#
# SPDX-FileCopyrightText: Copyright (c) 1993-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# 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 numpy as np
import tensorrt as trt

import pycuda.driver as cuda
import pycuda.autoinit


class WrapperBase:

    def close(self):
        pass


class SP_ONNXClassifierWrapper(WrapperBase):
    def __init__(
        self,
        file,
        target_dtype=np.float32,
        batch_size=1,
        max_keypoints=512,
        descriptor_dim=256,
    ):

        self.target_dtype = target_dtype
        self.load(file)

        self.stream = None
        self.max_keypoints = max_keypoints
        self.batch_size = batch_size
        self.descriptor_dim = descriptor_dim

    def load(self, file):
        f = open(file, "rb")
        self.runtime = trt.Runtime(trt.Logger(trt.Logger.WARNING))

        self.engine = self.runtime.deserialize_cuda_engine(f.read())
        self.context = self.engine.create_execution_context()

    def allocate_memory(self, batch):
        self.keypoints = np.empty(
            (self.batch_size, self.max_keypoints, 2), dtype=np.float32
        )
        self.scores = np.empty((self.batch_size, self.max_keypoints), dtype=np.float32)
        self.descriptors = np.empty(
            (self.batch_size, self.max_keypoints, self.descriptor_dim), dtype=np.float32
        )

        # Allocate device memory
        self.d_input = cuda.mem_alloc(1 * batch.nbytes)
        self.d_keypoints = cuda.mem_alloc(1 * self.keypoints.nbytes)
        self.d_scores = cuda.mem_alloc(1 * self.scores.nbytes)
        self.d_descriptors = cuda.mem_alloc(1 * self.descriptors.nbytes)

        tensor_names = [
            self.engine.get_tensor_name(i) for i in range(self.engine.num_io_tensors)
        ]

        self.context.set_tensor_address(tensor_names[0], int(self.d_input))
        self.context.set_tensor_address(tensor_names[1], int(self.d_keypoints))
        self.context.set_tensor_address(tensor_names[2], int(self.d_scores))
        self.context.set_tensor_address(tensor_names[3], int(self.d_descriptors))

        self.stream = cuda.Stream()

    def predict(self, batch):
        if self.stream is None:
            self.allocate_memory(batch)

        # Transfer input data to device
        cuda.memcpy_htod_async(self.d_input, batch, self.stream)
        # Execute model
        self.context.execute_async_v3(self.stream.handle)
        # Transfer predictions back
        cuda.memcpy_dtoh_async(self.keypoints, self.d_keypoints, self.stream)
        cuda.memcpy_dtoh_async(self.scores, self.d_scores, self.stream)
        cuda.memcpy_dtoh_async(self.descriptors, self.d_descriptors, self.stream)
        # Syncronize threads
        self.stream.synchronize()

        return self.keypoints, self.scores, self.descriptors