ = w_ * (int(c) ** (-0.5))
        w_ = torch.nn.functional.softmax(w_, dim=2)

        # attend to values
        v = v.reshape(b, c, h * w)
        w_ = w_.permute(0, 2, 1)  # b,hw,hw (first hw of k, second of q)
        h_ = torch.bmm(v, w_)  # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
        h_ = h_.reshape(b, c, h, w)

        h_ = self.proj_out(h_)

        return x + h_


class MemoryEfficientAttnBlock(nn.Module):
    """
    Uses xformers efficient implementation,
    see https://github.com/MatthieuTPHR/diffusers/blob/d80b531ff8060ec1ea982b65a1b8df70f73aa67c/src/diffusers/models/attention.py#L223
    Note: this is a single-head self-attention operation
    """

    #
    def __init__(self, in_channels):
        super().__init__()
        print(
            f"building MemoryEfficientAttnBlock (xformers) with {in_channels} in_channels"
        )
        self.in_channels = in_channels

        self.norm = Normalize(in_channels)
        self.q = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.k = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.v = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.proj_out = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.attention_op: Optional[Any] = None

    def forward(self, x):
        h_ = x
        h_ = self.norm(h_)
        q = self.q(h_)
        k = self.k(h_)
        v = self.v(h_)

        # compute attention
        B, C, H, W = q.shape
        q, k, v = map(lambda x: rearrange(x, "b c h w -> b (h w) c"), (q, k, v))

        q, k, v = map(
            lambda t: t.unsqueeze(3)
            .reshape(B, t.shape[1], 1, C)
            .permute(0, 2, 1, 3)
            .reshape(B * 1, t.shape[1], C)
            .contiguous(),
            (q, k, v),
        )
        out = Config.xformers.ops.memory_efficient_attention(
            q, k, v, attn_bias=None, op=self.attention_op
        )

        out = (
            out.unsqueeze(0)
            .reshape(B, 1, out.shape[1], C)
            .permute(0, 2, 1, 3)
            .reshape(B, out.shape[1], C)
        )
        out = rearrange(out, "b (h w) c -> b c h w", b=B, h=H, w=W, c=C)
        out = self.proj_out(out)
        return x + out


class SDPAttnBlock(nn.Module):

    def __init__(self, in_channels):
        super().__init__()
        print(f"building SDPAttnBlock (sdp) with {in_channels} in_channels")
        self.in_channels = in_channels

        self.norm = Normalize(in_channels)
        self.q = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.k = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.v = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )
        self.proj_out = torch.nn.Conv2d(
            in_channels, in_channels, kernel_size=1, stride=1, padding=0
        )

    def forward(self, x):
        h_ = x
        h_ = self.norm(h_)
        q = self.q(h_)
        k = self.k(h_)
        v = self.v(h_)

        # compute attention
        B, C, H, W = q.shape
        q, k, v = map(lambda x: rearrange(x, "b c h w -> b (h w) c"), (q, k, v))

        q, k, v = map(
            lambda t: t.unsqueeze(3)
            .reshape(B, t.shape[1], 1, C)
            .permute(0, 2, 1, 3)
            .reshape(B * 1, t.shape[1], C)
            .contiguous(),
            (q, k, v),
        )
        out = F.scaled_dot_product_attention(q, k, v)

        out = (
            out.unsqueeze(0)
            .reshape(B, 1, out.shape[1], C)
            .permute(0, 2, 1, 3)
            .reshape(B, out.shape[1], C)
        )
        out = rearrange(out, "b (h w) c -> b c h w", b=B, h=H, w=W, c=C)
        out = self.proj_out(out)
        return x + out


def make_attn(in_channels, attn_type="vanilla", attn_kwargs=None):
    assert attn_type in [
        "vanilla",
        "sdp",
        "xformers",
        "linear",
        "none",
    ], f"attn_type {attn_type} unknown"
    if attn_type == "vanilla":
        assert attn_kwargs is None
        return AttnBlock(in_channels)
    elif attn_type == "sdp":
        return SDPAttnBlock(in_channels)
    elif attn_type == "xformers":
        return MemoryEfficientAttnBlock(in_channels)
    elif attn_type == "none":
        return nn.Identity(in_channels)
    else:
        raise NotImplementedError()


