import torch
import torch.nn as nn
import torch.nn.functional as F
import math
from diffusers.models.attention_processor import Attention
from typing import Optional
from diffusers.models.embeddings import apply_rotary_emb


class FluxAttnProcessor2_0:
    """Attention processor used typically in processing the SD3-like self-attention projections."""

    def __init__(self, train_seq_len=512 + 64 * 64):
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError(
                "FluxAttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
            )
        self.train_seq_len = train_seq_len

    def __call__(
        self,
        attn: Attention,
        hidden_states: torch.FloatTensor,
        encoder_hidden_states: torch.FloatTensor = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        image_rotary_emb: Optional[torch.Tensor] = None,
        proportional_attention=False,
    ) -> torch.FloatTensor:
        batch_size, _, _ = (
            hidden_states.shape
            if encoder_hidden_states is None
            else encoder_hidden_states.shape
        )

        # `sample` projections.
        query = attn.to_q(hidden_states)
        key = attn.to_k(hidden_states)
        value = attn.to_v(hidden_states)

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        if attn.norm_q is not None:
            query = attn.norm_q(query)
        if attn.norm_k is not None:
            key = attn.norm_k(key)

        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
        if encoder_hidden_states is not None:
            # `context` projections.
            encoder_hidden_states_query_proj = attn.add_q_proj(encoder_hidden_states)
            encoder_hidden_states_key_proj = attn.add_k_proj(encoder_hidden_states)
            encoder_hidden_states_value_proj = attn.add_v_proj(encoder_hidden_states)

            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
                batch_size, -1, attn.heads, head_dim
            ).transpose(1, 2)
            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
                batch_size, -1, attn.heads, head_dim
            ).transpose(1, 2)
            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
                batch_size, -1, attn.heads, head_dim
            ).transpose(1, 2)

            if attn.norm_added_q is not None:
                encoder_hidden_states_query_proj = attn.norm_added_q(
                    encoder_hidden_states_query_proj
                )
            if attn.norm_added_k is not None:
                encoder_hidden_states_key_proj = attn.norm_added_k(
                    encoder_hidden_states_key_proj
                )

            # attention
            query = torch.cat([encoder_hidden_states_query_proj, query], dim=2)
            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)

        if image_rotary_emb is not None:
            query = apply_rotary_emb(query, image_rotary_emb)
            key = apply_rotary_emb(key, image_rotary_emb)

        if proportional_attention:
            attention_scale = math.sqrt(
                math.log(key.size(2), self.train_seq_len) / head_dim
            )
        else:
            attention_scale = math.sqrt(1 / head_dim)

        hidden_states = F.scaled_dot_product_attention(
            query, key, value, dropout_p=0.0, is_causal=False, scale=attention_scale
        )
        hidden_states = hidden_states.transpose(1, 2).reshape(
            batch_size, -1, attn.heads * head_dim
        )
        hidden_states = hidden_states.to(query.dtype)

        if encoder_hidden_states is not None:
            encoder_hidden_states, hidden_states = (
                hidden_states[:, : encoder_hidden_states.shape[1]],
                hidden_states[:, encoder_hidden_states.shape[1] :],
            )

            # linear proj
            hidden_states = attn.to_out[0](hidden_states)
            # dropout
            hidden_states = attn.to_out[1](hidden_states)
            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)

            return hidden_states, encoder_hidden_states
        else:
            return hidden_states


class FluxAttnAdaptationProcessor2_0(nn.Module):
    """Attention processor used typically in processing the SD3-like self-attention projections."""

