import numpy as np
import io
import os
import time
from collections import defaultdict, deque
import datetime

import torch
import torch.distributed as dist

def optimizer_to(optim, device):
    for param in optim.state.values():
        # Not sure there are any global tensors in the state dict
        if isinstance(param, torch.Tensor):
            param.data = param.data.to(device)
            if param._grad is not None:
                param._grad.data = param._grad.data.to(device)
        elif isinstance(param, dict):
            for subparam in param.values():
                if isinstance(subparam, torch.Tensor):
                    subparam.data = subparam.data.to(device)
                    if subparam._grad is not None:
                        subparam._grad.data = subparam._grad.data.to(device)
                        

class SmoothedValue(object):
    """Track a series of values and provide access to smoothed values over a
    window or the global series average.
    """

    def __init__(self, window_size=20, fmt=None):
        if fmt is None:
            fmt = "{median:.4f} ({global_avg:.4f})"
        self.deque = deque(maxlen=window_size)
        self.total = 0.0
        self.count = 0
        self.fmt = fmt

    def update(self, value, n=1):
        self.deque.append(value)
        self.count += n
        self.total += value * n

    def synchronize_between_processes(self):
        """
        Warning: does not synchronize the deque!
        """
        if not is_dist_avail_and_initialized():
            return
        t = torch.tensor([self.count, self.total], dtype=torch.float64, device='cuda')
        dist.barrier()
        dist.all_reduce(t)
        t = t.tolist()
        self.count = int(t[0])
        self.total = t[1]

    @property
    def median(self):
        d = torch.tensor(list(self.deque))
        return d.median().item()

    @property
    def avg(self):
        d = torch.tensor(list(self.deque), dtype=torch.float32)
        return d.mean().item()

    @property
    def global_avg(self):
        return self.total / self.count

    @property
    def max(self):
        return max(self.deque)

    @property
    def value(self):
        return self.deque[-1]

    def __str__(self):
        return self.fmt.format(
            median=self.median,
            avg=self.avg,
            global_avg=self.global_avg,
            max=self.max,
            value=self.value)


class MetricLogger(object):
    def __init__(self, delimiter="\t", accelerator=None):
        self.meters = defaultdict(SmoothedValue)
        self.delimiter = delimiter
        self.accelerator = accelerator

    def update(self, **kwargs):
        for k, v in kwargs.items():
            if isinstance(v, torch.Tensor):
                v = v.item()
            assert isinstance(v, (float, int))
            self.meters[k].update(v)

    def __getattr__(self, attr):
        if attr in self.meters:
            return self.meters[attr]
        if attr in self.__dict__:
            return self.__dict__[attr]
        raise AttributeError("'{}' object has no attribute '{}'".format(
            type(self).__name__, attr))

    def __str__(self):
        loss_str = []
        for name, meter in self.meters.items():
            loss_str.append(
                "{}: {}".format(name, str(meter))
            )
        return self.delimiter.join(loss_str)

    def global_avg(self):
        loss_str = []
        for name, meter in self.meters.items():
            loss_str.append(
                "{}: {:.4f}".format(name, meter.global_avg)
            )
        return self.delimiter.join(loss_str)    
    
    def synchronize_between_processes(self):
        for meter in self.meters.values():
            meter.synchronize_between_processes()

    def add_meter(self, name, meter):
        self.meters[name] = meter

    def log_every(self, iterable, print_freq, header=None):
        i = 0
        if not header:
            header = ''
        start_time = time.time()
        end = time.time()
        iter_time = SmoothedValue(fmt='{avg:.4f}')
        data_time = SmoothedValue(fmt='{avg:.4f}')
        space_fmt = ':' + str(len(str(len(iterable)))) + 'd'
        log_msg = [
            header,
            '[{0' + space_fmt + '}/{1}]',
            'eta: {eta}',
            '{meters}',
            'time: {time}',
            'data: {data}'
        ]
        if torch.cuda.is_available():
            log_msg.append('max mem: {memory:.0f}')
        log_msg = self.delimiter.join(log_msg)
        MB = 1024.0 * 1024.0

        if self.accelerator is not None:
            print_func = self.accelerator.print
        else:
            print_func = print
        for obj in iterable:
            data_time.update(time.time() - end)
            yield obj
            iter_time.update(time.time() - end)
            if i % print_freq == 0 or i == len(iterable) - 1:
                eta_seconds = iter_time.global_avg * (len(iterable) - i)
                eta_string = str(datetime.timedelta(seconds=int(eta_seconds)))
                if torch.cuda.is_available():
                    print_func(log_msg.format(
                        i, len(iterable), eta=eta_string,
                        meters=str(self),
                        time=str(iter_time), data=str(data_time),
                        memory=torch.cuda.max_memory_allocated() / MB))
                else:
                    print_func(log_msg.format(
                        i, len(iterable), eta=eta_string,
                        meters=str(self),
                        time=str(iter_time), data=str(data_time)))
            i += 1
            end = time.time()
        total_time = time.time() - start_time
        total_time_str = str(datetime.timedelta(seconds=int(total_time)))
        print_func('{} Total time: {} ({:.4f} s / it)'.format(
            header, total_time_str, total_time / len(iterable)))
        