class Encoder(nn.Module):
    def __init__(
        self,
        *,
        ch,
        out_ch,
        ch_mult=(1, 2, 4, 8),
        num_res_blocks,
        attn_resolutions,
        dropout=0.0,
        resamp_with_conv=True,
        in_channels,
        resolution,
        z_channels,
        double_z=True,
        use_linear_attn=False,
        **ignore_kwargs,
    ):
        super().__init__()
        ### setup attention type
        if Config.attn_mode == AttnMode.SDP:
            attn_type = "sdp"
        elif Config.attn_mode == AttnMode.XFORMERS:
            attn_type = "xformers"
        else:
            attn_type = "vanilla"
        if use_linear_attn:
            attn_type = "linear"
        self.ch = ch
        self.temb_ch = 0
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.in_channels = in_channels

        # downsampling
        self.conv_in = torch.nn.Conv2d(
            in_channels, self.ch, kernel_size=3, stride=1, padding=1
        )

        curr_res = resolution
        in_ch_mult = (1,) + tuple(ch_mult)
        self.in_ch_mult = in_ch_mult
        self.down = nn.ModuleList()
        for i_level in range(self.num_resolutions):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            block_in = ch * in_ch_mult[i_level]
            block_out = ch * ch_mult[i_level]
            for i_block in range(self.num_res_blocks):
                block.append(
                    ResnetBlock(
                        in_channels=block_in,
                        out_channels=block_out,
                        temb_channels=self.temb_ch,
                        dropout=dropout,
                    )
                )
                block_in = block_out
                if curr_res in attn_resolutions:
                    attn.append(make_attn(block_in, attn_type=attn_type))
            down = nn.Module()
            down.block = block
            down.attn = attn
            if i_level != self.num_resolutions - 1:
                down.downsample = Downsample(block_in, resamp_with_conv)
                curr_res = curr_res // 2
            self.down.append(down)

        # middle
        self.mid = nn.Module()
        self.mid.block_1 = ResnetBlock(
            in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,
        )
        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
        self.mid.block_2 = ResnetBlock(
            in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,
        )

        # end
        self.norm_out = Normalize(block_in)
        self.conv_out = torch.nn.Conv2d(
            block_in,
            2 * z_channels if double_z else z_channels,
            kernel_size=3,
            stride=1,
            padding=1,
        )

    def forward(self, x):
        # timestep embedding
        temb = None

        # downsampling
        hs = [self.conv_in(x)]
        for i_level in range(self.num_resolutions):
            for i_block in range(self.num_res_blocks):
                h = self.down[i_level].block[i_block](hs[-1], temb)
                if len(self.down[i_level].attn) > 0:
                    h = self.down[i_level].attn[i_block](h)
                hs.append(h)
            if i_level != self.num_resolutions - 1:
                hs.append(self.down[i_level].downsample(hs[-1]))

        # middle
        h = hs[-1]
        h = self.mid.block_1(h, temb)
        h = self.mid.attn_1(h)
        h = self.mid.block_2(h, temb)

        # end
        h = self.norm_out(h)
        h = nonlinearity(h)
        h = self.conv_out(h)
        return h


class Decoder(nn.Module):
    def __init__(
        self,
        *,
        ch,
        out_ch,
        ch_mult=(1, 2, 4, 8),
        num_res_blocks,
        attn_resolutions,
        dropout=0.0,
        resamp_with_conv=True,
        in_channels,
        resolution,
        z_channels,
        give_pre_end=False,
        tanh_out=False,
        use_linear_attn=False,
        **ignorekwargs,
    ):
        super().__init__()
        ### setup attention type
        if Config.attn_mode == AttnMode.SDP:
            attn_type = "sdp"
        elif Config.attn_mode == AttnMode.XFORMERS:
            attn_type = "xformers"
        else:
            attn_type = "vanilla"
        if use_linear_attn:
            attn_type = "linear"
        self.ch = ch
        self.temb_ch = 0
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.in_channels = in_channels
        self.give_pre_end = give_pre_end
        self.tanh_out = tanh_out

        # compute in_ch_mult, block_in and curr_res at lowest res
        in_ch_mult = (1,) + tuple(ch_mult)
        block_in = ch * ch_mult[self.num_resolutions - 1]
        curr_res = resolution // 2 ** (self.num_resolutions - 1)
        self.z_shape = (1, z_channels, curr_res, curr_res)