    def __init__(self, rank=16, dim=3072, to_out=False, train_seq_len=512 + 64 * 64):
        super().__init__()
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError(
                "FluxAttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
            )
        self.to_q_a = nn.Linear(dim, rank, bias=False)
        self.to_q_b = nn.Linear(rank, dim, bias=False)
        self.to_q_b.weight.data = torch.zeros_like(self.to_q_b.weight.data)
        self.to_k_a = nn.Linear(dim, rank, bias=False)
        self.to_k_b = nn.Linear(rank, dim, bias=False)
        self.to_k_b.weight.data = torch.zeros_like(self.to_k_b.weight.data)
        self.to_v_a = nn.Linear(dim, rank, bias=False)
        self.to_v_b = nn.Linear(rank, dim, bias=False)
        self.to_v_b.weight.data = torch.zeros_like(self.to_v_b.weight.data)
        if to_out:
            self.to_out_a = nn.Linear(dim, rank, bias=False)
            self.to_out_b = nn.Linear(rank, dim, bias=False)
            self.to_out_b.weight.data = torch.zeros_like(self.to_out_b.weight.data)
        self.train_seq_len = train_seq_len

    def __call__(
        self,
        attn: Attention,
        hidden_states: torch.FloatTensor,
        encoder_hidden_states: torch.FloatTensor = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        image_rotary_emb: Optional[torch.Tensor] = None,
        proportional_attention=False,
    ) -> torch.FloatTensor:
        batch_size, _, _ = (
            hidden_states.shape
            if encoder_hidden_states is None
            else encoder_hidden_states.shape
        )

        use_adaptation = True

        # `sample` projections.
        query = attn.to_q(hidden_states)
        key = attn.to_k(hidden_states)
        value = attn.to_v(hidden_states)

        if use_adaptation:
            query += self.to_q_b(self.to_q_a(hidden_states))
            key += self.to_k_b(self.to_k_a(hidden_states))
            value += self.to_v_b(self.to_v_a(hidden_states))

        inner_dim = key.shape[-1]
        head_dim = inner_dim // attn.heads

        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        if attn.norm_q is not None:
            query = attn.norm_q(query)
        if attn.norm_k is not None:
            key = attn.norm_k(key)

        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
        if encoder_hidden_states is not None:
            # `context` projections.
            encoder_hidden_states_query_proj = attn.add_q_proj(encoder_hidden_states)
            encoder_hidden_states_key_proj = attn.add_k_proj(encoder_hidden_states)
            encoder_hidden_states_value_proj = attn.add_v_proj(encoder_hidden_states)

            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
                batch_size, -1, attn.heads, head_dim
            ).transpose(1, 2)
            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
                batch_size, -1, attn.heads, head_dim
            ).transpose(1, 2)
            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
                batch_size, -1, attn.heads, head_dim
            ).transpose(1, 2)

            if attn.norm_added_q is not None:
                encoder_hidden_states_query_proj = attn.norm_added_q(
                    encoder_hidden_states_query_proj
                )
            if attn.norm_added_k is not None:
                encoder_hidden_states_key_proj = attn.norm_added_k(
                    encoder_hidden_states_key_proj
                )

            # attention
            query = torch.cat([encoder_hidden_states_query_proj, query], dim=2)
            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)

        if image_rotary_emb is not None:
            query = apply_rotary_emb(query, image_rotary_emb)
            key = apply_rotary_emb(key, image_rotary_emb)

        if proportional_attention:
            attention_scale = math.sqrt(
                math.log(key.size(2), self.train_seq_len) / head_dim
            )
        else:
            attention_scale = math.sqrt(1 / head_dim)

        hidden_states = F.scaled_dot_product_attention(
            query, key, value, dropout_p=0.0, is_causal=False, scale=attention_scale
        )
        hidden_states = hidden_states.transpose(1, 2).reshape(
            batch_size, -1, attn.heads * head_dim
        )
        hidden_states = hidden_states.to(query.dtype)

        if encoder_hidden_states is not None:
            encoder_hidden_states, hidden_states = (
                hidden_states[:, : encoder_hidden_states.shape[1]],
                hidden_states[:, encoder_hidden_states.shape[1] :],
            )

            # linear proj
            hidden_states = (
                (
                    attn.to_out[0](hidden_states)
                    + self.to_out_b(self.to_out_a(hidden_states))
                )
                if use_adaptation
                else attn.to_out[0](hidden_states)
            )
            # dropout
            hidden_states = attn.to_out[1](hidden_states)
            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)

            return hidden_states, encoder_hidden_states
        else:
            return hidden_states