import torch

from synthesizer import Transformer, SynthesizerDense, SynthesizerRandom, FactorizedSynthesizerDense, FactorizedSynthesizerRandom, MixtureSynthesizers, get_n_params, calculate_flops


def main():
    batch_size, channel_dim, sentence_length = 2, 1024, 32
    x = torch.randn([batch_size, sentence_length, channel_dim])

    vanilla = Transformer(channel_dim)
    out, attention_map = vanilla(x)
    print(out.size(), attention_map.size())
    n_params, flops = get_n_params(vanilla), calculate_flops(vanilla.children())
    print('vanilla, n_params: {}, flops: {}'.format(n_params, flops))

    dense_synthesizer = SynthesizerDense(channel_dim, sentence_length)
    out, attention_map = dense_synthesizer(x)
    print(out.size(), attention_map.size())
    n_params, flops = get_n_params(dense_synthesizer), calculate_flops(dense_synthesizer.children())
    print('dense_synthesizer, n_params: {}, flops: {}'.format(n_params, flops))

    random_synthesizer = SynthesizerRandom(channel_dim, sentence_length)
    out, attention_map = random_synthesizer(x)
    print(out.size(), attention_map.size())
    n_params, flops = get_n_params(random_synthesizer), calculate_flops(random_synthesizer.children())
    print('random_synthesizer, n_params: {}, flops: {}'.format(n_params, flops))

    random_synthesizer_fix = SynthesizerRandom(channel_dim, sentence_length, fixed=True)
    out, attention_map = random_synthesizer_fix(x)
    print(out.size(), attention_map.size())
    n_params, flops = get_n_params(random_synthesizer_fix), calculate_flops(random_synthesizer_fix.children())
    print('random_synthesizer_fix, n_params: {}, flops: {}'.format(n_params, flops))

    factorized_synthesizer_random = FactorizedSynthesizerRandom(channel_dim)
    out, attention_map = factorized_synthesizer_random(x)
    print(out.size(), attention_map.size())
    n_params, flops = get_n_params(factorized_synthesizer_random), calculate_flops(
        factorized_synthesizer_random.children())
    print('factorized_synthesizer_random, n_params: {}, flops: {}'.format(n_params, flops))

    factorized_synthesizer_dense = FactorizedSynthesizerDense(channel_dim, sentence_length)
    out, attention_map = factorized_synthesizer_dense(x)
    print(out.size(), attention_map.size())
    n_params, flops = get_n_params(factorized_synthesizer_dense), calculate_flops(
        factorized_synthesizer_dense.children())
    print('factorized_synthesizer_dense, n_params: {}, flops: {}'.format(n_params, flops))

    mixture_synthesizer = MixtureSynthesizers(channel_dim, sentence_length)
    out, attention_map = mixture_synthesizer(x)
    print(out.size(), attention_map.size())
    n_params, flops = get_n_params(mixture_synthesizer), calculate_flops(mixture_synthesizer.children())
    print('mixture_synthesizer, n_params: {}, flops: {}'.format(n_params, flops))


if __name__ == '__main__':
    main()