        # z to block_in
        self.conv_in = torch.nn.Conv2d(
            z_channels, block_in, kernel_size=3, stride=1, padding=1
        )

        # middle
        self.mid = nn.Module()
        self.mid.block_1 = ResnetBlock(
            in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,
        )
        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
        self.mid.block_2 = ResnetBlock(
            in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,
        )

        # upsampling
        self.up = nn.ModuleList()
        for i_level in reversed(range(self.num_resolutions)):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            block_out = ch * ch_mult[i_level]
            for i_block in range(self.num_res_blocks + 1):
                block.append(
                    ResnetBlock(
                        in_channels=block_in,
                        out_channels=block_out,
                        temb_channels=self.temb_ch,
                        dropout=dropout,
                    )
                )
                block_in = block_out
                if curr_res in attn_resolutions:
                    attn.append(make_attn(block_in, attn_type=attn_type))
            up = nn.Module()
            up.block = block
            up.attn = attn
            if i_level != 0:
                up.upsample = Upsample(block_in, resamp_with_conv)
                curr_res = curr_res * 2
            self.up.insert(0, up)  # prepend to get consistent order

        # end
        self.norm_out = Normalize(block_in)
        self.conv_out = torch.nn.Conv2d(
            block_in, out_ch, kernel_size=3, stride=1, padding=1
        )

    def forward(self, z):
        # assert z.shape[1:] == self.z_shape[1:]
        self.last_z_shape = z.shape

        # timestep embedding
        temb = None

        # z to block_in
        h = self.conv_in(z)

        # middle
        h = self.mid.block_1(h, temb)
        h = self.mid.attn_1(h)
        h = self.mid.block_2(h, temb)

        # upsampling
        for i_level in reversed(range(self.num_resolutions)):
            for i_block in range(self.num_res_blocks + 1):
                h = self.up[i_level].block[i_block](h, temb)
                if len(self.up[i_level].attn) > 0:
                    h = self.up[i_level].attn[i_block](h)
            if i_level != 0:
                h = self.up[i_level].upsample(h)

        # end
        if self.give_pre_end:
            return h

        h = self.norm_out(h)
        h = nonlinearity(h)
        h = self.conv_out(h)
        if self.tanh_out:
            h = torch.tanh(h)
        return h


class ConvEMA_Decoder(nn.Module):
    def __init__(
        self,
        *,
        ch,
        out_ch,
        ch_mult=(1, 2, 4, 8),
        num_res_blocks,
        attn_resolutions,
        dropout=0.0,
        resamp_with_conv=True,
        in_channels,
        resolution,
        z_channels,
        give_pre_end=False,
        tanh_out=False,
        use_linear_attn=False,
        ema_hidden_channels=64,
        ema_hidden_layers=2,
        ema_conv_kernel=3,
        ema_fusion_kernel=1,
        ema_skip_connection=True,
        detach_memory=True,
        **ignorekwargs,
    ):
        super().__init__()
        # Attention type
        if Config.attn_mode == AttnMode.SDP:
            attn_type = "sdp"
        elif Config.attn_mode == AttnMode.XFORMERS:
            attn_type = "xformers"
        else:
            attn_type = "vanilla"
        if use_linear_attn:
            attn_type = "linear"

        self.ch = ch
        self.temb_ch = 0
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.in_channels = in_channels
        self.give_pre_end = give_pre_end
        self.tanh_out = tanh_out
        self.detach_memory = detach_memory

        in_ch_mult = (1,) + tuple(ch_mult)
        block_in = ch * ch_mult[self.num_resolutions - 1]
        curr_res = resolution // 2 ** (self.num_resolutions - 1)
        self.z_shape = (1, z_channels, curr_res, curr_res)

        # z → block_in
        self.conv_in = torch.nn.Conv2d(z_channels, block_in, 3, 1, 1)

        # middle
        self.mid = nn.Module()
        self.mid.block_1 = ResnetBlock(in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,)
        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
        self.mid.block_2 = ResnetBlock(in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,)

