# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo

# SPDX-License-Identifier: Apache-2.0


import torch
from sageattention import sageattn

from sglang.multimodal_gen.runtime.layers.attention.backends.attention_backend import (  # FlashAttentionMetadata,
    AttentionBackend,
    AttentionImpl,
    AttentionMetadata,
)
from sglang.multimodal_gen.runtime.platforms import AttentionBackendEnum
from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger

logger = init_logger(__name__)


class SageAttentionBackend(AttentionBackend):
    accept_output_buffer: bool = True

    @staticmethod
    def get_supported_head_sizes() -> list[int]:
        return [32, 64, 96, 128, 160, 192, 224, 256]

    @staticmethod
    def get_enum() -> AttentionBackendEnum:
        return AttentionBackendEnum.SAGE_ATTN

    @staticmethod
    def get_impl_cls() -> type["SageAttentionImpl"]:
        return SageAttentionImpl


class SageAttentionImpl(AttentionImpl):

    def __init__(
        self,
        num_heads: int,
        head_size: int,
        causal: bool,
        softmax_scale: float,
        num_kv_heads: int | None = None,
        prefix: str = "",
        **extra_impl_args,
    ) -> None:
        self.causal = causal
        self.softmax_scale = softmax_scale
        self.dropout = extra_impl_args.get("dropout_p", 0.0)

    def forward(
        self,
        query: torch.Tensor,
        key: torch.Tensor,
        value: torch.Tensor,
        attn_metadata: AttentionMetadata,
        *,
        return_softmax_lse: bool = False,
    ) -> torch.Tensor:
        output = sageattn(
            query,
            key,
            value,
            # since input is (batch_size, seq_len, head_num, head_dim)
            tensor_layout="NHD",
            is_causal=self.causal,
            sm_scale=self.softmax_scale,
            return_lse=return_softmax_lse,
        )
        if return_softmax_lse:
            output, softmax_lse = output
            return output, softmax_lse
        return output