import torch
import torch.nn as nn
import numpy as np
from .transformer_block import TransformerBlock

class Decoder(nn.Module):
    """
    MAE Decoder based on Vision Transformer.
    """
    def __init__(
        self, 
        num_patches: int, 
        patch_size: int = 16, 
        in_chans: int = 3, 
        embed_dim: int = 768, 
        decoder_embed_dim: int = 512, 
        decoder_depth: int = 8, 
        decoder_num_heads: int = 16, 
        mlp_ratio: float = 4.0, 
        norm_layer: nn.Module = nn.LayerNorm
    ):
        super().__init__()
        
        self.decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True)
        self.mask_token = nn.Parameter(torch.zeros(1, 1, decoder_embed_dim))
        self.decoder_pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, decoder_embed_dim), requires_grad=False)
        
        self.decoder_blocks = nn.ModuleList([
            TransformerBlock(decoder_embed_dim, decoder_num_heads, mlp_ratio, qkv_bias=True, norm_layer=norm_layer)
            for _ in range(decoder_depth)
        ])
        
        self.decoder_norm = norm_layer(decoder_embed_dim)
        self.decoder_pred = nn.Linear(decoder_embed_dim, patch_size**2 * in_chans, bias=True)
        
        self.initialize_weights()

    def initialize_weights(self):
        # Initialize pos_embed with sin-cos embedding
        grid_size = int(self.decoder_pos_embed.shape[1] - 1)**.5
        pos_embed = self.get_2d_sincos_pos_embed(self.decoder_pos_embed.shape[-1], int(grid_size), cls_token=True)
        self.decoder_pos_embed.data.copy_(torch.from_numpy(pos_embed).float().unsqueeze(0))
        
        # Initialize mask_token
        torch.nn.init.normal_(self.mask_token, std=.02)
        
        # Initialize transformer blocks
        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            torch.nn.init.xavier_uniform_(m.weight)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)

    def get_2d_sincos_pos_embed(self, embed_dim, grid_size, cls_token=False):
        # Re-using the logic from Encoder (could be moved to utils for DRY)
        grid_h = np.arange(grid_size, dtype=np.float32)
        grid_w = np.arange(grid_size, dtype=np.float32)
        grid = np.meshgrid(grid_w, grid_h)
        grid = np.stack(grid, axis=0)
        grid = grid.reshape([2, 1, grid_size, grid_size])
        
        pos_embed = self.get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
        if cls_token:
            pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
        return pos_embed

    def get_2d_sincos_pos_embed_from_grid(self, embed_dim, grid):
        assert embed_dim % 2 == 0
        emb_h = self.get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0])
        emb_w = self.get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1])
        emb = np.concatenate([emb_h, emb_w], axis=1)
        return emb

    def get_1d_sincos_pos_embed_from_grid(self, embed_dim, pos):
        assert embed_dim % 2 == 0
        omega = np.arange(embed_dim // 2, dtype=np.float32)
        omega /= embed_dim / 2.
        omega = 1. / 10000**omega
        pos = pos.reshape(-1)
        out = np.einsum('m,d->md', pos, omega)
        emb_sin = np.sin(out)
        emb_cos = np.cos(out)
        emb = np.concatenate([emb_sin, emb_cos], axis=1)
        return emb

    def forward(self, x: torch.Tensor, ids_restore: torch.Tensor) -> torch.Tensor:
        # Project to decoder dimension
        x = self.decoder_embed(x)
        
        # Append mask tokens to sequence
        mask_tokens = self.mask_token.repeat(x.shape[0], ids_restore.shape[1] + 1 - x.shape[1], 1)
        x_ = torch.cat([x[:, 1:, :], mask_tokens], dim=1)  # no cls token
        x_ = torch.gather(x_, dim=1, index=ids_restore.unsqueeze(-1).repeat(1, 1, x.shape[2]))  # unshuffle
        x = torch.cat([x[:, :1, :], x_], dim=1)  # append cls token
        
        # Add pos embed
        x = x + self.decoder_pos_embed
        
        # Apply transformer blocks
        for block in self.decoder_blocks:
            x = block(x)
        x = self.decoder_norm(x)
        
        # Predict pixels
        x = self.decoder_pred(x)
        
        # Remove cls token
        x = x[:, 1:, :]
        
        return x