        # --- EMA memories ---
        self.ema_mem_block1_stab = None
        self.ema_mem_block1_unstab = None
        self.ema_mem_block2_stab = None
        self.ema_mem_block2_unstab = None
        self.ema_mem_convout_stab = None
        self.ema_mem_convout_unstab = None

        # --- EMA parameters per layer ---
        def make_ema_params(input_ch, output_ch):
            weights = nn.ParameterList()
            biases = nn.ParameterList()
            activations = nn.ModuleList()

            # First layer: input_ch -> hidden
            w = nn.Parameter(torch.empty(ema_hidden_channels, input_ch,
                                         ema_conv_kernel, ema_conv_kernel))
            b = nn.Parameter(torch.zeros(ema_hidden_channels))
            nn.init.kaiming_uniform_(w, a=0.2)
            weights.append(w)
            biases.append(b)
            activations.append(nn.LeakyReLU())

            # Hidden layers
            for _ in range(ema_hidden_layers):
                w = nn.Parameter(torch.empty(ema_hidden_channels, ema_hidden_channels,
                                             ema_conv_kernel, ema_conv_kernel))
                b = nn.Parameter(torch.zeros(ema_hidden_channels))
                nn.init.kaiming_uniform_(w, a=0.2)
                weights.append(w)
                biases.append(b)
                activations.append(nn.LeakyReLU())

            # Last layer: hidden -> output_ch
            w = nn.Parameter(torch.empty(output_ch, ema_hidden_channels,
                                         ema_conv_kernel, ema_conv_kernel))
            b = nn.Parameter(torch.full((output_ch,), -4.0))
            nn.init.xavier_uniform_(w)
            weights.append(w)
            biases.append(b)
            return weights, biases, activations

        self.ema1_weights, self.ema1_biases, self.ema1_activations = make_ema_params(
            input_ch=512*3, output_ch=512*(ema_fusion_kernel**2)
        ) 
        # input_ch: input + stabilized + unstabilized memories (so block_1 and block_2 = 512×3 = 1536)
        self.ema2_weights, self.ema2_biases, self.ema2_activations = make_ema_params(
            input_ch=512*3, output_ch=512*(ema_fusion_kernel**2)
        ) # h + mem_stab + mem_unstab (same as above)
        self.ema3_weights, self.ema3_biases, self.ema3_activations = make_ema_params(
            input_ch=3*3, output_ch=3*(ema_fusion_kernel**2)
        )

        self.ema_hidden_channels = ema_hidden_channels
        self.ema_hidden_layers = ema_hidden_layers
        self.ema_conv_kernel = ema_conv_kernel
        self.ema_fusion_kernel = ema_fusion_kernel
        self.ema_skip_connection = ema_skip_connection

        # upsampling
        self.up = nn.ModuleList()
        for i_level in reversed(range(self.num_resolutions)):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            block_out = ch * ch_mult[i_level]
            for i_block in range(self.num_res_blocks + 1):
                block.append(ResnetBlock(
                    in_channels=block_in,
                    out_channels=block_out,
                    temb_channels=self.temb_ch,
                    dropout=dropout,)
                )
                block_in = block_out
                if curr_res in attn_resolutions:
                    attn.append(make_attn(block_in, attn_type=attn_type))
            up = nn.Module()
            up.block = block
            up.attn = attn
            if i_level != 0:
                up.upsample = Upsample(block_in, resamp_with_conv)
                curr_res *= 2
            self.up.insert(0, up)

        # end
        self.norm_out = Normalize(block_in)
        self.conv_out = torch.nn.Conv2d(block_in, out_ch, 3, 1, 1)

    def _conv_ema(self, h, mem_stab, mem_unstab, weights, biases, activations):
        """
        h: [B, C, H, W] per-frame or per-chunk
        mem_stab, mem_unstab: previous memory tensors (same batch as h) or None
        weights, biases, activations: EMA layer parameters
        """

        # Ensure memory is initialized and matches batch size
        if mem_stab is None or mem_stab.shape[0] != h.shape[0]:
            mem_stab = torch.zeros_like(h)
            mem_unstab = torch.zeros_like(h)

        # Concatenate input with stabilized & unstabilized memory
        q = torch.cat([h, mem_stab, mem_unstab], dim=1)  # [B, 3*C, H, W]

