import math
from dataclasses import dataclass

import torch
from torch import Tensor, nn

import comfy.ops
import comfy.ldm.common_dit
from comfy.ldm.flux import layers
from comfy.ldm.flux.layers import (
    EmbedND,
    MLPEmbedder,
    RMSNorm,
    QKNorm,
    SelfAttention,
    ModulationOut,
    Modulation,
    timestep_embedding,
)
from comfy.ldm.flux.math import attention


class Approximator(nn.Module):
    def __init__(self, in_dim: int, out_dim: int, hidden_dim: int, n_layers = 4, operations=None):
        super().__init__()
        self.in_proj = operations.Linear(in_dim, hidden_dim, bias=True)
        self.layers = nn.ModuleList([MLPEmbedder( hidden_dim, hidden_dim, operations=operations) for x in range( n_layers)])
        self.norms = nn.ModuleList([RMSNorm( hidden_dim, operations=operations) for x in range( n_layers)])
        self.out_proj = operations.Linear(hidden_dim, out_dim)

    @property
    def device(self):
        # Get the device of the module (assumes all parameters are on the same device)
        return next(self.parameters()).device
    
    def forward(self, x: Tensor) -> Tensor:
        x = self.in_proj(x)

        for layer, norms in zip(self.layers, self.norms):
            x = x + layer(norms(x))

        x = self.out_proj(x)

        return x


class DoubleStreamBlock(layers.DoubleStreamBlock):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        del self.img_mod
        del self.txt_mod

    def forward(self, img: Tensor, txt: Tensor, pe: Tensor, distill_vec: Tensor | None=None):
        (img_mod1, img_mod2), (txt_mod1, txt_mod2) = distill_vec

        # prepare image for attention
        img_modulated = self.img_norm1(img)
        img_modulated = (1 + img_mod1.scale) * img_modulated + img_mod1.shift
        img_qkv = self.img_attn.qkv(img_modulated)
        img_q, img_k, img_v = img_qkv.view(img_qkv.shape[0], img_qkv.shape[1], 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
        img_q, img_k = self.img_attn.norm(img_q, img_k, img_v)

        # prepare txt for attention
        txt_modulated = self.txt_norm1(txt)
        txt_modulated = (1 + txt_mod1.scale) * txt_modulated + txt_mod1.shift
        txt_qkv = self.txt_attn.qkv(txt_modulated)
        txt_q, txt_k, txt_v = txt_qkv.view(txt_qkv.shape[0], txt_qkv.shape[1], 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
        txt_q, txt_k = self.txt_attn.norm(txt_q, txt_k, txt_v)

        # run actual attention
        attn = attention(torch.cat((txt_q, img_q), dim=2),
                         torch.cat((txt_k, img_k), dim=2),
                         torch.cat((txt_v, img_v), dim=2), pe=pe)

        txt_attn, img_attn = attn[:, : txt.shape[1]], attn[:, txt.shape[1] :]

        # calculate the img bloks
        img = img + img_mod1.gate * self.img_attn.proj(img_attn)
        img = img + img_mod2.gate * self.img_mlp((1 + img_mod2.scale) * self.img_norm2(img) + img_mod2.shift)

        # calculate the txt bloks
        txt += txt_mod1.gate * self.txt_attn.proj(txt_attn)
        txt += txt_mod2.gate * self.txt_mlp((1 + txt_mod2.scale) * self.txt_norm2(txt) + txt_mod2.shift)

        if txt.dtype == torch.float16:
            txt = torch.nan_to_num(txt, nan=0.0, posinf=65504, neginf=-65504)

        return img, txt


class SingleStreamBlock(layers.SingleStreamBlock):
    """
    A DiT block with parallel linear layers as described in
    https://arxiv.org/abs/2302.05442 and adapted modulation interface.
    """

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        del self.modulation

    def forward(self, x: Tensor, pe: Tensor, distill_vec: Tensor | None=None) -> Tensor:
        mod = distill_vec
        x_mod = (1 + mod.scale) * self.pre_norm(x) + mod.shift
        qkv, mlp = torch.split(self.linear1(x_mod), [3 * self.hidden_size, self.mlp_hidden_dim], dim=-1)

        q, k, v = qkv.view(qkv.shape[0], qkv.shape[1], 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
        q, k = self.norm(q, k, v)

        # compute attention
        attn = attention(q, k, v, pe=pe)
        # compute activation in mlp stream, cat again and run second linear layer
        output = self.linear2(torch.cat((attn, self.mlp_act(mlp)), 2))
        x += mod.gate * output
        if x.dtype == torch.float16:
            x = torch.nan_to_num(x, nan=0.0, posinf=65504, neginf=-65504)
        return x


class LastLayer(layers.LastLayer):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        del self.adaLN_modulation

    def forward(self, x: Tensor, distill_vec: Tensor | None=None) -> Tensor:
        shift, scale = distill_vec
        shift = shift.squeeze(1)
        scale = scale.squeeze(1)
        x = (1 + scale[:, None, :]) * self.norm_final(x) + shift[:, None, :]
        x = self.linear(x)
        return x