# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
import torch

try:
    print("calling import flash_attn_interface")
    import flash_attn_interface

    def is_hopper_gpu():
        print("is_hopper_gpu(): checking if not torch.cuda.is_available()")
        if not torch.cuda.is_available():
            print("is_hopper_gpu(): turch.cuda is not available, so this is not Hopper GPU")
            return False
        device_name = torch.cuda.get_device_name(0).lower()
        print(f"is_hopper_gpu(): device_name = {device_name}")
        return "h100" in device_name or "hopper" in device_name
    FLASH_ATTN_3_AVAILABLE = is_hopper_gpu()
except ModuleNotFoundError as e:
    print(f"Got a ModuleNotFoundError for Flash Attention 3: {e}")
    FLASH_ATTN_3_AVAILABLE = False
print(f"FLASH_ATTN_3_AVAILABLE ? -> {FLASH_ATTN_3_AVAILABLE}")

try:
    print("calling import flash_attn")
    import flash_attn
    FLASH_ATTN_2_AVAILABLE = True
except ModuleNotFoundError as e:
    print(f"Got a ModuleNotFoundError for Flash Attention 2: {e}")
    FLASH_ATTN_2_AVAILABLE = False
print(f"FLASH_ATTN_2_AVAILABLE ? -> {FLASH_ATTN_2_AVAILABLE}")


import warnings

__all__ = [
    'flash_attention',
    'attention',
]


def flash_attention(
    q,
    k,
    v,
    q_lens=None,
    k_lens=None,
    dropout_p=0.,
    softmax_scale=None,
    q_scale=None,
    causal=False,
    window_size=(-1, -1),
    deterministic=False,
    dtype=torch.bfloat16,
    version=None,
):
    """
    q:              [B, Lq, Nq, C1].
    k:              [B, Lk, Nk, C1].
    v:              [B, Lk, Nk, C2]. Nq must be divisible by Nk.
    q_lens:         [B].
    k_lens:         [B].
    dropout_p:      float. Dropout probability.
    softmax_scale:  float. The scaling of QK^T before applying softmax.
    causal:         bool. Whether to apply causal attention mask.
    window_size:    (left right). If not (-1, -1), apply sliding window local attention.
    deterministic:  bool. If True, slightly slower and uses more memory.
    dtype:          torch.dtype. Apply when dtype of q/k/v is not float16/bfloat16.
    """
    half_dtypes = (torch.float16, torch.bfloat16)
    assert dtype in half_dtypes
    assert q.device.type == 'cuda' and q.size(-1) <= 256

    # params
    b, lq, lk, out_dtype = q.size(0), q.size(1), k.size(1), q.dtype

    def half(x):
        return x if x.dtype in half_dtypes else x.to(dtype)

    # preprocess query
    if q_lens is None:
        q = half(q.flatten(0, 1))
        q_lens = torch.tensor(
            [lq] * b, dtype=torch.int32).to(
                device=q.device, non_blocking=True)
    else:
        q = half(torch.cat([u[:v] for u, v in zip(q, q_lens)]))

    # preprocess key, value
    if k_lens is None:
        k = half(k.flatten(0, 1))
        v = half(v.flatten(0, 1))
        k_lens = torch.tensor(
            [lk] * b, dtype=torch.int32).to(
                device=k.device, non_blocking=True)
    else:
        k = half(torch.cat([u[:v] for u, v in zip(k, k_lens)]))
        v = half(torch.cat([u[:v] for u, v in zip(v, k_lens)]))

    q = q.to(v.dtype)
    k = k.to(v.dtype)

    if q_scale is not None:
        q = q * q_scale

    if version is not None and version == 3 and not FLASH_ATTN_3_AVAILABLE:
        warnings.warn(
            'Flash attention 3 is not available, use flash attention 2 instead.'
        )