        # Forward through EMA conv layers
        q = nn.functional.conv2d(q, weights[0], biases[0], padding="same")
        q = activations[0](q)
        skip = q

        for i in range(self.ema_hidden_layers):
            q = nn.functional.conv2d(q, weights[i+1], biases[i+1], padding="same")
            q = activations[i+1](q)

        if self.ema_skip_connection:
            q = q + skip

        # Final conv
        shape = h.shape
        head = nn.functional.conv2d(q, weights[-1], biases[-1], padding="same")
        head = rearrange(head, "b (c p) h w -> b c p (h w)", c=shape[1])
        eta = torch.cat([head, torch.zeros_like(head[:, :, :1])], dim=2)
        eta = eta.softmax(dim=2)

        # Apply to stabilized memory
        mem_unf = nn.functional.unfold(mem_stab, kernel_size=self.ema_fusion_kernel, padding=self.ema_fusion_kernel // 2)
        mem_unf = rearrange(mem_unf, "b (c p) hw -> b c p hw", c=shape[1])
        h_flat = rearrange(h, "b c h w -> b c (h w)")
        h_out = (mem_unf * eta[:, :, :-1]).sum(dim=2) + eta[:, :, -1] * h_flat
        h_out = h_out.view(shape)

        # Update memory (detach to avoid gradients through previous frames)
        mem_stab = h_out.clone().detach()
        mem_unstab = h_out.clone().detach()

        return h_out, mem_stab, mem_unstab


    def forward(self, z):
        h = self.conv_in(z)

        # --- mid.block_1 + EMA ---
        h = self.mid.block_1(h, None)
        h = self.mid.attn_1(h)
        h, self.ema_mem_block1_stab, self.ema_mem_block1_unstab = self._conv_ema(
            h, self.ema_mem_block1_stab, self.ema_mem_block1_unstab,
            self.ema1_weights, self.ema1_biases, self.ema1_activations
        )

        # --- mid.block_2 + EMA ---
        h = self.mid.block_2(h, None)
        h, self.ema_mem_block2_stab, self.ema_mem_block2_unstab = self._conv_ema(
            h, self.ema_mem_block2_stab, self.ema_mem_block2_unstab,
            self.ema2_weights, self.ema2_biases, self.ema2_activations
        )

        # --- upsampling ---
        for i_level in reversed(range(self.num_resolutions)):
            for i_block in range(self.num_res_blocks + 1):
                h = self.up[i_level].block[i_block](h, None)
                if len(self.up[i_level].attn) > 0:
                    h = self.up[i_level].attn[i_block](h)
            if i_level != 0:
                h = self.up[i_level].upsample(h)

        # --- conv_out + EMA ---
        h = self.norm_out(h)
        h = nonlinearity(h)
        h = self.conv_out(h)
        h, self.ema_mem_convout_stab, self.ema_mem_convout_unstab = self._conv_ema(
            h, self.ema_mem_convout_stab, self.ema_mem_convout_unstab,
            self.ema3_weights, self.ema3_biases, self.ema3_activations
        )

        if self.tanh_out:
            h = torch.tanh(h)

        return h
    

class AutoencoderKL(nn.Module):

    def __init__(self, ddconfig, embed_dim):
        super().__init__()
        self.encoder = Encoder(**ddconfig)
        self.decoder = Decoder(**ddconfig)
        assert ddconfig["double_z"]
        self.quant_conv = torch.nn.Conv2d(2 * ddconfig["z_channels"], 2 * embed_dim, 1)
        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
        self.embed_dim = embed_dim

    def encode(self, x):
        h = self.encoder(x)
        moments = self.quant_conv(h)
        posterior = DiagonalGaussianDistribution(moments)
        return posterior

    def decode(self, z):
        z = self.post_quant_conv(z)
        dec = self.decoder(z)
        return dec

    def forward(self, input, sample_posterior=True):
        posterior = self.encode(input)
        if sample_posterior:
            z = posterior.sample()
        else:
            z = posterior.mode()
        dec = self.decode(z)
        return dec, posterior


class ConvEMA_AutoencoderKL(nn.Module):

    def __init__(self, ddconfig, embed_dim):
        super().__init__()
        self.encoder = Encoder(**ddconfig)
        self.decoder = ConvEMA_Decoder(**ddconfig)
        assert ddconfig["double_z"]
        self.quant_conv = torch.nn.Conv2d(2 * ddconfig["z_channels"], 2 * embed_dim, 1)
        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
        self.embed_dim = embed_dim