class AttrDict(dict):
    def __init__(self, *args, **kwargs):
        super(AttrDict, self).__init__(*args, **kwargs)
        self.__dict__ = self


def compute_acc(logits, label, reduction='mean'):
    ret = (torch.argmax(logits, dim=1) == label).float()
    if reduction == 'none':
        return ret.detach()
    elif reduction == 'mean':
        return ret.mean().item()

def compute_n_params(model, return_str=True):
    tot = 0
    for p in model.parameters():
        w = 1
        for x in p.shape:
            w *= x
        tot += w
    if return_str:
        if tot >= 1e6:
            return '{:.1f}M'.format(tot / 1e6)
        else:
            return '{:.1f}K'.format(tot / 1e3)
    else:
        return tot

def setup_for_distributed(is_master):
    """
    This function disables printing when not in master process
    """
    import builtins as __builtin__
    builtin_print = __builtin__.print

    def print(*args, **kwargs):
        force = kwargs.pop('force', False)
        if is_master or force:
            builtin_print(*args, **kwargs)

    __builtin__.print = print


def is_dist_avail_and_initialized():
    if not dist.is_available():
        return False
    if not dist.is_initialized():
        return False
    return True


def get_world_size():
    if not is_dist_avail_and_initialized():
        return 1
    return dist.get_world_size()


def get_rank():
    if not is_dist_avail_and_initialized():
        return 0
    return dist.get_rank()


def is_main_process():
    return get_rank() == 0


def save_on_master(*args, **kwargs):
    if is_main_process():
        torch.save(*args, **kwargs)


def init_distributed_mode(args):
    if 'RANK' in os.environ and 'WORLD_SIZE' in os.environ:
        args.rank = int(os.environ["RANK"])
        args.world_size = int(os.environ['WORLD_SIZE'])
        args.gpu = int(os.environ['LOCAL_RANK'])
    elif 'SLURM_PROCID' in os.environ:
        args.rank = int(os.environ['SLURM_PROCID'])
        args.gpu = args.rank % torch.cuda.device_count()
        print(args.gpu, os.environ['SLURM_LOCALID'], os.environ['SLURM_JOB_NODELIST'], os.environ['SLURM_STEP_GPUS'])
    else:
        print('Not using distributed mode')
        args.distributed = False
        return

    args.distributed = True

    torch.cuda.set_device(args.gpu)
    args.dist_backend = 'nccl'
    print('world_size', args.world_size, 'gpu', args.gpu, 'dist_url:', args.dist_url)
    print('| distributed init (rank {}): {}'.format(
        args.rank, args.dist_url), flush=True)
    torch.distributed.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
                                         world_size=args.world_size, rank=args.rank)
    print("init")
    torch.distributed.barrier()
    setup_for_distributed(args.rank == 0)


def init_distributed_mode_multinodes(args):
    import hostlist
    if 'RANK' in os.environ and 'WORLD_SIZE' in os.environ:
        args.rank = int(os.environ["RANK"])
        args.world_size = int(os.environ['WORLD_SIZE'])
        args.gpu = int(os.environ['LOCAL_RANK'])
    elif 'SLURM_PROCID' in os.environ:
        args.rank = int(os.environ['SLURM_PROCID'])
        args.gpu = args.rank % torch.cuda.device_count()
        print('slurm')
    else:
        print('Not using distributed mode')
        args.distributed = False
        return

    args.distributed = True
    # print(args.gpu, os.environ['SLURM_PROCID'], os.environ['SLURM_LOCALID'], os.environ['SLURM_JOB_NODELIST'], os.environ['SLURM_STEP_GPUS'])
    hostnames = hostlist.expand_hostlist(os.environ['SLURM_JOB_NODELIST'])
    os.environ['MASTER_ADDR'] = hostnames[0]
    gpu_ids = os.environ['SLURM_STEP_GPUS'].split(",")
    # os.environ['MASTER_PORT'] = str(12345 + int(min(gpu_ids))) # to avoid port conflict on the same node

    print(os.environ['MASTER_ADDR'], os.environ['MASTER_PORT'])

    torch.cuda.set_device(args.gpu)
    args.dist_backend = 'nccl'
    args.dist_url = 'tcp://'+os.environ['MASTER_ADDR']+':'+os.environ['MASTER_PORT']

    print('world_size', args.world_size, 'gpu', args.gpu)
    print('| distributed init (rank {}): {}'.format(
        args.rank, args.dist_url), flush=True)

