# coding=utf-8
# Copyright 2023 Harutatsu Akiyama and HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import random
import unittest

import numpy as np
import torch
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer

from diffusers import (
    AutoencoderKL,
    EulerDiscreteScheduler,
    UNet2DConditionModel,
)
from diffusers.image_processor import VaeImageProcessor
from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_instruct_pix2pix import (
    StableDiffusionXLInstructPix2PixPipeline,
)
from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, torch_device

from ..pipeline_params import (
    IMAGE_TO_IMAGE_IMAGE_PARAMS,
    TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS,
    TEXT_GUIDED_IMAGE_VARIATION_PARAMS,
)
from ..test_pipelines_common import (
    PipelineKarrasSchedulerTesterMixin,
    PipelineLatentTesterMixin,
    PipelineTesterMixin,
    SDXLOptionalComponentsTesterMixin,
)


enable_full_determinism()


class StableDiffusionXLInstructPix2PixPipelineFastTests(
    PipelineLatentTesterMixin,
    PipelineKarrasSchedulerTesterMixin,
    PipelineTesterMixin,
    SDXLOptionalComponentsTesterMixin,
    unittest.TestCase,
):
    pipeline_class = StableDiffusionXLInstructPix2PixPipeline
    params = TEXT_GUIDED_IMAGE_VARIATION_PARAMS - {"height", "width", "cross_attention_kwargs"}
    batch_params = TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS
    image_params = IMAGE_TO_IMAGE_IMAGE_PARAMS
    image_latents_params = IMAGE_TO_IMAGE_IMAGE_PARAMS

    def get_dummy_components(self):
        torch.manual_seed(0)
        unet = UNet2DConditionModel(
            block_out_channels=(32, 64),
            layers_per_block=2,
            sample_size=32,
            in_channels=8,
            out_channels=4,
            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
            # SD2-specific config below
            attention_head_dim=(2, 4),
            use_linear_projection=True,
            addition_embed_type="text_time",
            addition_time_embed_dim=8,
            transformer_layers_per_block=(1, 2),
            projection_class_embeddings_input_dim=80,  # 5 * 8 + 32
            cross_attention_dim=64,
        )

        scheduler = EulerDiscreteScheduler(
            beta_start=0.00085,
            beta_end=0.012,
            steps_offset=1,
            beta_schedule="scaled_linear",
            timestep_spacing="leading",
        )
        torch.manual_seed(0)
        vae = AutoencoderKL(
            block_out_channels=[32, 64],
            in_channels=3,
            out_channels=3,
            down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
            up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
            latent_channels=4,
            sample_size=128,
        )
        torch.manual_seed(0)
        text_encoder_config = CLIPTextConfig(
            bos_token_id=0,
            eos_token_id=2,
            hidden_size=32,
            intermediate_size=37,
            layer_norm_eps=1e-05,
            num_attention_heads=4,
            num_hidden_layers=5,
            pad_token_id=1,
            vocab_size=1000,
            # SD2-specific config below
            hidden_act="gelu",
            projection_dim=32,
        )
        text_encoder = CLIPTextModel(text_encoder_config)
        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")

        text_encoder_2 = CLIPTextModelWithProjection(text_encoder_config)
        tokenizer_2 = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")

        components = {
            "unet": unet,
            "scheduler": scheduler,
            "vae": vae,
            "text_encoder": text_encoder,
            "tokenizer": tokenizer,
            "text_encoder_2": text_encoder_2,
            "tokenizer_2": tokenizer_2,
        }
        return components

    def get_dummy_inputs(self, device, seed=0):
        image = floats_tensor((1, 3, 64, 64), rng=random.Random(seed)).to(device)
        image = image / 2 + 0.5
        if str(device).startswith("mps"):
            generator = torch.manual_seed(seed)
        else:
            generator = torch.Generator(device=device).manual_seed(seed)
        inputs = {
            "prompt": "A painting of a squirrel eating a burger",
            "image": image,
            "generator": generator,
            "num_inference_steps": 2,
            "guidance_scale": 6.0,
            "image_guidance_scale": 1,
            "output_type": "numpy",
        }
        return inputs

    def test_components_function(self):
        init_components = self.get_dummy_components()
        pipe = self.pipeline_class(**init_components)

        self.assertTrue(hasattr(pipe, "components"))
        self.assertTrue(set(pipe.components.keys()) == set(init_components.keys()))

    def test_inference_batch_single_identical(self):
        super().test_inference_batch_single_identical(expected_max_diff=3e-3)

    def test_attention_slicing_forward_pass(self):
        super().test_attention_slicing_forward_pass(expected_max_diff=2e-3)

    # Overwrite the default test_latents_inputs because pix2pix encode the image differently
    def test_latents_input(self):
        components = self.get_dummy_components()
        pipe = StableDiffusionXLInstructPix2PixPipeline(**components)
        pipe.image_processor = VaeImageProcessor(do_resize=False, do_normalize=False)
        pipe = pipe.to(torch_device)
        pipe.set_progress_bar_config(disable=None)

        out = pipe(**self.get_dummy_inputs_by_type(torch_device, input_image_type="pt"))[0]

        vae = components["vae"]
        inputs = self.get_dummy_inputs_by_type(torch_device, input_image_type="pt")

        for image_param in self.image_latents_params:
            if image_param in inputs.keys():
                inputs[image_param] = vae.encode(inputs[image_param]).latent_dist.mode()

        out_latents_inputs = pipe(**inputs)[0]

        max_diff = np.abs(out - out_latents_inputs).max()
        self.assertLess(max_diff, 1e-4, "passing latents as image input generate different result from passing image")

    def test_cfg(self):
        pass

    def test_save_load_optional_components(self):
        self._test_save_load_optional_components()