    def encode(self, x):
        h = self.encoder(x)
        moments = self.quant_conv(h)
        posterior = DiagonalGaussianDistribution(moments)
        return posterior

    def decode(self, z):
        z = self.post_quant_conv(z)
        dec = self.decoder(z)
        return dec

    def forward(self, input, sample_posterior=True):
        posterior = self.encode(input)
        if sample_posterior:
            z = posterior.sample()
        else:
            z = posterior.mode()
        dec = self.decode(z)
        return dec, posterior

# Attempt 3 - Go 3D convs for IP Lifting
class Decoder_3D(nn.Module):
    def __init__(
        self,
        *,
        ch,
        out_ch,
        ch_mult=(1, 2, 4, 8),
        num_res_blocks,
        attn_resolutions,
        dropout=0.0,
        resamp_with_conv=True,
        in_channels,
        resolution,
        z_channels,
        give_pre_end=False,
        tanh_out=False,
        use_linear_attn=False,
        **ignorekwargs,
    ):
        super().__init__()
        ### setup attention type
        if Config.attn_mode == AttnMode.SDP:
            attn_type = "sdp"
        elif Config.attn_mode == AttnMode.XFORMERS:
            attn_type = "xformers"
        else:
            attn_type = "vanilla"
        if use_linear_attn:
            attn_type = "linear"
        self.ch = ch
        self.temb_ch = 0
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.in_channels = in_channels
        self.give_pre_end = give_pre_end
        self.tanh_out = tanh_out

        # compute in_ch_mult, block_in and curr_res at lowest res
        in_ch_mult = (1,) + tuple(ch_mult)
        block_in = ch * ch_mult[self.num_resolutions - 1]
        curr_res = resolution // 2 ** (self.num_resolutions - 1)
        self.z_shape = (1, z_channels, curr_res, curr_res)

        # z to block_in
        self.conv_in = torch.nn.Conv2d(
            z_channels, block_in, kernel_size=3, stride=1, padding=1
        )

        # middle
        self.mid = nn.Module()
        self.mid.block_1 = ResnetBlock(
            in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,
        )
        self.temp_mid_block_1 = nn.Conv3d(
            in_channels=block_in,
            out_channels=block_in,
            kernel_size=(3,1,1), padding=(1,0,0), groups=block_in
        ) 
        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
        self.mid.block_2 = ResnetBlock(
            in_channels=block_in,
            out_channels=block_in,
            temb_channels=self.temb_ch,
            dropout=dropout,
        )
        self.temp_mid_block_2 = nn.Conv3d(
            in_channels=block_in,
            out_channels=block_in,
            kernel_size=(3,1,1), padding=(1,0,0), groups=block_in
        ) 

        # upsampling
        self.up = nn.ModuleList()
        for i_level in reversed(range(self.num_resolutions)):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            temp = nn.ModuleList()
            block_out = ch * ch_mult[i_level]
            for i_block in range(self.num_res_blocks + 1):
                block.append(
                    ResnetBlock(
                        in_channels=block_in,
                        out_channels=block_out,
                        temb_channels=self.temb_ch,
                        dropout=dropout,
                    )
                )
                block_in = block_out
                if curr_res in attn_resolutions:
                    temp.append(nn.Conv3d(block_in, block_in, kernel_size=(3,1,1), padding=(1,0,0), groups=block_in))
                    attn.append(make_attn(block_in, attn_type=attn_type))
            up = nn.Module()
            up.block = block
            up.temp = temp
            up.attn = attn
            if i_level != 0:
                up.upsample = Upsample(block_in, resamp_with_conv)
                curr_res = curr_res * 2
            self.up.insert(0, up)  # prepend to get consistent order

        # end
        self.norm_out = Normalize(block_in)
        self.conv_out = torch.nn.Conv2d(
            block_in, out_ch, kernel_size=3, stride=1, padding=1
        )
        self.temp_conv3d_out = nn.Conv3d(out_ch, out_ch, kernel_size=(3,1,1), padding=(1,0,0), groups=out_ch)

    def forward(self, z):
        # assert z.shape[1:] == self.z_shape[1:]
        self.last_z_shape = z.shape