    

    torch.distributed.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
                                         world_size=args.world_size, rank=args.rank)
    # torch.distributed.barrier()
    # setup_for_distributed(args.rank == 0)


def init_distributed_mode_multinodes_jz(args):
    import hostlist
    if args.jean_zay:
        hostnames = hostlist.expand_hostlist(os.environ['SLURM_JOB_NODELIST'])
        os.environ['MASTER_ADDR'] = hostnames[0]

        print(os.environ['MASTER_ADDR'], os.environ['MASTER_PORT'], os.environ['SLURM_PROCID'], os.environ['SLURM_NTASKS'], os.environ['SLURM_LOCALID'])
        args.gpu = int(os.environ['SLURM_LOCALID'])
        args.rank = int(os.environ['SLURM_PROCID'])
        args.world_size = int(os.environ['SLURM_NTASKS'])
        args.dist_url = 'env://'+os.environ['MASTER_ADDR']+':'+os.environ['MASTER_PORT']
        # args.dist_url = 'env://'
        print('jean zay')
    elif 'RANK' in os.environ and 'WORLD_SIZE' in os.environ:
        args.rank = int(os.environ["RANK"])
        args.world_size = int(os.environ['WORLD_SIZE'])
        args.gpu = int(os.environ['LOCAL_RANK'])
    elif 'SLURM_PROCID' in os.environ:
        args.rank = int(os.environ['SLURM_PROCID'])
        args.gpu = args.rank % torch.cuda.device_count()
        print('slurm')
    else:
        print('Not using distributed mode')
        args.distributed = False
        return

    args.distributed = True

    torch.cuda.set_device(args.gpu)
    args.dist_backend = 'nccl'
    print('world_size', args.world_size, 'gpu', args.gpu, 'rank', args.rank)
    print('| distributed init (rank {}): {}'.format(
        args.rank, args.dist_url), flush=True)
    torch.distributed.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
                                         world_size=args.world_size, rank=args.rank)


    torch.distributed.barrier()
    setup_for_distributed(args.rank == 0)










# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
"""
This file contains primitives for multi-gpu communication.
This is useful when doing distributed training.
"""

import functools
import logging
import numpy as np
import pickle
import torch
import torch.distributed as dist

import torch

_LOCAL_PROCESS_GROUP = None
"""
A torch process group which only includes processes that on the same machine as the current process.
This variable is set when processes are spawned by `launch()` in "engine/launch.py".
"""


def get_world_size() -> int:
    if not dist.is_available():
        return 1
    if not dist.is_initialized():
        return 1
    return dist.get_world_size()


def get_rank() -> int:
    if not dist.is_available():
        return 0
    if not dist.is_initialized():
        return 0
    return dist.get_rank()


def get_local_rank() -> int:
    """
    Returns:
        The rank of the current process within the local (per-machine) process group.
    """
    if not dist.is_available():
        return 0
    if not dist.is_initialized():
        return 0
    assert _LOCAL_PROCESS_GROUP is not None
    return dist.get_rank(group=_LOCAL_PROCESS_GROUP)


def get_local_size() -> int:
    """
    Returns:
        The size of the per-machine process group,
        i.e. the number of processes per machine.
    """
    if not dist.is_available():
        return 1
    if not dist.is_initialized():
        return 1
    return dist.get_world_size(group=_LOCAL_PROCESS_GROUP)


def is_main_process() -> bool:
    return get_rank() == 0


def synchronize():
    """
    Helper function to synchronize (barrier) among all processes when
    using distributed training
    """
    if not dist.is_available():
        return
    if not dist.is_initialized():
        return
    world_size = dist.get_world_size()
    if world_size == 1:
        return
    dist.barrier()


@functools.lru_cache()
def _get_global_gloo_group():
    """
    Return a process group based on gloo backend, containing all the ranks
    The result is cached.
    """
    if dist.get_backend() == "nccl":
        return dist.new_group(backend="gloo")
    else:
        return dist.group.WORLD


def _serialize_to_tensor(data, group):
    backend = dist.get_backend(group)
    assert backend in ["gloo", "nccl"]
    device = torch.device("cpu" if backend == "gloo" else "cuda")

    buffer = pickle.dumps(data)
    if len(buffer) > 1024 ** 3:
        logger = logging.getLogger(__name__)
        logger.warning(
            "Rank {} trying to all-gather {:.2f} GB of data on device {}".format(
                get_rank(), len(buffer) / (1024 ** 3), device
            )
        )
    storage = torch.ByteStorage.from_buffer(buffer)
    tensor = torch.ByteTensor(storage).to(device=device)
    return tensor


def _pad_to_largest_tensor(tensor, group):
    """
    Returns:
        list[int]: size of the tensor, on each rank
        Tensor: padded tensor that has the max size
    """
    world_size = dist.get_world_size(group=group)
    assert (
        world_size >= 1
    ), "comm.gather/all_gather must be called from ranks within the given group!"
    local_size = torch.tensor(
        [tensor.numel()], dtype=torch.int64, device=tensor.device)
    size_list = [
        torch.zeros([1], dtype=torch.int64, device=tensor.device)
        for _ in range(world_size)
    ]
    dist.all_gather(size_list, local_size, group=group)
    size_list = [int(size.item()) for size in size_list]

    max_size = max(size_list)