    # apply attention
    if (version is None or version == 3) and FLASH_ATTN_3_AVAILABLE:
        # Note: dropout_p, window_size are not supported in FA3 now.
        x = flash_attn_interface.flash_attn_varlen_func(
            q=q,
            k=k,
            v=v,
            cu_seqlens_q=torch.cat([q_lens.new_zeros([1]), q_lens]).cumsum(
                0, dtype=torch.int32).to(q.device, non_blocking=True),
            cu_seqlens_k=torch.cat([k_lens.new_zeros([1]), k_lens]).cumsum(
                0, dtype=torch.int32).to(q.device, non_blocking=True),
            max_seqlen_q=lq,
            max_seqlen_k=lk,
            softmax_scale=softmax_scale,
            causal=causal,
            deterministic=deterministic)[0].unflatten(0, (b, lq))
    elif FLASH_ATTN_2_AVAILABLE:
        x = flash_attn.flash_attn_varlen_func(
            q=q,
            k=k,
            v=v,
            cu_seqlens_q=torch.cat([q_lens.new_zeros([1]), q_lens]).cumsum(
                0, dtype=torch.int32).to(q.device, non_blocking=True),
            cu_seqlens_k=torch.cat([k_lens.new_zeros([1]), k_lens]).cumsum(
                0, dtype=torch.int32).to(q.device, non_blocking=True),
            max_seqlen_q=lq,
            max_seqlen_k=lk,
            dropout_p=dropout_p,
            softmax_scale=softmax_scale,
            causal=causal,
            window_size=window_size,
            deterministic=deterministic).unflatten(0, (b, lq))
    else:
        # Fallback to PyTorch's scaled_dot_product_attention when flash attention is not available
        if q_lens is not None or k_lens is not None:
            warnings.warn(
                'Padding mask is disabled when using scaled_dot_product_attention. It can have a significant impact on performance.'
            )
        
        print(f"DEBUG: Input shapes - q: {q.shape}, k: {k.shape}, v: {v.shape}")
        print(f"DEBUG: batch size: {b}, lq: {lq}, lk: {lk}")
        
        # Input format: q, k, v are already flattened to [total_seq_len, num_heads, head_dim]
        # Need to reshape to [B, num_heads, seq_len, head_dim] for scaled_dot_product_attention
        
        # Unflatten and transpose: [total_seq_len, H, C] -> [B, L, H, C] -> [B, H, L, C]
        q_reshaped = q.unflatten(0, (b, lq)).transpose(1, 2)
        k_reshaped = k.unflatten(0, (b, lk)).transpose(1, 2)  
        v_reshaped = v.unflatten(0, (b, lk)).transpose(1, 2)
        
        print(f"DEBUG: After reshape - q: {q_reshaped.shape}, k: {k_reshaped.shape}, v: {v_reshaped.shape}")
        
        x = torch.nn.functional.scaled_dot_product_attention(
            q_reshaped, k_reshaped, v_reshaped, 
            attn_mask=None, is_causal=causal, dropout_p=dropout_p)
        
        print(f"DEBUG: After attention - x: {x.shape}")
        
        # Transpose back: [B, H, L, C] -> [B, L, H, C]
        x = x.transpose(1, 2)
        print(f"DEBUG: After transpose - x: {x.shape}")
        
        # Flatten to [B*L, H, C] to match flash attention output format
        x = x.flatten(0, 1)
        print(f"DEBUG: Final output shape - x: {x.shape}")

    # output
    return x.type(out_dtype)


def attention(
    q,
    k,
    v,
    q_lens=None,
    k_lens=None,
    dropout_p=0.,
    softmax_scale=None,
    q_scale=None,
    causal=False,
    window_size=(-1, -1),
    deterministic=False,
    dtype=torch.bfloat16,
    fa_version=None,
):
    if FLASH_ATTN_2_AVAILABLE or FLASH_ATTN_3_AVAILABLE:
        return flash_attention(
            q=q,
            k=k,
            v=v,
            q_lens=q_lens,
            k_lens=k_lens,
            dropout_p=dropout_p,
            softmax_scale=softmax_scale,
            q_scale=q_scale,
            causal=causal,
            window_size=window_size,
            deterministic=deterministic,
            dtype=dtype,
            version=fa_version,
        )
    else:
        if q_lens is not None or k_lens is not None:
            warnings.warn(
                'Padding mask is disabled when using scaled_dot_product_attention. It can have a significant impact on performance.'
            )
        attn_mask = None

        q = q.transpose(1, 2).to(dtype)
        k = k.transpose(1, 2).to(dtype)
        v = v.transpose(1, 2).to(dtype)

        out = torch.nn.functional.scaled_dot_product_attention(
            q, k, v, attn_mask=attn_mask, is_causal=causal, dropout_p=dropout_p)

        out = out.transpose(1, 2).contiguous()
        return out