        # timestep embedding
        temb = None

        # z to block_in
        h = self.conv_in(z)

        # middle
        h = self.mid.block_1(h, temb)
        print("[+] After mid_block_1: ", h.size())
        ###############
        h_3d = h.view(1, h.size(1), -1, h.size(2), h.size(3)) # 1, C, T, H, W
        print("[+] Reshape for first 3d conv: ", h_3d.size())
        h_3d = self.temp_mid_block_1(h_3d) # 1, C_out, T_out, H_out, W_out
        print("[+] After first 3d conv: ", h_3d.size())
        h_3d = h_3d.view(-1, h.size(1), h.size(3), h.size(4)) 
        print("[+] Reshape for residual addition after first 3d conv: ", h_3d.size())
        h = h + h_3d
        ###############
        
        h = self.mid.attn_1(h)
        print("[+] After mid_attn_1: ", h.size())
        h = self.mid.block_2(h, temb)
        print("[+] After mid_block_2: ", h.size())

        ###############
        h_3d2 = h.view(1, h.size(1), -1, h.size(2), h.size(3)) # 1, T, C, H, W
        print("[+] Reshape after mid_block_2 for 2nd conv3d: ", h_3d2.size())
        h_3d2 = self.temp_mid_block_2(h_3d2)
        print("[+] After 2nd conv3d: ", h_3d2.size())
        h_3d2 = h_3d2.view(-1, h.size(1), h.size(3), h.size(4))
        print("[+] Reshape h3d2 for residual addition: ", h_3d2.size())
        h = h + h_3d2
        ###############

        # upsampling
        for i_level in reversed(range(self.num_resolutions)):
            for i_block in range(self.num_res_blocks + 1):
                h = self.up[i_level].block[i_block](h, temb)
                print(f"i_block: {i_block} h.shape:", h.size())
                if len(self.up[i_level].attn) > 0:
                    ###############
                    h_res = h.view(1, h.size(1), -1, h.size(2), h.size(3))
                    print("[+] h reshaped for 3d conv:", h_res.size())
                    h_res = self.up[i_level].temp[i_block](h_res)
                    print("[+] h after 3d conv:", h_res.size())
                    h_res = h_res.view(-1, h.size(1), h.size(3), h.size(4))
                    print("[+] h reshaped for residual addn:", h_res.size())
                    h = h + h_res 
                    ###############
                    h = self.up[i_level].attn[i_block](h)

            if i_level != 0:
                h = self.up[i_level].upsample(h)

        # end
        if self.give_pre_end:
            return h

        h = self.norm_out(h)
        h = nonlinearity(h)
        h = self.conv_out(h)
        print("[+] After conv_out shape:", h.size())
        ###############
        h_out = h.view(1, h.size(1), -1, h.size(2), h.size(3)) # 1, T, C, H, W
        print("[+] Reshape after conv_out:", h.size())
        h_out = self.temp_conv3d_out(h_out)
        print("[+] After last 3d conv:", h.size())
        h_out = h_out.view(-1, h.size(1), h.size(3), h.size(4))
        print("[+] Reshape after last 3d conv:", h.size())
        h = h + h_out
        ###############

        if self.tanh_out:
            h = torch.tanh(h)
        return h


class Decoder3D_AutoencoderKL(nn.Module):
    def __init__(self, ddconfig, embed_dim):
        super().__init__()
        self.encoder = Encoder(**ddconfig)
        self.decoder = Decoder_3D(**ddconfig)
        assert ddconfig["double_z"]
        self.quant_conv = torch.nn.Conv2d(2 * ddconfig["z_channels"], 2 * embed_dim, 1)
        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
        self.embed_dim = embed_dim

    def encode(self, x):
        h = self.encoder(x)
        moments = self.quant_conv(h)
        posterior = DiagonalGaussianDistribution(moments)
        return posterior

    def decode(self, z):
        z = self.post_quant_conv(z)
        dec = self.decoder(z)
        return dec

    def forward(self, input, sample_posterior=True):
        posterior = self.encode(input)
        if sample_posterior:
            z = posterior.sample()
        else:
            z = posterior.mode()
        dec = self.decode(z)
        return dec, posterior