    # we pad the tensor because torch all_gather does not support
    # gathering tensors of different shapes
    if local_size != max_size:
        padding = torch.zeros(
            (max_size - local_size,), dtype=torch.uint8, device=tensor.device
        )
        tensor = torch.cat((tensor, padding), dim=0)
    return size_list, tensor


def all_gather(data, group=None):
    """
    Run all_gather on arbitrary picklable data (not necessarily tensors).
    Args:
        data: any picklable object
        group: a torch process group. By default, will use a group which
            contains all ranks on gloo backend.
    Returns:
        list[data]: list of data gathered from each rank
    """
    if get_world_size() == 1:
        return [data]
    if group is None:
        group = _get_global_gloo_group()
    if dist.get_world_size(group) == 1:
        return [data]

    tensor = _serialize_to_tensor(data, group)

    size_list, tensor = _pad_to_largest_tensor(tensor, group)
    max_size = max(size_list)

    # receiving Tensor from all ranks
    tensor_list = [
        torch.empty((max_size,), dtype=torch.uint8, device=tensor.device)
        for _ in size_list
    ]
    dist.all_gather(tensor_list, tensor, group=group)

    data_list = []
    for size, tensor in zip(size_list, tensor_list):
        buffer = tensor.cpu().numpy().tobytes()[:size]
        data_list.append(pickle.loads(buffer))

    return data_list


def gather(data, dst=0, group=None):
    """
    Run gather on arbitrary picklable data (not necessarily tensors).
    Args:
        data: any picklable object
        dst (int): destination rank
        group: a torch process group. By default, will use a group which
            contains all ranks on gloo backend.
    Returns:
        list[data]: on dst, a list of data gathered from each rank. Otherwise,
            an empty list.
    """
    if get_world_size() == 1:
        return [data]
    if group is None:
        group = _get_global_gloo_group()
    if dist.get_world_size(group=group) == 1:
        return [data]
    rank = dist.get_rank(group=group)

    tensor = _serialize_to_tensor(data, group)
    size_list, tensor = _pad_to_largest_tensor(tensor, group)

    # receiving Tensor from all ranks
    if rank == dst:
        max_size = max(size_list)
        tensor_list = [
            torch.empty((max_size,), dtype=torch.uint8, device=tensor.device)
            for _ in size_list
        ]
        dist.gather(tensor, tensor_list, dst=dst, group=group)

        data_list = []
        for size, tensor in zip(size_list, tensor_list):
            buffer = tensor.cpu().numpy().tobytes()[:size]
            data_list.append(pickle.loads(buffer))
        return data_list
    else:
        dist.gather(tensor, [], dst=dst, group=group)
        return []


def shared_random_seed():
    """
    Returns:
        int: a random number that is the same across all workers.
            If workers need a shared RNG, they can use this shared seed to
            create one.
    All workers must call this function, otherwise it will deadlock.
    """
    ints = np.random.randint(2 ** 31)
    all_ints = all_gather(ints)
    return all_ints[0]



def reduce_dict(input_dict, average=True):
    """
    Reduce the values in the dictionary from all processes so that process with rank
    0 has the reduced results.
    Args:
        input_dict (dict): inputs to be reduced. (values not necessarily tensors).
        average (bool): whether to do average or sum
    Returns:
        a dict with the same keys as input_dict, after reduction.
    """

    world_size = get_world_size()
    if world_size < 2:
        return input_dict

    with torch.no_grad():

        # Convert to CUDA Tensor for dist.reduce()
        input_dict_cuda_vals = {}
        for k, v in input_dict.items():
            if type(v) == torch.Tensor:
                input_dict_cuda_vals[k] = v.to('cuda')
            else:
                input_dict_cuda_vals[k] = torch.tensor(v, device='cuda')

        names = []
        values = []
        for k, v in sorted(input_dict_cuda_vals.items()):
            names.append(k)
            values.append(v)
        values = torch.stack(values, dim=0)
        dist.reduce(values, dst=0) # reduce to gpu 0

        if dist.get_rank() == 0 and average:
            # only main process gets accumulated, so only divide by
            # world_size in this case
            values /= world_size
        reduced_dict = {k: v for k, v in zip(names, values)}
    return reduced_dict